Re: [PATCH v3 10/10] media: platform: mtk-mdp3: decompose hardware-related information in shared memory

[AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>]

Il 03/11/22 07:48, Moudy Ho ha scritto:
> The communication between the MDP3 kernel driver and SCP is to
> pass a shared memory through the cooperation of "mtk-mdp3-vpu.c" and
> remoteproc driver.
> The data structure of this shared memory is defined in "mtk-img-ipi.h",
> as shown below:
> 
> vpu->work_addr -> +-----------------------------------------+
>                    |                                         |
>                    | To SCP : Input frame parameters         |
>                    |          (struct img_ipi_frameparam)    |
>                    |                                         |
>       vpu->pool -> +-----------------------------------------+
>                    |                                         |
>                    | From SCP : Output component config pool |
>                    |            (struct img_config)          |
>                    |                                         |
>                    |           *struct img_confg 1           |
>                    |                    |                    |
>                    |                    |                    |
>                    |                    v                    |
>                    |           *struct img_config N          |
>                    |            (N = MDP_CONFIG_POOL_SIZE)   |
>                    +-----------------------------------------+
> 
> One output component configuration contains the components
> currently used by the pipeline, and has the register settings
> that each component needs to set.
> 
> Since the quantity, type and function of components on each chip
> will vary, the effect is that the size of the "struct img_config"
> and its substructures will be different on each chip.
> In addition, all chips will have to update their SCP firmware for
> every change if the output component config structure is defined
> and shared by a common header.
> 
> Therefore, all functions that operate on "struct img_config" and
> its substructures must be separated by chips and so are the
> relevant definations.
> 
> Signed-off-by: Moudy Ho <moudy.ho@mediatek.com>

Hi Moudy,
thanks for this much needed rework of the IMG-IPI parameter passing architecture!

I can for sure go for a review of the code that you've currently pushed, but I
would prefer that you also push support for MT8192 and/or MT8195 (requiring the
different IPI structures and alignment), as previously discussed.

That will make us able to actually perform validation and will make us able to
give you a better code review.

Since this series is definitely big (hence, a bit difficult to review, but that's
fine, as there are no alternatives!), you can push support for the new chip(s) in
a separate series, dependent on this one, so that we also won't block this rework
for SoC-specific implementation code reviews.

Many thanks again!

P.S.: There's a typo in this commit message `*struct img_confg` :-)

Cheers,
Angelo