Re: [PATCH 07/14] dt-bindings: media: i2c: Add A31 MIPI CSI-2 bindings documentation

From: Paul Kocialkowski
Date: Wed Nov 04 2020 - 05:48:37 EST


Hi,

On Tue 27 Oct 20, 19:44, Maxime Ripard wrote:
> On Tue, Oct 27, 2020 at 10:52:21AM +0100, Paul Kocialkowski wrote:
> > Hi,
> >
> > On Mon 26 Oct 20, 17:14, Maxime Ripard wrote:
> > > i2c? :)
> >
> > Oops, good catch!
> >
> > > On Fri, Oct 23, 2020 at 07:45:39PM +0200, Paul Kocialkowski wrote:
> > > > This introduces YAML bindings documentation for the A31 MIPI CSI-2
> > > > controller.
> > > >
> > > > Signed-off-by: Paul Kocialkowski <paul.kocialkowski@xxxxxxxxxxx>
> > > > ---
> > > > .../media/allwinner,sun6i-a31-mipi-csi2.yaml | 168 ++++++++++++++++++
> > > > 1 file changed, 168 insertions(+)
> > > > create mode 100644 Documentation/devicetree/bindings/media/allwinner,sun6i-a31-mipi-csi2.yaml
> > > >
> > > > diff --git a/Documentation/devicetree/bindings/media/allwinner,sun6i-a31-mipi-csi2.yaml b/Documentation/devicetree/bindings/media/allwinner,sun6i-a31-mipi-csi2.yaml
> > > > new file mode 100644
> > > > index 000000000000..9adc0bc27033
> > > > --- /dev/null
> > > > +++ b/Documentation/devicetree/bindings/media/allwinner,sun6i-a31-mipi-csi2.yaml
> > > > @@ -0,0 +1,168 @@
> > > > +# SPDX-License-Identifier: GPL-2.0
> > > > +%YAML 1.2
> > > > +---
> > > > +$id: http://devicetree.org/schemas/media/allwinner,sun6i-a31-mipi-csi2.yaml#
> > > > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > > +
> > > > +title: Allwinner A31 MIPI CSI-2 Device Tree Bindings
> > > > +
> > > > +maintainers:
> > > > + - Paul Kocialkowski <paul.kocialkowski@xxxxxxxxxxx>
> > > > +
> > > > +properties:
> > > > + compatible:
> > > > + oneOf:
> > > > + - const: allwinner,sun6i-a31-mipi-csi2
> > > > + - items:
> > > > + - const: allwinner,sun8i-v3s-mipi-csi2
> > > > + - const: allwinner,sun6i-a31-mipi-csi2
> > > > +
> > > > + reg:
> > > > + maxItems: 1
> > > > +
> > > > + interrupts:
> > > > + maxItems: 1
> > > > +
> > > > + clocks:
> > > > + items:
> > > > + - description: Bus Clock
> > > > + - description: Module Clock
> > > > +
> > > > + clock-names:
> > > > + items:
> > > > + - const: bus
> > > > + - const: mod
> > > > +
> > > > + phys:
> > > > + items:
> > > > + - description: MIPI D-PHY
> > > > +
> > > > + phy-names:
> > > > + items:
> > > > + - const: dphy
> > > > +
> > > > + resets:
> > > > + maxItems: 1
> > > > +
> > > > + # See ./video-interfaces.txt for details
> > > > + ports:
> > > > + type: object
> > > > +
> > > > + properties:
> > > > + port@0:
> > > > + type: object
> > > > + description: Input port, connect to a MIPI CSI-2 sensor
> > > > +
> > > > + properties:
> > > > + reg:
> > > > + const: 0
> > > > +
> > > > + endpoint:
> > > > + type: object
> > > > +
> > > > + properties:
> > > > + remote-endpoint: true
> > > > +
> > > > + bus-type:
> > > > + const: 4
> > > > +
> > > > + clock-lanes:
> > > > + maxItems: 1
> > > > +
> > > > + data-lanes:
> > > > + minItems: 1
> > > > + maxItems: 4
> > > > +
> > > > + required:
> > > > + - bus-type
> > > > + - data-lanes
> > > > + - remote-endpoint
> > > > +
> > > > + additionalProperties: false
> > > > +
> > > > + required:
> > > > + - endpoint
> > > > +
> > > > + additionalProperties: false
> > > > +
> > > > + port@1:
> > > > + type: object
> > > > + description: Output port, connect to a CSI controller
> > > > +
> > > > + properties:
> > > > + reg:
> > > > + const: 1
> > > > +
> > > > + endpoint:
> > > > + type: object
> > > > +
> > > > + properties:
> > > > + remote-endpoint: true
> > > > +
> > > > + bus-type:
> > > > + const: 4
> > >
> > > That one seems a bit weird. If the input and output ports are using the
> > > same format, what is that "bridge" supposed to be doing?
> >
> > Fair enough. What this represents is the internal link (likely a FIFO) between
> > the two controllers. It is definitely not a MIPI CSI-2 bus but there's no
> > mbus type for an internal link (probably because it's not a bus after all).
> >
> > Note that on the CSI controller side, we need the bus-type to be set to 4 for it
> > to properly select the MIPI CSI-2 input. So it just felt more logical to have
> > the same on the other side of the endpoint. On the other hand, we can just
> > remove it on the MIPI CSI-2 controller side since it won't check it and have it
> > fallback to the unknown mbus type.
> >
> > But that would make the types inconsistent on the two sides of the link.
> > I don't think V4L2 will complain about it at the moment, but it would also make
> > sense that it does eventually.
> >
> > What do you think?
>
> There's still the same issue though, it doesn't make any sense that a
> bridge doesn't change the bus type. If it really did, we wouldn't need
> that in the first place.

Yes I agreee.

> What you want to check in your driver is whether the subdev you're
> connected to has a sink pad that uses MIPI-CSI

I'm not really sure that's possible, but if it is it would indeed be the most
appropriate solution. If it's not, we still need to know that we need to feed
from MIPI CSI-2 so I don't see any other option than report MIPI CSI-2 on both
ends of MIPI CSI-2 controller.

But there's still the question of what media bus type should be reported for
the CSI <-> MIPI CSI-2 link. I'm fine with unknown but we could also add a
generic internal bus type for this case.

Paul

> Maxime
>
> > > > + additionalProperties: false
> > > > +
> > > > + required:
> > > > + - endpoint
> > > > +
> > > > + additionalProperties: false
> > > > +
> > > > +required:
> > > > + - compatible
> > > > + - reg
> > > > + - interrupts
> > > > + - clocks
> > > > + - clock-names
> > > > + - resets
> > > > +
> > > > +additionalProperties: false
> > > > +
> > > > +examples:
> > > > + - |
> > > > + #include <dt-bindings/interrupt-controller/arm-gic.h>
> > > > + #include <dt-bindings/clock/sun8i-v3s-ccu.h>
> > > > + #include <dt-bindings/reset/sun8i-v3s-ccu.h>
> > > > +
> > > > + mipi_csi2: mipi-csi2@1cb1000 {
> > >
> > > The unit name should be pretty standard, with the list here:
> > >
> > > https://github.com/devicetree-org/devicetree-specification/blob/master/source/chapter2-devicetree-basics.rst#generic-names-recommendation
> > >
> > > there's nothing really standing out for us in that list, but given that
> > > there's dsi, we should stick with csi
> >
> > Then what really surprises me is that the CSI controllers are called "camera",
> > not "csi". If "camera" is supposed to cover both image sensor and camera sensor
> > interfaces, it would probably fit MIPI CSI-2 as well.
> >
> > I see lots of names with -controller for controllers with specific devices
> > attached, like "nand-controller" or "lcd-controller". Maybe using
> > "camera-controller" for the CSI and MIPI CSI-2 controllers would make the most
> > sense, while keeping "camera" for the actual image sensors.
> >
> > What do you think?
>
> If you really want to discuss this, feel free to open a PR for the DT
> spec and add it. However, I still think this csi would be best here:
> it's neither a camera nor a camera controller
>
> maxime



--
Paul Kocialkowski, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

Attachment: signature.asc
Description: PGP signature