Re: [PATCH 3/4] dt-bindings: reset: uniphier: Add AHCI core reset description

[Kunihiko Hayashi]

Hi Philipp,

On Mon, 12 Nov 2018 15:21:46 +0100 <p.zabel@xxxxxxxxxxxxxx> wrote:

> Hi,
> 
> On Mon, 2018-11-12 at 21:02 +0900, Kunihiko Hayashi wrote:
> > Hi,
> > 
> > Thank you for some comments and pointing out.
> > 
> > On Sat, 10 Nov 2018 01:14:06 +0900 <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> > 
> > > On Sat, Nov 10, 2018 at 12:02 AM Philipp Zabel <p.zabel@xxxxxxxxxxxxxx> wrote:
> > > > 
> > > > Hi Kunihiko,
> > > > 
> > > > On Fri, 2018-11-09 at 10:42 +0900, Kunihiko Hayashi wrote:
> > > > > Add compatible strings for reset control of AHCI core implemented in
> > > > > UniPhier SoCs. The reset control belongs to AHCI glue layer.
> > > > > 
> > > > > Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@xxxxxxxxxxxxx>
> > > > > ---
> > > > >  Documentation/devicetree/bindings/reset/uniphier-reset.txt | 3 +++
> > > > >  1 file changed, 3 insertions(+)
> > > > > 
> > > > > diff --git a/Documentation/devicetree/bindings/reset/uniphier-reset.txt b/Documentation/devicetree/bindings/reset/uniphier-reset.txt
> > > > > index f63c511..ea00517 100644
> > > > > --- a/Documentation/devicetree/bindings/reset/uniphier-reset.txt
> > > > > +++ b/Documentation/devicetree/bindings/reset/uniphier-reset.txt
> > > > > @@ -133,6 +133,9 @@ Required properties:
> > > > >      "socionext,uniphier-pxs2-usb3-reset" - for PXs2 SoC USB3
> > > > >      "socionext,uniphier-ld20-usb3-reset" - for LD20 SoC USB3
> > > > >      "socionext,uniphier-pxs3-usb3-reset" - for PXs3 SoC USB3
> > > > > +    "socionext,uniphier-pro4-ahci-reset" - for Pro4 SoC AHCI
> > > > > +    "socionext,uniphier-pxs2-ahci-reset" - for PXs2 SoC AHCI
> > > > > +    "socionext,uniphier-pxs3-ahci-reset" - for PXs3 SoC AHCI
> > > > 
> > > > Since the driver behaves identically for "socionext,uniphier-pro4-usb3-
> > > > reset" and "socionext,uniphier-pro4-ahci-reset", would it make sense to
> > > > add a common compatible?
> > > 
> > > As far as I could guess, he just happened to find the same driver code
> > > could be reused for other hardware.
> 
> Ok, in that case never mind. I just assumed that this could be a case of
> glue layer building blocks being reused by the hardware engineers, since
> the driver reuses the same clock names for USB3 and AHCI both on Pro4
> and on PXs2.
> 
> > > Theoretically, this can happen anywhere since
> > > a reset controller is just a set of registers
> > > each bit of which is connected to a reset line.
> > > 
> > > If you added a super-generic compatible like "simple-reset",
> > > I would agree with
> > > "socionext,uniphier-pro4-usb3-reset", "simple-reset"
> > > since this is a pattern.
> > 
> > I think it's more generic to define simple-reset with parent clock/reset
> > control without both SoC and device names.
> > 
> > However, such parent clocks/resets strongly depends on SoC, 
> > so I think it's difficut to lead generic definition in this case.
> > 
> > If we add generic compatible string, I also add "simple-reset".
> 
> There is no "simple-reset" binding definition. As soon as there are SoC
> specific clocks, it's not really generic anymore.

I see. I understand "simple-reset" doesn't need.

> > > However,
> > > "socionext,uniphier-pro4-glue-reset" is kind of a halfway house
> > > where it is SoC-specific, but still ambiguous.
> > 
> > Surely, it might be hard to understand that pro4-glue-reset is SoC-specific
> > but for generic-device.
> 
> I agree.
> 
> > > > Something like:
> > > > "socionext,uniphier-pro4-usb3-reset", "socionext,uniphier-pro4-glue-reset" - for USB3 SoC AHCI
> > > > "socionext,uniphier-pro4-ahci-reset", "socionext,uniphier-pro4-glue-reset" - for Pro4 SoC AHCI
> > > > 
> > > > That way if more places turn up where the glue layer reset is used,
> > > > you can add them without patching the driver every time.
> > > 
> > > 
> > > This is a trade-off between "patch the driver"
> > > and "potential change of the binding".
> > 
> > Adding "glue-reset" is usable for devices having same parent clocks/resets as
> > usb3/ahci, and we can add the devices without patches.
> > 
> > However, if the device needs other parent clocks/resets for the same SoC,
> > we can't add "glue-reset" and we might add patches for the device as a result.
> > In this case, the "glue-reset" will become difficult to understand.
> 
> Then let's not bother with it.
> I made the suggestion without knowing the full picture.

This might confuse you without bigger picture.

> > > There is no real hardware like pro4-glue-reset.
> > > 
> > > I am guessing this is a part of syscon or something,
> > > but I cannot find any explanation in a bigger picture.
> > > 
> > > So, I cannot judge this further more.
> > 
> > Since it's hard to lead the best result,
> > currently I'd like to suggest the current compatibles,
> 
> That is fine with me.
> 
> > with "simple-reset" if necessary.
> 
> Not necessary.

Okay, I don't add it.

Thank you,

---
Best Regards,
Kunihiko Hayashi