Hi Rob,

On 1/25/21 8:47 PM, Rob Herring wrote:

On Mon, Jan 25, 2021 at 6:16 PM Suman Anna [off-list ref] wrote:

quoted

Hi Rob,

On 1/25/21 6:04 PM, Rob Herring wrote:

quoted

On Fri, Jan 15, 2021 at 02:58:19PM -0600, Suman Anna wrote:

quoted

The '#address-cells' property looks to be a required property for
interrupt controller nodes as indicated by a warning message seen
when building dtbs with W=2. Adding the property to the PRUSS INTC
dts nodes though fails the dtbs_check. Add this property to the
PRUSS INTC binding to make it compliant with both dtbs_check and
building dtbs.

Signed-off-by: Suman Anna <redacted>
---
Hi Rob,

This patch is also part of our effort to get rid of the warnings seen
around interrupt providers on TI K3 dtbs [1]. I needed this in the PRUSS
INTC bindings to not get a warning with dtbs_check while also ensuring
no warnings while building dtbs with W=2.

I would have expected the '#address-cells' requirement to be inherited
automatically. And looking through the schema files, I actually do not
see the interrupt-controller.yaml included automatically anywhere. You
had asked us to drop the inclusion in this binding in our first version
with YAML [3]. Am I missing something, and how do we ensure that this
is enforced automatically for everyone?

interrupt-controller.yaml is applied to any node named
'interrupt-controller'. More generally, if 'compatible' is not present,
then we look at $nodename for the default 'select'. In your case, you
didn't name the node appropriately.

Thanks for the clarification. Yeah, I didn't add anything specifically, since
the expectation is interrupt-controller. Should I be adding that to this binding?

No, either interrupt-controller.yaml needs to learn a new node name or
your node names need to be fixed. I prefer the latter, but if you have
more than 1 and don't have a unit-address (and in turn a 'reg' prop)
we'd have to do the former. How are the interrupts controllers
accessed if there's no way to address them?

The PRUSS INTC will always have a unit-address, so we won't have the issues with
having to maintain unique names. All my examples already have the nodes in the
form 'interrupt-controller@<addr>'. Anyway, I will drop this patch, and post a
new patch adding the $nodename to the binding.

quoted

quoted

We can't check this in interrupt-controller.yaml because #address-cells
is not always 0. GICv3 is one notable exception.

quoted

regards
Suman

[1] https://patchwork.kernel.org/project/linux-arm-kernel/patch/20210115083003.27387-1-lokeshvutla@ti.com/

I've commented on this thread now in regards to #address-cells.

I suppose I still need this patch to be defined to unblock the ICSSG nodes
getting accepted by our dts maintainer. Care to give your Reviewed-by for the
change? Or I can spin a v2 with $nodename added as well if that's needed too.

No, I don't think you have to add #address-cells. We need to fix the
warning in dtc.

Thank you for clarifying this.

regards
Suman