Re: [PATCH] dt-bindings: pci: brcmstb: Add rp1-nexus node to fix DTC warning

[+cc Herve]

On 18:58 Tue 09 Dec     , Andrea della Porta wrote:

Hi Rob,

On 11:09 Thu 04 Dec     , Rob Herring wrote:

quoted

On Mon, Sep 1, 2025 at 3:48 AM Andrea della Porta [off-list ref] wrote:

quoted

Hi Krzysztof,

On 08:50 Fri 22 Aug     , Krzysztof Kozlowski wrote:

quoted

On 21/08/2025 17:22, Andrea della Porta wrote:

quoted

Hi Krzysztof,

On 10:55 Tue 12 Aug     , Krzysztof Kozlowski wrote:

quoted

On 12/08/2025 10:50, Andrea della Porta wrote:

quoted

The devicetree compiler is complaining as follows:

arch/arm64/boot/dts/broadcom/rp1-nexus.dtsi:3.11-14.3: Warning (unit_address_vs_reg): /axi/pcie@1000120000/rp1_nexus: node has a reg or ranges property, but no unit name
/home/andrea/linux-torvalds/arch/arm64/boot/dts/broadcom/bcm2712-rpi-5-b.dtb: pcie@1000120000: Unevaluated properties are not allowed ('rp1_nexus' was unexpected)

Please trim the paths.

Ack.

quoted

quoted

Add the optional node that fix this to the DT binding.

Reported-by: kernel test robot <redacted>
Closes: https://lore.kernel.org/oe-kbuild-all/202506041952.baJDYBT4-lkp@intel.com/ (local)
Signed-off-by: Andrea della Porta <andrea.porta@suse.com>
---
 Documentation/devicetree/bindings/pci/brcm,stb-pcie.yaml | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/Documentation/devicetree/bindings/pci/brcm,stb-pcie.yaml b/Documentation/devicetree/bindings/pci/brcm,stb-pcie.yaml
index 812ef5957cfc..7d8ba920b652 100644
--- a/Documentation/devicetree/bindings/pci/brcm,stb-pcie.yaml
+++ b/Documentation/devicetree/bindings/pci/brcm,stb-pcie.yaml

@@ -126,6 +126,15 @@ required:
 allOf:
   - $ref: /schemas/pci/pci-host-bridge.yaml#
   - $ref: /schemas/interrupt-controller/msi-controller.yaml#
+  - if:
+      properties:
+        compatible:
+          contains:
+            const: brcm,bcm2712-pcie
+    then:
+      properties:
+        rp1_nexus:

No, you cannot document post-factum... This does not follow DTS coding
style.

I think I didn't catch what you mean here: would that mean that
we cannot resolve that warning since we cannot add anything to the
binding?

I meant, you cannot use a warning from the code you recently introduced
as a reason to use incorrect style.

Fixing warning is of course fine and correct, but for the code recently
introduced and which bypassed ABI review it is basically like new review
of new ABI.

This needs standard review practice, so you need to document WHY you
need such node. Warning is not the reason here why you are doing. If
this was part of original patchset, like it should have been, you would
not use some imaginary warning as reason, right?

So provide reason why you need here this dedicated child, what is that
child representing.

Ack.

quoted

Otherwise I can suggest: drop the child and DTSO, this also solves the
warning...

This would not fix the issue: it's the non overlay that needs the specific
node. But I got the point, and we have a solution for that (see below).

quoted

quoted

Regarding rp1_nexus, you're right I guess it should be
rp1-nexus as per DTS coding style.

quoted

Also:

Node names should be generic. See also an explanation and list of
examples (not exhaustive) in DT specification:
https://devicetree-specification.readthedocs.io/en/latest/chapter2-devicetree-basics.html#generic-names-recommendation

In this case it could be difficult: we need to search for a DT node

Search like in driver? That's wrong, you should be searching by compatible.

Thanks for the hint. Searching by compatble is the solution.

No, it is not.

This is partly true, indeed. On one side there's the need to avoid a
specific node name ('rp1_nexus'), so the only other unique identifier would
be the compatible string ('pci1de4,1' in this case, which identifies that specific
device). Unfortunately, the same compatible string is also assigned to the pci
endpoint node filled automatically by enabling CONFIG_PCI_DYNAMIC_OF_NODES.
We would end up with two nodes with the same compatible, which is not unique
anymore. 
This applies only when using 'full' dtb (bcm2712-rpi-5-b.dtb) *and* you enable
CONFIG_PCI_DYNAMIC_OF_NODES, the latter being not necessary since the overlay dtb
(...-ovl-rp1.dtb) is not in use here. To overcome this problem, the solutions
I can think of are the following:

1- Just disable CONFIG_PCI_DYNAMIC_OF_NODES should work, but only when using the
   full dtb version. However, if the user enable that option for debug or to use
   the overlay dtb version, he better be sure not to use teh full dtb or it won't
   work.
   This solution seems really weak.

2- Add another compatible string other than 'pci1de4,1', so it will be really
   unique.

quoted

quoted

quoted

quoted

starting from the DT root and using generic names like pci@0,0 or
dev@0,0 could possibly led to conflicts with other peripherals.
That's why I chose a specific name.

Dunno, depends what can be there, but you do not get a specific
(non-generic) device node name for a generic PCI device or endpoint.

I would use 'port' instead of rp1-nexus. Would it work for you?

Do you still plan to fix this? This is broken far worse than just the node name.

Yes, if we want to get rid of that nasty warning and comply with DT guidelines,
I think I really need to fix that.

quoted

The 'rp1_nexus' node is applied to the PCI host bridge. That's wrong
unless this is PCI rather than PCIe. There's the root port device in
between.

The clue that things are wrong are start in the driver here:

rp1_node = of_find_node_by_name(NULL, "rp1_nexus");
if (!rp1_node) {
  rp1_node = dev_of_node(dev);
  skip_ovl = false;
}

You should not need to do this nor care what the node name is. The PCI
core should have populated pdev->dev.of_node for you. If not, your DT
is wrong. Turn on CONFIG_PCI_DYNAMIC_OF_NODES and look at the
resulting PCI nodes. They should also match what the hierarchy looks
like with lspci. I don't recommend you rely on
CONFIG_PCI_DYNAMIC_OF_NODES, but statically populate the nodes in the
DT. First, CONFIG_PCI_DYNAMIC_OF_NODES is an under development thing
and I hope to get rid of the config option. Second, your case is
static (i.e. not a PCIe card in a slot) so there is no issue
hardcoding the DT.

Are you planning to get rid of the CONFIG_PCI_DYNAMIC_OF_NODES features?
There could be other drivers relying on that besides RP1 in overlay mode,
e.g. LAN966x for instance.

Thanks,
Andrea

The 'full' dtb (bcm2712-rpi-5-b.dtb) is indeed statically populated.
CONFIG_PCI_DYNAMIC_OF_NODES is needed only if you use the overlay approach
(bcm2712-rpi-5-b-ovl-rp1.dtb) and in that case the node will be
added to the correct device node at runtime, and there won't be any node
labeled as rp1_nexus.
That conditional just check for teh presence of the rp1_nexus node to
choose if the overlay should be loaded at runtime (if rp1_nexus is present,
then we are in the static case so no overlay need to be loaded).

quoted

As far as the node name, I don't care so much as long as the driver
doesn't care and you don't use '_' or 'pcie?' (that's for PCI
bridges).

And why do we have drivers/misc/rp1/rp1-pci.dtso and a .dtso in
arch/arm64? There should not be both.

drivers/misc/rp1/rp1-pci.dtso is just a wrapper over rp1-common.dtsi,
which is to be compiled in as binary blob in the driver and loaded at
runtime.
rp1.dtbo is optional, it's there just for completeness in case anyone want
to load teh overlay from the fw, as explained here:
https://lore.kernel.org/all/ab9ab3536baf5fdf6016f2a01044f00034189291.1742418429.git.andrea.porta@suse.com/ (local)
If it causes confusion, we can probably get rid of it with no penalty.

Many thanks,
Andrea

quoted

Rob