Re: [PATCH v6 12/16] dt-bindings: display: mediatek: color: add compatible for MT8195

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Il 28/09/23 18:49, Conor Dooley ha scritto:
On Thu, Sep 28, 2023 at 02:52:23AM +0000, Moudy Ho (何宗原) wrote:
On Wed, 2023-09-27 at 10:47 +0100, Conor Dooley wrote:
On Wed, Sep 27, 2023 at 07:19:28AM +0000, Moudy Ho (何宗原) wrote:
On Fri, 2023-09-22 at 16:51 +0100, Conor Dooley wrote:
On Fri, Sep 22, 2023 at 04:49:14PM +0100, Conor Dooley wrote:
On Fri, Sep 22, 2023 at 03:21:12PM +0800, Moudy Ho wrote:
Add a compatible string for the COLOR block in MediaTek
MT8195
that
is controlled by MDP3.

Signed-off-by: Moudy Ho <moudy.ho@xxxxxxxxxxxx>
---
  .../devicetree/bindings/display/mediatek/mediatek,color.yaml
| 1 +
  1 file changed, 1 insertion(+)

diff --git
a/Documentation/devicetree/bindings/display/mediatek/mediatek
,col
or.yaml
b/Documentation/devicetree/bindings/display/mediatek/mediatek
,col
or.yaml
index f21e44092043..b886ca0d89ea 100644
---
a/Documentation/devicetree/bindings/display/mediatek/mediatek
,col
or.yaml
+++
b/Documentation/devicetree/bindings/display/mediatek/mediatek
,col
or.yaml
@@ -26,6 +26,7 @@ properties:
            - mediatek,mt2701-disp-color
            - mediatek,mt8167-disp-color
            - mediatek,mt8173-disp-color
+          - mediatek,mt8195-mdp3-color

How come this one is a "mdp3" not a "disp"?

I don't know what mdp3 means & googling gives me no answers.
What's
the
"disp" one controlled by, since it isn't controlled by mdp3?


Mediatek's Media Data Path ver.3 (MDP3) is associated with MMSYS
and
acts as an independent driver that operates between VDEC and DISP.
By controlling multiple components, it carries out tasks like
converting color formats, resizing, and applying specific Picture
Quality (PQ) effects.
The driver can be found at "driver/media/platform/mediatek/mdp3".
Since the same hardware components are configured in both MDP3 and
DISP, considering previous discussions, I attemped to integrate
into a
single binding, named after the controlling user.

I'm still kinda struggling to understand this. Do you mean that the
hardware can be controlled by either of the disp and mdp3 drivers,
and
a compatible containing "disp" would use one driver, and one
containing
"mdp3" would use another?


Sorry for any confusion caused by the software information. In the
video pipeline, after decoding, the data flows sequentially through two
subsystems: MDP and DISP. Each subsystems has multiple IPs, with some
serving the same functionality as COLOR mentioned here. However, these
IPs cannot be controlled by different subsystems. Therefore, I included
the name of the subsystem after SoC to identify the configuration's
location. Is this approach feasible?

I'll have to leave things to the likes of Laurent to comment here I
think. I don't understand this hardware well enough to have a useful
opinion. It would seem like a different part of the datapath is a
different device that should be documented separately, but I don't know
enough to say for sure, sorry.

Hardware speaking, it's not a different device: those all reside in the
same block, except they are configured to route their I/O *either* to the
display pipeline, *or* to the MDP3 pipeline.

I would agree though in that this could be more flexible, as in, not
having a requirement to say "mdp3" or "disp", and managing the COLOR
blocks generically and letting the drivers to choose the actual path
transparently from what the devicetree compatible is, but there's no
practical point in doing this in the end, because there is an enough
number of (for example, COLOR) blocks such that one can be completely
reserved to MDP3 and one completely reserved to DISP.

So, we don't *need* this flexibility, but would be nice to have for
different (unexistant, basically) usecases...

The thing is, if we go for the maximum flexibility, the drawback is
that we'd see a number of nodes like

shared_block: something@somewhere {
	compatible = "mediatek,something";
}

mdp3: dma-controller@14001000 {
	......
	mediatek,color = <&color0>;
	mediatek,stitch = <&stitch0>;
	mediatek,hdr = <&hdr0>;
	mediatek,aal = <&aal0>;
	....
	long list of another 10 components
}

display: something@somewhere {
	......
	an even longer list than the MDP3 one
}

...or perhaps even a graph, which is even longer in the end.

I'm not against this kind of structure, but I wonder if it's worth it.

Cheers,
Angelo



[Index of Archives]     [Linux Input]     [Video for Linux]     [Gstreamer Embedded]     [Mplayer Users]     [Linux USB Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Yosemite Backpacking]

  Powered by Linux