Re: [PATCH] arm64: dts: mediatek: Remove undocumented, unapproved aliases
From: AngeloGioacchino Del Regno
Date: Wed Nov 18 2026 - 07:30:52 EST
On 11/18/26 13:09, Krzysztof Kozlowski wrote:
On 17/11/2026 15:06, AngeloGioacchino Del Regno wrote:
On 11/17/26 14:05, Krzysztof Kozlowski wrote:
The ovl, aal, color, dsi, gce, rdma, split and other aliases were never
approved by Devicetree maintainers, because they were never documented
as actual ABI. Drop them, because no new custom aliases are allowed for
years now.
This might break users of this undocumented ABI, so these should be
updated to use only documented and approved ABI.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxxxxxxxx>
That'll just break everything.
True, almost everything. I suppose few aliases are not used, but rest are.
Would still create regressions, and that's not acceptable - the cleanup can be
done only - and exclusively - after addressing the situation elsewhere so that
no regressions are seen.
Still, a question mark remains on "new DT on old kernel" compatibility which would
miserably break - that may not be something to care about in this case, and I don't
think that it's time to start thinking about that.
Well, it's a start of a discussion. :)
Yes, I like this spirit.
I do have a mediatek-drm restructuring series that is partially resolving this
situation, at least for some SoCs it's getting 0 and 1 hw instance id dynamically
recognized, but with some caveats that could be unacceptable (as in, if 0 is
disabled and 1 is enabled, 1 becomes 0, so it's not actually resolving anything
for real, it's just making it tick right when stars in the sky align in the
right way, which for some SoCs is reliably possible without side effects).
In any case, all of the ones that have only instance 0 (and all of the instance
0s) can be removed even right now already (still needs some testing to make sure
that there's no regression), but everything else can't, at least not until the
whole restructuring thing is merged as that'd break things.
Worst part is we have two new SoCs (MT8196, MT8189) ready to be upstreamed here,
which took almost a year of development... and it also does rely on the aliases.
I was in the process of slowly cleaning up those (as I said, there's already a
restructuring that makes things a little better in this regard), but at this point
that'll need to be done with the highest possible priority.
That's unfortunate and still new DTS/DTSI should not come with these
aliases.
Indeed.
Since conversations went on in the other thread, though, if we have to spend any
other word on this topic, I think it's a better idea to continue on the other one
( 94ed1340-94bd-49ee-ba31-9bb63164c89a@xxxxxxxxxx ) and let this one go.
Cheers!
Angelo