Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH] iio: adc: stm32-adc: fix runtime autosuspend delay when
 slow polling
From:   Fabrice Gasnier <fabrice.gasnier@st.com>
To:     Jonathan Cameron <jic23@kernel.org>,
        "Rafael J. Wysocki" <rjw@rjwysocki.net>, <ulf.hansson@linaro.org>
CC:     <linux-arm-kernel@lists.infradead.org>,
        <linux-kernel@vger.kernel.org>, <mcoquelin.stm32@gmail.com>,
        <alexandre.torgue@st.com>, <olivier.moysan@st.com>,
        <linux-iio@vger.kernel.org>, <lars@metafoo.de>, <knaack.h@gmx.de>,
        <pmeerw@pmeerw.net>, <linux-stm32@st-md-mailman.stormreply.com>,
        <linux-pm@vger.kernel.org>, <khilman@kernel.org>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Benjamin GAIGNARD <benjamin.gaignard@st.com>
References: <1579854369-7972-1-git-send-email-fabrice.gasnier@st.com>
 <20200202153354.3dae5863@archlinux>
 <d30cb29b-d15c-a9fe-8c95-7ce59ce15062@st.com>
Message-ID: <563e72b2-9bef-f00d-429c-db77d7689cc8@st.com>
Date:   Fri, 6 Mar 2020 09:22:25 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.9.0
MIME-Version: 1.0
In-Reply-To: <d30cb29b-d15c-a9fe-8c95-7ce59ce15062@st.com>
Content-Type: text/plain; charset="utf-8"
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

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);
>>>  
>>