[PATCH V4 3/3] powerpc/fsl-pci: Unify pci/pcie initialization code

[Scott Wood]

On 08/07/2012 10:57 PM, Jia Hongtao-B38951 wrote:
> 
> 
>> -----Original Message-----
>> From: Wood Scott-B07421
>> Sent: Tuesday, August 07, 2012 11:20 PM
>> To: Jia Hongtao-B38951
>> Cc: Wood Scott-B07421; Kumar Gala; linuxppc-dev at lists.ozlabs.org; Li
>> Yang-R58472
>> Subject: Re: [PATCH V4 3/3] powerpc/fsl-pci: Unify pci/pcie
>> initialization code
>>
>> On 08/07/2012 01:23 AM, Jia Hongtao-B38951 wrote:
>>>> -----Original Message-----
>>>> From: Wood Scott-B07421
>>>> Sent: Monday, August 06, 2012 11:16 PM
>>>> To: Jia Hongtao-B38951
>>>> Cc: Wood Scott-B07421; Kumar Gala; linuxppc-dev at lists.ozlabs.org; Li
>>>> Yang-R58472
>>>> Subject: Re: [PATCH V4 3/3] powerpc/fsl-pci: Unify pci/pcie
>>>> initialization code
>>>>
>>>> On 08/05/2012 09:39 PM, Jia Hongtao-B38951 wrote:
>>>>>
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Wood Scott-B07421
>>>>>> Sent: Saturday, August 04, 2012 12:04 AM
>>>>>> To: Jia Hongtao-B38951
>>>>>> Cc: Kumar Gala; linuxppc-dev at lists.ozlabs.org; Wood Scott-B07421; Li
>>>>>> Yang-R58472
>>>>>> Subject: Re: [PATCH V4 3/3] powerpc/fsl-pci: Unify pci/pcie
>>>>>> initialization code
>>>>>>
>>>>>> On 08/02/2012 10:39 PM, Jia Hongtao-B38951 wrote:
>>>>>>>
>>>>>>>
>>>>>>>> -----Original Message-----
>>>>>>>> From: Kumar Gala [mailto:galak at kernel.crashing.org]
>>>>>>>> Sent: Thursday, August 02, 2012 8:24 PM
>>>>>>>> To: Jia Hongtao-B38951
>>>>>>>> Cc: linuxppc-dev at lists.ozlabs.org; Wood Scott-B07421; Li
>>>>>>>> Yang-R58472
>>>>>>>> Subject: Re: [PATCH V4 3/3] powerpc/fsl-pci: Unify pci/pcie
>>>>>>>> initialization code
>>>>>>>>
>>>>>>>> You need to convert all boards to use fsl_pci_init before this
>> patch.
>>>>>>>> Otherwise we'll end up with PCI getting initialized twice on
>> boards.
>>>>>>>>
>>>>>>>> - k
>>>>>>>
>>>>>>> If we covert all boards with platform driver in this patch PCI will
>>>>>>> be initialized only once without converting all boards to use
>>>>>>> fsl_pci_init first.
>>>>>>
>>>>>> Then we'd have to pick apart core changes from board changes when
>>>>>> reviewing.
>>>>>>
>>>>>>> If we convert all boards to use fsl_pci_init before this patch and
>>>>>>> convert them to use platform driver again after this patch. Then
>>>>>>> between this patch and next pci will be initialized twice too.
>>>>>>
>>>>>> Why?  That one patch should both create the platform driver and
>>>>>> remove the init from fsl_pci_init() -- except things like primary
>> bus
>>>>>> detection which has to happen globally.
>>>>>>
>>>>>> -Scott
>>>>>
>>>>> "One patch both create the platform driver and remove the init from
>>>>> fsl_pci_init()" means we should create platform driver and applied to
>>>>> all boards. If so why not just directly convert all boards using
>>>>> platform driver?
>>>>
>>>> Because it's harder to review when you have a bunch of board code in
>> the
>>>> patch in addition to core changes.
>>>>
>>>> Because you might want people to actually test on the boards in
>> question
>>>> when converting, especially given the change in how primary buses are
>>>> determined, and that some boards may need to provide their own
>>>> alternative.
>>>>
>>>> -Scott
>>>
>>> But if we separate the core changes and the boards update, between this
>> two
>>> patches PCI will be initialized twice.
>>
>> As I said earlier, you can remove the initcall and require boards to
>> manually call fsl_pci_init() until all boards are converted.
>>
>> -Scott
> 
> As I said earlier, I can do this but it does not solve the twice-init problem.

I must have missed it.  Why does it not solve the problem?  If a board
doesn't call fsl_pci_init(), the platform driver doesn't get registered.

> If I do this first and then add platform driver we also have to convert all
> boards using platform driver in the same patch.
> 
> We finally using the platform driver so Why do you keep insisting on converting
> all boards using fsl_pci_init() first even it does no improvement.

What we're asking for is bisectability (don't have any intermediate
stages where PCI gets initialized twice), and the ability to have a
smooth transition where boards can be converted as people are able to
test them and look into their individual needs regarding primary bus.

-Scott