Re: [PATCH 13/13] media: qcom: camss: Add support for VFE hardware version Titan 780

[Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>]

Hi Depeng.

On 8/16/24 15:42, Depeng Shao wrote:
> Hi Vladimir,
> 
> On 8/16/2024 5:31 AM, Vladimir Zapolskiy wrote:
>> Hi Depeng.
>>
>> On 8/15/24 18:43, Depeng Shao wrote:
>>> Hi Vladimir,
>>>
>>>>>
>>>>> Thanks for the confirmation, even though I add the rup_update and
>>>>> buf_done function in later commits, it is still called in platform
>>>>> specific code(camss-vfe-780.c), so I will keep as it is done today.
>>>>
>>>> let it be so.
>>>>
>>>> I have another ask about it, please move new camss_reg_update() out from
>>>> camss.c into camss-csid.c, and camss_buf_done() from camss.c into camss-
>>>> vfe.c
>>>>
>>>
>>> The cross direct call has been removed by below commit, so it looks
>>> strange if I add the cross direct call.
>>>
>>> media: qcom: camss: Decouple VFE from CSID
>>> https://lore.kernel.org/lkml/20240522154659.510-9-
>>> quic_grosikop@quicinc.com/
>>
>> This I don't understand, please elaborate. I don't ask for a "cross direct
>> call", but you do introduce a CSID specific function in the generic camss.c
>> and another VFE specific function in the same camss.c
>>
> 
> CSID calls vfe_get/vfe_put to power up/reset vfe hw in old code, but
> above decouple commit removes this cross direct call, this commit has
> been merged recently.

Apparently I was imprecise, this is not the thing, which I don't understand,
for me it was the wording of "cross direct call".

>> What I ask is just move the current versions of camss_buf_done() and
>> camss_reg_update() out from camss.c to the files, which are related to the
>> sub-IP blocks, and of course move the function declarations from camss.h
>> into camss-vfe.h and camss-csid.h respectively.
>>
>> If possible there shall be no CSID or VFE specific specific code in
>> camss.c,
>> and that fact is that it's possible.
>>
> 
> Yes, I understand what you mean. Let's take camss_buf_done as example,
> if we move camss_buf_done to camss-vfe.c, but this function is called in
> csid csid driver, so here will have a cross direct call again,


> camss_reg_update is same. Since the cross call is removed in above
> commit, then it will be strange if I do this again.

It might be strange, but what I ask is to make the code way less convoluted.

> So, I moved them to camss.c

I'm still missing the essence of having two layers of indirection instead
of just one. Previous code was a function call from csid to vfe, now it's
csid to camss to vfe, I don't understand why there is a need to introduce
just an additional layer, it greatly complicates the code, also it slightly
drops the performance.

Previously there was no 'struct vfe_device *' or 'struct csid_device *'
types in the generic camss.c, now these "new" types leaked from camss-csid.h
and camss-vfe.h into camss.c, and the reason why remains unknown.

Okay, please ignore this one review request, let it be kept as is for a while.

As a side note, generally there might be various reasons to revert the code
changes or to return to the previous logic.

Thank you.

--
Best wishes,
Vladimir