On Fri, 8 Dec 2023 at 14:21, Krzysztof Kozlowski
<krzysztof.kozlowski@xxxxxxxxxx> wrote:
>
> On 08/12/2023 13:17, Dmitry Baryshkov wrote:
> >>>>>> Anyway, I was thinking this should be rather argument to phy-cells.
> >>>>> I'm not sure I'm for this, because the results would be:
> >>>>>
> >>>>> --- device.dts ---
> >>>>> &dp_controller0 {
> >>>>> phys = <&dp_phy0 PHY_EDP>;
> >>>>> };
> >>>>>
> >>>>> &dp_controller1 {
> >>>>> phys = <&dp_phy1 PHY_DP>;
> >>>>> };
> >>>>> ------------------
> >>>>>
> >>>>> as opposed to:
> >>>>>
> >>>>> --- device.dts ---
> >>>>> &dp_phy0 {
> >>>>> phy-type <PHY_EDP>;
> >>>>> };
> >>>>>
> >>>>> &dp_phy1 {
> >>>>> phy-type = <PHY_DP>;
> >>>>> };
> >>>>> ------------------
> >>>>
> >>>> Which is exactly what I proposed/wanted to see.
> >>>>
> >>>>>
> >>>>> i.e., we would be saying "this board is connected to this phy
> >>>>> instead" vs "this phy is of this type on this board".
> >>>>>
> >>>>> While none of them really fit the "same hw, different config"
> >>>>> situation, I'd vote for the latter one being closer to the
> >>>>> truth
> >>>>
> >>>> Then maybe I miss the bigger picture, but commit msg clearly says:
> >>>> "multiple PHYs that can work in both eDP or DP mode"
> >>>>
> >>>> If this is not the case, describe the hardware correctly in the commit
> >>>> msg, so people will not ask stupid questions...
> >>>
> >>> There are multiple PHYs (each of them at its own address space). Each
> >>> of the PHYs in question can be used either for the DisplayPort output
> >>> (directly or through the USB-C) or to drive the eDP panel.
> >>>
> >>> Same applies to the displayport-controller. It can either drive the DP
> >>> or eDP output, hardware-wise it is the same.
> >>
> >> Therefore what I proposed was correct - the block which uses the phy
> >> configures its mode. Because this part:
> >> "this phy is of this type on this board".
> >> is not true. The phy is both types.
> >
> > But hopefully you don't mean using #phy-cells here. There are no
> > sub-PHYs or anything like that.
>
> I am exactly talking about phy-cells. Look at first example from Abel's
> code.
I always had an impression that #foo-cells means that there are
different units within the major handler. I.e. #clock-cells mean that
there are several different clocks, #reset-cells mean that there are
several resets, etc.
Ok, maybe this is not a perfect description. We need cells to identify
a particular instance within the major block. Maybe that sounds more
correct.
For the USB+DP PHY we use #phy-cells to select between USB3 and DP
PHYs. But for these PHYs we do not have sub-devices, sub-blocks, etc.
There is a single PHY which works in either of the modes.
Last, but not least, using #phy-cells in this way would create
asymmetry with all the other PHYs (and especially other QMP PHYs)
present on these platforms.
If you feel that phy-type is not an appropriate solution, I'd vote for
not having the type in DT at all, letting the DP controller determine
the proper mode on its own.
--
With best wishes
Dmitry
