Hi Kevin,

Sorry for the delay.

On Mon, 2026-03-09 at 11:41 -0700, Kevin Tung wrote:

On Tue, Mar 3, 2026 at 6:41 PM Andrew Jeffery
[off-list ref] wrote:

quoted

Hi Kevin,

Sorry for the patchy replies so far, but this series bothers me and
other priorities keep bumping it down the list.

On Mon, 2026-02-23 at 19:17 +0800, Kevin Tung wrote:

quoted

Two different power monitor devices, using different drivers, reuse
I2C addresses 0x40 and 0x45 on bus 10 across Yosemite5 board variants.
Defining these devices statically in the DTS can lead to incorrect
driver binding on newer boards when the wrong device is instantiated.

There are effective methods of maintaining devicetrees for variants.
Why are we choosing to remove information about the platform rather
than use existing techniques to properly describe them?

Hi Andrew,

This is due to hardware design changes during earlier development
stages, and the fix is expected to remain stable as the design has
matured.
Could you guide me on the best way to maintain devicetrees for
variants? Thank you :)

My expectation is your platforms move through several design phases
prior to (mass?) production. My suspicion is that you have sent a
devicetree for the pre-production design phases, and you're trying to
evolve that one devicetree to match the design for whatever current
phase you're in.

So, ideally: Send a devicetree only for the finalised design. Don't
send devicetrees for pre-production designs.

If you feel you can't do that for some reason, an alternative is to
have a separate .dts file for each phase in the design process.

This may sound tedious but it doesn't have to be a burden to maintain.

For instance, you can use one or more .dtsi files to describe the
common components and relationships for your platform. These .dtsi
files are then #included into .dts files as usual. Often .dtsi files
are used to isolate different hardware scopes (SoC vs board, for
instance), but we're not limited to that, we can use them for the
purpose outlined above too.

If there are only (very) minor differences, there's also the option of
#including another .dts file. From there you can adjust properties or
even delete nodes where it makes sense. For example, we maintain a .dts
file for the latest revision of the AST2600-EVB, but we also have a
separate .dts for the A1 revision with a different regulator setup:

- https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/arm/boot/dts/aspeed/aspeed-ast2600-evb.dts?h=v7.0-rc5
- https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/arm/boot/dts/aspeed/aspeed-ast2600-evb-a1.dts?h=v7.0-rc5

Any of these are better options than this current approach of trying to
justify incompatible changes against unclear design boundaries.

Andrew