On a Keenetic KN-1012 (MT7981) the EN8811H sits behind a port of an MT7531 switch. The driver has to be built in there: as a module it loads after DSA has validated the port against the generic PHY driver and failed to set it up. Built in, it probes when the Ethernet MDIO bus is registered, before the root filesystem is mounted, so request_firmware_direct() fails with -ENOENT, the PHY stays without a driver and the port never comes up, although the bootloader has already loaded and started the MD32 firmware. When the firmware files cannot be read and the MCU reports ready, keep the running firmware and let the first .config_init() restart it, so the PHY starts from a freshly started MCU as after a download rather than from whatever state the bootloader left. Nothing changes when the files are there. A board that relies on this must not give phylib a reset-gpios line, as the pulse at registration stops the MCU. The AN8811HB path is unchanged. Tested on the KN-1012: with the bootloader load the PHY binds and links; without it 0x8009 reads 0x0 and the probe fails as before; a rebind with the files present downloads them as before. Suggested-by: Andrew Lunn Signed-off-by: Aleksei Sviridkin --- v2: read the status into a variable instead of wrapping the comparison (Jakub). v1: https://lore.kernel.org/netdev/20261017180251.3618606-1-f@lex.la/ This replaces "net: phylink: wait for a PHY that probes after the MAC" https://lore.kernel.org/netdev/20261006124731.1528963-1-f@lex.la/ following Andrew's suggestion in https://lore.kernel.org/netdev/b175c750-104a-4859-a137-09de78a6d8bb@lunn.ch/ Developed with LLM assistance; tested on hardware as described above. drivers/net/phy/air_en8811h.c | 23 +++++++++++++++++++---- 1 file changed, 19 insertions(+), 4 deletions(-) diff --git a/drivers/net/phy/air_en8811h.c b/drivers/net/phy/air_en8811h.c index a03d63dd4ac9..d1de58c3c294 100644 --- a/drivers/net/phy/air_en8811h.c +++ b/drivers/net/phy/air_en8811h.c @@ -518,6 +518,19 @@ static int an8811hb_load_firmware(struct phy_device *phydev) return en8811h_wait_mcu_ready(phydev); } +/* No firmware files, but the bootloader may have started the MCU */ +static int en8811h_use_running_fw(struct phy_device *phydev, int err) +{ + int status; + + status = phy_read_mmd(phydev, MDIO_MMD_VEND1, EN8811H_PHY_FW_STATUS); + if (status != EN8811H_PHY_READY) + return err; + + en8811h_print_fw_version(phydev); + return 1; +} + static int en8811h_load_firmware(struct phy_device *phydev) { struct device *dev = &phydev->mdio.dev; @@ -526,11 +539,13 @@ static int en8811h_load_firmware(struct phy_device *phydev) ret = request_firmware_direct(&fw1, EN8811H_MD32_DM, dev); if (ret < 0) - return ret; + return en8811h_use_running_fw(phydev, ret); ret = request_firmware_direct(&fw2, EN8811H_MD32_DSP, dev); - if (ret < 0) + if (ret < 0) { + ret = en8811h_use_running_fw(phydev, ret); goto en8811h_load_firmware_rel1; + } ret = air_phy_buckpbus_reg_write(phydev, EN8811H_FW_CTRL_1, EN8811H_FW_CTRL_1_START); @@ -1214,8 +1229,8 @@ static int en8811h_probe(struct phy_device *phydev) if (ret < 0) return ret; - /* mcu has just restarted after firmware load */ - priv->mcu_needs_restart = false; + /* drop whatever state the bootloader left in a running mcu */ + priv->mcu_needs_restart = ret > 0; /* MDIO_DEVS1/2 empty, so set mmds_present bits here */ phydev->c45_ids.mmds_present |= MDIO_DEVS_PMAPMD | MDIO_DEVS_AN; -- 2.53.0