Hi Rob,
On Thu, Jun 13, 2019 at 11:32:30AM -0600, Rob Herring wrote:
> On Thu, Jun 13, 2019 at 7:25 AM Maxime Ripard <maxime.ripard@xxxxxxxxxxx> wrote:
> > On Mon, Jun 10, 2019 at 12:59:29PM -0600, Rob Herring wrote:
> > > On Mon, Jun 10, 2019 at 8:31 AM Andrew Lunn <andrew@xxxxxxx> wrote:
> > > >
> > > > > +required:
> > > > > + - compatible
> > > > > + - reg
> > > > > + - interrupts
> > > > > + - clocks
> > > > > + - phy
> > > > > + - allwinner,sram
> > > >
> > > > Quoting ethernet.txt:
> > > >
> > > > - phy: the same as "phy-handle" property, not recommended for new bindings.
> > > >
> > > > - phy-handle: phandle, specifies a reference to a node representing a PHY
> > > > device; this property is described in the Devicetree Specification and so
> > > > preferred;
> > > >
> > > > Can this be expressed in Yaml? Accept phy, but give a warning. Accept
> > > > phy-handle without a warning? Enforce that one or the other is
> > > > present?
> > >
> > > The common schema could have 'phy: false'. This works as long as we've
> > > updated (or plan to) all the dts files to use phy-handle. The issue is
> > > how far back do you need kernels to work with newer dtbs.
> >
> > I guess another question being raised by this is how hard do we want
> > to be a deprecating things, and should the DT validation be a tool to
> > enforce that validation.
> >
> > For example, you've used in you GPIO meta-schema false for anything
> > ending with -gpio, since it's deprecated. This means that we can't
> > convert any binding using a deprecated property without introducing a
> > build error in the schemas, which in turn means that you'll have a lot
> > of friction to support schemas, since you would have to convert your
> > driver to support the new way of doing things, before being able to
> > have a schema for your binding.
>
> I've err'ed on the stricter side. We may need to back off on some
> things to get to warning free builds. Really, I'd like to have levels
> to separate checks for existing bindings, new bindings, and pedantic
> checks.
That would be awesome. Do you have a plan for that already though? I
can't really think of a way to implement it at the moment.
> For '-gpio', we may be okay because the suffix is handled in the GPIO
> core. It should be safe to update the binding to use the preferred
> form.
It might require a bit of work though in drivers, since the fallback
is only handled if you're using the gpiod API, and not the legacy one.
> > And then, we need to agree on how to express the deprecation. I guess
> > we could allow the deprecated keyword that will be there in the
> > draft-8, instead of ad-hoc solutions?
>
> Oh, nice! I hadn't seen that. Seems like we should use that. We can
> start even without draft-8 support because unknown keywords are
> ignored (though we probably have to add it to our meta-schema). Then
> at some point we can add a 'disallow deprecated' flag to the tool.
So, in the generic ethernet binding, we would have:
properties:
phy-handle:
$ref: /schemas/types.yaml#definitions/phandle
description:
Specifies a reference to a node representing a PHY device.
phy:
$ref: "#/properties/phy-handle"
deprecated: true
phy-device:
$ref: "#/properties/phy-handle"
deprecated: true
Does that sound good?
Now, how do we handle the case above, in the device specific binding?
We just require the non-deprecated one, or the three?
Thanks!
Maxime
--
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
Attachment:
signature.asc
Description: PGP signature
