Hi Rob
Cc Geert
> > Sorry to bother you but I wonder if you could help me understand why the
> > dtc checker warns for the issues I tried to work around in this patch,
> > and if possible how I can improve my solution to get rid of the
> > warnings.
> >
> > This patch addresses the same problem for a few different devices. I
> > will focus on the last one (/soc/isp@fed00000/ports/port@0) for my
> > question, but all warnings here have the same issue.
> >
> > I have a port node the represents a sink for a video input. This sink
> > can either be connected to source A or source B, but not both at the
> > same time. So each possible source is represented by an endpoint in the
> > port node. Each endpoint have specific register address on the port bus
> > that is described in the bindings as they map to different physical pins
> > on the hardware.
> >
> > The issue I have is that not all hardware configurations have both
> > source A and B populated. All combinations of A, B and C are possible
> > depending on the platform.
> >
> > A)
> > ports {
> > ...
> > port@0 {
> > ...
> > sourceA: endpoint@0 {
> > reg = <0>
> > };
> > sourceB: endpoint@1 {
> > reg = <1>
> > };
> > };
> > };
> >
> > B)
> > ports {
> > ...
> > port@0 {
> > ...
> > sourceA: endpoint@0 {
> > reg = <0>
> > };
> > };
> > };
> >
> > C)
> > ports {
> > ...
> > port@0 {
> > ...
> > sourceB: endpoint@1 {
> > reg = <1>
> > };
> > };
> > };
> >
> > For option A and C the checker is happy, but for option B the warnings
> > this patch tries to address are triggered. While reading the
> > dtc/checks.c I find check_graph_child_address() is the one that is
> > triggering the warning. And this function explicitly checks for port
> > buses with a single endpoint with a register value of 0.
> >
> > This check was added way back in 2018 in commit df536831d02c ("checks:
> > add graph binding checks"). But I can't find any information on the
> > specifics. Is this design a bad idea for port buses for some reason I
> > don't understand? AFIU this design is possible on other type of buses?
> > And do you have any guidance on how I can dig myself out of this hole?
>
> Don't.
>
> The check is only with W=1. It is for cases where there is never more
> than 1 endpoint. dtc can't distinguish when that is the case, so there's
> going to be cases to ignore. Perhaps we could demote it W=2, but I'd
> prefer not to. Making W=1 warning free may be a platform goal, but
> that's not an overall go. If we fix something everywhere, then the check
> is promoted.
I get this warning, but I'm not using W=1 option.
I'm using v6.13-rcX, but which kernel version hide it in normal compile
(= without W=1) ?
${linux}/arch/arm64/boot/dts/renesas/r8a779a0.dtsi:1189.12-1199.7:
Warning (graph_child_address): /soc/video@e6ef0000/ports/port@2:
graph node has single child node 'endpoint@0', #address-cells/#size-cells
are not necessary
Thank you for your help !!
Best regards
---
Kuninori Morimoto
