linux-pm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* Re: [PATCH] iio: adc: stm32-adc: fix runtime autosuspend delay when slow polling
       [not found] <1579854369-7972-1-git-send-email-fabrice.gasnier@st.com>
@ 2020-02-02 15:33 ` Jonathan Cameron
  2020-02-12 14:02   ` Fabrice Gasnier
  0 siblings, 1 reply; 3+ messages in thread
From: Jonathan Cameron @ 2020-02-02 15:33 UTC (permalink / raw)
  To: Fabrice Gasnier
  Cc: linux-arm-kernel, linux-kernel, mcoquelin.stm32,
	alexandre.torgue, olivier.moysan, linux-iio, lars, knaack.h,
	pmeerw, linux-stm32, Rafael J. Wysocki, linux-pm

On Fri, 24 Jan 2020 09:26:09 +0100
Fabrice Gasnier <fabrice.gasnier@st.com> wrote:

> When the ADC is runtime suspended and starting a conversion, the stm32-adc
> driver calls pm_runtime_get_sync() that gets cascaded to the parent
> (e.g. runtime resume of stm32-adc-core driver). This also kicks the
> autosuspend delay (e.g. 2s) of the parent.
> Once the ADC is active, calling pm_runtime_get_sync() again (upon a new
> capture) won't kick the autosuspend delay for the parent (stm32-adc-core
> driver) as already active.
> 
> Currently, this makes the stm32-adc-core driver go in suspend state
> every 2s when doing slow polling. As an example, doing a capture, e.g.
> cat in_voltageY_raw at a 0.2s rate, the auto suspend delay for the parent
> isn't refreshed. Once it expires, the parent immediately falls into
> runtime suspended state, in between two captures, as soon as the child
> driver falls into runtime suspend state:
> - e.g. after 2s, + child calls pm_runtime_put_autosuspend() + 100ms
>   autosuspend delay of the child.
> - stm32-adc-core switches off regulators, clocks and so on.
> - They get switched on back again 100ms later in this example (at 2.2s).
> 
> So, add an explicit call to pm_runtime_mark_last_busy() for the parent
> driver (stm32-adc-core), synchronously with the child driver (stm32-adc),
> to avoid this.
> 
> Fixes: 9bdbb1139ca1 ("iio: adc: stm32-adc: add power management support")
> 
> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>

Hi Fabrice,

Whilst this will clearly work, it seems like a somewhat adhoc solution.
Power management specialists (cc'd):  Is this what we should be doing, or
have Fabrice and I both missed something that we should be doing here?

Thanks,

Jonathan

> ---
>  drivers/iio/adc/stm32-adc.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/iio/adc/stm32-adc.c b/drivers/iio/adc/stm32-adc.c
> index 3b291d7..670157e 100644
> --- a/drivers/iio/adc/stm32-adc.c
> +++ b/drivers/iio/adc/stm32-adc.c
> @@ -1157,6 +1157,7 @@ static int stm32_adc_single_conv(struct iio_dev *indio_dev,
>  
>  	stm32_adc_conv_irq_disable(adc);
>  
> +	pm_runtime_mark_last_busy(dev->parent);
>  	pm_runtime_mark_last_busy(dev);
>  	pm_runtime_put_autosuspend(dev);
>  
> @@ -1278,6 +1279,7 @@ static int stm32_adc_update_scan_mode(struct iio_dev *indio_dev,
>  	adc->num_conv = bitmap_weight(scan_mask, indio_dev->masklength);
>  
>  	ret = stm32_adc_conf_scan_seq(indio_dev, scan_mask);
> +	pm_runtime_mark_last_busy(dev->parent);
>  	pm_runtime_mark_last_busy(dev);
>  	pm_runtime_put_autosuspend(dev);
>  
> @@ -1329,6 +1331,7 @@ static int stm32_adc_debugfs_reg_access(struct iio_dev *indio_dev,
>  	else
>  		*readval = stm32_adc_readl(adc, reg);
>  
> +	pm_runtime_mark_last_busy(dev->parent);
>  	pm_runtime_mark_last_busy(dev);
>  	pm_runtime_put_autosuspend(dev);
>  
> @@ -1451,6 +1454,7 @@ static int __stm32_adc_buffer_postenable(struct iio_dev *indio_dev)
>  err_clr_trig:
>  	stm32_adc_set_trig(indio_dev, NULL);
>  err_pm_put:
> +	pm_runtime_mark_last_busy(dev->parent);
>  	pm_runtime_mark_last_busy(dev);
>  	pm_runtime_put_autosuspend(dev);
>  
> @@ -1487,6 +1491,7 @@ static void __stm32_adc_buffer_predisable(struct iio_dev *indio_dev)
>  	if (stm32_adc_set_trig(indio_dev, NULL))
>  		dev_err(&indio_dev->dev, "Can't clear trigger\n");
>  
> +	pm_runtime_mark_last_busy(dev->parent);
>  	pm_runtime_mark_last_busy(dev);
>  	pm_runtime_put_autosuspend(dev);
>  }
> @@ -1874,6 +1879,7 @@ static int stm32_adc_probe(struct platform_device *pdev)
>  		goto err_hw_stop;
>  	}
>  
> +	pm_runtime_mark_last_busy(dev->parent);
>  	pm_runtime_mark_last_busy(dev);
>  	pm_runtime_put_autosuspend(dev);
>  


^ permalink raw reply	[flat|nested] 3+ messages in thread
* Re: [PATCH] iio: adc: stm32-adc: fix runtime autosuspend delay when slow polling
  2020-02-02 15:33 ` [PATCH] iio: adc: stm32-adc: fix runtime autosuspend delay when slow polling Jonathan Cameron
@ 2020-02-12 14:02   ` Fabrice Gasnier
  2020-03-06  8:22     ` Fabrice Gasnier
  0 siblings, 1 reply; 3+ messages in thread
From: Fabrice Gasnier @ 2020-02-12 14:02 UTC (permalink / raw)
  To: Jonathan Cameron, Rafael J. Wysocki
  Cc: linux-arm-kernel, linux-kernel, mcoquelin.stm32,
	alexandre.torgue, olivier.moysan, linux-iio, lars, knaack.h,
	pmeerw, linux-stm32, linux-pm, ulf.hansson, khilman

On 2/2/20 4:33 PM, Jonathan Cameron wrote:
> On Fri, 24 Jan 2020 09:26:09 +0100
> Fabrice Gasnier <fabrice.gasnier@st.com> wrote:
> 
>> When the ADC is runtime suspended and starting a conversion, the stm32-adc
>> driver calls pm_runtime_get_sync() that gets cascaded to the parent
>> (e.g. runtime resume of stm32-adc-core driver). This also kicks the
>> autosuspend delay (e.g. 2s) of the parent.
>> Once the ADC is active, calling pm_runtime_get_sync() again (upon a new
>> capture) won't kick the autosuspend delay for the parent (stm32-adc-core
>> driver) as already active.
>>
>> Currently, this makes the stm32-adc-core driver go in suspend state
>> every 2s when doing slow polling. As an example, doing a capture, e.g.
>> cat in_voltageY_raw at a 0.2s rate, the auto suspend delay for the parent
>> isn't refreshed. Once it expires, the parent immediately falls into
>> runtime suspended state, in between two captures, as soon as the child
>> driver falls into runtime suspend state:
>> - e.g. after 2s, + child calls pm_runtime_put_autosuspend() + 100ms
>>   autosuspend delay of the child.
>> - stm32-adc-core switches off regulators, clocks and so on.
>> - They get switched on back again 100ms later in this example (at 2.2s).
>>
>> So, add an explicit call to pm_runtime_mark_last_busy() for the parent
>> driver (stm32-adc-core), synchronously with the child driver (stm32-adc),
>> to avoid this.
>>
>> Fixes: 9bdbb1139ca1 ("iio: adc: stm32-adc: add power management support")
>>
>> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>
> 
> Hi Fabrice,
> 
> Whilst this will clearly work, it seems like a somewhat adhoc solution.
> Power management specialists (cc'd):  Is this what we should be doing, or
> have Fabrice and I both missed something that we should be doing here?

Hi all, PM specialists,

As per my understanding, pm_runtime_mark_last_busy() doesn't cascade to
the parent device:

- in pm_runtime.h:
static inline void pm_runtime_mark_last_busy(struct device *dev)
{
	WRITE_ONCE(dev->power.last_busy, ktime_get_mono_fast_ns());
}

STM32 ADC driver uses a model with an autosupsend delay for
- a parent driver to handle common resources, registers etc.
- child drivers for each ADC.

So the question is on how to fix the behavior I described:
1: Child  activity with "short" autosuspend_delay
2: Parent activity with "longer" autosuspend_delay
     _     _     _     _     _       _     _     _
1: _| |___| |___| |___| |___| |_..._| |___| |___| |_...

    v v   v v   v v   v v   v v ... v v   v v   v v
    | |                                     |   |
    | +- pm_runtime_mark_last_busy()        |   |
    | +- pm_runtime_put_autosuspend()       v   |
    |                                       |   |
    +--- pm_runtime_get_sync()              |   v
    |                                       |   |
    +---> expires after autosuspend_delay   |   |
    |                                       |   |
    v                                       v   v
     _______________________________________     ___...
2: _|                           ...         |___|

Glitches on parent dev near autosuspend_delay ^

- does the child driver needs to "kick" parent driver with
pm_runtime_mark_last_busy(), as proposed in current patch ?

- or is it something that should be done by PM runtime core routines ?
e.g. make pm_runtime_mark_last_busy() recursive or something else ?

Please advise
Best regards,
Fabrice

> 
> Thanks,
> 
> Jonathan
> 
>> ---
>>  drivers/iio/adc/stm32-adc.c | 6 ++++++
>>  1 file changed, 6 insertions(+)
>>
>> diff --git a/drivers/iio/adc/stm32-adc.c b/drivers/iio/adc/stm32-adc.c
>> index 3b291d7..670157e 100644
>> --- a/drivers/iio/adc/stm32-adc.c
>> +++ b/drivers/iio/adc/stm32-adc.c
>> @@ -1157,6 +1157,7 @@ static int stm32_adc_single_conv(struct iio_dev *indio_dev,
>>  
>>  	stm32_adc_conv_irq_disable(adc);
>>  
>> +	pm_runtime_mark_last_busy(dev->parent);
>>  	pm_runtime_mark_last_busy(dev);
>>  	pm_runtime_put_autosuspend(dev);
>>  
>> @@ -1278,6 +1279,7 @@ static int stm32_adc_update_scan_mode(struct iio_dev *indio_dev,
>>  	adc->num_conv = bitmap_weight(scan_mask, indio_dev->masklength);
>>  
>>  	ret = stm32_adc_conf_scan_seq(indio_dev, scan_mask);
>> +	pm_runtime_mark_last_busy(dev->parent);
>>  	pm_runtime_mark_last_busy(dev);
>>  	pm_runtime_put_autosuspend(dev);
>>  
>> @@ -1329,6 +1331,7 @@ static int stm32_adc_debugfs_reg_access(struct iio_dev *indio_dev,
>>  	else
>>  		*readval = stm32_adc_readl(adc, reg);
>>  
>> +	pm_runtime_mark_last_busy(dev->parent);
>>  	pm_runtime_mark_last_busy(dev);
>>  	pm_runtime_put_autosuspend(dev);
>>  
>> @@ -1451,6 +1454,7 @@ static int __stm32_adc_buffer_postenable(struct iio_dev *indio_dev)
>>  err_clr_trig:
>>  	stm32_adc_set_trig(indio_dev, NULL);
>>  err_pm_put:
>> +	pm_runtime_mark_last_busy(dev->parent);
>>  	pm_runtime_mark_last_busy(dev);
>>  	pm_runtime_put_autosuspend(dev);
>>  
>> @@ -1487,6 +1491,7 @@ static void __stm32_adc_buffer_predisable(struct iio_dev *indio_dev)
>>  	if (stm32_adc_set_trig(indio_dev, NULL))
>>  		dev_err(&indio_dev->dev, "Can't clear trigger\n");
>>  
>> +	pm_runtime_mark_last_busy(dev->parent);
>>  	pm_runtime_mark_last_busy(dev);
>>  	pm_runtime_put_autosuspend(dev);
>>  }
>> @@ -1874,6 +1879,7 @@ static int stm32_adc_probe(struct platform_device *pdev)
>>  		goto err_hw_stop;
>>  	}
>>  
>> +	pm_runtime_mark_last_busy(dev->parent);
>>  	pm_runtime_mark_last_busy(dev);
>>  	pm_runtime_put_autosuspend(dev);
>>  
> 

^ permalink raw reply	[flat|nested] 3+ messages in thread
* Re: [PATCH] iio: adc: stm32-adc: fix runtime autosuspend delay when slow polling
  2020-02-12 14:02   ` Fabrice Gasnier
@ 2020-03-06  8:22     ` Fabrice Gasnier
  0 siblings, 0 replies; 3+ messages in thread
From: Fabrice Gasnier @ 2020-03-06  8:22 UTC (permalink / raw)
  To: Jonathan Cameron, Rafael J. Wysocki, ulf.hansson
  Cc: linux-arm-kernel, linux-kernel, mcoquelin.stm32,
	alexandre.torgue, olivier.moysan, linux-iio, lars, knaack.h,
	pmeerw, linux-stm32, linux-pm, khilman, Vincent Guittot,
	Benjamin GAIGNARD

On 2/12/20 3:02 PM, Fabrice Gasnier wrote:
> On 2/2/20 4:33 PM, Jonathan Cameron wrote:
>> On Fri, 24 Jan 2020 09:26:09 +0100
>> Fabrice Gasnier <fabrice.gasnier@st.com> wrote:
>>
>>> When the ADC is runtime suspended and starting a conversion, the stm32-adc
>>> driver calls pm_runtime_get_sync() that gets cascaded to the parent
>>> (e.g. runtime resume of stm32-adc-core driver). This also kicks the
>>> autosuspend delay (e.g. 2s) of the parent.
>>> Once the ADC is active, calling pm_runtime_get_sync() again (upon a new
>>> capture) won't kick the autosuspend delay for the parent (stm32-adc-core
>>> driver) as already active.
>>>
>>> Currently, this makes the stm32-adc-core driver go in suspend state
>>> every 2s when doing slow polling. As an example, doing a capture, e.g.
>>> cat in_voltageY_raw at a 0.2s rate, the auto suspend delay for the parent
>>> isn't refreshed. Once it expires, the parent immediately falls into
>>> runtime suspended state, in between two captures, as soon as the child
>>> driver falls into runtime suspend state:
>>> - e.g. after 2s, + child calls pm_runtime_put_autosuspend() + 100ms
>>>   autosuspend delay of the child.
>>> - stm32-adc-core switches off regulators, clocks and so on.
>>> - They get switched on back again 100ms later in this example (at 2.2s).
>>>
>>> So, add an explicit call to pm_runtime_mark_last_busy() for the parent
>>> driver (stm32-adc-core), synchronously with the child driver (stm32-adc),
>>> to avoid this.
>>>
>>> Fixes: 9bdbb1139ca1 ("iio: adc: stm32-adc: add power management support")
>>>
>>> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>
>>
>> Hi Fabrice,
>>
>> Whilst this will clearly work, it seems like a somewhat adhoc solution.
>> Power management specialists (cc'd):  Is this what we should be doing, or
>> have Fabrice and I both missed something that we should be doing here?
> 
> Hi all, PM specialists,
> 
> As per my understanding, pm_runtime_mark_last_busy() doesn't cascade to
> the parent device:
> 
> - in pm_runtime.h:
> static inline void pm_runtime_mark_last_busy(struct device *dev)
> {
> 	WRITE_ONCE(dev->power.last_busy, ktime_get_mono_fast_ns());
> }

Hi all, PM specialists,

Gentle reminder on this topic.
There maybe several solutions to address this. Not sure what would be
an acceptable / preferred fix ?

Just to summarize:
---
A - Current post: I'm wondering if this is suitable to call
    pm_runtime_mark_last_busy() every time, for the parent device from
    child driver ?

---
B - Is it suitable to update the PM runtime, so it's recursive, to
    refresh "last_busy" for the parent(s) ? e.g. something like:

 static inline void pm_runtime_mark_last_busy(struct device *dev)
 {
        WRITE_ONCE(dev->power.last_busy, ktime_get_mono_fast_ns());
+       if (dev->parent)
+               pm_runtime_mark_last_busy(dev->parent);
 }

---
C - I find few drivers that rely on the rmp_idle() callback in the
    parent to call pm_runtime_mark_last_busy(), like:
  -> dwc3_runtime_idle()
  -> gb_bundle_idle()

  -> that may lead in the stm32-adc-core driver to add:

+static int stm32_adc_core_runtime_idle(struct device *dev)
+{
+       pm_runtime_mark_last_busy(dev);
+
+       return 0;
+}

 static const struct dev_pm_ops stm32_adc_core_pm_ops = {
        SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
                                pm_runtime_force_resume)
        SET_RUNTIME_PM_OPS(stm32_adc_core_runtime_suspend,
                           stm32_adc_core_runtime_resume,
-                          NULL)
+                          stm32_adc_core_runtime_idle)
 };


I've test all these solutions. Would you have some recommendations ?

Please advise,
Best regards,
Fabrice

> 
> STM32 ADC driver uses a model with an autosupsend delay for
> - a parent driver to handle common resources, registers etc.
> - child drivers for each ADC.
> 
> So the question is on how to fix the behavior I described:
> 1: Child  activity with "short" autosuspend_delay
> 2: Parent activity with "longer" autosuspend_delay
>      _     _     _     _     _       _     _     _
> 1: _| |___| |___| |___| |___| |_..._| |___| |___| |_...
> 
>     v v   v v   v v   v v   v v ... v v   v v   v v
>     | |                                     |   |
>     | +- pm_runtime_mark_last_busy()        |   |
>     | +- pm_runtime_put_autosuspend()       v   |
>     |                                       |   |
>     +--- pm_runtime_get_sync()              |   v
>     |                                       |   |
>     +---> expires after autosuspend_delay   |   |
>     |                                       |   |
>     v                                       v   v
>      _______________________________________     ___...
> 2: _|                           ...         |___|
> 
> Glitches on parent dev near autosuspend_delay ^
> 
> - does the child driver needs to "kick" parent driver with
> pm_runtime_mark_last_busy(), as proposed in current patch ?
> 
> - or is it something that should be done by PM runtime core routines ?
> e.g. make pm_runtime_mark_last_busy() recursive or something else ?
> 
> Please advise
> Best regards,
> Fabrice
> 
>>
>> Thanks,
>>
>> Jonathan
>>
>>> ---
>>>  drivers/iio/adc/stm32-adc.c | 6 ++++++
>>>  1 file changed, 6 insertions(+)
>>>
>>> diff --git a/drivers/iio/adc/stm32-adc.c b/drivers/iio/adc/stm32-adc.c
>>> index 3b291d7..670157e 100644
>>> --- a/drivers/iio/adc/stm32-adc.c
>>> +++ b/drivers/iio/adc/stm32-adc.c
>>> @@ -1157,6 +1157,7 @@ static int stm32_adc_single_conv(struct iio_dev *indio_dev,
>>>  
>>>  	stm32_adc_conv_irq_disable(adc);
>>>  
>>> +	pm_runtime_mark_last_busy(dev->parent);
>>>  	pm_runtime_mark_last_busy(dev);
>>>  	pm_runtime_put_autosuspend(dev);
>>>  
>>> @@ -1278,6 +1279,7 @@ static int stm32_adc_update_scan_mode(struct iio_dev *indio_dev,
>>>  	adc->num_conv = bitmap_weight(scan_mask, indio_dev->masklength);
>>>  
>>>  	ret = stm32_adc_conf_scan_seq(indio_dev, scan_mask);
>>> +	pm_runtime_mark_last_busy(dev->parent);
>>>  	pm_runtime_mark_last_busy(dev);
>>>  	pm_runtime_put_autosuspend(dev);
>>>  
>>> @@ -1329,6 +1331,7 @@ static int stm32_adc_debugfs_reg_access(struct iio_dev *indio_dev,
>>>  	else
>>>  		*readval = stm32_adc_readl(adc, reg);
>>>  
>>> +	pm_runtime_mark_last_busy(dev->parent);
>>>  	pm_runtime_mark_last_busy(dev);
>>>  	pm_runtime_put_autosuspend(dev);
>>>  
>>> @@ -1451,6 +1454,7 @@ static int __stm32_adc_buffer_postenable(struct iio_dev *indio_dev)
>>>  err_clr_trig:
>>>  	stm32_adc_set_trig(indio_dev, NULL);
>>>  err_pm_put:
>>> +	pm_runtime_mark_last_busy(dev->parent);
>>>  	pm_runtime_mark_last_busy(dev);
>>>  	pm_runtime_put_autosuspend(dev);
>>>  
>>> @@ -1487,6 +1491,7 @@ static void __stm32_adc_buffer_predisable(struct iio_dev *indio_dev)
>>>  	if (stm32_adc_set_trig(indio_dev, NULL))
>>>  		dev_err(&indio_dev->dev, "Can't clear trigger\n");
>>>  
>>> +	pm_runtime_mark_last_busy(dev->parent);
>>>  	pm_runtime_mark_last_busy(dev);
>>>  	pm_runtime_put_autosuspend(dev);
>>>  }
>>> @@ -1874,6 +1879,7 @@ static int stm32_adc_probe(struct platform_device *pdev)
>>>  		goto err_hw_stop;
>>>  	}
>>>  
>>> +	pm_runtime_mark_last_busy(dev->parent);
>>>  	pm_runtime_mark_last_busy(dev);
>>>  	pm_runtime_put_autosuspend(dev);
>>>  
>>

^ permalink raw reply	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2020-03-06  8:23 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <1579854369-7972-1-git-send-email-fabrice.gasnier@st.com>
2020-02-02 15:33 ` [PATCH] iio: adc: stm32-adc: fix runtime autosuspend delay when slow polling Jonathan Cameron
2020-02-12 14:02   ` Fabrice Gasnier
2020-03-06  8:22     ` Fabrice Gasnier

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).