Re: [PATCH v8 5/5] PCI: of: Create device-tree PCI host bridge node

[Herve Codina <herve.codina@bootlin.com>]

Hi Claudiu,

On Fri, 13 Jun 2025 16:36:16 +0300
Claudiu Beznea <claudiu.beznea@tuxon.dev> wrote:

...

> I pointed to the wrong function. It's not of_pci_make_host_bridge_node()
> [1] but of_pci_make_dev_node() which creates a node with a similar naming
> and makes things not working on my side.
> 
> [1] https://elixir.bootlin.com/linux/v6.15/source/drivers/pci/of.c#L694

Ok, so your issue is not related patches applied from the "PCI: of: Create
device-tree PCI host bridge node" series.
  https://lore.kernel.org/all/20250224141356.36325-6-herve.codina@bootlin.com/

Indeed, this series add the node creation for the host bridge with
of_pci_make_host_bridge_node() but you pointed now of_pci_make_dev_node()
which is the creation for PCI device node and this function was not modify by the
series.

of_pci_make_host_bridge_node() should not create anything. Can you confirm on your
side that it doesn't create any nodes.

If so, maybe the problem comes from of_irq_parse_raw() or similar.

...

> 
> > 
> > On this system, I didn't observed any issues but of course, the PCIe drivers are
> > different.
> > Also, on my system, no node were created by of_pci_make_host_bridge_node().  
> 
> Sorry for the confusion, it is of_pci_make_dev_node() on my side which
> creates the node.
> 
> > 
> > To be honest, I didn't re-test recently to see if something has been broken.
> > I can do that on my side with my system.

I have re-tested and I confirm that I have no issue on my system.

> > 
> > On your side, maybe you can have look at the Armada PCIe driver and see if
> > something could explain your behavior. I am not sure that you need to add the
> > pci@0,0 node in your DT.  
> 
> I can't find a driver that uses the approach I'm trying in my patches. This
> approach was suggested in the review process [2] by Rob who mentioned that
> now we should be able drop legacy interrupt controller nodes. There are
> some Apple device trees that points the interrupt-map to the port node (the
> way I tried in my workaround) [3], but I can't find more than that.
> 
> The topology in my case is:
> 
> root@smarc-rzg3s:~# lspci -t
> -[0000:00]---00.0-[01]----00.0
> 
> root@smarc-rzg3s:~# lspci
> 00:00.0 PCI bridge: Renesas Technology Corp. Device 0033
> 01:00.0 Non-Volatile memory controller: Micron Technology Inc 2550 NVMe SSD
> (DRAM-less) (rev 01)
> 
> When not working pci@0,0 is exported as follows in rootfs:
> 
> root@smarc-rzg3s:~# ls /sys/firmware/devicetree/base/soc/pcie@11e40000 -l
> -r--r--r--    1 root     root             4 Jan 12 10:28 #address-cells
> -r--r--r--    1 root     root             4 Jan 12 10:28 #interrupt-cells
> -r--r--r--    1 root     root             4 Jan 12 10:28 #size-cells
> -r--r--r--    1 root     root             8 Jan 12 10:28 bus-range
> -r--r--r--    1 root     root            13 Jan 12 10:28 clock-names
> -r--r--r--    1 root     root            24 Jan 12 10:28 clocks
> -r--r--r--    1 root     root            26 Jan 12 10:28 compatible
> -r--r--r--    1 root     root             4 Jan 12 10:28 device-id
> -r--r--r--    1 root     root             4 Jan 12 10:28 device_type
> -r--r--r--    1 root     root            28 Jan 12 10:28 dma-ranges
> -r--r--r--    1 root     root             0 Jan 12 10:28 interrupt-controller
> -r--r--r--    1 root     root           144 Jan 12 10:28 interrupt-map
> -r--r--r--    1 root     root            16 Jan 12 10:28 interrupt-map-mask
> -r--r--r--    1 root     root           164 Jan 12 10:28 interrupt-names
> -r--r--r--    1 root     root             4 Jan 12 10:28 interrupt-parrent

Why parrent instead of parent in interrupt-parrent ?

> -r--r--r--    1 root     root           192 Jan 12 10:28 interrupts
> -r--r--r--    1 root     root             5 Jan 12 10:28 name
> -r--r--r--    1 root     root             4 Jan 12 10:28 num-lanes
> drwxr-xr-x    2 root     root             0 Jan 12 10:17 pci@0,0
> -r--r--r--    1 root     root             4 Jan 12 10:28 phandle
> -r--r--r--    1 root     root             4 Jan 12 10:28 pinctrl-0
> -r--r--r--    1 root     root             8 Jan 12 10:28 pinctrl-names
> -r--r--r--    1 root     root             4 Jan 12 10:28 power-domains
> -r--r--r--    1 root     root            28 Jan 12 10:28 ranges
> -r--r--r--    1 root     root            16 Jan 12 10:28 reg
> -r--r--r--    1 root     root             4 Jan 12 10:28 renesas,sysc
> -r--r--r--    1 root     root            63 Jan 12 10:28 reset-names
> -r--r--r--    1 root     root            56 Jan 12 10:28 resets
> -r--r--r--    1 root     root             5 Jan 12 10:28 status
> -r--r--r--    1 root     root             4 Jan 12 10:28 vendor-id
> root@smarc-rzg3s:~#
> root@smarc-rzg3s:~# ls
> /sys/firmware/devicetree/base/soc/pcie@11e40000/pci@0,0 -l
> -r--r--r--    1 root     root             4 Jan 12 10:17 #address-cells
> -r--r--r--    1 root     root             4 Jan 12 10:17 #interrupt-cells
> -r--r--r--    1 root     root             4 Jan 12 10:17 #size-cells
> -r--r--r--    1 root     root             8 Jan 12 10:17 bus-range
> -r--r--r--    1 root     root            41 Jan 12 10:17 compatible
> -r--r--r--    1 root     root             4 Jan 12 10:17 device_type
> -r--r--r--    1 root     root           144 Jan 12 10:17 interrupt-map
> -r--r--r--    1 root     root            16 Jan 12 10:17 interrupt-map-mask
> -r--r--r--    1 root     root            32 Jan 12 10:17 ranges
> -r--r--r--    1 root     root            20 Jan 12 10:17 reg
> root@smarc-rzg3s:~#
> root@smarc-rzg3s:~#
> root@smarc-rzg3s:~#
> root@smarc-rzg3s:~#
> root@smarc-rzg3s:~# cat
> /sys/firmware/devicetree/base/soc/pcie@11e40000/pci@0,0/compatible
> pci1912,33pciclass,060400pciclass,0604root@smarc-rzg3s:~#
> root@smarc-rzg3s:~#
> root@smarc-rzg3s:~#
> 
> In case I describe a port in device tree, it works because the pci@0,0 is
> not created anymore when device is enumerated and thus the interrupt
> parsing is working.
> 
> Herve: do you have some hints?

First interrupt-parrent in your /sys/firmware/devicetree/base/soc/pcie@11e40000
files.

If it is just a typo in this email, maybe the interrupt parsing itself.

Can you provide an extract for the DT with nodes created at runtime.
I mean can you run 'dtc -I dtb -O dts /proc/device-tree' and provide the output
related to PCI nodes including the PCIe controller ?

> 
> Rob: do you know some device trees where the interrupt-map points to the
> node itself as suggested in [2] so that I can check is something is missing
> on my side?
> 
> Thank you,
> Claudiu
> 
> [2] https://lore.kernel.org/all/20250509210800.GB4080349-robh@kernel.org/
> [3]
> https://elixir.bootlin.com/linux/v6.15/source/arch/arm64/boot/dts/apple/t8112.dtsi#L951
> 

Best regards,
Hervé

-- 
Hervé Codina, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com