Hi Heikki and Rob, (trimming text): On Mon, Jun 15, 2020 at 04:22:07PM +0300, Heikki Krogerus wrote: > On Fri, Jun 12, 2020 at 10:34:06AM -0700, Prashant Malani wrote: > > Hi Rob, > > > Yes, but let's stop calling it a mux. It's a "USB Type C signal routing blob". > > > > Ack. > > > > Let's go with "-switch" ? That's what the connector class uses and it > > conveys the meaning (unless that is a reserved keyword in DT). > > Just as a clarification here, we should not be even talking about > signal routing here. We are talking about functions that an external > components perform from the connector's perspective. It depends on the > platform does that function require changing the routing of the lines > from the connector. For example, data role swapping does not require > muxing on platforms that have single dual-role USB controller, but > platforms that have separate IPs for the USB host and USB device > controllers will need a mux. > > Note, that it is even possible that switching from USB to DisplayPort > mode does not require any pin reconfiguration from the mux, even if > the platform has one, because the PHY can be shared between USB3 and > DP. Then the PHY just needs to be told that it is now in DP mode when > DP alt mode is entered instead of the mux. > > > > > Would this block the addition of the "*-switch" properties? IIUC the > > > > two are related but not dependent on each other. > > > > > > > > The *-switch properties are phandles which the Type C connector class > > > > framework expects (and uses to get handles to those switches). > > > > These would point to the "mux" or "group of mux" abstractions as noted earlier. > > > > > > You don't need them though. Walk the graph. You get the connector > > > port@1 remote endpoint and then get its parent. > > > > > > > I see; would it be something along the lines of this? (DT example > > follows; search for "example_end" to jump to bottom): > > I just realized that you have in practice placed the mux-agent into > the graph below, right? That we can not do, because it is not > physically connected to the connector. Is this a requirement? I read through the graph.txt file [1] and I couldn't find anything stating that a physical connection between two devices was required (but I may be misinterpreting that document) [1]: https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/graph.txt > > > <example_start> > > > > <example_end> > > > > Would this be conformant to OF graph and usb-connector bindings > > requirements? We'll certainly send out a format PATCH/RFC series for > > this, but I was hoping to gauge whether we're thinking along the right lines. > > > > So, in effect this would mean: > > - New bindings(and compatible strings) to be added for: > > typec-{orientation,data-role,mode}-switch. > > - Handling in Type C connector class to parse switches from OF graph. > > - Handling in Type C connector class for distinct switches for port@1 > > (SS lines) and port@2 (SBU lines). > > > > The only thing I'm confused about is how we can define these switch > > remote-endpoint bindings in usb-connector.yaml; the port can have an > > remote-endpoint, but can we specify what the parent of the remote-endpoint > > should have as a compatible string? Or do we not need to? > > thanks, > > -- > heikki Best regards, -Prashant
