Re: [PATCH 3/3] MAINTAINERS: Refactor device tree maintainership

[Stephen Warren <swarren@wwwdotorg.org>]

On 07/23/2013 11:55 AM, Ian Campbell wrote:
> On Tue, 2013-07-23 at 13:44 -0500, Rob Herring wrote:
>> On 07/23/2013 01:09 PM, Ian Campbell wrote:
>>> On Tue, 2013-07-23 at 10:59 -0700, Stephen Warren wrote:
>>>
>>>> I think the solution is to introduce some new shared/common location for
>>>> shared/common *.dtsi files, into the kernel tree, in the interim.
>>>>
>>>> When *.dts move out of the kernel, this common location can simply be
>>>> consumed as part of the DT tree re-organization.
>>>>
>>>> Or perhaps, we could move *.dts around in the kernel to match the
>>>> proposed DT tree structure before that point in time?
>>>
>>> FWIW I can easily handle any transformation as part of the automated
>>> extraction into the device-tree.git. If it can expressed as a sed script
>>> then so much the better, e.g. the current rules are
>>> http://xenbits.xen.org/gitweb/?p=people/ianc/device-tree-rebasing.git;a=blob;f=scripts/rewrite-paths.sed;h=f7a157d1b486bac058f50e42cf7bedc8630e54ff;hb=HEAD.
>>> If it gets too complicated for sed I can always switch to something
>>> else.
>>>
>>> I'm already pending a complete rebuild of the export to add in the
>>> Documentation/devicetree/bindings sub tree but since it takes an age to
>>> run I was waiting for the output of this conversation before kicking
>>> that off.
>>
>> I'd doubt we could completely script this with a generic rule without a
>> bunch of manual transformations.
> 
> Right. The advantage of scripting, even if that script is just a big
> long list of manual rules, is that we can preserve the history from
> linux.git in device-tree.git, so git annotate etc say something useful.
> 
> But if I'm going to write a big long list of such rules I'd like to only
> do the bulk of the work once, so we need to decide on the layout first.
> 
>>  So I think either restructuring in the
>> kernel or when we move them out of the kernel makes more sense. We know
>> the problem is coming, but it is not yet a major, pressing issue.
>>
>> OTOH, you could see how far you get by putting dts files in directories
>> by their board level compatible string vendor and put any include files
>> where ever they are included from. Of course, that is just my proposed
>> layout. I haven't heard any opinions on that layout.
> 
> I don't really have a strong opinion on the layout myself, I'm happy to
> implement whatever works for people.
> 
> The board level compatible string vendor is the top-level compatible,
> right? e.g.:
>         / {
>                 model = "Calxeda Highbank";
>                 compatible = "calxeda,highbank";
> "calxeda,highbank" in this case. So you propose s|,|/| on that so the
> file ends up in calxeda/highbank.dts?
> 
> For the included ecx-common.dtsi I think you are proposing to find all
> the files which include it and then find the
> deepest-common-subdirectory? In this case the other include is
> ecx-2000.dts which is compatible calxeda,ecx-2000
> (->calxeda/ecx2000.dts) so the common subdir is calxeda and dtsi would
> become calxeda/ecs-2000.dtsi. That works, I should probably have picked
> an example which didn't rely on files you are so involved with in case
> there is a selection bias at work ;-)

That layout sounds pretty reasonable. Are all those patches underneath
dt/ or is calxeda/ a top-level directory?

I guess that means that common files like skeleton.dtsi will simply end
up at the top-level (or in "top-level" dt/). That's probably fine.

I'd like the filenames to stay the same; to some degree I consider that
part of the DT ABI. At least for Tegra, and I think most/many other SoCs
follow this scheme..., we have named the files ${soc}-${board}.dts, and
made sure that the ${soc} and ${board} values match U-Boot's values, so
U-Boot can just "load mmc 0:1 ${soc}-${board}.dts" and hence use the
exact same U-Boot script for all Tegra boards (and potentially for any
board that defines ${soc} and ${board}, and doing that is easy).