| Message ID | 20240906144905.591508-1-o.rempel@pengutronix.de |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | [v1] dt-bindings: net: ethernet-phy: Add forced-master/slave properties for SPE PHYs | expand |
| Context | Check | Description |
|---|---|---|
| robh/checkpatch | success | |
| robh/patch-applied | success | |
| robh/dtbs-check | warning | build log |
| robh/dt-meta-schema | success |
On Fri, Sep 06, 2024 at 04:49:05PM +0200, Oleksij Rempel wrote: > Add two new properties, `forced-master` and `forced-slave`, to the > ethernet-phy binding. These properties are intended for Single Pair > Ethernet (1000/100/10Base-T1) PHYs, where each PHY and product may have > a predefined link role (master or slave). Typically, these roles are set > by hardware strap pins, but in some cases, device tree configuration is > necessary. > > Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> > --- > .../devicetree/bindings/net/ethernet-phy.yaml | 22 +++++++++++++++++++ > 1 file changed, 22 insertions(+) > > diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > index d9b62741a2259..af7a1eb6ceff6 100644 > --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml > +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > @@ -158,6 +158,28 @@ properties: > Mark the corresponding energy efficient ethernet mode as > broken and request the ethernet to stop advertising it. > > + forced-master: > + $ref: /schemas/types.yaml#/definitions/flag > + description: > + If set, forces the PHY to operate as a master. This is used in Single Pair > + Ethernet (1000/100/10Base-T1) where each PHY and product has a predefined > + link role (master or slave). This property is board-specific, as the role > + is usually configured by strap pins but can be set through the device tree > + if needed. > + This property is mutually exclusive with 'forced-slave'; only one of them > + should be used. DT reviewers tend to complain about such mutually exclusive properties. What you are effectively adding is support for the ethtool: ethtool -s [master-slave preferred-master|preferred-slave|forced-master|forced-slave] 10Base-T1 often does not have autoneg, so preferred-master & preferred-slave make non sense in this context, but i wounder if somebody will want these later. An Ethernet switch is generally preferred-master for example, but the client is preferred-slave. Maybe make the property a string with supported values 'forced-master' and 'forced-slave', leaving it open for the other two to be added later. I've not seen the implementation yet, but i don't think there is much driver specific here. We already have phydev->master_slave_set, it just needs to be set from this property. Can it be done in phylib core somewhere? Andrew
On Fri, Sep 06, 2024 at 05:11:54PM +0200, Andrew Lunn wrote: > On Fri, Sep 06, 2024 at 04:49:05PM +0200, Oleksij Rempel wrote: > > Add two new properties, `forced-master` and `forced-slave`, to the > > ethernet-phy binding. These properties are intended for Single Pair > > Ethernet (1000/100/10Base-T1) PHYs, where each PHY and product may have > > a predefined link role (master or slave). Typically, these roles are set > > by hardware strap pins, but in some cases, device tree configuration is > > necessary. > > > > Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> > > --- > > .../devicetree/bindings/net/ethernet-phy.yaml | 22 +++++++++++++++++++ > > 1 file changed, 22 insertions(+) > > > > diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > index d9b62741a2259..af7a1eb6ceff6 100644 > > --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > @@ -158,6 +158,28 @@ properties: > > Mark the corresponding energy efficient ethernet mode as > > broken and request the ethernet to stop advertising it. > > > > + forced-master: > > + $ref: /schemas/types.yaml#/definitions/flag > > + description: > > + If set, forces the PHY to operate as a master. This is used in Single Pair > > + Ethernet (1000/100/10Base-T1) where each PHY and product has a predefined > > + link role (master or slave). This property is board-specific, as the role > > + is usually configured by strap pins but can be set through the device tree > > + if needed. > > + This property is mutually exclusive with 'forced-slave'; only one of them > > + should be used. > > DT reviewers tend to complain about such mutually exclusive > properties. Yes, at this point i was uncertain. > What you are effectively adding is support for the ethtool: > > ethtool -s [master-slave preferred-master|preferred-slave|forced-master|forced-slave] ack > 10Base-T1 often does not have autoneg, so preferred-master & > preferred-slave make non sense in this context, but i wounder if > somebody will want these later. An Ethernet switch is generally > preferred-master for example, but the client is preferred-slave. Good point. > Maybe make the property a string with supported values 'forced-master' > and 'forced-slave', leaving it open for the other two to be added > later. > > I've not seen the implementation yet, but i don't think there is much > driver specific here. We already have phydev->master_slave_set, it > just needs to be set from this property. Can it be done in phylib core > somewhere? Yes, this is the idea.
On Fri, 6 Sep 2024 17:11:54 +0200 Andrew Lunn <andrew@lunn.ch> wrote: [...] > 10Base-T1 often does not have autoneg, so preferred-master & > preferred-slave make non sense in this context, but i wounder if > somebody will want these later. An Ethernet switch is generally > preferred-master for example, but the client is preferred-slave. > > Maybe make the property a string with supported values 'forced-master' > and 'forced-slave', leaving it open for the other two to be added > later. My two cents, don't take it as a nack or any strong disagreement, my experience with SPE is still limited. I agree that for SPE, it's required that PHYs get their role assigned as early as possible, otherwise the link can't establish. I don't see any other place but DT to put that info, as this would be required for say, booting over the network. This to me falls under 'HW representation', as we could do the same with straps. However for preferred-master / preferred-slave, wouldn't we be crossing the blurry line of "HW description => system configuration in the DT" ? Maxime
> > 10Base-T1 often does not have autoneg, so preferred-master & > > preferred-slave make non sense in this context, but i wounder if > > somebody will want these later. An Ethernet switch is generally > > preferred-master for example, but the client is preferred-slave. > > > > Maybe make the property a string with supported values 'forced-master' > > and 'forced-slave', leaving it open for the other two to be added > > later. > > My two cents, don't take it as a nack or any strong disagreement, my > experience with SPE is still limited. I agree that for SPE, it's > required that PHYs get their role assigned as early as possible, > otherwise the link can't establish. I don't see any other place but DT > to put that info, as this would be required for say, booting over the > network. This to me falls under 'HW representation', as we could do the > same with straps. > > However for preferred-master / preferred-slave, wouldn't we be crossing > the blurry line of "HW description => system configuration in the DT" ? Yes, we are somewhere near the blurry line. This is why i gave the example of an Ethernet switch, vs a client. Again, it could be done with straps, so following your argument, it could be considered HW representation. But if it is set wrong, it probably does not matter, auto-neg should still work. Except for a very small number of PHYs whos random numbers are not random... But this is also something we don't actually need to resolve now. The design allows for it, but we don't really need to decided if it is acceptable until somebody actually posts a patch. Andrew
On 9/6/24 09:11, Andrew Lunn wrote: >>> 10Base-T1 often does not have autoneg, so preferred-master & >>> preferred-slave make non sense in this context, but i wounder if >>> somebody will want these later. An Ethernet switch is generally >>> preferred-master for example, but the client is preferred-slave. >>> >>> Maybe make the property a string with supported values 'forced-master' >>> and 'forced-slave', leaving it open for the other two to be added >>> later. >> >> My two cents, don't take it as a nack or any strong disagreement, my >> experience with SPE is still limited. I agree that for SPE, it's >> required that PHYs get their role assigned as early as possible, >> otherwise the link can't establish. I don't see any other place but DT >> to put that info, as this would be required for say, booting over the >> network. This to me falls under 'HW representation', as we could do the >> same with straps. >> >> However for preferred-master / preferred-slave, wouldn't we be crossing >> the blurry line of "HW description => system configuration in the DT" ? > > Yes, we are somewhere near the blurry line. This is why i gave the > example of an Ethernet switch, vs a client. Again, it could be done > with straps, so following your argument, it could be considered HW > representation. But if it is set wrong, it probably does not matter, > auto-neg should still work. Except for a very small number of PHYs > whos random numbers are not random... Having had to deal with an Ethernet PHY that requires operating in slave mode "preferably" in order to have a correct RXC duty cycle, if you force both sides of the link to "slave", auto-negotiation will fail, however thanks to auto-negotiation you can tell that there was a master/slave resolution failure. (This reminds me I need to send the patch for that PHY errata at some point). In the case that Oleksij seems to be after, there is no auto-negotiation (is that correct?), so it seems to me that the Device Tree is coming to the rescue of an improperly strapped HW, and is used as a way to change the default HW configuration so as to have a fighting chance of having a functional link. That is not unprecedented, but it is definitively a bit blurry...
On Fri, Sep 06, 2024 at 09:22:29AM -0700, Florian Fainelli wrote: > On 9/6/24 09:11, Andrew Lunn wrote: > > > > 10Base-T1 often does not have autoneg, so preferred-master & > > > > preferred-slave make non sense in this context, but i wounder if > > > > somebody will want these later. An Ethernet switch is generally > > > > preferred-master for example, but the client is preferred-slave. > > > > > > > > Maybe make the property a string with supported values 'forced-master' > > > > and 'forced-slave', leaving it open for the other two to be added > > > > later. > > > > > > My two cents, don't take it as a nack or any strong disagreement, my > > > experience with SPE is still limited. I agree that for SPE, it's > > > required that PHYs get their role assigned as early as possible, > > > otherwise the link can't establish. I don't see any other place but DT > > > to put that info, as this would be required for say, booting over the > > > network. This to me falls under 'HW representation', as we could do the > > > same with straps. > > > > > > However for preferred-master / preferred-slave, wouldn't we be crossing > > > the blurry line of "HW description => system configuration in the DT" ? > > > > Yes, we are somewhere near the blurry line. This is why i gave the > > example of an Ethernet switch, vs a client. Again, it could be done > > with straps, so following your argument, it could be considered HW > > representation. But if it is set wrong, it probably does not matter, > > auto-neg should still work. Except for a very small number of PHYs > > whos random numbers are not random... > > Having had to deal with an Ethernet PHY that requires operating in slave > mode "preferably" in order to have a correct RXC duty cycle, if you force > both sides of the link to "slave", auto-negotiation will fail, however > thanks to auto-negotiation you can tell that there was a master/slave > resolution failure. (This reminds me I need to send the patch for that PHY > errata at some point). > > In the case that Oleksij seems to be after, there is no auto-negotiation (is > that correct?), so it seems to me that the Device Tree is coming to the > rescue of an improperly strapped HW, and is used as a way to change the > default HW configuration so as to have a fighting chance of having a > functional link. That is not unprecedented, but it is definitively a bit > blurry... Yes, there is no auto-negotiation on T1 PHY variants, so the DT property is to fix broken or not existing for some reason HW straps.
diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml index d9b62741a2259..af7a1eb6ceff6 100644 --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml @@ -158,6 +158,28 @@ properties: Mark the corresponding energy efficient ethernet mode as broken and request the ethernet to stop advertising it. + forced-master: + $ref: /schemas/types.yaml#/definitions/flag + description: + If set, forces the PHY to operate as a master. This is used in Single Pair + Ethernet (1000/100/10Base-T1) where each PHY and product has a predefined + link role (master or slave). This property is board-specific, as the role + is usually configured by strap pins but can be set through the device tree + if needed. + This property is mutually exclusive with 'forced-slave'; only one of them + should be used. + + forced-slave: + $ref: /schemas/types.yaml#/definitions/flag + description: + If set, forces the PHY to operate as a slave. This is used in Single Pair + Ethernet (1000/100/10Base-T1) where each PHY and product has a predefined + link role (master or slave). This property is board-specific, as the role + is usually configured by strap pins but can be set through the device tree + if needed. + This property is mutually exclusive with 'forced-master'; only one of them + should be used. + pses: $ref: /schemas/types.yaml#/definitions/phandle-array maxItems: 1
Add two new properties, `forced-master` and `forced-slave`, to the ethernet-phy binding. These properties are intended for Single Pair Ethernet (1000/100/10Base-T1) PHYs, where each PHY and product may have a predefined link role (master or slave). Typically, these roles are set by hardware strap pins, but in some cases, device tree configuration is necessary. Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> --- .../devicetree/bindings/net/ethernet-phy.yaml | 22 +++++++++++++++++++ 1 file changed, 22 insertions(+) -- 2.39.2