[PATCH v5 5/9] drm/bridge: anx7625: Add typec_mux_set callback function

[Pin-yen Lin]

On Thu, Jul 7, 2022 at 8:17 AM Stephen Boyd <swboyd at chromium.org> wrote:
>
> Quoting Prashant Malani (2022-07-06 11:26:19)
> >
> > Stephen, any pending concerns?
>
> No more pending concerns.
>
> > If not,I will post a v6 series with the suggested changes:
> > - Drop typec-switch binding; instead add a new top-level port with
> > end-points for each Type-C connector's switch.
> > - Drop it6505 patches.
> > - Squash anx7625 driver patches into one patch.
> > - Add a comment mentioning that we aren't registering the orientation-switch.

We've been working on these changes, and the new DT node looks like this:

```
    anx_bridge_dp: anx7625-dp at 58 {
        [...]
        mode-switch;
        ports {
            [...]
            typec_switches: port at 2 {
                #adderss-cells = <1>;
                #size-cells = <0>;
                reg = <2>;

                anx_typec0: endpoint at 0 {
                    reg = <0>;
                    remote-endpoint = <&typec_port0>;
                };
                anx_typec1: endpoint at 1 {
                    reg = <1>;
                    remote-endpoint = <&typec_port1>;
                };
            };
        };
```

However we found some issues with that approach:
1. The current typec mux API does not allow us to put muxes into
`ports` directly.
`fwnode_typec_mux_get` searches for the parent node behind the port(s)
nodes, so we cannot register the muxes with the port nodes unless we
change the interface.
2. We need a compatible string between the `endpoint` nodes and the
parent node (anx7625-dp at 58).
This is because when the driver core builds the device links, they
only add links on nodes with a compatible string for `remote-endpoint`
properties[1].
Without a compatible string, the parent node of `typec_port0`
(cros-ec-typec in our case) has to be probed before anx7625, but this
leads to a deadlock because cros-ec-typec requires anx7625 to register
the typec_mux drivers first. I'm not sure if this is cros-ec-typec
specific, though.
*Any* compatible string fixes this issue, and it doesn't have to be
"typec-switch".

--

Alternatively, can we split the two muxes into two sub-nodes, like the
following snippet?

```
    anx_bridge_dp: anx7625-dp at 58 {
        [...]
        mode-switch;

        anx_mux0 {
            compatible = "typec-switch";
            reg = <0>;

            port {
                anx_typec0: endpoint {
                    remote-endpoint = <&typec_port0>;
                };
            };
        };

        anx_mux1 {
            compatible = "typec-switch";
            reg = <1>;

            port {
                anx_typec1: endpoint {
                    remote-endpoint = <&typec_port1>;
                };
            };
        };
```

This eliminates the additional "switches" node in the devicetree. The
sub-nodes also describe our hardware design, which split the DP lanes
of anx7625 to two type-c ports.

[1]: The `node_not_dev` property searches for a node with a compatible
string: https://elixir.bootlin.com/linux/latest/source/drivers/of/property.c#L1390



>
> Ok. I'll take a look on v6.