Re: [PATCH 2/2] dt-bindings: net: dsa: mediatek,mt7530: document MDIO-bus

[Andrew Lunn <andrew@lunn.ch>]

On Sun, Apr 30, 2023 at 10:54:50PM +0300, Arınç ÜNAL wrote:
> On 30.04.2023 21:48, Andrew Lunn wrote:
> > > > Try setting ds->slave_mii_bus to the MDIO bus you register via
> > > > of_mdiobus_register().
> > > 
> > > That seems to be the case already, under mt7530_setup_mdio():
> > > 
> > > 	bus = devm_mdiobus_alloc(dev);
> > > 	if (!bus)
> > > 		return -ENOMEM;
> > > 
> > > 	ds->slave_mii_bus = bus;
> > > 
> > > The bus is registered with devm_of_mdiobus_register(), if that matters. (My
> > > current knowledge about OF or OF helpers for MDIO is next to nothing.)
> > > 
> > > The same behaviour is there.
> > 
> > Maybe take a look at what is going on in dsa_slave_phy_setup() and
> > dsa_slave_phy_connect().
> > 
> > The way i understand it, is it first looks in DT to see if there is a
> > phy-handle, and if there is, it uses it. If not, it assumes there is a
> > 1:1 mapping between port number and PHY address, and looks to see if a
> > PHY has been found on ds->slave_mii_bus at that address, and uses it.
> > 
> > So i don't think you need to list the PHY, the fallback should be
> > used.
> 
> Thanks for pointing me in the right direction Andrew.
> 
> I applied this diff:
> 
> diff --git a/drivers/net/phy/mdio_bus.c b/drivers/net/phy/mdio_bus.c
> index 389f33a12534..19d0c209e7e9 100644
> --- a/drivers/net/phy/mdio_bus.c
> +++ b/drivers/net/phy/mdio_bus.c
> @@ -117,8 +117,12 @@ struct phy_device *mdiobus_get_phy(struct mii_bus *bus, int addr)
>  	mdiodev = bus->mdio_map[addr];
> -	if (!mdiodev)
> +	if (!mdiodev) {
> +		dev_info(&bus->dev, "mdio device doesn't exist\n");
>  		return NULL;
> +	}
> +
> +	dev_info(&bus->dev, "mdio device exists\n");
>  	if (!(mdiodev->flags & MDIO_DEVICE_FLAG_PHY))
>  		return NULL;
> diff --git a/net/dsa/slave.c b/net/dsa/slave.c
> index 165bb2cb8431..0be408e32a76 100644
> --- a/net/dsa/slave.c
> +++ b/net/dsa/slave.c
> @@ -2487,6 +2487,7 @@ static int dsa_slave_phy_setup(struct net_device *slave_dev)
>  		/* We could not connect to a designated PHY or SFP, so try to
>  		 * use the switch internal MDIO bus instead
>  		 */
> +		netdev_err(slave_dev, "using switch's internal MDIO bus\n");
>  		ret = dsa_slave_phy_connect(slave_dev, dp->index, phy_flags);
>  	}
>  	if (ret) {
> 
> With or without this patch, the switch's internal MDIO bus is used to set
> up the PHYs.
> 
> DT that defines ethphy0 only, without this patch applied:
> 
> [    4.660784] mt7530-mdio mdio-bus:1f wan (uninitialized): using switch's internal MDIO bus
> [    4.669026] mdio_bus mt7530-0: mdio device exists
> [    4.677693] mt7530-mdio mdio-bus:1f wan (uninitialized): PHY [mt7530-0:00] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.693238] mt7530-mdio mdio-bus:1f lan0 (uninitialized): using switch's internal MDIO bus
> [    4.701589] mdio_bus mt7530-0: mdio device exists
> [    4.707101] mt7530-mdio mdio-bus:1f lan0 (uninitialized): PHY [mt7530-0:01] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.718550] mt7530-mdio mdio-bus:1f lan1 (uninitialized): using switch's internal MDIO bus
> [    4.726856] mdio_bus mt7530-0: mdio device exists
> [    4.732384] mt7530-mdio mdio-bus:1f lan1 (uninitialized): PHY [mt7530-0:02] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.743822] mt7530-mdio mdio-bus:1f lan2 (uninitialized): using switch's internal MDIO bus
> [    4.752154] mdio_bus mt7530-0: mdio device exists
> [    4.757662] mt7530-mdio mdio-bus:1f lan2 (uninitialized): PHY [mt7530-0:03] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.769099] mt7530-mdio mdio-bus:1f lan3 (uninitialized): using switch's internal MDIO bus
> [    4.781872] mdio_bus mt7530-0: mdio device exists
> [    4.787413] mt7530-mdio mdio-bus:1f lan3 (uninitialized): PHY [mt7530-0:04] driver [MediaTek MT7530 PHY] (irq=POLL)
> 
> Same DT but with this patch applied:
> 
> [    4.621547] mt7530-mdio mdio-bus:1f: configuring for fixed/trgmii link mode
> [    4.631524] mt7530-mdio mdio-bus:1f wan (uninitialized): using switch's internal MDIO bus
> [    4.639764] mdio_bus mt7530-0: mdio device exists
> [    4.647770] mt7530-mdio mdio-bus:1f wan (uninitialized): PHY [mt7530-0:00] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.663898] mt7530-mdio mdio-bus:1f lan0 (uninitialized): using switch's internal MDIO bus
> [    4.672253] mdio_bus mt7530-0: mdio device doesn't exist
> [    4.677597] mt7530-mdio mdio-bus:1f lan0 (uninitialized): no phy at 1
> [    4.684053] mt7530-mdio mdio-bus:1f lan0 (uninitialized): failed to connect to PHY: -ENODEV
> [    4.692435] mt7530-mdio mdio-bus:1f lan0 (uninitialized): error -19 setting up PHY for tree 0, switch 0, port 1
> [    4.703087] mt7530-mdio mdio-bus:1f lan1 (uninitialized): using switch's internal MDIO bus
> [    4.711408] mdio_bus mt7530-0: mdio device doesn't exist
> [    4.716731] mt7530-mdio mdio-bus:1f lan1 (uninitialized): no phy at 2
> [    4.723214] mt7530-mdio mdio-bus:1f lan1 (uninitialized): failed to connect to PHY: -ENODEV
> [    4.731597] mt7530-mdio mdio-bus:1f lan1 (uninitialized): error -19 setting up PHY for tree 0, switch 0, port 2
> [    4.742199] mt7530-mdio mdio-bus:1f lan2 (uninitialized): using switch's internal MDIO bus
> [    4.755431] mdio_bus mt7530-0: mdio device doesn't exist
> [    4.760793] mt7530-mdio mdio-bus:1f lan2 (uninitialized): no phy at 3
> [    4.767263] mt7530-mdio mdio-bus:1f lan2 (uninitialized): failed to connect to PHY: -ENODEV
> [    4.775632] mt7530-mdio mdio-bus:1f lan2 (uninitialized): error -19 setting up PHY for tree 0, switch 0, port 3
> [    4.786270] mt7530-mdio mdio-bus:1f lan3 (uninitialized): using switch's internal MDIO bus
> [    4.794591] mdio_bus mt7530-0: mdio device doesn't exist
> [    4.799944] mt7530-mdio mdio-bus:1f lan3 (uninitialized): no phy at 4
> [    4.806397] mt7530-mdio mdio-bus:1f lan3 (uninitialized): failed to connect to PHY: -ENODEV
> [    4.814782] mt7530-mdio mdio-bus:1f lan3 (uninitialized): error -19 setting up PHY for tree 0, switch 0, port 4
> 
> DT without the mdio node defined, with this patch applied:
> 
> [    4.650766] mt7530-mdio mdio-bus:1f: configuring for fixed/trgmii link mode
> [    4.660687] mt7530-mdio mdio-bus:1f wan (uninitialized): using switch's internal MDIO bus
> [    4.668937] mdio_bus mt7530-0: mdio device exists
> [    4.677787] mt7530-mdio mdio-bus:1f wan (uninitialized): PHY [mt7530-0:00] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.693165] mt7530-mdio mdio-bus:1f lan0 (uninitialized): using switch's internal MDIO bus
> [    4.701517] mdio_bus mt7530-0: mdio device exists
> [    4.707029] mt7530-mdio mdio-bus:1f lan0 (uninitialized): PHY [mt7530-0:01] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.718469] mt7530-mdio mdio-bus:1f lan1 (uninitialized): using switch's internal MDIO bus
> [    4.726773] mdio_bus mt7530-0: mdio device exists
> [    4.732322] mt7530-mdio mdio-bus:1f lan1 (uninitialized): PHY [mt7530-0:02] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.743793] mt7530-mdio mdio-bus:1f lan2 (uninitialized): using switch's internal MDIO bus
> [    4.752143] mdio_bus mt7530-0: mdio device exists
> [    4.757662] mt7530-mdio mdio-bus:1f lan2 (uninitialized): PHY [mt7530-0:03] driver [MediaTek MT7530 PHY] (irq=POLL)
> [    4.769105] mt7530-mdio mdio-bus:1f lan3 (uninitialized): using switch's internal MDIO bus
> [    4.781905] mdio_bus mt7530-0: mdio device exists
> [    4.787459] mt7530-mdio mdio-bus:1f lan3 (uninitialized): PHY [mt7530-0:04] driver [MediaTek MT7530 PHY] (irq=POLL)
> 
> This is how I define it, mind you no phandles.
> 
> switch@1f {
> 	...
> 	mdio {
> 		#address-cells = <0x01>;
> 		#size-cells = <0x00>;
> 
> 		ethernet-phy@0 {
> 			reg = <0x00>;
> 		};
> 	};
> };
> 
> Like you said, if the mdio node is not defined, the driver will assume 1:1
> mapping. If not, it will need all the PHYs to be defined on the mdio node
> along with on the ports node. Hence back to my original statement, we can
> either force defining the PHYs on the mdio node which would break the ABI,
> or forget about doing PHY muxing this way.
> 
> There are no MDIO operations needed on the PHYs for the PHY muxing anyway,
> so I'd rather do this some other way.

The problem is __of_mdiobus_register() :-(

If there is a node in DT, the scanning of all addresses is
disabled. It looks at just the addresses listed in DT. Plain
mdiobus_register() by default looks at all addresses to discover if a
PHY is there.

Hence when slave.c tries to connect to the PHY, only the PHY you
listed in DT has been found.

I don't see a way around this, at least not without adding a new
variant of of_mdiobus_register which combined DT with a bus scan.

	Andrew