On 7/15/25 11:20 AM, Vladimir Zapolskiy wrote:

On 7/15/25 12:01, Konrad Dybcio wrote:

quoted

On 7/15/25 8:35 AM, Vladimir Zapolskiy wrote:

quoted

On 7/15/25 03:13, Bryan O'Donoghue wrote:

quoted

On 14/07/2025 16:30, Vladimir Zapolskiy wrote:

quoted

quoted

I think that is genuinely something we should handle in camss-csid.c
maybe with some meta-data inside of the ports/endpoints..

This is a CSIPHY property, a CSIPHY hardware configuration and a wiring
of sensors to a CSIPHY. Where is the relation to CSID here? There is no.

All the PHY really needs to know is the # of lanes in aggregate, which
physical lanes to map to which logical lanes and the pixel clock.

We should add additional support to the Kernel's D-PHY API parameters
mechanism to support that physical-to-logical mapping but, that's not
required for this series or for any currently know upstream user of CAMSS.

quoted

Please share at least a device tree node description, which supports
a connection of two sensors to a single CSIPHY, like it shall be done
expectedly.

&camss {
       port@0 {
           csiphy0_lanes01_ep: endpoint0 {
               data-lanes = <0 1>;
               remote-endpoint = <&sensor0_ep>;
           };

           csiphy0_lanes23_ep: endpoint0 {
               data-lanes = <2 3>;
               remote-endpoint = <&sensor1_ep>;
           };
        };
};

Don't you understand that this is broken?.. That's no good.

Please listen and reread the messages given to you above, your proposed
"solution" does not support by design a valid hardware setup of two
sensors connected to the same CSIPHY.

I would propose to stop force pushing an uncorrectable dt scheme, it
makes no sense.

If all you're asking for is an ability to grab an of_graph reference
from the camss (v4l2) driver, you can simply do something along the
lines of of_graph_get_remote_port(phy->dev->of_node)

It's not about the driver specifics, my comment is about a proper
hardware description in dts notation, please see the device tree node
names.

I'm a little lost on what you're trying to argue for..

I could make out:

1. "the phy should be a multimedia device"
2. "There is no ports at all, which makes the device tree node unusable,
  since you can not provide a way to connect any sensors to the phy."

I don't really understand #1.. maybe that could be the case if the PHY
has a multitude of tunables (which I don't know if it does, but wouldn't
be exactly surprised if it did) that may be usecase/pipeline-specific

As for #2, I do think it makes sense to connect the sensors to the PHY,
as that's a representation of electrical signals travelling from the
producer to the consumer (plus the data passed in e.g. data-lanes is
directly related to the PHY and necessarily consumed by its driver)

Konrad