RFC net: mdio: an MDIO device driver for the Airoha EN8811H

Last version, not for application, sorry for the noise. The other option was to drop this quietly. I would rather have the reviewed state on lore for whoever has a board that needs it; I don't.

The board it was written for does not need it. With the EN8811H described the ordinary way, the phylink half [1] carries the port: the PHY driver is a module that arrives after the switch has probed, and the firmware sits on the squashfs, readable by the time the module loads.

The second reason v2 [2] gave, a PHY node whose detach wipes the loaded image, is wrong. I tried it: reset-gpios on the PHY node, the 10 ms assert delay the Bananapi BPI-R3 Mini and OpenWrt One describe, a switch unbind that reaches phy_detach(). The MD32 image stays. A marker written into its RAM reads back unchanged, and the port comes back after rebind. v2 states the opposite in two places.

What is left is the one-shot probe: -ENOENT at .probe() is not retried, so a board whose firmware is not readable at that moment either loads the driver after the filesystem or embeds 144 KB of blobs. That is the case this driver is for, and I have no such board. Rather stop than send code I cannot run.

For whoever searches for this later: in-tree the chip is on the OpenWrt One (mt7981b-openwrt-one) and the Bananapi BPI-R3 Mini (mt7986a-bananapi-bpi-r3-mini), both with the PHY on a MAC that attaches at ifup, so neither is that board. Nothing
builds AIR_EN8811H_PHY in by default, so the case is a custom config with the driver built in and the firmware on a filesystem mounted later. Out of tree the chip also sits behind the MT7530 DSA bus on EN7581, mdio-airoha on AN7583 and the SiFlower xgmac; a DSA port losing a module that arrives late is what the phylink half is for. The AN8811HB shares the download code touched here and I have none.

The series describes the chip as an MDIO device that owns the download and the reset, and publishes the PHY on a child bus once the firmware runs. Patch 4 leaves a running MD32 alone, so a board using the PHY driver by itself keeps working as before.

Andrew, your points from v2 are answered in that thread. Two things changed since. I said the next round would go without the RFC tag; it keeps it, because this one is not for application. And the two questions I left you there, whether to swap the Kconfig direction and whether to add a MAINTAINERS record over both files, need no answer now.

Tested on a Netcraze NC-1012, also sold as the Keenetic KN-1012 (MT7981B, EN8811H on a 2500base-x switch port), OpenWrt 6.18.44, the series carried as target patches:

- The MCU driver takes mdio-bus:0d, the PHY appears as mdio-bus:0d:0d from /soc/ethernet@15100000/mdio-bus/mcu@d/mdio/ethernet-phy@d, DSA attaches it to lan4 with irq=15, the interrupt fires, link up at 1Gbps. No bind or unbind in the driver's sysfs directory, rmmod of either module refused while the PHY is attached.
- Download and late arrival, on an image with no firmware in the
squashfs: the driver binds, publishes nothing, warns once at 61s of accumulated waiting naming both files and -ENOENT. Files copied in at 103s are taken on the next poll at 128s, the PHY is published, DSA attaches it with irq=15, link up at 160s.
- A PHY node at an address the MCU does not answer: one error naming the node and both addresses, -EINVAL, no child bus.
- Adoption of an already running MD32 has no hardware evidence. This board comes up needing firmware on every boot, warm or cold.

Changes since v2:
- mdiodev_lock() helpers dropped, 7 patches became 6
- wrong-address and empty-node checks moved into probe, so they fail once instead of on every retry
- child bus lock taken once around the whole programming phase, the firmware read stays outside it
- a failed bus registration keeps retrying on -EPROBE_DEFER
- resume comment and commit text had the order backwards; the MCU resumes before the PHY
- the detach-wipes-firmware reason is withdrawn

[1] https://lore.kernel.org/netdev/[email protected]/
[2] https://lore.kernel.org/netdev/[email protected]/

Aleksei Sviridkin (6):
dt-bindings: net: add Airoha EN8811H PHY MCU
net: phy: air: type the buckpbus core on the mdio device
net: phy: air: move the EN8811H firmware download into the library
net: phy: air: skip the download when the MD32 is already running
net: mdio: add Airoha EN8811H MDIO device driver
net: mdio: en8811h: add the nested bus

 .../bindings/net/airoha,en8811h-mcu.yaml      | 127 +++++
 MAINTAINERS                                   |   8 +
 drivers/net/mdio/Kconfig                      |  13 +
 drivers/net/mdio/Makefile                     |   1 +
 drivers/net/mdio/mdio-airoha-en8811h.c        | 378 +++++++++++++++
 drivers/net/phy/air_en8811h.c                 | 155 +-----
 drivers/net/phy/air_phy_lib.c                 | 440 ++++++++++++++++--
 drivers/net/phy/air_phy_lib.h                 |  27 ++
 include/linux/mdio/mdio-airoha-en8811h.h      |  29 ++
 9 files changed, 999 insertions(+), 179 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/net/airoha,en8811h-mcu.yaml
 create mode 100644 drivers/net/mdio/mdio-airoha-en8811h.c
 create mode 100644 include/linux/mdio/mdio-airoha-en8811h.h

base-commit: ab217fbb9b2169ce677b09a66558d5c3adcfbb76

2.53.0