Re: ixgbe - only presenting one out of four interfaces on 5.9

[Jakub Kicinski]

On Sat, 17 Oct 2020 00:39:11 +0200 Ian Kumlien wrote:
> On Fri, Oct 16, 2020 at 9:21 PM Jakub Kicinski <k...@kernel.org> wrote:
> > > > You can actually see it dmesg... And i tried some basic looking at
> > > > changes to see if it was obvious.... but..  
> >
> > Showing a full dmesg may be helpful, but looking at what you posted it
> > seems like the driver gets past the point where netdev gets registered,
> > so the only thing that could fail after that point AFIACT is
> > mdiobus_register(). Could be some breakage in MDIO.  
> 
> Humm... interesting, will have a look at it
> 
> > Any chance you could fire up perf, bpftrace and install a kretprobe to
> > see what mdiobus_register() returns? You can rebind the device to the
> > driver through sysfs.  
> 
> Do you need a difference between the kernels?

Nope, we can safely assume that it return 0 on kernels where things
work.

Looking at the changes in this area now - it's gotta be:

commit 09ef193fef7e ("net: ethernet: ixgbe: check the return value of 
ixgbe_mii_bus_init()")

This should make things work again for you:

diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c 
b/drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c
index f77fa3e4fdd1..ac5bfc2b5f81 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c
@@ -922,7 +922,7 @@ s32 ixgbe_mii_bus_init(struct ixgbe_hw *hw)
        case IXGBE_DEV_ID_X550EM_A_1G_T:
        case IXGBE_DEV_ID_X550EM_A_1G_T_L:
                if (!ixgbe_x550em_a_has_mii(hw))
-                       return -ENODEV;
+                       return 0;
                bus->read = &ixgbe_x550em_a_mii_bus_read;
                bus->write = &ixgbe_x550em_a_mii_bus_write;
                break;