Re: [PATCH 2/3] dt-bindings: usb: mtk-xhci: Allow middle optional clocks to be missing

From: Nícolas F. R. A. Prado
Date: Mon Jun 20 2022 - 11:51:14 EST


On Mon, Jun 20, 2022 at 10:50:57AM +0200, Krzysztof Kozlowski wrote:
> On 20/06/2022 08:59, Chunfeng Yun wrote:
> > On Sun, 2022-06-19 at 14:05 +0200, Krzysztof Kozlowski wrote:
> >> On 19/06/2022 09:46, Chunfeng Yun wrote:
> >>> On Fri, 2022-06-17 at 18:25 -0700, Krzysztof Kozlowski wrote:
> >>>> On 17/06/2022 15:29, Nícolas F. R. A. Prado wrote:
> >>>>> The current clock list in the binding doesn't allow for one of
> >>>>> the
> >>>>> optional clocks to be missing and a subsequent clock to be
> >>>>> present.
> >>>>> An
> >>>>> example where this is an issue is in mt8192.dtsi, which has
> >>>>> "sys_ck",
> >>>>> "ref_ck", "xhci_ck" and would cause dtbs_check warnings.
> >>>>>
> >>>>> Change the clock list in a way that allows the middle optional
> >>>>> clocks to
> >>>>> be missing, while still guaranteeing a fixed order. The
> >>>>> "ref_ck" is
> >>>>> kept
> >>>>> as a const even though it is optional for simplicity, since it
> >>>>> is
> >>>>> present in all current dts files.
> >>>>>
> >>>>> Signed-off-by: Nícolas F. R. A. Prado <nfraprado@xxxxxxxxxxxxx>
> >>>>> ---
> >>>>>
> >>>>> .../devicetree/bindings/usb/mediatek,mtk-xhci.yaml | 9
> >>>>> +++++++--
> >>>>> 1 file changed, 7 insertions(+), 2 deletions(-)
> >>>>>
> >>>>> diff --git
> >>>>> a/Documentation/devicetree/bindings/usb/mediatek,mtk-
> >>>>> xhci.yaml b/Documentation/devicetree/bindings/usb/mediatek,mtk-
> >>>>> xhci.yaml
> >>>>> index 63cbc2b62d18..99a1b233ec90 100644
> >>>>> --- a/Documentation/devicetree/bindings/usb/mediatek,mtk-
> >>>>> xhci.yaml
> >>>>> +++ b/Documentation/devicetree/bindings/usb/mediatek,mtk-
> >>>>> xhci.yaml
> >>>>> @@ -80,8 +80,13 @@ properties:
> >>>>> items:
> >>>>> - const: sys_ck # required, the following ones are
> >>>>> optional
> >>>>> - const: ref_ck
> >>>>> - - const: mcu_ck
> >>>>> - - const: dma_ck
> >>>>> + - enum:
> >>>>> + - mcu_ck
> >>>>> + - dma_ck
> >>>>> + - xhci_ck
> >>>>> + - enum:
> >>>>> + - dma_ck
> >>>>> + - xhci_ck
> >>>>> - const: xhci_ck
> >>>>
> >>>> You allow now almost any order here, including incorrect like
> >>>> sys,ref,xhci,xhci,xhci.
> >>>>
> >>>> The order of clocks has to be fixed and we cannot allow
> >>>> flexibility.
> >>>> Are
> >>>> you sure that these clocks are actually optional (not wired to
> >>>> the
> >>>> device)?
> >>>
> >>> In fact, these optional clocks are fixed, due to no gates are
> >>> provided,
> >>> SW can't control them by CCF;
> >>> In this case, I usually use a fixed clock, or ignore it.
> >>
> >> But in some versions these clocks are controllable or not?
> > Some SoCs are controllable, some ones are not (fixed clock).
>
> Thanks for confirming. Then I would prefer to make these clocks required
> (not optional) and always provide them - via common clock framework or
> fixed-clock.

Hi Krzysztof and Chunfeng,

thank you both for the feedback.

Since the solution I proposed in this patch is not acceptable I see two options:
1. Split the clocks in several if blocks matched by compatibles
2. Make the clocks required and use fixed-clock nodes for the missing clocks in
the DT

My understanding is that 1 is the desirable solution if the clock is really
missing in some hardware variants, while 2 is desirable if all hardware variants
really receive all the clocks, only that on some variants they're fixed and not
controlable by SW.

>From what I'm reading of this discussion it seems that the latter is the case
here and thus we should go for 2. Is this correct?

Also Chunfeng, do you have information on whether the same is true for the MMC
HW block? I recently submitted some changes to that binding [1] but I followed
approach 1 there instead. However if all the clocks are present in the HW level
there as well it would make more sense for me to change it to follow approach 2.

Thanks,
Nícolas

[1] https://lore.kernel.org/all/20220617230114.2438875-1-nfraprado@xxxxxxxxxxxxx