Re: [PATCH v2] docs: devicetree: overlay-notes: recommend top-level compatible in DTSO

David Gibson <[email protected]> Mon, 8 Sep 2025 14:02:45 +1000

Newsgroups	org.kernel.vger.devicetree-spec,org.kernel.vger.linux-devicetree,org.kernel.vger.linux-doc,org.kernel.vger.linux-kernel
Message-ID	<aL5VZfOoy1g2uyAH@zatzit>

On Thu, Sep 04, 2025 at 10:40:31AM -0400, Raymond Mao wrote:
> Hi David,
> 
> On Wed, 3 Sept 2025 at 22:58, David Gibson <[email protected]> wrote:
> >
> > On Tue, Sep 02, 2025 at 10:43:50AM -0700, Raymond Mao wrote:
> > > When managing multiple base device trees and overlays in a structured
> > > way (e.g. bundled in firmware or tools), it is helpful to identify the
> > > intended target base DT for each overlay, which can be done via a
> > > top-level compatible string in the overlay.
> > >
> > > This provides a way to identify which overlays should be applied once the
> > > DT is selected for the case when a device have a common firmware binary
> > > which only differs on the DT and overlays.
> > >
> > > This patch updates the document with a note and example for this
> > > practice.
> > > For more information on this firmware requirement, please see [1].
> > >
> > > [1] https://github.com/FirmwareHandoff/firmware_handoff/pull/74
> >
> > I think this idea is probably useful enough to be a good idea anyway.
> > However, note that it leans in to an existing ugliness of the overlay format:
> >
> > Overlay dtbs kind of mix "in band" information - the actual new
> > content for the tree - with "out of band" information - how to apply
> > the overlay itself.  Whether a given property is data or metadata is
> > determined by it's place in the tree in a moderately complex and not
> > super obvious way.
> >
> > About the clearest divide that exists is that generally the root and
> > first-level subnodes are information only for overlay application,
> > everything under that is data to be applied to the tree.  This all
> > tends to have names that would be unlikely (though not strictly
> > impossible) in a fully applied tree.
> >
> > Putting 'compatible' at the root of the overlay is putting something
> > that looks very much like a regular device tree property in a place
> > and with a function that's purely about applying / validating the
> > overlay itself.
> >
> 
> Since all information at the root of an overlay is considered as
> metadata (out-of-band),
> If you think 'compatible' is confused, I can change it to
> 'overlay-compatible' - which should be 'unlikely' to exist in a full
> tree.

No, as I said, I think the advantages of this proposal still outweigh
the disadvantages.  Just pointing out that this is highlighting some
of the ugliness in the current way overlays are designed, which is
relevant in the context of concurrent discussions about connectors and
the like.

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson

signature.asc (application/pgp-signature, 833 B)