Re: [PATCH 1/2] mfd: pm8x41: add support for Qualcomm 8x41 PMICs

[Frank Rowand <frowand.list@gmail.com>]

On 4/25/2014 5:40 PM, Courtney Cavin wrote:
> On Sat, Apr 26, 2014 at 02:28:06AM +0200, Frank Rowand wrote:
>> On 4/23/2014 6:19 AM, Ivan T. Ivanov wrote:
> [...]
>>>> +static int pm8x41_probe(struct spmi_device *sdev)
>>>> +{
>>>> +	struct regmap *regmap;
>>>> +
>>>> +	regmap = devm_regmap_init_spmi_ext(sdev, &pm8x41_regmap_config);
>>>> +	if (IS_ERR(regmap)) {
>>>> +		dev_dbg(&sdev->dev, "regmap creation failed.\n");
>>>> +		return PTR_ERR(regmap);
>>>> +	}
>>>> +
>>>> +	return of_platform_populate(sdev->dev.of_node, NULL, NULL, &sdev->dev);
>>>
>>> I think that this will not going to work. For example in this particular
>>> case, both controllers have "qcom,qpnp-revid" peripheral which is
>>> located at offset 0x100.
>>>
>>> And the result is:
>>>
>>> [    0.963944] sysfs: cannot create duplicate filename '/bus/platform/devices/100.revid'
>>
>> The duplicate filename error is because pm8x41_probe() is calling of_platform_populate()
>> directly.
>>
>> If that call is removed then there is no attempt to create a revid file in
>> /sys/bus/platform/devices.  If I add your pm8841@4 node to my dts file, then
>> the 100.revid file is properly created at
>>
>>   /sys/firmware/devicetree/base/soc/qcom,spmi@fc4cf000/pm8841@4/qcom,revid@100
>>
>> and no attempt is made to create /sys/bus/platform/devices/100.revid
>>
>> This appears to be correct to me, because I do not think revid should be created as
>> a platform device.
>>
> 
> Wait, what?  That's the entire point of this driver.
> 
> [...] 
>>> Any suggestions?
>>
>> Remove of_platform_populate() from pm8x41_probe().  Do you know of any
>> other reason it can not be removed?
> 
> I do!  If you remove it, the entire driver becomes a useless pile of
> nothing!
> 
> The PMICs targeted by this driver are mostly made up of independent
> blocks which should be able to be written as standalone device drivers.
> The design of this driver is to create a simple bus-like layer between
> SPMI and these independent blocks.  of_platform_populate() is what makes
> it work.
> 
> -Courtney
> 

Hmmmm, yet another device tree conceptual thing for me to grok.  I'll go
off and see if a bus-like layer can exist without being a platform device.

-Frank