domain off

domain off

[Sriram V]

Hello,

I am trying to power down power domains which are not used but i am
unable to do so unless i write manually into the registers.

Can anyone help me on the standard way to power down power domains on
a exynos 5422.

I  tried marking status /bts-status to disable in the device tree file
for pd-disp1, but that does not help.

What is the standard way of powering down if they are not used. Do we
need to do using dt or somewhere in the kernel board file?

-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozlowski]

2016-01-09 19:19 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>
> Hello,
>
> I am trying to power down power domains which are not used but i am
> unable to do so unless i write manually into the registers.
>
> Can anyone help me on the standard way to power down power domains on
> a exynos 5422.
>
> I  tried marking status /bts-status to disable in the device tree file
> for pd-disp1, but that does not help.
>
> What is the standard way of powering down if they are not used. Do we
> need to do using dt or somewhere in the kernel board file?

Hi,

The domains defined in DT (and enabled in DT... status=disabled marks
the *node* as disabled, not the power domain) are automatically turned
off when attached devices are runtime suspended. Use
PM_DEBUG+PM_ADVANCED_DEBUG and then:
cat /sys/kernel/debug/pm_genpd/pm_genpd_summary

Domains not present there are not defined so they are left in default
status set by bootloader (probably on).

BR,
Krzysztof

Re: domain off

[Sriram V]

Hi Krzysztof,

I see that in pm_genpd_sumarry, the domains are off. However, i still
see the clocks are on.
For example; isp and cam is turned off in pm_genpd_summary, however
cam clocks are still on.

Should'nt the clocks be turned off first and then the domain.






On Sun, Jan 10, 2016 at 11:45 AM, Krzysztof Kozlowski
<k.kozlowski@samsung.com> wrote:
> 2016-01-09 19:19 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>
>> Hello,
>>
>> I am trying to power down power domains which are not used but i am
>> unable to do so unless i write manually into the registers.
>>
>> Can anyone help me on the standard way to power down power domains on
>> a exynos 5422.
>>
>> I  tried marking status /bts-status to disable in the device tree file
>> for pd-disp1, but that does not help.
>>
>> What is the standard way of powering down if they are not used. Do we
>> need to do using dt or somewhere in the kernel board file?
>
> Hi,
>
> The domains defined in DT (and enabled in DT... status=disabled marks
> the *node* as disabled, not the power domain) are automatically turned
> off when attached devices are runtime suspended. Use
> PM_DEBUG+PM_ADVANCED_DEBUG and then:
> cat /sys/kernel/debug/pm_genpd/pm_genpd_summary
>
> Domains not present there are not defined so they are left in default
> status set by bootloader (probably on).
>
> BR,
> Krzysztof



-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozlowski]

2016-01-10 22:49 GMT+09:00 Sriram V <vshrirama@gmail.com>:
> Hi Krzysztof,
>
> I see that in pm_genpd_sumarry, the domains are off. However, i still
> see the clocks are on.
> For example; isp and cam is turned off in pm_genpd_summary, however
> cam clocks are still on.
>
> Should'nt the clocks be turned off first and then the domain.

No, the clocks in clock hierarchy are usually not disabled as a part
of domain power off. However respective drivers attached to a power
domain, when entering runtime suspend, should disable these clocks.

Best regards,
Krzysztof

Re: domain off

[Sriram V]

Hi Krzystof,



On Mon, Jan 11, 2016 at 4:57 AM, Krzysztof Kozlowski
<k.kozlowski@samsung.com> wrote:
> 2016-01-10 22:49 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>> Hi Krzysztof,
>>
>> I see that in pm_genpd_sumarry, the domains are off. However, i still
>> see the clocks are on.
>> For example; isp and cam is turned off in pm_genpd_summary, however
>> cam clocks are still on.
>>
>> Should'nt the clocks be turned off first and then the domain.
>
> No, the clocks in clock hierarchy are usually not disabled as a part
> of domain power off. However respective drivers attached to a power
> domain, when entering runtime suspend, should disable these clocks.
>

That is true, runtime suspend framework should disable the clocks.
Hence, my question is if the clocks are active, Will disabling the
domain take effect? Because when i print the status - they show that
the domain is OFF.

On my system, what i see is camera clocks are active, But the domain
is OFF. Hence, i am not sure if camera subsystem is consuming power or
not since clocks are on.

>From my understanding, the domain status should not indicate OFF if
any clocks are still on.



> Best regards,
> Krzysztof



-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozlowski]

On 11.01.2016 10:48, Sriram V wrote:
> Hi Krzystof,
> 
> 
> 
> On Mon, Jan 11, 2016 at 4:57 AM, Krzysztof Kozlowski
> <k.kozlowski@samsung.com> wrote:
>> 2016-01-10 22:49 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>> Hi Krzysztof,
>>>
>>> I see that in pm_genpd_sumarry, the domains are off. However, i still
>>> see the clocks are on.
>>> For example; isp and cam is turned off in pm_genpd_summary, however
>>> cam clocks are still on.
>>>
>>> Should'nt the clocks be turned off first and then the domain.
>>
>> No, the clocks in clock hierarchy are usually not disabled as a part
>> of domain power off. However respective drivers attached to a power
>> domain, when entering runtime suspend, should disable these clocks.
>>
> 
> That is true, runtime suspend framework should disable the clocks.
> Hence, my question is if the clocks are active, Will disabling the
> domain take effect? Because when i print the status - they show that
> the domain is OFF.

Yes, the domain will be disabled.

> 
> On my system, what i see is camera clocks are active, But the domain
> is OFF. Hence, i am not sure if camera subsystem is consuming power or
> not since clocks are on.
> 
>>From my understanding, the domain status should not indicate OFF if
> any clocks are still on.

Clocks and power domains in Linux are (mostly) independent subsystems so
the domain will be disabled even though some clocks are active. Some
clocks may be even missing in Linux (not defined) - it doesn't matter...

It doesn't matter except in all the cases when this is a bug. :) The
clocks should be disabled and some muxes reparented to oscclk. One of
the cases we had was hang on reading clk_summary when power domain is off.

Best regards,
Krzysztof

Re: domain off

[Sriram V]

Hi Krzyszof,



On Mon, Jan 11, 2016 at 7:51 AM, Krzysztof Kozlowski
<k.kozlowski@samsung.com> wrote:
> On 11.01.2016 10:48, Sriram V wrote:
>> Hi Krzystof,
>>
>>
>>
>> On Mon, Jan 11, 2016 at 4:57 AM, Krzysztof Kozlowski
>> <k.kozlowski@samsung.com> wrote:
>>> 2016-01-10 22:49 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>>> Hi Krzysztof,
>>>>
>>>> I see that in pm_genpd_sumarry, the domains are off. However, i still
>>>> see the clocks are on.
>>>> For example; isp and cam is turned off in pm_genpd_summary, however
>>>> cam clocks are still on.
>>>>
>>>> Should'nt the clocks be turned off first and then the domain.
>>>
>>> No, the clocks in clock hierarchy are usually not disabled as a part
>>> of domain power off. However respective drivers attached to a power
>>> domain, when entering runtime suspend, should disable these clocks.
>>>
>>
>> That is true, runtime suspend framework should disable the clocks.
>> Hence, my question is if the clocks are active, Will disabling the
>> domain take effect? Because when i print the status - they show that
>> the domain is OFF.
>
> Yes, the domain will be disabled.
>
>>
>> On my system, what i see is camera clocks are active, But the domain
>> is OFF. Hence, i am not sure if camera subsystem is consuming power or
>> not since clocks are on.
>>
>>>From my understanding, the domain status should not indicate OFF if
>> any clocks are still on.
>
> Clocks and power domains in Linux are (mostly) independent subsystems so
> the domain will be disabled even though some clocks are active. Some
> clocks may be even missing in Linux (not defined) - it doesn't matter...
>
> It doesn't matter except in all the cases when this is a bug. :) The
> clocks should be disabled and some muxes reparented to oscclk. One of
> the cases we had was hang on reading clk_summary when power domain is off.
>

Thanks for your pointers.

When i boot the kernel, I see that the following domains are currently
on. I see that the following domains are currently on, Linux does not
know about these domains and they are permanently on.

pd-fsys, pd-psgen, pd-peric, pd-wcore

Is there a way of finding out if these can be shut off. i dont think
if clocks for the IP's serviced by these domains and the domain itself
is correctly tied-in in runtime-pm. As these domains are always on.

Also, what kernel are you using?

Is there a way of finding out if these can be shut off. i dont think
if clocks for the IP's serviced by these domains
and the domain itself is correctly tied-in in runtime-pm.

Also, what kernel are you using?



> Best regards,
> Krzysztof
>



-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozlowski]

On 20.01.2016 18:52, Sriram V wrote:
> Hi Krzyszof,
> 
> 
> 
> On Mon, Jan 11, 2016 at 7:51 AM, Krzysztof Kozlowski
> <k.kozlowski@samsung.com> wrote:
>> On 11.01.2016 10:48, Sriram V wrote:
>>> Hi Krzystof,
>>>
>>>
>>>
>>> On Mon, Jan 11, 2016 at 4:57 AM, Krzysztof Kozlowski
>>> <k.kozlowski@samsung.com> wrote:
>>>> 2016-01-10 22:49 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>>>> Hi Krzysztof,
>>>>>
>>>>> I see that in pm_genpd_sumarry, the domains are off. However, i still
>>>>> see the clocks are on.
>>>>> For example; isp and cam is turned off in pm_genpd_summary, however
>>>>> cam clocks are still on.
>>>>>
>>>>> Should'nt the clocks be turned off first and then the domain.
>>>>
>>>> No, the clocks in clock hierarchy are usually not disabled as a part
>>>> of domain power off. However respective drivers attached to a power
>>>> domain, when entering runtime suspend, should disable these clocks.
>>>>
>>>
>>> That is true, runtime suspend framework should disable the clocks.
>>> Hence, my question is if the clocks are active, Will disabling the
>>> domain take effect? Because when i print the status - they show that
>>> the domain is OFF.
>>
>> Yes, the domain will be disabled.
>>
>>>
>>> On my system, what i see is camera clocks are active, But the domain
>>> is OFF. Hence, i am not sure if camera subsystem is consuming power or
>>> not since clocks are on.
>>>
>>> >From my understanding, the domain status should not indicate OFF if
>>> any clocks are still on.
>>
>> Clocks and power domains in Linux are (mostly) independent subsystems so
>> the domain will be disabled even though some clocks are active. Some
>> clocks may be even missing in Linux (not defined) - it doesn't matter...
>>
>> It doesn't matter except in all the cases when this is a bug. :) The
>> clocks should be disabled and some muxes reparented to oscclk. One of
>> the cases we had was hang on reading clk_summary when power domain is off.
>>
> 
> Thanks for your pointers.
> 
> When i boot the kernel, I see that the following domains are currently
> on. I see that the following domains are currently on, Linux does not
> know about these domains and they are permanently on.
> 
> pd-fsys, pd-psgen, pd-peric, pd-wcore
> 
> Is there a way of finding out if these can be shut off. i dont think
> if clocks for the IP's serviced by these domains and the domain itself
> is correctly tied-in in runtime-pm. As these domains are always on.

The datasheet/reference manual in the "Local power control" describes
conditions whether they could be disabled or not. It also documents what
is required to power off them.

> 
> Also, what kernel are you using?

Usually mainline and linux-next.

In some of the projects I use different ones, e.g. these on tizen.org
(https://review.tizen.org/git/?p=platform/kernel/linux-exynos.git;a=summary).

Best regards,
Krzysztof

Re: domain off

[Sriram V]

Hi Krzysztof,


Are you saying that only power domains under "Local Power Control" can
be attached to runtime PM and can be turned off?

In that case, i dont see other domains such as peris, psgen, wcore in
there, Hence these need to be always on?

is my understanding correct?


Thanks


On Wed, Jan 20, 2016 at 3:42 PM, Krzysztof Kozlowski
<k.kozlowski@samsung.com> wrote:
> On 20.01.2016 18:52, Sriram V wrote:
>> Hi Krzyszof,
>>
>>
>>
>> On Mon, Jan 11, 2016 at 7:51 AM, Krzysztof Kozlowski
>> <k.kozlowski@samsung.com> wrote:
>>> On 11.01.2016 10:48, Sriram V wrote:
>>>> Hi Krzystof,
>>>>
>>>>
>>>>
>>>> On Mon, Jan 11, 2016 at 4:57 AM, Krzysztof Kozlowski
>>>> <k.kozlowski@samsung.com> wrote:
>>>>> 2016-01-10 22:49 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>>>>> Hi Krzysztof,
>>>>>>
>>>>>> I see that in pm_genpd_sumarry, the domains are off. However, i still
>>>>>> see the clocks are on.
>>>>>> For example; isp and cam is turned off in pm_genpd_summary, however
>>>>>> cam clocks are still on.
>>>>>>
>>>>>> Should'nt the clocks be turned off first and then the domain.
>>>>>
>>>>> No, the clocks in clock hierarchy are usually not disabled as a part
>>>>> of domain power off. However respective drivers attached to a power
>>>>> domain, when entering runtime suspend, should disable these clocks.
>>>>>
>>>>
>>>> That is true, runtime suspend framework should disable the clocks.
>>>> Hence, my question is if the clocks are active, Will disabling the
>>>> domain take effect? Because when i print the status - they show that
>>>> the domain is OFF.
>>>
>>> Yes, the domain will be disabled.
>>>
>>>>
>>>> On my system, what i see is camera clocks are active, But the domain
>>>> is OFF. Hence, i am not sure if camera subsystem is consuming power or
>>>> not since clocks are on.
>>>>
>>>> >From my understanding, the domain status should not indicate OFF if
>>>> any clocks are still on.
>>>
>>> Clocks and power domains in Linux are (mostly) independent subsystems so
>>> the domain will be disabled even though some clocks are active. Some
>>> clocks may be even missing in Linux (not defined) - it doesn't matter...
>>>
>>> It doesn't matter except in all the cases when this is a bug. :) The
>>> clocks should be disabled and some muxes reparented to oscclk. One of
>>> the cases we had was hang on reading clk_summary when power domain is off.
>>>
>>
>> Thanks for your pointers.
>>
>> When i boot the kernel, I see that the following domains are currently
>> on. I see that the following domains are currently on, Linux does not
>> know about these domains and they are permanently on.
>>
>> pd-fsys, pd-psgen, pd-peric, pd-wcore
>>
>> Is there a way of finding out if these can be shut off. i dont think
>> if clocks for the IP's serviced by these domains and the domain itself
>> is correctly tied-in in runtime-pm. As these domains are always on.
>
> The datasheet/reference manual in the "Local power control" describes
> conditions whether they could be disabled or not. It also documents what
> is required to power off them.
>
>>
>> Also, what kernel are you using?
>
> Usually mainline and linux-next.
>
> In some of the projects I use different ones, e.g. these on tizen.org
> (https://review.tizen.org/git/?p=platform/kernel/linux-exynos.git;a=summary).
>
> Best regards,
> Krzysztof



-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozłowski]

2016-01-20 19:35 GMT+09:00 Sriram V <vshrirama@gmail.com>:
> Hi Krzysztof,
>
>
> Are you saying that only power domains under "Local Power Control" can
> be attached to runtime PM and can be turned off?
>
> In that case, i dont see other domains such as peris, psgen, wcore in
> there, Hence these need to be always on?
>
> is my understanding correct?

I don't know what kind of documentation you have. Judging by the
question above I think that you are asking me about interpretation of
datasheet... Really, your question is very strange... My Exynos5422
datasheet is quite specific on this topic.
Anyway you should ask for that to the vendor of the SoC - Samsung LSI.
It looks like you are looking for a support which should be given not
here, but by the vendor (because they sold you the SoC along with the
code).
Just to make it clear: I am not related anyhow to the Samsung LSI. I
am speaking only for myself.

As for the domains, you cannot independently control (under runtime
PM) all of the domains. Only some of them. Datasheet explicitly
describes what can be controlled. Other stuff is being turned off when
entering deeper sleep modes (like AFTR, SLEEP) - you can look for this
in cpuidle or suspend related code (open-source or vendor's).

Best regards,
Krzysztof

Re: domain off

[Sriram V]

Hi Krzystof,

Thanks for the clarifications.

On Wed, Jan 20, 2016 at 7:00 PM, Krzysztof Kozłowski
<k.kozlowski.k@gmail.com> wrote:
> 2016-01-20 19:35 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>> Hi Krzysztof,
>>
>>
>> Are you saying that only power domains under "Local Power Control" can
>> be attached to runtime PM and can be turned off?
>>
>> In that case, i dont see other domains such as peris, psgen, wcore in
>> there, Hence these need to be always on?
>>
>> is my understanding correct?
>
> I don't know what kind of documentation you have. Judging by the
> question above I think that you are asking me about interpretation of
> datasheet... Really, your question is very strange... My Exynos5422
> datasheet is quite specific on this topic.
> Anyway you should ask for that to the vendor of the SoC - Samsung LSI.
> It looks like you are looking for a support which should be given not
> here, but by the vendor (because they sold you the SoC along with the
> code).
> Just to make it clear: I am not related anyhow to the Samsung LSI. I
> am speaking only for myself.
>
> As for the domains, you cannot independently control (under runtime
> PM) all of the domains. Only some of them. Datasheet explicitly
> describes what can be controlled. Other stuff is being turned off when
> entering deeper sleep modes (like AFTR, SLEEP) - you can look for this
> in cpuidle or suspend related code (open-source or vendor's).


I could not find in the suspend code where it turns off all the domains on 5422
in the vendor and linux-next kernel. Am i missing something?

Or are you referring to some code?


Thanks and Regards,
Sriram



>
> Best regards,
> Krzysztof



-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozłowski]

2016-01-25 18:54 GMT+09:00 Sriram V <vshrirama@gmail.com>:
> Hi Krzystof,
>
> Thanks for the clarifications.
>
> On Wed, Jan 20, 2016 at 7:00 PM, Krzysztof Kozłowski
> <k.kozlowski.k@gmail.com> wrote:
>> 2016-01-20 19:35 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>> Hi Krzysztof,
>>>
>>>
>>> Are you saying that only power domains under "Local Power Control" can
>>> be attached to runtime PM and can be turned off?
>>>
>>> In that case, i dont see other domains such as peris, psgen, wcore in
>>> there, Hence these need to be always on?
>>>
>>> is my understanding correct?
>>
>> I don't know what kind of documentation you have. Judging by the
>> question above I think that you are asking me about interpretation of
>> datasheet... Really, your question is very strange... My Exynos5422
>> datasheet is quite specific on this topic.
>> Anyway you should ask for that to the vendor of the SoC - Samsung LSI.
>> It looks like you are looking for a support which should be given not
>> here, but by the vendor (because they sold you the SoC along with the
>> code).
>> Just to make it clear: I am not related anyhow to the Samsung LSI. I
>> am speaking only for myself.
>>
>> As for the domains, you cannot independently control (under runtime
>> PM) all of the domains. Only some of them. Datasheet explicitly
>> describes what can be controlled. Other stuff is being turned off when
>> entering deeper sleep modes (like AFTR, SLEEP) - you can look for this
>> in cpuidle or suspend related code (open-source or vendor's).
>
>
> I could not find in the suspend code where it turns off all the domains on 5422
> in the vendor and linux-next kernel. Am i missing something?
>
> Or are you referring to some code?

Here:
http://lxr.free-electrons.com/source/arch/arm/mach-exynos/pmu.c#L443
The exynos5420_pmu_config contains values for PMU registers applied
when entering AFTR/LPA (cpuidle) or SLEEP (suspend to RAM) modes.
Around line #534 the configuration for power domain starts:
http://lxr.free-electrons.com/source/arch/arm/mach-exynos/pmu.c#L534

The array should configure all power domains to certain state (off or on).

Best regards,
Krzysztof

Re: domain off

[Sriram V]

Hi Krzystof,

Thanks, right now we are seeing that the board hangs whenever the
those domains are list of domains
are modified by the cpuidle framework.

Also, I have a related question, devfreq also updates the clocks as
well depending on bus load conditions

If a particular domains are OFF. Wont the devfreq update produce a
crash. However, this does not happen.
I do see lots of dout_aclk print which modifies the freq of clocks of
different modules But that does not produce a crash.

Am i missing something here?


Any idea why is this?



On Mon, Jan 25, 2016 at 5:10 PM, Krzysztof Kozłowski
<k.kozlowski.k@gmail.com> wrote:
> 2016-01-25 18:54 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>> Hi Krzystof,
>>
>> Thanks for the clarifications.
>>
>> On Wed, Jan 20, 2016 at 7:00 PM, Krzysztof Kozłowski
>> <k.kozlowski.k@gmail.com> wrote:
>>> 2016-01-20 19:35 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>>>> Hi Krzysztof,
>>>>
>>>>
>>>> Are you saying that only power domains under "Local Power Control" can
>>>> be attached to runtime PM and can be turned off?
>>>>
>>>> In that case, i dont see other domains such as peris, psgen, wcore in
>>>> there, Hence these need to be always on?
>>>>
>>>> is my understanding correct?
>>>
>>> I don't know what kind of documentation you have. Judging by the
>>> question above I think that you are asking me about interpretation of
>>> datasheet... Really, your question is very strange... My Exynos5422
>>> datasheet is quite specific on this topic.
>>> Anyway you should ask for that to the vendor of the SoC - Samsung LSI.
>>> It looks like you are looking for a support which should be given not
>>> here, but by the vendor (because they sold you the SoC along with the
>>> code).
>>> Just to make it clear: I am not related anyhow to the Samsung LSI. I
>>> am speaking only for myself.
>>>
>>> As for the domains, you cannot independently control (under runtime
>>> PM) all of the domains. Only some of them. Datasheet explicitly
>>> describes what can be controlled. Other stuff is being turned off when
>>> entering deeper sleep modes (like AFTR, SLEEP) - you can look for this
>>> in cpuidle or suspend related code (open-source or vendor's).
>>
>>
>> I could not find in the suspend code where it turns off all the domains on 5422
>> in the vendor and linux-next kernel. Am i missing something?
>>
>> Or are you referring to some code?
>
> Here:
> http://lxr.free-electrons.com/source/arch/arm/mach-exynos/pmu.c#L443
> The exynos5420_pmu_config contains values for PMU registers applied
> when entering AFTR/LPA (cpuidle) or SLEEP (suspend to RAM) modes.
> Around line #534 the configuration for power domain starts:
> http://lxr.free-electrons.com/source/arch/arm/mach-exynos/pmu.c#L534
>
> The array should configure all power domains to certain state (off or on).
>
> Best regards,
> Krzysztof



-- 
Regards,
Sriram

Re: domain off

[Krzysztof Kozłowski]

2016-02-04 21:21 GMT+09:00 Sriram V <vshrirama@gmail.com>:
>
> Hi Krzystof,
>
> Thanks, right now we are seeing that the board hangs whenever the
> those domains are list of domains
> are modified by the cpuidle framework.
>
> Also, I have a related question, devfreq also updates the clocks as
> well depending on bus load conditions
>
> If a particular domains are OFF. Wont the devfreq update produce a
> crash. However, this does not happen.
> I do see lots of dout_aclk print which modifies the freq of clocks of
> different modules But that does not produce a crash.
>
> Am i missing something here?

This is not related to cpuidle. For Exynos usually the cpuidle enters
system sleep modes, so entire system is sleeping and devfreq does not
work in that moment.

However when power domains are off because of runtime PM, changing
clocks may be completely ok. Many clocks are not part of the power
domain itself. In the same time devfreq usually changes top level
clocks which is also quite safe.

Best regards,
Krzysztof