Hi Tomi, On Fri, Mar 25, 2022 at 12:42:15PM +0200, Tomi Valkeinen wrote: > On 24/03/2022 16:23, Maxime Ripard wrote: > > On Thu, Mar 24, 2022 at 03:43:42PM +0200, Laurent Pinchart wrote: > > > On Thu, Mar 24, 2022 at 09:18:19AM +0100, Maxime Ripard wrote: > > > > On Wed, Mar 23, 2022 at 10:38:19PM +0200, Laurent Pinchart wrote: > > > > > Hi Maxime, > > > > > > > > > > (CC'ing Sakari) > > > > > > > > > > Thank you for the patch. > > > > > > > > > > On Wed, Mar 23, 2022 at 04:48:23PM +0100, Maxime Ripard wrote: > > > > > > MIPI-DSI devices, if they are controlled through the bus itself, have to > > > > > > be described as a child node of the controller they are attached to. > > > > > > > > > > > > Thus, there's no requirement on the controller having an OF-Graph output > > > > > > port to model the data stream: it's assumed that it would go from the > > > > > > parent to the child. > > > > > > > > > > > > However, some bridges controlled through the DSI bus still require an > > > > > > input OF-Graph port, thus requiring a controller with an OF-Graph output > > > > > > port. This prevents those bridges from being used with the controllers > > > > > > that do not have one without any particular reason to. > > > > > > > > > > > > Let's drop that requirement. > > > > > > > > > > I'm sure this won't come as a surprise, I'm very much opposed to this > > > > > change, for two reasons. > > > > > > > > > > First, ports are part of the hardware, even if they're not connected. It > > > > > thus simplifies handling in drivers if they're always present. > > > > > > > > > > Then, and that's the most important reason, I think it's a mistake not > > > > > to model the DSI data connection using OF graph unconditionally, even > > > > > when the DSI sink device is also controlled through the DSI bus (using > > > > > DCS) and is in that case a child of the DSI source device in the DT > > > > > hierarchy. > > > > > > > > That's the way we do for any other device though. You never addressed > > > > that comment, but it's very much the same that occurs for i2c or spi > > > > controllers and their device. They all get their data from the parent > > > > bus. I don't see you advocate for using OF-Graph for those devices. > > > > > > Those are different, there's no data stream independent of the control > > > communications. > > > > Fine, then you have Ethernet PHYs, or any MMIO device that does DMA. > > Have those devices had the need for OF graphs? No, they don't. It's kind of my point actually :) > For display and capture we have a clear need. I don't think we should > sometimes use OF graphs and sometimes not, but rather use them > consistently at least in any new driver. There's a clear need when the data don't follow the obvious path, ie when there's a direct FIFO between a CRTC and its connector. If it was going through that obvious path, like an Ethernet PHY, then we wouldn't need it. A device being controlled through the DSI bus has that obvious path. If it's taking its data through the DSI bus, but is being controlled by an i2c bus, then it needs an OF-graph node... > > > > > The device tree describes a control hierarchy between devices. OF graph > > > > > overlays on top of that a data transfer graph. The two are different > > > > > concepts, and the fact that DSI can sometimes be used as a control bus > > > > > doesn't change the concept. Using OF graph unconditionally to describe > > > > > the data connections for DSI leads to less variation in the device tree > > > > > structure, and thus less complexity in the implementation. We're > > > > > suffering from the fact we haven't made it a requirement in the first > > > > > place, which can't be fixed due to ABI breakage constraints, but let's > > > > > not acknowledge it as a good idea. > > > > > > > > Honestly, it doesn't matter one bit. > > > > > > > > We have a huge discrepancy here today, and only a couple of bridges have > > > > that arbitrary restriction. The situation you don't want to acknowledge > > > > is the de-facto standard, by the generic binding and by what all the > > > > bridges and panels are implementing. Even panel-simple-dsi is doing it. > > > > So it's very much there already. > > > > > > It's here, and I think we should move away from it for new DSI sinks. > > > I'd like OF graph to be used consistently for new drivers. We can't > > > change existing DT bindings and drivers to drop support for the > > > non-OF-graph description due to ABI stability, but we can avoid > > > repeating the mistake going forward. > > > > > > > What I'm trying to address here is that some controllers that do > > > > everything right can't be used because that restriction is completely > > > > arbitrary and in opposition to the consensus. And they can't be used > > > > *today*. > > > > > > > > If we want to change that consensus, fine, but we should still have one. > > > > Having some bridges enforcing custom rules for no reason is very much > > > > unacceptable. > > > > > > > > And changing that consensus won't happen overtime, we'll have to take > > > > care of the backward compatibility, etc. So it won't fix the issue that > > > > we can't use any bridge with any controller any time soon. > > > > > > I don't think that's the issue at hand here. You can still use a > > > non-OF-graph DT event if the nodes for the two bridges affected by this > > > patch define a port@0. It can just be left unconnected. > > > > > > I do agree it will cause some DT bindings for DCS-based DSI sinks to > > > have ports will others won't. If your concern is that all DT bindings > > > should be coherent, would you be OK with a patch that makes the sink > > > port mandatory in all DT bindings for DSI bridges and panels (and fixes > > > the mainline DT sources accordingly to make sure they validate) ? The > > > port would not be connected of course (at least when used with DSI > > > source drivers that don't use OF graph today). That would make DT > > > bindings coherent, and would be a first step towards using OF graph > > > everywhere. > > > > I'm trying to fix a (recent) mistake/cargo-cult in new bindings. That > > discussion is not going to be fairly controversial and I don't see how > > that can be solved quickly. So, as a second step, why not. But this one > > needs to come first. > > I feel like I don't quite understand the problem and the discussion. > > What's the problem this fixes? DT validation? Why not just fix the dts files > which use these devices (although I didn't see any in mainline), by adding > the port nodes? > > Or is the argument that we should also support "implicit ports" when the > control and data busses are the same? ... But the discussion really is two-fold. The problem this patch addresses is that there's two odd bridges that have an inconsistent binding with how the rest of the similar devices now. It was a mistake that slipped through review, we should address it. Then, there's the larger discussion of whether or not we want to push the OF-graph for everyone. But that will require some kind of consensus, some work on the MIPI-DSI encoders that don't have that kind of support, in a backward compatible manner, etc. It's not going to happen over-night, especially since it seems like it's something that would be nice to have, but no-one is really interested in actually making it happen. And we shouldn't wait for that to happen to fix the inconsistency mentioned above. Maxime
Attachment:
signature.asc
Description: PGP signature
