Re: [PATCH v1 1/2] of: add devm_ functions for populate and depopulate

[Rob Herring]

On Fri, Feb 24, 2017 at 9:26 AM, Benjamin Gaignard
<benjamin.gaignard@xxxxxxxxxx> wrote:
> 2017-02-24 16:20 GMT+01:00 Rob Herring <robh+dt@xxxxxxxxxx>:
>> On Fri, Feb 24, 2017 at 9:13 AM, Benjamin Gaignard
>> <benjamin.gaignard@xxxxxxxxxx> wrote:
>>> 2017-02-24 15:17 GMT+01:00 Rob Herring <robh+dt@xxxxxxxxxx>:
>>>> On Fri, Feb 17, 2017 at 8:31 AM, Benjamin Gaignard
>>>> <benjamin.gaignard@xxxxxxxxxx> wrote:
>>>>> Lost of calls to of_platform_populate() are not unbalanced by a call
>>>>
>>>> s/Lost/Lots/
>>>>
>>>>> to of_platform_depopulate(). This create issues while drivers are
>>>>> bind/unbind.
>>>>>
>>>>> In way to solve those issues is to add devm_of_platform_populate()
>>>>> which will call of_platform_depopulate() when the device is unbound
>>>>> from the bus.
>>>>
>>>> One complication is of_platform_populate is designed to be called
>>>> multiple times. We call it with the default match table and then
>>>> platforms can call it again with a custom match table for example.
>>>
>>> I do not plan to use it every where but only in some drivers probe()
>>> to avoid adding goto to call of_platform_depopulate() for each error
>>> cases which may occur after calling populate.
>>>
>>>>
>>>>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@xxxxxxxxxx>
>>>>> ---
>>>>>  drivers/of/platform.c       | 77 +++++++++++++++++++++++++++++++++++++++++++++
>>>>>  include/linux/of_platform.h | 20 ++++++++++++
>>>>>  2 files changed, 97 insertions(+)
>>>>>
>>>>> diff --git a/drivers/of/platform.c b/drivers/of/platform.c
>>>>> index b8064bc..3dbebf7 100644
>>>>> --- a/drivers/of/platform.c
>>>>> +++ b/drivers/of/platform.c
>>>>> @@ -571,6 +571,83 @@ void of_platform_depopulate(struct device *parent)
>>>>>  }
>>>>>  EXPORT_SYMBOL_GPL(of_platform_depopulate);
>>>>>
>>>>> +static void devm_of_platform_populate_release(struct device *dev, void *res)
>>>>> +{
>>>>> +       of_platform_depopulate(*(struct device **)res);
>>>>> +}
>>>>> +
>>>>> +/**
>>>>> + * devm_of_platform_populate() - Populate platform_devices from device tree data
>>>>> + * @dev: device that requested to populate from device tree data
>>>>> + * @root: parent of the first level to probe or NULL for the root of the tree
>>>>> + * @matches: match table, NULL to use the default
>>>>
>>>> NULL is no match table, not the default which means only populate
>>>> immediate children.
>>>
>>> I have copy of_platform_populate() description...
>>> I replace it by: @matches: match table (could be NULL)
>>>
>>>>
>>>>> + * @lookup: auxdata table for matching id and platform_data with device nodes
>>>>> + * @parent: parent to hook devices from, NULL for toplevel
>>>>
>>>> I think this needs to be a bit different args as the use is limited.
>>>> root and parent must not be NULL. dev should be the same as parent.
>>>> lookup was for legacy, so drop that.
>>>
>>> So function prototype will become:
>>> int devm_of_platform_populate(struct device *dev, struct device_node
>>> *root, const struct of_device_id *matches)
>>
>> I just noticed something else too. dev->of_node should equal root I
>> think, so we can get rid of root param.
>
> match parameter is very often set to NULL since this function will
> have a limited
> scope why not also remove it and only keep dev ?

That's fine with me.

Rob