Hi Laurent, Looks like I didn't follow-up here. On Tue 15 Feb 22, 12:16, Laurent Pinchart wrote: > Hi Paul, > > On Tue, Feb 15, 2022 at 11:10:52AM +0100, Paul Kocialkowski wrote: > > On Mon 14 Feb 22, 19:09, Laurent Pinchart wrote: > > > On Mon, Feb 14, 2022 at 05:18:07PM +0100, Paul Kocialkowski wrote: > > > > On Mon 07 Feb 22, 17:51, Laurent Pinchart wrote: > > > > > On Sat, Feb 05, 2022 at 07:54:24PM +0100, Paul Kocialkowski wrote: > > > > > > This introduces YAML bindings documentation for the Allwinner A31 Image > > > > > > Signal Processor (ISP). > > > > > > > > > > > > Signed-off-by: Paul Kocialkowski > > > > > > --- > > > > > > .../media/allwinner,sun6i-a31-isp.yaml | 117 ++++++++++++++++++ > > > > > > 1 file changed, 117 insertions(+) > > > > > > create mode 100644 Documentation/devicetree/bindings/media/allwinner,sun6i-a31-isp.yaml > > > > > > > > > > > > diff --git a/Documentation/devicetree/bindings/media/allwinner,sun6i-a31-isp.yaml b/Documentation/devicetree/bindings/media/allwinner,sun6i-a31-isp.yaml > > > > > > new file mode 100644 > > > > > > index 000000000000..2d87022c43ce > > > > > > --- /dev/null > > > > > > +++ b/Documentation/devicetree/bindings/media/allwinner,sun6i-a31-isp.yaml > > > > > > @@ -0,0 +1,117 @@ > > > > > > +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause) > > > > > > +%YAML 1.2 > > > > > > +--- > > > > > > +$id: http://devicetree.org/schemas/media/allwinner,sun6i-a31-isp.yaml# > > > > > > +$schema: http://devicetree.org/meta-schemas/core.yaml# > > > > > > + > > > > > > +title: Allwinner A31 Image Signal Processor Driver (ISP) Device Tree Bindings > > > > > > + > > > > > > +maintainers: > > > > > > + - Paul Kocialkowski > > > > > > + > > > > > > +properties: > > > > > > + compatible: > > > > > > + enum: > > > > > > + - allwinner,sun6i-a31-isp > > > > > > + - allwinner,sun8i-v3s-isp > > > > > > + > > > > > > + reg: > > > > > > + maxItems: 1 > > > > > > + > > > > > > + interrupts: > > > > > > + maxItems: 1 > > > > > > + > > > > > > + clocks: > > > > > > + items: > > > > > > + - description: Bus Clock > > > > > > + - description: Module Clock > > > > > > + - description: DRAM Clock > > > > > > > > > > That's interesting, does the ISP have a dedicated DRAM ? > > > > > > > > It doesn't, it actually uses the main DRAM with the "mbus" interconnect. > > > > The clock is probably for the DMA engine. > > > > > > > > > > + > > > > > > + clock-names: > > > > > > + items: > > > > > > + - const: bus > > > > > > + - const: mod > > > > > > + - const: ram > > > > > > + > > > > > > + resets: > > > > > > + maxItems: 1 > > > > > > + > > > > > > + ports: > > > > > > + $ref: /schemas/graph.yaml#/properties/ports > > > > > > + > > > > > > + properties: > > > > > > + port@0: > > > > > > + $ref: /schemas/graph.yaml#/$defs/port-base > > > > > > + description: CSI0 input port > > > > > > + > > > > > > + properties: > > > > > > + reg: > > > > > > + const: 0 > > > > > > + > > > > > > + endpoint: > > > > > > + $ref: video-interfaces.yaml# > > > > > > + unevaluatedProperties: false > > > > > > > > > > If no other property than remote-endpoint are allowed, I'd write > > > > > > > > > > endpoint: > > > > > $ref: video-interfaces.yaml# > > > > > remote-endpoint: true > > > > > additionalProperties: false > > > > > > > > > > Same below. > > > > > > > > > > > + > > > > > > + additionalProperties: false > > > > > > + > > > > > > + port@1: > > > > > > + $ref: /schemas/graph.yaml#/$defs/port-base > > > > > > + description: CSI1 input port > > > > > > + > > > > > > + properties: > > > > > > + reg: > > > > > > + const: 0 > > > > > > > > > > This should be 1. > > > > > > > > Correct, thanks! > > > > > > > > > > + > > > > > > + endpoint: > > > > > > + $ref: video-interfaces.yaml# > > > > > > + unevaluatedProperties: false > > > > > > + > > > > > > + additionalProperties: false > > > > > > + > > > > > > + anyOf: > > > > > > + - required: > > > > > > + - port@0 > > > > > > + - required: > > > > > > + - port@1 > > > > > > > > > > As ports are an intrinsic property of the ISP, both should be required, > > > > > but they don't have to be connected. > > > > > > > > Well the ISP does have the ability to source from either CSI0 and CSI1 > > > > but I don't really get the point of declaring both ports when only one > > > > of the two controllers is present. > > > > > > If it's within an SoC I don't mind too much. What I usually insist on is > > > declaring all ports even when no external devices are connected on the > > > board. It may however be easier to implement things on the driver side > > > when all the ports are declared, even for internal devices. I won't > > > insist either way here. > > > > > > > > By the way, how do you select at runtime which CSI-2 RX the ISP gets its > > > > > image stream from ? Is it configured through registers of the ISP ? > > > > > > > > Actually what the ISP gets is fully dependent on what is received by the > > > > CSI controller it is connected to (which can be the mipi csi-2 controller > > > > or its direct parallel pins), so the configuration happens on the CSI side. > > > > > > OK, then how do you select at runtime which CSI the ISP gets its image > > > stream from ? :-) > > > > What is done in the driver is that all available csi(s) entities pads are linked > > to a single csi sink media pad, which allows userspace to enable one or the > > other. If there's only one, it's enabled by default. > > > > The actual stream source (isp_dev->proc.source) is selected at link_validate > > time and the source bit is set in sun6i_isp_proc_enable. > > > > I hope this answers your question! > > Yes it does, thank you. > > While this works, it makes life a bit more complicated for userspace, as > switching between the two sources require disabling the link first and > then enabling the new one. This is something that caused issues in the > libcamera simple pipeline handler, I ended up having to implement a > workaround. That surprises me a bit, I thought this was a typical use-case for links. So the fact that it's a two-step process causes issues somehow? > Could you instead have two sink pads for the ISP, and select the sensor > at stream on time instead of link validation time by checking which link > is enabled ? If no links or both links are enabled, you can then return > an error. Yes that's totally doable. There's a similar situation with the sun6i-csi bridge where the source pad has two possible links: one for routing to sun6i-csi capture (video device) and one for routing to the isp entity. Would that also be best represented as two pads? > Ideally I'd say such internal routing should use the new V4L2 subdev > routing API that is currently being implemented (see [1]), but I don't > know when it will land, and I don't want to delay your patch series. > > [1] https://lore.kernel.org/linux-media/20211130141536.891878-28-tomi.valkeinen@ideasonboard.com I'm still a bit confused what problem this is trying to solve. My understanding was that the current pad/link API allows representing complex topologies and switching different paths with link enable/disable. Cheers, Paul > > > > > > + > > > > > > +required: > > > > > > + - compatible > > > > > > + - reg > > > > > > + - interrupts > > > > > > + - clocks > > > > > > + - clock-names > > > > > > + - resets > > > > > > + > > > > > > +additionalProperties: false > > > > > > + > > > > > > +examples: > > > > > > + - | > > > > > > + #include > > > > > > + #include > > > > > > + #include > > > > > > + > > > > > > + isp: isp@1cb8000 { > > > > > > + compatible = "allwinner,sun8i-v3s-isp"; > > > > > > + reg = <0x01cb8000 0x1000>; > > > > > > + interrupts = ; > > > > > > + clocks = <&ccu CLK_BUS_CSI>, > > > > > > + <&ccu CLK_CSI1_SCLK>, > > > > > > + <&ccu CLK_DRAM_CSI>; > > > > > > + clock-names = "bus", "mod", "ram"; > > > > > > + resets = <&ccu RST_BUS_CSI>; > > > > > > + > > > > > > + ports { > > > > > > + #address-cells = <1>; > > > > > > + #size-cells = <0>; > > > > > > + > > > > > > + port@0 { > > > > > > + reg = <0>; > > > > > > + > > > > > > + isp_in_csi0: endpoint { > > > > > > + remote-endpoint = <&csi0_out_isp>; > > > > > > + }; > > > > > > + }; > > > > > > + }; > > > > > > + }; > > > > > > + > > > > > > +... > > -- > Regards, > > Laurent Pinchart -- Paul Kocialkowski, Bootlin Embedded Linux and kernel engineering https://bootlin.com