[PATCH openbmc v2] Re-enable out of tree device trees

[Joel Stanley]

On Wed, May 25, 2016 at 5:38 AM, Patrick Williams <patrick at stwcx.xyz> wrote:
> On Thu, May 19, 2016 at 11:26:01AM +0930, Joel Stanley wrote:
>> Hi Brad,
>>
>> On Wed, May 18, 2016 at 12:00 AM, OpenBMC Patches
>> <openbmc-patches at stwcx.xyz> wrote:
>> > From: Brad Bishop <bradleyb at fuzziesquirrel.com>
>> >
>> > This snippet was mistakenly removed with 8ef9fee.
>>
>> It was intentionally removed.
>>
>> > Note that this isn't a matter of policy, it simply enables
>> > the option to do it.  The in-tree device tree is still the default.
>> >
>> > This enables things like workbook or schematic generated device
>> > trees.  It also addresses the inevitable fact that the upstream kernel
>> > will probably not want device trees for every board in the world.
>>
>> As we do development the device tree is strongly coupled to the
>> kernel. I think we want to keep the dts in the kernel tree, so that
>> changes are made together.
>>
>> Secondly, when testing kernels the developer may or may not be using
>> the poky environment. This means that the developer writing and
>> supporting your kernel is not testing the same code that is running on
>> the board.
>>
>> We can add every device trees to our kernel tree. As the upstream
>> maintainer of the aspeed port, I will also send them all upstream.
>
> I can think of almost a dozen openpower boxes, plus the IBM branded FSP
> boxes.  That is before we even start to support x86 and ARM servers.  Do
> you really think we are going to want to submit every single device tree
> for every single openbmc supported box upstream?  That sounds really
> cumbersome for both us and the Linux community.

Yes, I do think this is what we want to do.

By adding it to the kernel tree we ensure we stay in sync with what
the code expects. It also allows a level of review and quality
control.

> Yocto already has a mechanism for creating device trees from snippets.
> As we start to use the MRW data from the system engineers, I would like to
> see us generate many aspects of the device tree from those directly.