Hi Paul,
> Am 28.09.2021 um 14:06 schrieb H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>:
>
> Hi Paul,
>
>> Am 28.09.2021 um 12:21 schrieb H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>:
>>
>>>> @@ -1492,10 +1555,16 @@ static int ingenic_drm_init(void)
>>>> {
>>>> int err;
>>>> + if (IS_ENABLED(CONFIG_DRM_INGENIC_DW_HDMI)) {
>>>> + err = platform_driver_register(ingenic_dw_hdmi_driver_ptr);
>>>> + if (err)
>>>> + return err;
>>>> + }
>>>
>>> I don't see why you need to register the ingenic-dw-hdmi driver here. Just register it in the ingenic-dw-hdmi driver.
>>
>> Ok, I never though about this (as the code was not from me). We apparently just followed the IPU code pattern (learning by example).
>>
>> It indeed looks not necessary and would also avoid the ingenic_dw_hdmi_driver_ptr dependency.
>>
>> But: what is ingenic_ipu_driver_ptr then good for?
>>
>> If we can get rid of this as well, we can drop patch 1/10 ("drm/ingenic: Fix drm_init error path if IPU was registered") completely.
>
> A quick test shows that it *is* required. At least if I configure everything as modules.
> But like you I can't explain why.
>
> Well, just a very rough idea (may be wrong): the bridge chain is not like an i2c bus and
> clients are not automatically loaded/probed if linked in the device tree. Therefore the
> consumer (ingenic_drm_drv) must register the "clients" like IPU and HDMI.
Any suggestion how to proceed here for v5?
BR,
Nikolaus
