Re: [PATCH net-next v2 1/3] net: macb: fix SGMII with inband aneg disabled

From: Charles Perry

Date: Thu Mar 05 2026 - 10:19:45 EST

On Thu, Mar 05, 2026 at 02:19:06PM +0000, Conor Dooley wrote:
> On Thu, Mar 05, 2026 at 05:47:34AM -0800, Charles Perry wrote:
> > On Wed, Mar 04, 2026 at 06:39:38PM +0000, Russell King (Oracle) wrote:
> > > On Wed, Mar 04, 2026 at 06:06:20PM +0000, Conor Dooley wrote:
> > > > On Wed, Mar 04, 2026 at 04:55:37PM +0000, Russell King (Oracle) wrote:
> > > > > On Wed, Mar 04, 2026 at 04:23:30PM +0000, Conor Dooley wrote:
> > > > > > On Wed, Mar 04, 2026 at 06:59:35AM -0800, Charles Perry wrote:
> > > > > > > On Wed, Mar 04, 2026 at 11:15:43AM +0000, Conor Dooley wrote:
> > > > > > > > On Tue, Feb 24, 2026 at 12:28:52PM -0800, Charles Perry wrote:
> > > > > > > > > Make it possible to connect a PHY which does not use inband
> > > > > > > > > autoneg to a gem MAC using phylink's information.
> > > > > > > > >
> > > > > > > > > The previous implementation relied on whether or not the link
> > > > > > > > > was a fixed-link to disable SGMII autoneg. This commit extend
> > > > > > > > > this to all link which are not configured for inband
> > > > > > > > > autonegotiation.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Charles Perry <charles.perry@xxxxxxxxxxxxx>
> > > > > > > >
> > > > > > > > This breaks the macb on mpfs-icicle-kit, I get stuck with:
> > > > > > > >
> > > > > > > > [ 7.189102] mpfs-sys-controller syscontroller: Registered MPFS system controller
> > > > > > > > [ 7.260946] macb 20110000.ethernet eth0: PHY [20112000.ethernet-ffffffff:08] driver [Vitesse VSC8662] (irq=POLL)
> > > > > > > > [ 7.273881] macb 20110000.ethernet eth0: configuring for phy/sgmii link mode
> > > > > > > > [ 7.296580] macb 20110000.ethernet: gem-ptp-timer ptp clock registered.
> > > > > > > > [ 7.345782] macb 20112000.ethernet eth1: PHY [20112000.ethernet-ffffffff:09] driver [Vitesse VSC8662] (irq=POLL)
> > > > > > > > [ 7.358082] macb 20112000.ethernet eth1: configuring for phy/sgmii link mode
> > > > > > > > [ 7.380479] macb 20112000.ethernet: gem-ptp-timer ptp clock registered.
> > > > > > > > [ 11.376763] macb 20110000.ethernet eth0: Link is Up - 1Gbps/Full - flow control off
> > > > > > > > [ 11.398403] Sending DHCP requests .
> > > > > > > > [ 11.472699] macb 20112000.ethernet eth1: Link is Up - 1Gbps/Full - flow control off
> > > > > > > > [ 13.938425] ..... timed out!
> > > > > > > > [ 93.598491] macb 20110000.ethernet eth0: Link is Down
> > > > > > > > [ 93.641823] macb 20110000.ethernet: gem-ptp-timer ptp clock unregistered.
> > > > > > > > [ 93.659433] macb 20112000.ethernet eth1: Link is Down
> > > > > > > > [ 93.691724] macb 20112000.ethernet: gem-ptp-timer ptp clock unregistered.
> > > > > > > > [ 93.703977] IP-Config: Retrying forever (NFS root)...
> > > > > > > > [ 93.758382] macb 20110000.ethernet eth0: PHY [20112000.ethernet-ffffffff:08] driver [Vitesse VSC8662] (irq=POLL)
> > > > > > > > [ 93.770655] macb 20110000.ethernet eth0: configuring for phy/sgmii link mode
> > > > > > > > [ 93.786497] macb 20110000.ethernet eth0: Link is Up - 1Gbps/Full - flow control off
> > > > > > > > [ 93.795840] macb 20110000.ethernet: gem-ptp-timer ptp clock registered.
> > > > > > > > [ 93.844481] macb 20112000.ethernet eth1: PHY [20112000.ethernet-ffffffff:09] driver [Vitesse VSC8662] (irq=POLL)
> > > > > > > > [ 93.856769] macb 20112000.ethernet eth1: configuring for phy/sgmii link mode
> > > > > > > > [ 93.870926] macb 20112000.ethernet eth1: Link is Up - 1Gbps/Full - flow control off
> > > > > > > > [ 93.880302] macb 20112000.ethernet: gem-ptp-timer ptp clock registered.
> > > > > > > >
> > > > > > >
> > > > > > > Hello Conor,
> > > > > > >
> > > > > > > I checked the driver for the VSC8662 and it doesn't have the
> > > > > > > ->inband_caps() and ->config_inband() callbacks so Linux leaves whatever
> > > > > > > the bootloader puts or uses the defaults. Looking at the datasheet, this
> > > > > > > should be register 23 (Extended PHY Control Set 1) bit 13 (MAC interface
> > > > > > > auto-negotiation)
> > > > > > >
> > > > > > > My guess is that this bit is set and since this patch disable inband
> > > > > > > autonegotiation (because phylink decides it), there is a mismatch.
> > > > > > >
> > > > > > > Can you add 'managed = "in-band-status"' in your device tree under the macb
> > > > > > > node? That's not necessarily the fix, I just want to confirm my theory.
> > > > > >
> > > > > > No, it just produces a different error:
> > > > > > [ 5.769864] mpfs-sys-controller syscontroller: Registered MPFS system controller
> > > > > > [ 5.829146] macb 20110000.ethernet eth0: Could not attach PHY (-19)
> > > > > > [ 5.854232] IP-Config: Failed to open eth0
> > > > > > [ 5.897152] macb 20112000.ethernet eth1: Could not attach PHY (-19)
> > > > > > [ 5.921061] IP-Config: Failed to open eth1
> > > > > > [ 5.925592] IP-Config: No network devices available
> > > > > > [ 5.938800] clk: Disabling unused clocks
> > > > > > [ 5.944156] PM: genpd: Disabling unused power domains
> > > > > > [ 5.961029] check access for rdinit=/usr/sbin/init failed: -2, ignoring
> > > > >
> > > > > -19 is -ENODEV (I wish everyone would use %pe so we get english
> > > > > error messages rather than having to look up errno codes in the
> > > > > header files.)
> > > > >
> > > > > macb uses either phylink_of_phy_connect() or phylink_connect_phy().
> > > > > I don't think phylink_connect_phy() would return -ENODEV, but
> > > > > phylink_of_phy_connect() would - but I can't see that adding
> > > > > 'managed = "in-band-status";' to DT would cause that. The only
> > > > > case I can see is that fwnode_phy_find_device() fails to find the
> > > > > phydev, but there is a PHY node specified in DT, but that would
> > > > > fail without in-band-status.
> > > >
> > > > It's as you say, and fwnode_phy_find_device() is the source.
> > > > fwnode_mdio_find_device() is what fails, returning NULL.
> > >
> > > Ah, I've just found the reason:
> > >
> > > /* With fixed-link, we don't need to register the MDIO bus,
> > > * except if we have a child named "mdio" in the device tree.
> > > * In that case, some devices may be attached to the MACB's MDIO bus.
> > > */
> > > mdio_np = of_get_child_by_name(np, "mdio");
> > > if (!mdio_np && of_phy_is_fixed_link(np))
> > > return macb_mii_probe(bp->dev);
> > >
> > > of_phy_is_fixed_link() will return true as soon as you add that managed
> > > property, and as the mac1 node just lists the PHYs without using a
> > > mdio child:
> > >
> > > &mac1 {
> > > phy-mode = "sgmii";
> > > phy-handle = <&phy1>;
> > > status = "okay";
> > >
> > > phy1: ethernet-phy@9 {
> > > reg = <9>;
> > > };
> > >
> > > phy0: ethernet-phy@8 {
> > > reg = <8>;
> > > };
> > > };
> > >
> > > it means mdio_np is NULL above. Hence, no MDIO bus will be created.
> >
> > Thank you Russell for finding this.
> >
> > Conor, can you try with this:
> >
> > ````
> > &mac0 {
> > phy-mode = "sgmii";
> > phy-handle = <&phy0>;
> > status = "okay";
> > managed = "in-band-status";
> > };
> >
> > &mac1 {
> > phy-mode = "sgmii";
> > phy-handle = <&phy1>;
> > status = "okay";
> > managed = "in-band-status";
> >
> > mdio {
> > phy1: ethernet-phy@9 {
> > reg = <9>;
> > };
> >
> > phy0: ethernet-phy@8 {
> > reg = <8>;
> > };
> > };
> > };
> > ````
> >
> > I think we're almost there.
>
> With this, I can boot. Is that enough information for you to resolve the
> problem?

Yes, thanks alot Conor!

The problem is that all the SGMII setups using GEM probably currently work
with autonegotiation enabled while phylink doesn't know about it. This is
because the PCSAUTONEG bit (12) in the PCSCNTRL register (0x200) is set by
default and starting from e276e5e40e92 ("net: macb: Disable PCS
auto-negotiation for SGMII fixed-link mode"), the bit is forced to 1 (for
non fixed-link setups). So people have just worked a phy config with
bootstrap resistors or bootloader code to enable it on the PHY. Now that
this patch made the PCSAUTONEG bit follow what phylink knows, we end up
misconfiguring the link because most PHYs probably don't have the
->inband_caps() and ->config_inband() callbacks.

I'm wondering if the "bool default_an_inband" from "struct phylink_config"
could help us here. The comment reads: "if true, defaults to MLO_AN_INBAND
rather than MLO_AN_PHY. A fixed-link specification will override." Could
we use this in GEM to signify "all the legacy setups uses inband aneg so
please used this". Russell, can you comment on this?

Patching PHY drivers to add ->inband_caps() and ->config_inband() is
another solution but there are most likely other PHYs than the VSC8662 that
would need this.

Adding 'managed = "in-band-status"' is another solution but it involves
some changes on all the device tree using GEM and SGMII.

Thanks,
Charles