Re: [PATCH v5 08/12] dt-bindings: mediatek: Change the binding for mmsys clocks

[Stephen Boyd]

Quoting Matthias Brugger (2018-11-21 09:09:52)
> 
> 
> On 21/11/2018 17:46, Stephen Boyd wrote:
> > Quoting Rob Herring (2018-11-19 11:15:16)
> >> On Sun, Nov 18, 2018 at 11:12 AM Matthias Brugger
> >> <matthias.bgg@xxxxxxxxx> wrote:
> >>> On 11/17/18 12:15 AM, Rob Herring wrote:
> >>>> On Fri, Nov 16, 2018 at 01:54:45PM +0100, matthias.bgg@xxxxxxxxxx wrote:
> >>>>> -    #clock-cells = <1>;
> >>>>> +
> >>>>> +    mmsys_clk: clock-controller@14000000 {
> >>>>> +            compatible = "mediatek,mt2712-mmsys-clk";
> >>>>> +            #clock-cells = <1>;
> >>>>
> >>>> This goes against the general direction of not defining separate nodes
> >>>> for providers with no resources.
> >>>>
> >>>> Why do you need this and what does it buy if you have to continue to
> >>>> support the existing chips?
> >>>>
> >>>
> >>> It would show explicitly that the mmsys block is used to probe two
> >>> drivers, one for the gpu and one for the clocks. Otherwise that is
> >>> hidden in the drm driver code. I think it is cleaner to describe that in
> >>> the device tree.
> >>
> >> No, that's maybe cleaner for the driver implementation in the Linux
> >> kernel. What about other OS's or when Linux drivers and subsystems
> >> needs change? Cleaner for DT is design bindings that reflect the h/w.
> >> Hardware is sometimes just messy.
> >>
> > 
> > I agree. I fail to see what this patch series is doing besides changing
> > driver probe and device creation methods and making a backwards
> > incompatible change to DT. Is there any other benefit here?
> > 
> 
> You are referring whole series?
> Citing the cover letter:
> "MMSYS in Mediatek SoCs has some registers to control clock gates (which is
> used in the clk driver) and some registers to set the routing and enable
> the differnet (sic!) blocks of the display subsystem.
> 
> Up to now both drivers, clock and drm are probed with the same device tree
> compatible. But only the first driver get probed, which in effect breaks
> graphics on mt8173 and mt2701.

Ouch!

> 
> This patch uses a platform device registration in the DRM driver, which
> will trigger the probe of the corresponding clock driver. It was tested on the
> bananapi-r2 and the Acer R13 Chromebook."

Alright, please don't add nodes in DT just to make device drivers probe.
Instead, register clks from the drm driver or create a child platform
device for the clk bits purely in the drm driver and have that probe the
associated clk driver from there.

> 
> DT is broken right now, because two drivers rely on the same node, which gets
> consumed just once. The new DT introduced does not break anything because it is
> only used for boards that: "[..] are not available to the general public
> (mt2712e) or only have the mmsys clock driver part implemented (mt6797)."

Ok, so backwards compatibility is irrelevant then. Sounds fine to me.