Re: [PATCH v4 02/10] property: Add device_get_child_node_count_named()

[Matti Vaittinen <mazziesaccount@gmail.com>]

On 27/02/2025 16:49, Andy Shevchenko wrote:
> On Thu, Feb 27, 2025 at 10:01:49AM +0200, Matti Vaittinen wrote:
>> On 26/02/2025 16:11, Andy Shevchenko wrote:
>>> On Wed, Feb 26, 2025 at 04:04:02PM +0200, Matti Vaittinen wrote:
>>>> On 25/02/2025 15:59, Andy Shevchenko wrote:
>>>>> On Tue, Feb 25, 2025 at 03:29:17PM +0200, Matti Vaittinen wrote:
>>>>>> On 25/02/2025 12:39, Andy Shevchenko wrote:
>>>>>>> On Tue, Feb 25, 2025 at 12:29:31PM +0200, Matti Vaittinen wrote:
>>>>>>>> On 25/02/2025 12:21, Andy Shevchenko wrote:
>>>>>>>>> On Tue, Feb 25, 2025 at 11:40:16AM +0200, Heikki Krogerus wrote:
> 
> ...
> 
>>>>>>>>>>
>>>>>>>>>> I did not check how many users are you proposing for this, but if
>>>>>>>>>> there's only one, then IMO this should not be a global function yet.
>>>>>>>>>> It just feels to special case to me. But let's see what the others
>>>>>>>>>> think.
>>>>>>>>>
>>>>>>>>> The problem is that if somebody hides it, we might potentially see
>>>>>>>>> a duplication in the future. So I _slightly_ prefer to publish and
>>>>>>>>> then drop that after a few cycles if no users appear.
>>>>>>>>
>>>>>>>> After taking a very quick grep I spotted one other existing place where we
>>>>>>>> might be able to do direct conversion to use this function.
>>>>>>>>
>>>>>>>> drivers/net/ethernet/freescale/gianfar.c
>>>>>>>>
>>>>>>>> That'd be 2 users.
>>>>>>>
>>>>>>> I haven't checked myself, I believe your judgement,
>>>>>>
>>>>>> I took a better look and you obviously shouldn't believe :) The gianfar used
>>>>>> of_node instead of the fwnode. So, it'd be a single caller at starters.
>>>>>
>>>>> ...which is the same as dev_of_node(), which means that you can use your
>>>>> function there.
>>>>
>>>> I'm unsure what you mean. The proposed function
>>>> device_get_child_node_count_named() takes device pointer. I don't see how
>>>> dev_of_node() helps converting node to device?
>>>
>>> dev_of_node() takes the device pointer and dev_fwnode() takes that as well,
>>> it means that there is no difference which one to use OF-centric or fwnode
>>
>> The proposed device_get_child_node_count_named() takes a device pointer. I
>> don't see how dev_of_node() helps if there is just of_node and no device
>> pointer available in the calling code.
> 
> ???
> 
> The loops are working on
> 
> 	struct device_node *np = pdev->dev.np;
> 
> which is the equivalent to
> 
> 	struct device_node *np = dev_of_node(&pdev->dev);
> 
> which takes device pointer.
> 
>> (Well, as I wrote below, I could
>> alter the gianfar code by dropping the gfar_of_group_count(), so that I have
>> the device pointer in caller). Anyways, I don't see how dev_of_node() should
>> help unless you're proposing I add a of_get_child_node_count_named() or
>> somesuch - which I don't think makes sense.
> 
> Are you forbidding yourself to change the function prototype to take a device
> pointer instead of device_node one? :-)
> 

This is our point of misunderstanding. As I wrote, and as you can see 
from the prototype, the function _is_ taking the device pointer. Hence I 
didn't understand how dev_of_node() should help us.

>>> API in this particular case. Just make sure that the function (and there
>>> is also a second loop AFAICS) takes struct device *dev instead of struct
>>> device_node *np as a parameter.
>>
>> I think I lost the track here :)
> 
> Make gfar_of_group_count() to take device pointer. As simple as that.

that'd just make the gfar_of_group_count() a wrapper of the 
of_get_child_node_count_named(). I prefer killing whole 
gfar_of_group_count().


Yours,
	-- Matti