[U-Boot] [PATCH] sunxi: allow CONFIG_DEFAULT_FDT_FILE override

[Martin Kelly <mkelly@xevo.com>]

On 05/31/2018 12:58 PM, klaus.goger at theobroma-systems.com wrote:
> 
> 
>> On 31.05.2018, at 21:48, Martin Kelly <mkelly@xevo.com> wrote:
>>
>> On 05/31/2018 11:15 AM, klaus.goger at theobroma-systems.com wrote:
>>> Hi Martin,
>>>> On 31.05.2018, at 19:49, Martin Kelly <mkelly@xevo.com> wrote:
>>>>
>>>> On 05/23/2018 11:34 AM, Martin Kelly wrote:
>>>>> On 05/17/2018 05:01 AM, Maxime Ripard wrote:
>>>>>> On Tue, May 01, 2018 at 05:46:41PM -0700, Martin Kelly wrote:
>>>>>>> Currently, sunxi-common.h ignores CONFIG_DEFAULT_FDT_FILE and assumes
>>>>>>> the kernel fdtfile and the u-boot devicetree names are the same.
>>>>>>> Although this is typically the case, sometimes you might want to
>>>>>>> customize one of these differently, so it's useful to allow them to be
>>>>>>> different.
>>>>>>>
>>>>>>> Add logic in sunxi-common.h to respect CONFIG_DEFAULT_FDT_FILE, if set,
>>>>>>> and default to the values it currently uses.
>>>>>>>
>>>>>>> Signed-off-by: Martin Kelly <mkelly@xevo.com>
>>>>>>
>>>>>> Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
>>>>>>
>>>>>> Maxime
>>>>>>
>>>>> (ping)
>>>>
>>>> (ping again)
>>>>
>>>> Can we get this patch merged?
>>> The else branch will never be selected.
>>> CONFIG_DEFAULT_FDT_FILE will be set to “” if not set by your defconfig.
>>
>> Thanks for noticing this; unfortunately you are right. I thought I tested this, but it was in the context of a larger build and I missed it.
>>
>> It looks like we have two options:
>>
>> (a) Test for empty string in preprocessor language. This is quite ugly and has some gotchas but should be doable (e.g. https://stackoverflow.com/questions/3781520/how-to-test-if-preprocessor-symbol-is-defined-but-has-no-value).
>>
>> (b) Always use the value of CONFIG_DEFAULT_FDT_FILE and stop using CONFIG_DEFAULT_DEVICE_TREE as a default. In each board file, explicitly set CONFIG_DEFAULT_FDT_FILE to its current default. This is what other boards are doing.
>>
>> I favor (b). If others are in favor, I'll submit a patch for it.
> 
> I run into the same issue for the rockchip architecture[1]. After giving up on finding a
> rock-solid solution for detecting the emtpy string I decided to go with your option (b).
> Another solution would be to hide CONFIG_DEFAULT_DEVICE_TREE behind a
> Kconfig bool so it would not be defined until something like CONFIG_FDT_FILE is
> selected. But in favour of having U-Boot devicetree decoupled from any OS fdt I
> still prefer option (b).
> 
> [1] https://patchwork.ozlabs.org/patch/920785/
> 
> 

Thanks, I'll go ahead with that solution and issue a follow-up patch.

Jagan, should I make the patch against this one, or against master?