Re: [PATCH 2/3] ARM: sunxi: Add an ahci-platform compatible AHCI driver for the Allwinner SUNXi series of SoCs

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 12/06/2013 10:12 AM, Oliver Schinagl wrote:
>
> On 06-12-13 10:01, Thomas Petazzoni wrote:
>> Dear Tejun Heo,
>>
>> On Wed, 4 Dec 2013 08:23:12 -0500, Tejun Heo wrote:
>>
>>>> But again, point me (for dummies ;) in the right direction and I'll
>>>> work on it with some help.
>>> Richard and Shawn recently worked on ahci_imx.  Can you guys please
>>> talk with each other and figure out what can be done to share as much
>>> as possible among these new platform-specific drivers?  I'd really
>>> like to see the common things factored out as much as possible with
>>> only the actual hardware differences described for each device.
>> Also, please Cc me on such discussions. I have a pending AHCI platform
>> driver for another ARM SoC family. It is very similar to ahci_platform,
>> but needs to do a few more things that are SoC specific (map an
>> additional register area, and do some SoC-specific stuff with them).
>>
>> For the moment, we're left with two approaches:
>>
>>   * Do what Oliver did, where the ahci_<foo> driver will do its own
>>     SoC-specific stuff, and then will register an additional
>>     platform_device to trigger the ->probe() of the generic
>>     ahci_platform driver. I must say I don't really like this solution,
>>     since it involves having two platform_device registered for the same
>>     piece of hardware (one platform_device to trigger the ->probe of
>>     ahci_<foo>, and another one to trigger the ->probe of ahci_platform).
>>
>>   * Duplicate in ahci_<foo> the (relatively small) amount of code that
>>     is present in ahci_platform.
>>
>>  From my point of view, ahci_platform should be turned into a small
>> "library", that provides an API for ahci_<foo> drivers to 1/ do their
>> own custom stuff and 2/ do the common ahci_platform stuff.
>>
>> This way we avoid the registration of two platform_device for the same
>> piece of hardware, and we avoid the duplication of code.
>>
>> Want me to propose a RFC for this idea?
> I've started to do what sdhci does with their pltfrm driver, assuming that's the right approach. Since i'm only dabbling and not always 100% sure what should or shouldn't be done, it may take a little while, but looks promising from my end ;)
>
> So is the sdhci-pltfrm approach the correct one? We still have ahci_* drivers, but ahci_platform.c won't be a driver in the sense that it is now anymore.

Sounds good to me. May I suggest simply adding a new ahci_pltfrm driver for
this and leaving the existing ahci_platform alone? Of course in the end
we want the old ahci_platform to go away, but it is probably best to
introduce the new one in parallel and then port things over 1 by 1
by people who can actually test the port :)

Regards,

Hans