Re: [PATCH net-next v6 0/3] net: phylink: wait for a PHY that probes after the MAC

On Wed, Oct 14, 2026 at 12:04:36PM +0200, Paolo Abeni wrote:

On 10/13/26 17:12, Aleksei Sviridkin wrote:

quoted

On 10/13/26 09:53, Paolo Abeni wrote:

quoted

The above looks like a significant amount of complexity for OoT driver,
and I understand it could be solved with a different kconf?

The PHY driver is in mainline, air_en8811h. Only the board's device
tree lives in OpenWrt.

A config change can avoid it, but it is expensive here. Building
air_en8811h in is not enough, since it loads its firmware with
request_firmware_direct() from the root filesystem, which is not mounted
yet when the switch probes. It also needs CONFIG_EXTRA_FIRMWARE with
144 KB of blobs, in a kernel that OpenWrt shares between every board
of the subtarget.

Is that the expected answer for a board like this one, or would you
take a smaller change in phylink?

To me this series still looks too much churn for something that has
alternative options. I would like to hear a 2nd opinion from Russell,
Andrew or Heiner.

I think the hardware design is poor, it would of made everybody life
easier spending an extra $1 on an EEPROM for the PHY firmware.

One alternative is to put the firmware download into the
bootloader. That means it is not in the initramfs, which seems to be
the major reason for doing this. By the time Linux gets to drive the
hardware, the hardware is actually usable. That makes our life
simpler.

	Andrew