Hi Maxime
On Sat, 14 Dec 2019 at 20:26, Maxime Ripard <maxime@xxxxxxxxxx> wrote:
>
> On Fri, Dec 13, 2019 at 03:04:35PM +0000, Mike Leach wrote:
> > > > + type: object
> > > > + description:
> > > > + A trigger connections child node which describes the trigger signals
> > > > + between this CTI and another hardware device. This device may be a CPU,
> > > > + CoreSight device, any other hardware device or simple external IO lines.
> > > > + The connection may have both input and output triggers, or only one or the
> > > > + other.
> > > > +
> > > > + properties:
> > > > +
> > > > + arm,trig-in-sigs:
> > > > + allOf:
> > > > + - $ref: /schemas/types.yaml#/definitions/uint32-array
> > > > + minItems: 1
> > > > + maxItems: 32
> > > > + description:
> > > > + List of CTI trigger in signal numbers in use by a trig-conns node.
> > > > +
> > > > + arm,trig-in-types:
> > > > + allOf:
> > > > + - $ref: /schemas/types.yaml#/definitions/uint32-array
> > > > + minItems: 1
> > > > + maxItems: 32
> > > > + description:
> > > > + List of constants representing the types for the CTI trigger in
> > > > + signals. Types in this array match to the corresponding signal in the
> > > > + arm,trig-in-sigs array. If the -types array is smaller, or omitted
> > > > + completely, then the types will default to GEN_IO.
> > > > +
> > > > + arm,trig-out-sigs:
> > > > + allOf:
> > > > + - $ref: /schemas/types.yaml#/definitions/uint32-array
> > > > + minItems: 1
> > > > + maxItems: 32
> > > > + description:
> > > > + List of CTI trigger out signal numbers in use by a trig-conns node.
> > > > +
> > > > + arm,trig-out-types:
> > > > + allOf:
> > > > + - $ref: /schemas/types.yaml#/definitions/uint32-array
> > > > + minItems: 1
> > > > + maxItems: 32
> > > > + description:
> > > > + List of constants representing the types for the CTI trigger out
> > > > + signals. Types in this array match to the corresponding signal
> > > > + in the arm,trig-out-sigs array. If the "-types" array is smaller,
> > > > + or omitted completely, then the types will default to GEN_IO.
> > > > +
> > > > + arm,trig-filters:
> > > > + allOf:
> > > > + - $ref: /schemas/types.yaml#/definitions/uint32-array
> > > > + minItems: 1
> > > > + maxItems: 32
> > > > + description:
> > > > + List of CTI trigger out signals that will be blocked from becoming
> > > > + active, unless filtering is disabled on the driver.
> > > > +
> > > > + arm,trig-conn-name:
> > > > + allOf:
> > > > + - $ref: /schemas/types.yaml#/definitions/string
> > > > + description:
> > > > + Defines a connection name that will be displayed, if the cpu or
> > > > + arm,cs-dev-assoc properties are not being used in this connection.
> > > > + Principle use for CTI that are connected to non-CoreSight devices, or
> > > > + external IO.
> > > > +
> > > > + anyOf:
> > > > + - required:
> > > > + - arm,trig-in-sigs
> > > > + - required:
> > > > + - arm,trig-out-sigs
> > > > + oneOf:
> > > > + - required:
> > > > + - arm,trig-conn-name
> > > > + - required:
> > > > + - cpu
> > > > + - required:
> > > > + - arm,cs-dev-assoc
> > >
> > > This would mean that either arm,trig-conn-name, cpu xor
> > > arm,cs-dev-assoc is required in the trig_conn child nodes, but they
> > > don't seem to be defined at this level but in the parent node?
> > >
> >
> > cpu and rm,cs-dev-assoc can appear in the parent node if the
> > arm,coresight-cti-v8-arch compatible string is used - hence declared
> > at that level. See the examples for the v8 compatible layout.
> >
> > > > +
> > > > +required:
> > > > + - compatible
> > > > + - reg
> > > > + - clocks
> > > > + - clock-names
> > > > +
> > > > +examples:
> > > > + # minimum CTI definition. DEVID register used to set number of triggers.
> > > > + - |
> > > > + cti@20020000 {
> > > > + compatible = "arm,coresight-cti", "arm,primecell";
> > > > + reg = <0x20020000 0x1000>;
> > > > +
> > > > + clocks = <&soc_smc50mhz>;
> > > > + clock-names = "apb_pclk";
> > > > + };
> > > > + # v8 architecturally defined CTI - CPU + ETM connections generated by the
> > > > + # driver according to the v8 architecture specification.
> > > > + - |
> > > > + cti@859000 {
> > > > + compatible = "arm,coresight-cti-v8-arch", "arm,coresight-cti",
> > > > + "arm,primecell";
> > > > + reg = <0x859000 0x1000>;
> > > > +
> > > > + clocks = <&soc_smc50mhz>;
> > > > + clock-names = "apb_pclk";
> > > > +
> > > > + cpu = <&CPU1>;
> > > > + arm,cs-dev-assoc = <&etm1>;
> > >
> > > and it looks like arm,cs-dev-assoc and cpu can be combined?
> > >
> > Absolutely - a CTI can and is connected to both the CPU and an
> > optional ETM attached to the CPU in a v8 architecture system.
>
> That's not what
>
> > > > + oneOf:
> > > > + - required:
> > > > + - arm,trig-conn-name
> > > > + - required:
> > > > + - cpu
> > > > + - required:
> > > > + - arm,cs-dev-assoc
>
> Is saying though. oneOf is a xor, you can have one of the three
> schemas that are valid, but not more than that.
>
This required schema only applies to the trig-conns@ child nodes. So
each trig-conns can have one only of the three attributes - as each
trig-conns node represents a single connection between the CTI and a
component, and defines the trigger signals that exist in that
connections.
At the cti@ level, if the compatible is the arm,coresight-cti-v8-arch
type, then both cpu and arm,cs-dev-assoc can appear at this level as
we only need to know the devices it is connected to - the individual
trigger signals in the connections are defined by the architecture and
do not need repeating here.
> > > > + };
> > > > + # Implementation defined CTI - CPU + ETM connections explicitly defined..
> > > > + # Shows use of type constants from dt-bindings/arm/coresight-cti-dt.h
> > > > + - |
> > > > + #include <dt-bindings/arm/coresight-cti-dt.h>
> > > > +
> > > > + cti@858000 {
> > > > + compatible = "arm,coresight-cti", "arm,primecell";
> > > > + reg = <0x858000 0x1000>;
> > > > +
> > > > + clocks = <&soc_smc50mhz>;
> > > > + clock-names = "apb_pclk";
> > > > +
> > > > + arm,cti-ctm-id = <1>;
> > > > +
> > > > + trig-conns@0 {
> > >
> > > So, what I think happened here is that your patternProperties doesn't
> > > match anything (underscore vs dash), and since you don't have
> > > additionalProperties set to false, it doesn't warn that there's
> > > properties that aren't defined in the binding. Meaning that none of
> > > the child nodes in the bindings are effectively validated.
> > >
> >
> > OK - fixing the name should fix this.
> > I can't use additionalProperties as this is prohibited for bindings
> > that use ref: (according to the example-schema.yaml)
>
> Ah right, I missed that one, sorry. In this case, you can use the
> keyword unevaluatedProperties instead. As its name suggests, it will
> report an error if there's a property that hasn't been validated by
> any schemas.
>
> Note that this is a keyword introduced by the latest schemas spec
> (draft 2019-09) which isn't supported by the DT tools yet. But putting
> it into your schema will at least allow to have that check when the
> tools will support it.
>
> > > > + arm,trig-in-sigs = <4 5 6 7>;
> > > > + arm,trig-in-types = <ETM_EXTOUT
> > > > + ETM_EXTOUT
> > > > + ETM_EXTOUT
> > > > + ETM_EXTOUT>;
> > > > + arm,trig-out-sigs = <4 5 6 7>;
> > > > + arm,trig-out-types = <ETM_EXTIN
> > > > + ETM_EXTIN
> > > > + ETM_EXTIN
> > > > + ETM_EXTIN>;
> > > > + arm,cs-dev-assoc = <&etm0>;
> > > > + };
> > >
> > > Nodes with unit-address must have a matching reg property. This will
> > > generate a dtc warning too.
> >
> > That's interesting - I don't get any dtc warnings when compiling or
> > when running make dt_binding_checl
>
> Linux disables a lot of DTC warnings. You can see all (I think?) the
> warnings by passing W=1 in the environment when you compile the device
> trees.
>
Thanks - the W=12 starts outputting warnings for lack of reg in child nodes.
I'll build the requirement into the next version so it is explicitly
called out in the check.
> > Is this rule documented anywhere? I cannot see it in the 0.2 spec
> > version from devicetree.org or any of the examples / docs in the
> > kernel tree.
>
> The commit adding it to DTC is this one
> https://git.kernel.org/pub/scm/utils/dtc/dtc.git/commit/?id=852e9ecbe1976927057402f8a8f71ee8e8a49098
I was referring to the requirement for reg = <N> in node@<N> child
nodes - though I have now spotted this in the early part of the
0.3-rc2 spec - which appears to be the latest I have access to.
So I can add in the reg<> parameters and fix this up.
>
> It just seems that while valid, it's against best practices.
>
> Maxime
Thanks
Mike
--
Mike Leach
Principal Engineer, ARM Ltd.
Manchester Design Centre. UK
