Re: [PATCH v2 07/16] drm/imx: Add i.MX8qxp Display Controller pixel engine

[Krzysztof Kozlowski <krzk@kernel.org>]

On 30/07/2024 11:42, Liu Ying wrote:
> On 07/30/2024, Krzysztof Kozlowski wrote:
>> On 30/07/2024 08:55, Liu Ying wrote:
>>> On 07/28/2024, Dmitry Baryshkov wrote:
>>>> On Fri, Jul 12, 2024 at 05:32:34PM GMT, Liu Ying wrote:
>>>>> i.MX8qxp Display Controller pixel engine consists of all processing
>>>>> units that operate in the AXI bus clock domain.  Add drivers for
>>>>> ConstFrame, ExtDst, FetchLayer, FetchWarp and LayerBlend units, as
>>>>> well as a pixel engine driver, so that two displays with primary
>>>>> planes can be supported.  The pixel engine driver as a master binds
>>>>> those unit drivers as components.  While at it, the pixel engine
>>>>> driver is a component to be bound with the upcoming DRM driver.
>>>>
>>>> Same question / comment: create subnodes directly, without going
>>>> through the subdevices. A lot of small functions that would benefit
>>>> being inlined.
>>>
>>> Like I replied in patch 06/16, I can't create sub devices directly.
>>>
>>> Can you please point out typical ones for those small functions if
>>> the comment still stands?
>>>
>>>>
>>>>> +static int dc_cf_bind(struct device *dev, struct device *master, void *data)
>>>>> +{
>>>>> +	struct platform_device *pdev = to_platform_device(dev);
>>>>> +	struct dc_drm_device *dc_drm = data;
>>>>> +	struct dc_pe *pe = dc_drm->pe;
>>>>> +	struct dc_cf_priv *priv;
>>>>> +	int id;
>>>>> +
>>>>> +	priv = drmm_kzalloc(&dc_drm->base, sizeof(*priv), GFP_KERNEL);
>>>>> +	if (!priv)
>>>>> +		return -ENOMEM;
>>>>> +
>>>>> +	priv->reg_cfg = devm_platform_ioremap_resource_byname(pdev, "cfg");
>>>>> +	if (IS_ERR(priv->reg_cfg))
>>>>> +		return PTR_ERR(priv->reg_cfg);
>>>>> +
>>>>> +	id = of_alias_get_id(dev->of_node, "dc0-constframe");
>>>>
>>>> Is it documented? Acked?
>>>
>>> Like I replied in patch 06/16, I can add aliases nodes to examples,
>>> if needed.
>>>
>>> No Nak from DT maintainers I'd say, but I hope there will be direct
>>> Ack(s).
>>>
>>
>> It was not Acked, because there was no documentation added for it.
> 
> I may add aliases nodes in examples in next version, if no objections.

Example is just example. It is not a documentation. You must explain it
in the binding, e.g. description.

> 
>> Anyway, naming is quite cryptic, e.g. "0" in "dc0" is quite confusing.
>> Do you expect different aliases for dc1 or dc9? But anyway, aliases for
> 
> Yes, I do.  If the alias approach is used, DC instance ids need to be
> specified in aliases.

Really? Uh, that does not look good. I tend to like this binding less
and less.


Best regards,
Krzysztof