On 22.11.2023 16:37, Krzysztof Kozlowski wrote:
On 22/11/2023 16:32, Rafał Miłecki wrote:
On 22.11.2023 16:00, Krzysztof Kozlowski wrote:
On 22/11/2023 15:52, Rafał Miłecki wrote:
+maintainers:
+ - Rafał Miłecki <rafal@xxxxxxxxxx>
+
+allOf:
+ - $ref: serial.yaml#
+
+properties:
+ compatible:
+ items:
+ - enum:
+ - brcm,bcm4908-hs-uart
+ - brcm,bcm4912-hs-uart
+ - brcm,bcm6756-hs-uart
+ - brcm,bcm6813-hs-uart
+ - brcm,bcm6846-hs-uart
+ - brcm,bcm6855-hs-uart
+ - brcm,bcm6856-hs-uart
+ - brcm,bcm6858-hs-uart
+ - brcm,bcm6878-hs-uart
+ - brcm,bcm47622-hs-uart
+ - brcm,bcm63138-hs-uart
+ - brcm,bcm63146-hs-uart
+ - brcm,bcm63158-hs-uart
+ - brcm,bcm63178-hs-uart
+ - const: brcm,bcmbca-hs-uart
git grep did not find driver for this compatible. Is it in separate
patchset?
No. My project based on BCMBCA has been canceled and I don't work on it
full time anymore. I just wanted to fill empty bits I can afford
handling in my free time and complete hardware description in DTS.
I may still work on some BCMBCA drivers from time to time but as a side
project.
This means we cannot use driver to verify whether the fallback is
actually suitable. Considering that existing UART bindings do not
fallback (brcm,bcm6345-uart, brcm,bcm7271-uart), I don't understand what
is the benefit here.
I believed the rule for maintaining bindings and DTS files was to
describe hardware no matter what/if system needs it.
For example a year ago I added binding for BCMBCA SoC timer without
actual driver, see e112f2de151b ("dt-bindings: timer: Add Broadcom's
BCMBCA timers").
I'm not sure if we're going to agree on this, but personally I like
describing hardware as much as I can. So it's well documented /
understood and people may eventually write drivers for it. Maybe it's
partially because I come from Broadcom's world that isn't well known
for upstream efforts in general.
The problem is that "brcm,bcmbca-hs-uart" is not describing hardware. It
is saying that all these devices have similar (compatible) programming
model, so the OS can use just one compatible. This goes away from pure
hardware description into interpretation.
Rob already commented on such non-SoC compatibles multiple times. I do
not see any reason here to not use specific compatible as fallback.
Do I get it right we should rather have some base specific compatible
like: "brcm,bcm63138-hs-uart" and then if anything use fallback to it
like: "brcm,bcm4908-hs-uart", "brcm,bcm63138-hs-uart"; ?
