Re: [PATCH v5 2/3] dt-bindings: net: mscc-vsc8531: add optional clock properties

[Heiko Stuebner <heiko@sntech.de>]

Am Freitag, 19. Juni 2020, 07:01:58 CEST schrieb Florian Fainelli:
> 
> On 6/18/2020 5:11 AM, Heiko Stuebner wrote:
> > From: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
> > 
> > Some mscc ethernet phys have a configurable clock output, so describe the
> > generic properties to access them in devicetrees.
> > 
> > Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
> > ---
> >  Documentation/devicetree/bindings/net/mscc-phy-vsc8531.txt | 2 ++
> >  1 file changed, 2 insertions(+)
> > 
> > diff --git a/Documentation/devicetree/bindings/net/mscc-phy-vsc8531.txt b/Documentation/devicetree/bindings/net/mscc-phy-vsc8531.txt
> > index 5ff37c68c941..67625ba27f53 100644
> > --- a/Documentation/devicetree/bindings/net/mscc-phy-vsc8531.txt
> > +++ b/Documentation/devicetree/bindings/net/mscc-phy-vsc8531.txt
> > @@ -1,6 +1,8 @@
> >  * Microsemi - vsc8531 Giga bit ethernet phy
> >  
> >  Optional properties:
> > +- clock-output-names	: Name for the exposed clock output
> > +- #clock-cells		: should be 0
> >  - vsc8531,vddmac	: The vddmac in mV. Allowed values is listed
> >  			  in the first row of Table 1 (below).
> >  			  This property is only used in combination
> > 
> 
> With that approach, you also need to be careful as a driver writer to
> ensure that you have at least probed the MDIO bus to ensure that the PHY
> device has been created (and therefore it is available as a clock
> provider) if that same Ethernet MAC is a consumer of that clock (which
> it appears to be). Otherwise you may just never probe and be trapped in
> a circular dependency.

Yep - although without anything like this, the phy won't emit any clock
at all. Even when enabling the clock output in u-boot already, when the
kernel starts that config is lost,  so no existing board should break.


As you can see in the discussion about patch 3/3 the wanted solution
is not so clear cut as well. With Rob suggesting this clock-provider way
and Russell strongly encouraging taking a second look.

[My first iteration (till v4) was doing it like other phys by specifying
a property to just tell the phy what frequency to output]

I don't really have a preference for one or the other, so
maybe you can also give a vote over there ;-)

Heiko