On 8/12/26 14:53, David Heidelberg wrote:
On 12/08/2026 14:22, Neil Armstrong wrote:
On 8/12/26 14:15, David Heidelberg wrote:
On 12/08/2026 13:52, Neil Armstrong wrote:
On 8/12/26 10:46, David Heidelberg wrote:
On 12/08/2026 07:21, Esteban Urrutia wrote:
On 8/11/26 9:05 AM, David Heidelberg wrote:
Hi Esteban,
Usually the used pattern is vendor,ddic+panel-model
novatek,nt37703-motorola-bronco
...but motorola bronco isn't the model of the panel.
My recommendation - look at some video of screen replacement, stop at moment
when FLEX cable from panel is visible and that's where the panel model usually
is.
I somewhat doubt the panel model would be located in the FPC cable.
I did search anyway but didn't find anything of interest. And I'm not
willing to risk damaging the display flex or the battery of my
development device just for a possibility either.
No-one asks you to disassemble the device :)
Sadly, I haven't found any video doing display replacement either.
In worst case such as this I assume it make sense keep the phone model in the
compatible until panel codename is found.
For the compatible pattern I used the implemented pattern for the
NT36523 TDDI, which is also used for other ICs such as the RM69380.
So, I believe it should be okay to leave the compatible as is.
I would like to hear a word from Neil first. In similar cases like this would
make sense to have one way defining these compatible strings used in phones.
bit I missed in the review the part:
+ compatible = "motorola,bronco-tianma-v2-nt37703",
"novatek,nt37703";
Unless the driver is able to auto-detect the connected panel (unlikely),
novatek,nt37703 doesn't make sense. The DDIC without the panel is useless.
For example, in sdm845-oneplus-fajita and enchilada, we have the generic
compatible because it was introduced before, so it's kept for compatibility
reasons. There is no need to do that here.
Yeah today the norm is to add the ddic as second compatible entry so the first
compatible just describes the pane, so here it should be:
compatible = "motorola,bronco-tianma-v2", "novatek,nt37703";
So let's me present a situation which can happen, let say I'll be upstreaming:
DDIC (novatek,nt37700) + panel (samsung,amb630qy).
For the example I choose samsung,amb630qy because here the panel has phone
model in it, which avoids the possible issue I'll be describing with this
schema.
Vendor can pair this panel with different DDIC. Thus it would be
compatible = "samsung,amb630qy", "novatek,nt37700"; // new one
compatible = "samsung,amb630qy", "novatek,nt37703"; // existing
The reality the driver picks only the first compatible, and that's all causing
to load wrong driver now.
No, the core will pick the first compatible that matches, so...
Solution here would be differentiate between
a) DDIC + panel
b) module name (consisting DDIC + panel).
We could say module is motorola,bronco-tianma-v2 and contains [ DDIC + unknown
panel ], which doesn't even have to be mentioned in device-tree.
While it may seems I overthinking, in this case it doesn't matter that much, I
already have patchsets for DDIC which can be combined with three types of
panels (and using different schema) and I would like to have good solution for
longterm contributions and development.
... in theory we could have:
compatible = "samsung,amb630qy", "novatek,nt37700"; // new one
compatible = "samsung,amb630qy", "novatek,nt37703"; // existing
and drivers would only have:
"novatek,nt37700"
and
"novatek,nt37703"
as compatible and then a separate match table with :
"samsung,amb630qy" and "samsung,amb630qy" which associates with the panel data.
If it doesn't match, return -ENODEV and it would work like before,
I mean it's valid, it's a matter of implementation, if the same module can have
different DDIC then it should be described like that.
In this case, the bindings should be for the module and not for the DDIC.
Definitely make sense in the context. This would mean I'll need to adapt my
patchset to-be-sent :)
Btw the disccusion slowly mentally gets me closer to the Camera module model,
where you have multiple components such as
module (sunny-XY) -> camera (sony,imx519)
-> focus (dongwoon,dw9714)
-> OIS
-> fixed components as optics
where as we have here only
"module" (motorola-bronco-tianma-v2) -> DDIC (nt37703)
↕
-> panel (unknown)
-> touchscreen (sometimes, depend on the
phone build)
Sometimes init sequences and stuff depends on each other, thus touchscreen is
part of the module in some cases.
Would it make sense to somehow model this into DT?
Since we doing all the stuff in the panel node, it could be something like
display@0 {
compatible = "novatek,nt37703"; // first thing we interact with here
module = "motorola-bronco-tianma-v2"; // DDIC driver picks the module data
vddio-supply = "...";
backlight = <&pmi8998_wled>;
...
// I already have combination where two panels has different supplies
// Samsung S9 vs Pixel 3 XL. One wiring is over DDIC it seems, second
// directly to the panel
panel { // REAL panel node
compatible = "samsung,ams622xy" // maybe as a docs?
vddneg-supply = <&lab>;
vddpos-supply = <&ibb>;
}
}
Just dumping my brain, feel free to ignore if it's complete nonsense.
Ultimately, we should describe everything, if we can, most of the time we
can't as we don't have the DDIC and only timings and a bunch of DSI
commands. So we need to keep the good ol' model for a while.
But yeah perhaps it would nice to experiment and implement a full DDIC
support modeled as a DRM bridge which has a panel subnode handled by
a panel-simple with only the panel resources (timings, regulators,
backlight regulator if not handled by DDIC).
Feel free to experiment and throw some RFC so we can discuss.
BTW I'll be hosting 2 BoFs about modern supports of DRM panels at the next
Linux Plumbers and Embedded Linux Conferences in Prague, and this is one
of the subject I'll speak about.
Neil
David
Neil
What do you think?
David
Neil
Thanks
David
David Heidelberg
Regards,
Esteban
