[U-Boot] [PATCH] dm: core: Kconfig: set OF_TRANSLATE default value to n

[Stephen Warren <swarren@wwwdotorg.org>]

On 11/06/2015 05:19 AM, Stefan Roese wrote:
> +Stephan Warren & Thomas Chou
>
> On 06.11.2015 13:08, Simon Glass wrote:
>> +Stefan
>>
>> Hi Mugunthan,
>>
>> On 4 November 2015 at 07:25, Mugunthan V N <mugunthanvnm@ti.com> wrote:
>>> Based on the OF_TRANSLATE Kconfig description, this is required
>>> only on platforms with complex "ranges" in the DT nodes. For
>>> simpler platforms using SIMPLE_BUS should be sufficient. Since
>>> this a set to default y, simple bus is never used in any
>>> platform. So make the default value as "n" and enable
>>> OF_TRANSLATE only on required platform.
>>
>> How do you know it is not used? This is surprising because Stefan
>> presumably added it to fix something.
>
> I assume Mugunthan meant, that simple bus is never used. This
> is not quite correct, as its at least used per default in the
> SPL (because of size reasons).
>
> And yes, this translation via fdt_translate_address() is needed.
> At least for some platforms, like mvebu & nios2 (IIRC). Stephen
> also made some quite striking comments about this a few weeks
> ago. And I really think that we should make use of this function
> per default. Platforms that know for sure that they don't need
> this functionality can always disable it selectively.
>
> Mugunthan, why do you want to disable this functionality? Does
> it cause problems on your platform? If yes, which ones?

Processing of ranges is a fundamental part of handling DT in general. It 
should be enabled by default. If there are well-researched cases where 
(a) including the code is problematic (e.g. due to code size 
restrictions) and (b) someone is watching the DTs processed on those 
platforms like a hawk to ensure no ranges properties are introduced into 
the DT, *then* we could make a special-case exception to disable the 
feature in those specific cases. As such, I think enable-by-default, 
disable-when-chosen is the appropriate default.