[RFC PATCH 0/1] Categorize ARM dts directory

Ansuel Smith ansuelsmth at gmail.com
Tue Mar 29 15:56:02 AEDT 2022

On Tue, Mar 29, 2022 at 03:20:18PM +0200, Krzysztof Kozlowski wrote:
> On 28/03/2022 02:09, Ansuel Smith wrote:
> > Hi,
> > as the title say, the intention of this ""series"" is to finally categorize
> > the ARM dts directory in subdirectory for each oem.
> > 
> > The main reason for this is that it became unpractical to handle 2600
> > dts files and try to even understand/edit/check the situation for a
> > specific target.
> > 
> > In arm64 we already have this kind of separation and I honestly think
> > that this was never proposed for ARM due to the fact that there are
> > 2600+ files to sort and the fact that it will be a mess to merge this
> > entirely but IMHO with a little bit of effort we can finally solve this
> > problem and have a well organized directory just like arm64.
> > 
> > Some prerequisite on how this work was done:
> > - This comes entirely from a python script created by me for the task.
> >   linked here [1]
> > - I had to manually categorize all the different arch in the makefile
> >   based on the oem. I searched every arch on the internet trying to
> >   understand the correct oem. I hope they are correct but I would love
> >   some comments about them.
> > - This current ""series"" is all squashed in one big commit to better
> >   receive comments for this. The final version ideally would have all
> >   changes in separate commits. The script can already do this, it's just
> >   commented.
> > 
> > Here is a list of some discoveries while doing all the sorting.
> > These are totally additional reason why we need this.
> > 
> > While creating the script I discovered some funny things:
> > - We have orphan dts! There are dts that are never compiled and are
> >   there just for reference. We would never have noticed this without this
> >   change and probably nobody noticed it. They are currently all listed
> >   in the python script.
> > - We have dtsi shared across different oem. My current solution for them
> >   is: NOT SORT THEM and leave them in the generic directory and create a
> >   link in each oem dts that points to these dtsi. This is to try in
> >   every way possible to skip any additional changes to the dts.
> >   Current dtsi that suffers from this are only 3. (listed in the script)
> > - arm64 dts and dtsi reference ARM dts. Obviously this change would cause
> >   broken include for these special dtsi. The script creates a dependency
> >   table of the entire arm64 directory and fix every broken dependency
> >   (hoping they all use a sane include logic... regex is used to parse
> >   all the different dependency)
> > 
> > So in short the script does the following steps:
> > 1. Enumerate all the action to do... (dts to move, scan dependency for
> >    the dts...)
> > 2. Generate the arm64 dependency
> > 3. Creates the Makefile
> > 4. Generate the Makefiles for the current oem
> > 5. Move all the related dts and dtsi for the current oem
> > 6. Check broken dependency and fix them by editing the dts and writing
> >    the correct include (or fix any symbolic link)
> > 
> > This is an output that describes all the things done by the script [2]
> > 
> > I really hope I didn't commit any logic mistake in the script but most
> > of the work should be done.
> > 
> +Cc Arnd and Olof,
> Ansuel,
> Thanks for you patch. Please cc the SoC maintainers in such submissions.
> It seems that you got some quite nice discussion, but still the core
> folks are not Cced, so no one would be able to take your patch...

I had some problem with gmail and sending mail too much users. I put Rob
and You and all the various list to try to workaround the "gmail spam

> I am pretty sure we were discussing such split idea in the past and it
> did not get traction, but I cannot recall the exact discussion.

I think the main issue here is how to handle bot and how problematic is
to merge this. As written in the cover letter the final version of this
should be a big series of 50+ patch with every commit specific to each
oem. In theory we should be able to merge the different oem separately
and try to at least start the categorization. 
Another idea I got to at least have a "migration path" is to convert
every dts in the dts/ directory to a symbolic link that target the dts
in the correct oem. But I assume that would fix only part of the problem
and git am will still be problematic.

> To me the idea is good but will cause huge `git am` conflicts.
> Cherry-picks, backports and merges should nicely detect path renames,
> but git am (and b4 am) I think cannot.

I know but we should really consider this kind of change. The current
state of the dts/ directory is embarassing and keeping it that way cause
only more problems and makes this even more difficult.
Hope we find a solution and fix this for good!

> Best regards,
> Krzysztof


More information about the Linux-aspeed mailing list