Re: [PATCH v3 0/6] Add Synopsys DesignWare HDMI RX Controller

From: Krzysztof Kozlowski
Date: Thu Apr 04 2024 - 04:22:52 EST


On 04/04/2024 10:19, Heiko Stübner wrote:
> Am Donnerstag, 4. April 2024, 08:15:50 CEST schrieb Krzysztof Kozlowski:
>> On 04/04/2024 00:48, Heiko Stübner wrote:
>>> Am Mittwoch, 3. April 2024, 13:24:05 CEST schrieb Krzysztof Kozlowski:
>>>> On 03/04/2024 13:20, Shreeya Patel wrote:
>>>>> On Wednesday, April 03, 2024 15:51 IST, Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> wrote:
>>>>>
>>>>>> On 03/04/2024 11:24, Shreeya Patel wrote:
>>>>>>> On Thursday, March 28, 2024 04:20 IST, Shreeya Patel <shreeya.patel@xxxxxxxxxxxxx> wrote:
>>>>>>>
>>>>>>>> This series implements support for the Synopsys DesignWare
>>>>>>>> HDMI RX Controller, being compliant with standard HDMI 1.4b
>>>>>>>> and HDMI 2.0.
>>>>>>>>
>>>>>>>
>>>>>>> Hi Mauro and Hans,
>>>>>>>
>>>>>>> I haven't received any reviews so far. Hence, this is just a gentle reminder to review this patch series.
>>>>>>
>>>>>> Why did you put clk changes here? These go via different subsystem. That
>>>>>> might be one of obstacles for your patchset.
>>>>>>
>>>>>
>>>>> I added clock changes in this patch series because HDMIRX driver depends on it.
>>>>> I thought it is wrong to send the driver patches which don't even compile?
>>>>
>>>> Hm, why HDMIRX driver depends on clock? How? This sounds really wrong.
>>>> Please get it reviewed internally first.
>>>
>>> For the change in question, the clock controller on the soc also handles
>>> the reset controls (hence its name CRU, clock-and-reset-unit) .
>>>
>>> There are at least 660 reset lines in the unit and it seems the hdmi-rx one
>>> was overlooked on the initial submission, hence patches 1+2 add the
>>> reset-line.
>>>
>>> Of course, here only the "arm64: dts:" patch depends on the clock
>>> change, is it references the new reset-id.
>>
>> Wait, that's expected, but it is not what was written. Claim was HDMIRX
>> driver depends *build time* ("don't even compile").
>
> Trying to do a full build (kernel + dts) will fail, because the the
> device-tree patch references the newly added reset-id .
>
> So you end up with a dtc error. Same with the binding.

Which is quite expected, nothing special, most patchsets have exactly
the same dependency. It's not a HDMIRX driver dependency. It's DTS and
clock provider on the binding header, not clock consumer.

We solved it many times and different SoC subsystems have their own
guidelines. Putting here media is not the right approach and not justified.

Best regards,
Krzysztof