[PATCH 4/4] ARM: dts: vf610-zii-dev-rev-b: add interrupts for 88e1545 PHY

[Russell King - ARM Linux]

On Thu, Dec 21, 2017 at 04:20:34PM -0800, Florian Fainelli wrote:
> On 12/21/2017 04:14 PM, Russell King - ARM Linux wrote:
> > On Thu, Dec 21, 2017 at 11:53:47PM +0100, Linus Walleij wrote:
> >> On Thu, Dec 21, 2017 at 6:32 PM, Russell King - ARM Linux
> >> <linux at armlinux.org.uk> wrote:
> >>
> >>> What we have here is _really_ a shared interrupt between four
> >>> separate devices, and we need a way to sanely describe resources
> >>> shared between several device instances to pinmux.  Unfortunately,
> >>> it seems pinmux is designed around one device having exclusive use
> >>> of a resource, which makes it hard to describe shared interrupts in
> >>> DT.
> >>>
> >>> Given that DT should be a description of the hardware, and should be
> >>> independent of the OS implementation, I'd say this is a pinmux bug,
> >>> because pinmux gets in the way of describing the hardware correctly.
> >>> ;)
> >>
> >> Hm that would be annoying. But when I look at it I think it would
> >> actually work. Did you try just assigning the same pin control
> >> state to all the PHY's and see what happens?
> >>
> >> Just set
> >> pinctrl-names = "default";
> >> pinctrl-0 = <&pinctrl_mv88e1545>;
> >>
> >> on all of them?
> > 
> > It was tried, DT was happy, but the kernel on boot complained because
> > pinctrl objected, which caused the drivers to fail to bind:
> > 
> > libphy: mdio: probed
> > vf610-pinctrl 40048000.iomuxc: pin VF610_PAD_PTB0 already requested by !mdio-mux!mdio at 4!switch at 0!mdio:00; cannot claim for !mdio-mux!mdio at 4!switch at 0!mdio:01
> > vf610-pinctrl 40048000.iomuxc: pin-22 (!mdio-mux!mdio at 4!switch at 0!mdio:01) status -22
> > vf610-pinctrl 40048000.iomuxc: could not request pin 22 (VF610_PAD_PTB0) from group pinctrl-mv88e1545  on device 40048000.iomuxc
> > Marvell 88E1545 !mdio-mux!mdio at 4!switch at 0!mdio:01: Error applying setting, reverse things back
> > Marvell 88E1545: probe of !mdio-mux!mdio at 4!switch at 0!mdio:01 failed with error -22
> > vf610-pinctrl 40048000.iomuxc: pin VF610_PAD_PTB0 already requested by !mdio-mux!mdio at 4!switch at 0!mdio:00; cannot claim for !mdio-mux!mdio at 4!switch at 0!mdio:02
> > vf610-pinctrl 40048000.iomuxc: pin-22 (!mdio-mux!mdio at 4!switch at 0!mdio:02) status -22
> > vf610-pinctrl 40048000.iomuxc: could not request pin 22 (VF610_PAD_PTB0) from group pinctrl-mv88e1545  on device 40048000.iomuxc
> > Marvell 88E1545 !mdio-mux!mdio at 4!switch at 0!mdio:02: Error applying setting, reverse things back
> > Marvell 88E1545: probe of !mdio-mux!mdio at 4!switch at 0!mdio:02 failed with error -22
> > 
> 
> You could also see it another way, because this is a quad PHY in a
> single package, you could theoretically have a representation that
> exposes a node container for the 4 PHYs, and that container node
> requests the pinmux/pinctrl. Of course, this would not work with the
> MDIO code which would not go one level down, and would expect the PHYs
> to be at the same level as the container node...

It would actually - we have other devices that sit on buses that take
several addresses, and we describe the "first" main device or use MFD
for it.

For example, in the case of the TDA998x HDMI encoder, these are two
devices merged into one package - the HDMI encoder at one address, and
a TDA9950 at another address.  Both addresses are related, so if you
tie the address configuration pins, the offset is added to both base
addresses.  We represent the TDA998x in DT, and have the TDA998x
driver create a separate device itself for the TDA9950.

What we could do for any multi-package PHY is describe the first PHY
as a multi-package PHY in DT, extend the phy binding to include a PHY
package index, and have the PHY driver create the MDIO devices for
the other PHYs.  Eg,

switch {
...
	ports {
		port at 0 {
			reg = <0>;
			label = "lan6";
			phy-handle = <&switch2phy 0>;
		};
		port at 1 {
			reg = <1>;
			label = "lan6";
			phy-handle = <&switch2phy 1>;
		};
		port at 2 {
			reg = <2>;
			label = "lan6";
			phy-handle = <&switch2phy 2>;
		};
	};

	mdio {
		#address-cells = <1>;
		#size-cells = <0>;

		switch2phy: phy at 0 {
			compatible = "marvell,88e1545", "ethernet-phy-ieee802.3-c22";
			interrupt-parent = <&gpio0>;
			interrupts = <22 IRQ_TYPE_LEVEL_LOW>;
			reg = <0>;
			pinctrl-names = "default";
			pinctrl-0 = <&pinctrl_mv88e1545>;
		};
	};
};

The "marvell,88e1545" driver would be responsible for creating the
PHY devices for the other MDIO bus addresses (iow 1 to 3.)

This would be an accurate respresentation of the hardware in DT,
probably more so than the trap we seem to have fallen into by
describing the individual PHYs - which we've fallen into because
that's how our current implementation requires us to describe them.

Since DT is supposed to be a hardware description, I think the question
we ought to ask is: if we were starting afresh, how would we describe
these packages that contain multiple PHYs?

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up