Re: [PATCH 1/3] iio: core: Add devm_ APIs for iio_channel_{get,release}

[Jonathan Cameron]

On 10/04/16 18:35, Laxman Dewangan wrote:
> 
> On Sunday 10 April 2016 07:35 PM, Jonathan Cameron wrote:
>> On 06/04/16 15:58, Laxman Dewangan wrote:
>>> Hi Daniel,
>>>
>>>
>>> On Wednesday 06 April 2016 07:19 PM, Daniel Baluta wrote:
>>>> On Wed, Apr 6, 2016 at 1:31 PM, Laxman Dewangan <ldewangan@xxxxxxxxxx> wrote:
>>>>> Some of kernel driver uses the IIO framework to get the sensor
>>>>> value via ADC or IIO HW driver. The client driver get iio channel
>>>>> by iio_channel_get() and release it by calling iio_channel_release().
>>>>>
>>>>> Add resource managed version (devm_*) of these APIs so that if client
>>>>> calls the devm_iio_channel_get() then it need not to release it explicitly,
>>>>> it can be done by managed device framework when driver get un-binded.
>>>>>
>>>>> This reduces the code in error path and also need of .remove callback in
>>>>> some cases.
>>>>>
>>>> Please provide at least one example of code that uses this API.
>>> Most of client for this APIs are in other subsystem.
>>> When I was working on the patch
>>> [PATCH 2/2] thermal: generic-adc: Add ADC based thermal sensor driver
>>>
>>> if I have devm_iio_channel_get() then I can get .remove callback at all.
>>>
>>> I did not use this new APIs in my patch because they are in different subsystem.
>> It's actually worse than that having taken a quick look at the generic-adc thermal patch
>> you reference above.
>> (perhaps worth cc'ing linux-iio for next version of that).
> Sure. I will CC.
> 
>>
>> Without this devm function set you have a race in remove in which I think you can
>> get attempts to access the channels after they have been released...
> Yaah, possibly race for very small time possible.
> 
> The limitation of devm_ api usage is that, we can keep using this
> till we have devm_ api continuous and if some resource are not there
> for devm_ then we can not use further. Possibly, I need to wait for
> the devm_iio_channel_get() to merge and available for all subsystem
> to use (next release) and then only I can use
> devm_thermal_zone_of_sensor_register().
> 
The alternative would be to merge this devm_ support as a prerequisite for your thermal
patches and have it go through that tree.  As it's self contained I have
no particular problem with that if you'd prefer to do it that way.

Otherwise, you will need to do as you say above (not use
devm_thermal_zone_of_sensor_register) to make sure it isn't broken in the meantime.

Jonathan