> -----Original Message----- > From: Richard Patel <[email protected]> > Sent: Saturday, September 5, 2026 7:33 PM > To: [email protected] > Cc: Nguyen, Anthony L <[email protected]>; Kitszel, Przemyslaw > <[email protected]>; Andrew Lunn <[email protected]>; Steve Douthit > <[email protected]>; [email protected]; > [email protected]; > Richard Patel <[email protected]> > Subject: [PATCH iwl-net v2] ixgbe: fix MDIO bus lifetime > > Plugging in an unsupported SFP module caused ixgbe NICs to get > wedged permanently. > > ixgbe unregisters the netdev when rejecting an SFP transceiver as > unsupported. > > ixgbe 0000:3b:00.0: failed to initialize because an unsupported SFP+ module > type was detected. > ixgbe 0000:3b:00.0: Reload the driver after installing a supported module. > ixgbe 0000:3b:00.0: removed PHC on enp59s0 > > This code path failed to unregister an MDIO bus owned by that > netdev, so when unregister_netdev() tries to tear it down > (via device_del() -> devres_release_all() -> mdiobus_free()), > it trips over the leaked registration. > > ixgbe-mdio-0000:3b:00.0: not in UNREGISTERED state > WARNING: drivers/net/phy/mdio_bus_provider.c:711 at mdiobus_free+0x3d/0x50, > CPU#21: kworker/u128:0/12 > > This prevents ixgbe from reloading, even after PCIe remove/rescan: > > sysfs: cannot create duplicate filename > '/class/mdio_bus/ixgbe-mdio-0000:3b:00.0' > > Switch to devm_mdiobus_register(), drop custom unregister calls. > This is safe because the bus is already devm-allocated, so unregister > always happens before free. > > Assisted-by: LLM > Fixes: 8fa10ef01260 ("ixgbe: register a mdiobus") > Suggested-by: Andrew Lunn <[email protected]> > Signed-off-by: Richard Patel <[email protected]> > --- > v1 -> v2: use devm_mdiobus_register() instead of another > mdiobus_unregister() call > v1: https://lore.kernel.org/netdev/[email protected]/ > > drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 9 ++------- > drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c | 2 +- > 2 files changed, 3 insertions(+), 8 deletions(-)
Tested-by: Sunitha Mekala <[email protected]> (A Contingent worker at Intel)
