Skip to contentSkip to conversation

mvneta: add FIXED_PHY dependency

PATCH In review Arnd Bergmann 4 participants 10 messages Nov 9, 2015

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mvneta: add FIXED_PHY dependency

The fixed_phy infrastructure is done in a way that is optional, by providing 'static inline' helper functions doing nothing in include/linux/phy_fixed.h for all its APIs. However, three out of the four users (DSA, BCMGENET, and SYSTEMPORT) always 'select FIXED_PHY', presumably because they need that. MVNETA is the fourth one, and if that is built-in but FIXED_PHY is configured as a loadable module, we get a link error:

drivers/built-in.o: In function `mvneta_fixed_link_update':
fpga-mgr.c:(.text+0x33ed80): undefined reference to `fixed_phy_update_state'

Presumably this driver has the same dependency as the others, so this patch also uses 'select' to ensure that the fixed-phy support is built-in.

Fixes 898b2970e2c9 ("mvneta: implement SGMII-based in-band link state signaling")

---
Found using ARM randconfig tests. An alternative here would be to use 'depends on FIXED_PHY || FIXED_PHY=n', I picked the 'select' approach for consistency.

Should we perhaps make 'FIXED_PHY' a silent option and remove the inline helpers, based on the assumption that a driver that wants these will not work without them?

drivers/net/ethernet/marvell/Kconfig +1 -0
index 80af9ffce5ea..a1c862b4664d 100644
@@ -44,6 +44,7 @@ config MVNETA
 	tristate "Marvell Armada 370/38x/XP network interface support"
 	depends on PLAT_ORION
 	select MVMDIO
+	select FIXED_PHY
 	---help---
 	  This driver supports the network interface units in the
 	  Marvell ARMADA XP, ARMADA 370 and ARMADA 38x SoC family.
maintainer
#1 lore raw

From: Arnd Bergmann <[email protected]>
Date: Mon, 09 Nov 2015 15:08:57 +0100

Quoted 17 lines

The fixed_phy infrastructure is done in a way that is optional, by providing 'static inline' helper functions doing nothing in include/linux/phy_fixed.h for all its APIs. However, three out of the four users (DSA, BCMGENET, and SYSTEMPORT) always 'select FIXED_PHY', presumably because they need that. MVNETA is the fourth one, and if that is built-in but FIXED_PHY is configured as a loadable module, we get a link error:

drivers/built-in.o: In function `mvneta_fixed_link_update':
fpga-mgr.c:(.text+0x33ed80): undefined reference to `fixed_phy_update_state'

Presumably this driver has the same dependency as the others, so this patch also uses 'select' to ensure that the fixed-phy support is built-in.

Fixes 898b2970e2c9 ("mvneta: implement SGMII-based in-band link state signaling")
Quoted 8 lines

---
Found using ARM randconfig tests. An alternative here would be to use 'depends on FIXED_PHY || FIXED_PHY=n', I picked the 'select' approach for consistency.

Should we perhaps make 'FIXED_PHY' a silent option and remove the inline helpers, based on the assumption that a driver that wants these will not work without them?

This seems reasonable to fix the problem for the time being, applied.

Thanks Arnd.

#2 lore raw
In reply to Arnd Bergmann
Quoted 5 lines · Arnd Bergmann

The fixed_phy infrastructure is done in a way that is optional, by providing 'static inline' helper functions doing nothing in include/linux/phy_fixed.h for all its APIs. However, three out of the four users (DSA, BCMGENET, and SYSTEMPORT) always 'select FIXED_PHY', presumably because they need that.

Hi Arnd

Need is probably too strong, it could be considered an optional feature. If you don't have a fixed_phy property in your DT blob, you don't need fixed phy support in your image.

Quoted 9 lines

MVNETA is the fourth one, and if that is built-in but FIXED_PHY is configured as a loadable module, we get a link error:

drivers/built-in.o: In function `mvneta_fixed_link_update':
fpga-mgr.c:(.text+0x33ed80): undefined reference to `fixed_phy_update_state'

Presumably this driver has the same dependency as the others, so this patch also uses 'select' to ensure that the fixed-phy support is built-in.

This will work, and is uniform with the other instances. But maybe a more correct fix is to ensure fixed-phy is never a module when there is a builtin user.

Quoted 3 lines

Should we perhaps make 'FIXED_PHY' a silent option and remove the inline helpers, based on the assumption that a driver that wants these will not work without them?

I suppose it comes down to, are we allowed to optionally implement part of the DT binding?

Andrew

#3 lore raw
Quoted 7 lines · On Monday 09 November 2015 17:42:32 Andrew Lunn wrote:
Quoted 5 lines · Arnd Bergmann

The fixed_phy infrastructure is done in a way that is optional, by providing 'static inline' helper functions doing nothing in include/linux/phy_fixed.h for all its APIs. However, three out of the four users (DSA, BCMGENET, and SYSTEMPORT) always 'select FIXED_PHY', presumably because they need that.

Quoted 5 lines

Hi Arnd

Need is probably too strong, it could be considered an optional feature. If you don't have a fixed_phy property in your DT blob, you don't need fixed phy support in your image.

Ok, I see.

Quoted 6 lines
Quoted 6 lines

MVNETA is the fourth one, and if that is built-in but FIXED_PHY is configured as a loadable module, we get a link error:

drivers/built-in.o: In function `mvneta_fixed_link_update':
fpga-mgr.c:(.text+0x33ed80): undefined reference to `fixed_phy_update_state'

Quoted 7 lines
Quoted 3 lines

Presumably this driver has the same dependency as the others, so this patch also uses 'select' to ensure that the fixed-phy support is built-in.

This will work, and is uniform with the other instances. But maybe a more correct fix is to ensure fixed-phy is never a module when there is a builtin user.

That is hard to express with Kconfig. The alternative I listed instead guarantees that CONFIG_MVNETA cannot be set to 'y' whenever FIXED_PHY=m. For all practical purposes that has the same effect.

The fixed-phy support isn't very big (around 2KB), so I wonder how relevant that optimization is.

Quoted 6 lines
Quoted 3 lines

Should we perhaps make 'FIXED_PHY' a silent option and remove the inline helpers, based on the assumption that a driver that wants these will not work without them?

I suppose it comes down to, are we allowed to optionally implement part of the DT binding?

I'm not sure what you are asking. A lot of DT bindings have both optional and mandatory properties. For mvneta, the "phy" and "phy-mode" properties are listed as mandatory, so the driver can safely assume that they are always present. If there are reasons to leave them out, and for the driver to handle that case correctly, the binding should be updated to mark them as optional.

Arnd

#4 lore raw
Quoted 6 lines · Arnd Bergmann

I'm not sure what you are asking. A lot of DT bindings have both optional and mandatory properties. For mvneta, the "phy" and "phy-mode" properties are listed as mandatory, so the driver can safely assume that they are always present. If there are reasons to leave them out, and for the driver to handle that case correctly, the binding should be updated to mark them as optional.

They are "optional" because when you're using a DSA switch, you don't specify a PHY (because, there isn't one). For example, this is what I'm using with an Armada 388 board with a Marvell DSA switch. The DSA does not appear as a PHY, and no node in the DSA stanza can be referenced for a phy entry in the ethernet device's stanza.

                        eth1: ethernet at 30000 {
                                compatible = "marvell,armada-370-neta";
                                reg = <0x30000 0x4000>;
                                interrupts-extended = <&mpic 10>;
                                clocks = <&gateclk 3>;
                                managed = "in-band-status";
                                phy-mode = "sgmii";
                                status = "okay";
                        };

        dsa at 0 {
                compatible = "marvell,dsa";
                dsa,ethernet = <&eth1>;
                dsa,mii-bus = <&mdio>;
                pinctrl-0 = <&clearfog_dsa0_clk_pins &clearfog_dsa0_pins>;
                pinctrl-names = "default";
                #address-cells = <2>;
                #size-cells = <0>;

                switch at 0 {
		...
		};
	};
Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#5 lore raw
Quoted 6 lines · On Monday 09 November 2015 17:08:34 Russell King - ARM Linux wrote:

They are "optional" because when you're using a DSA switch, you don't specify a PHY (because, there isn't one). For example, this is what I'm using with an Armada 388 board with a Marvell DSA switch. The DSA does not appear as a PHY, and no node in the DSA stanza can be referenced for a phy entry in the ethernet device's stanza.

Quoted 11 lines
                        eth1: ethernet at 30000 {
                                compatible = "marvell,armada-370-neta";
                                reg = <0x30000 0x4000>;
                                interrupts-extended = <&mpic 10>;
                                clocks = <&gateclk 3>;
                                managed = "in-band-status";
                                phy-mode = "sgmii";
                                status = "okay";
                        };

Ok, then it would be nice to change the binding to reflect that, and also document the "managed" property there.

Arnd

#6 lore raw
Quoted 19 lines · Arnd Bergmann
Quoted 6 lines · On Monday 09 November 2015 17:08:34 Russell King - ARM Linux wrote:

They are "optional" because when you're using a DSA switch, you don't specify a PHY (because, there isn't one). For example, this is what I'm using with an Armada 388 board with a Marvell DSA switch. The DSA does not appear as a PHY, and no node in the DSA stanza can be referenced for a phy entry in the ethernet device's stanza.

Quoted 11 lines
                        eth1: ethernet at 30000 {
                                compatible = "marvell,armada-370-neta";
                                reg = <0x30000 0x4000>;
                                interrupts-extended = <&mpic 10>;
                                clocks = <&gateclk 3>;
                                managed = "in-band-status";
                                phy-mode = "sgmii";
                                status = "okay";
                        };
Quoted 2 lines

Ok, then it would be nice to change the binding to reflect that, and also document the "managed" property there.

"managed" is already documented. See Documentation/devicetree/bindings/net/ethernet.txt:

- managed: string, specifies the PHY management type. Supported values are:
"auto", "in-band-status". "auto" is the default, it usess MDIO for management if fixed-link is not specified.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#7 lore raw
In reply to Arnd Bergmann
Quoted 9 lines
Quoted 2 lines

I suppose it comes down to, are we allowed to optionally implement part of the DT binding?

I'm not sure what you are asking. A lot of DT bindings have both optional and mandatory properties. For mvneta, the "phy" and "phy-mode" properties are listed as mandatory, so the driver can safely assume that they are always present. If there are reasons to leave them out, and for the driver to handle that case correctly, the binding should be updated to mark them as optional.

Hi Arnd

You are looking at it from the perspective of the driver. I was meaning from the perspective of the DT blob. Can be blob assume the driver implements all of the binding, all of the time?

You use fixed-phy when the MAC is connected to a switch, not a phy. Or when the MAC is connected to an SFP module. The driver can currently be built to not implement the fixed-phy party of the binding. Is that O.K. from the perspective of the DT blob? Or should the driver always implement all of the binding, in which these NOP stubs should be removed and fixed phy always be enabled for the drivers that use it.

Andrew

#8 lore raw
Quoted 10 lines · On Monday 09 November 2015 18:08:49 Andrew Lunn wrote:
Quoted 9 lines
Quoted 2 lines

I suppose it comes down to, are we allowed to optionally implement part of the DT binding?

I'm not sure what you are asking. A lot of DT bindings have both optional and mandatory properties. For mvneta, the "phy" and "phy-mode" properties are listed as mandatory, so the driver can safely assume that they are always present. If there are reasons to leave them out, and for the driver to handle that case correctly, the binding should be updated to mark them as optional.

Quoted 5 lines

Hi Arnd

You are looking at it from the perspective of the driver. I was meaning from the perspective of the DT blob. Can be blob assume the driver implements all of the binding, all of the time?

That question is not really relevant: the DT describes the hardware, it doesn't matter whether there are drivers for all the bits or whether all properties are read.

Quoted 6 lines

You use fixed-phy when the MAC is connected to a switch, not a phy. Or when the MAC is connected to an SFP module. The driver can currently be built to not implement the fixed-phy party of the binding. Is that O.K. from the perspective of the DT blob? Or should the driver always implement all of the binding, in which these NOP stubs should be removed and fixed phy always be enabled for the drivers that use it.

Sure, that is ok.

Arnd

#9 lore raw
In reply to Andrew Lunn
Quoted 2 lines · Andrew Lunn

You use fixed-phy when the MAC is connected to a switch, not a phy. Or when the MAC is connected to an SFP module.

... hopefully not for much longer. Once -rc1 is out, I'll sort out posting my phylink and SFP module hotplug support as a RFC.

Signature · 2 lines
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.