Re: [PATCH] ARM: pxa: fix build failure for regulator consumer in em-x270.c

[Paul Gortmaker <paul.gortmaker@windriver.com>]

On 12-03-29 10:28 AM, Igor Grinberg wrote:
> On 03/28/12 18:59, Paul Gortmaker wrote:
>> On 12-03-28 12:13 PM, Mark Brown wrote:
>>> On Wed, Mar 28, 2012 at 11:59:41AM -0400, Paul Gortmaker wrote:
>>>> On 12-03-28 11:27 AM, Mark Brown wrote:
>>>
>>>> static struct platform_device em_x270_gps_userspace_consumer = {
>>>>         .name           = "reg-userspace-consumer",
>>>>         .id             = 0,
>>>
>>>> static struct platform_device em_x270_gprs_userspace_consumer = {
>>>>         .name           = "reg-userspace-consumer",
>>>>         .id             = 1,
>>>
>>>> Note that the existing names currently don't incorporate the .id
>>>> field as a suffix, and so never were unique.
>>>
>>> No, this is just a basic part of how platform devices work - the device
>>> name is always the same and if you've got more than one of them they get
>>> different .ids.  dev_name() returns name.id, or just name if id is set
>>> to -1 indicating that there's onyl one device of a given type.
>>
>> OK, so Igor - can you simply retest the v2 patch, but make the
>> two trivial changes:
>>
>> -REGULATOR_CONSUMER(ldo3, "reg-userspace-consumer", "vcc gps");
>> +REGULATOR_CONSUMER(ldo3, "reg-userspace-consumer.0", "vcc gps");
>>
>> -REGULATOR_CONSUMER(ldo19, "reg-userspace-consumer", "vcc gprs");
>> +REGULATOR_CONSUMER(ldo19, "reg-userspace-consumer.1", "vcc gprs");
> 
> Well, I thought of this solution, but I don't like it, as it makes
> the whole thing very fragile and sensitive to the reg-userspace-consumer
> platform device registration order and count, isn't it?

Why would it be order dependent?  You can see in the above structs
that they explicitly call out an ID value -- it is _not_ chosen at
the time of registration.  So it will always be:

reg-userspace-consumer.0  -->  gps consumer
reg-userspace-consumer.1  -->  gprs consumer

At least that is my (limited) understanding, based on what
Mark was saying.

Paul.
--

> (That's why I proposed to use NULL...).
> 
> So, Mark, how do you think the above issues can be handled without
> putting NULL into the dev_name?
>