Re: [PATCH v13 00/10] lan966x pci device: Add support for SFPs, PCI part

[Abdurrahman Hussain]

Hi Herve,

On Mon, Oct 12, 2026 at 04:28:14PM +0200, Herve Codina wrote:
> Patches 8 and 9 fix fw_devlink issues specific to PCI and the
> device-tree nodes created during enumeration.

We hit exactly this on x86 switch platforms with
CONFIG_PCI_DYNAMIC_OF_NODES. A driver applies a device-tree overlay
onto each Xilinx FPGA PCI endpoint and populates I2C controllers
(xiic), watchdogs and GPIO blocks. Their interrupts go through small
chained irqchips that are themselves fed by the FPGA's MSI vectors.

Without this series, fw_devlink drops every supplier link under the
synthesized PCI root bridge node ("Not linking ... might never become
dev"), so the irqchips end up with no consumers. Unbinding one of them
frees IRQ descriptors that the I2C controllers and the watchdog still
hold. We get a remove_proc_entry "removing non-empty directory 'irq/N'"
WARNING per IRQ, and the I2C controllers are dead until reboot.

With the series applied on v7.3, the device links are created (up to
32 consumers per irqchip), and unbinding an irqchip now unbinds its
consumers first. I tested on two platforms, with 2 and 3 FPGAs (4 and
7 irqchips). Repeated unload/reload of the irqchip driver was clean
every time, and everything probed again afterwards.

Tested-by: Abdurrahman Hussain <abdurrahman@xxxxxxxxxx>

Thanks,
Abdurrahman