| Message ID | 20240909124342.2838263-2-o.rempel@pengutronix.de (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: phy: Support master-slave config via device tree | expand |
On 9/9/24 05:43, Oleksij Rempel wrote: > Introduce a new `master-slave` string property in the ethernet-phy > binding to specify the link role for Single Pair Ethernet > (1000/100/10Base-T1) PHYs. This property supports the values > `forced-master` and `forced-slave`, which allow the PHY to operate in a > predefined role, necessary when hardware strap pins are unavailable or > wrongly set. > > Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> > --- > changes v2: > - use string property instead of multiple flags > --- > .../devicetree/bindings/net/ethernet-phy.yaml | 14 ++++++++++++++ > 1 file changed, 14 insertions(+) > > diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > index d9b62741a2259..025e59f6be6f3 100644 > --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml > +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > @@ -158,6 +158,20 @@ properties: > Mark the corresponding energy efficient ethernet mode as > broken and request the ethernet to stop advertising it. > > + master-slave: > + $ref: /schemas/types.yaml#/definitions/string > + enum: > + - forced-master > + - forced-slave > + description: | > + Specifies the predefined link role for the PHY in Single Pair Ethernet > + (1000/100/10Base-T1). This property is required for setups where the link > + role must be assigned by the device tree due to limitations in using > + hardware strap pins. Nit: the way this is implemented right now, this is also applicable to 1000BaseT which is fine, just needs to be called out explicitly IMHO.
On Mon, Sep 09, 2024 at 02:43:40PM +0200, Oleksij Rempel wrote: > Introduce a new `master-slave` string property in the ethernet-phy > binding to specify the link role for Single Pair Ethernet > (1000/100/10Base-T1) PHYs. This property supports the values > `forced-master` and `forced-slave`, which allow the PHY to operate in a > predefined role, necessary when hardware strap pins are unavailable or > wrongly set. > > Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> > --- > changes v2: > - use string property instead of multiple flags > --- > .../devicetree/bindings/net/ethernet-phy.yaml | 14 ++++++++++++++ > 1 file changed, 14 insertions(+) > > diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > index d9b62741a2259..025e59f6be6f3 100644 > --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml > +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > @@ -158,6 +158,20 @@ properties: > Mark the corresponding energy efficient ethernet mode as > broken and request the ethernet to stop advertising it. > > + master-slave: Outdated terminology and kind of vague what it is for... The usual transformation to 'controller-device' would not make much sense though. I think a better name would be "spe-link-role" or "spe-link-mode". > + $ref: /schemas/types.yaml#/definitions/string > + enum: > + - forced-master > + - forced-slave > + description: | > + Specifies the predefined link role for the PHY in Single Pair Ethernet > + (1000/100/10Base-T1). This property is required for setups where the link > + role must be assigned by the device tree due to limitations in using > + hardware strap pins. > + > + - 'forced-master': The PHY is forced to operate as a master. > + - 'forced-slave': The PHY is forced to operate as a slave. > + > pses: > $ref: /schemas/types.yaml#/definitions/phandle-array > maxItems: 1 > -- > 2.39.2 >
On Mon, Sep 09, 2024 at 11:20:09AM -0500, Rob Herring wrote: > On Mon, Sep 09, 2024 at 02:43:40PM +0200, Oleksij Rempel wrote: > > Introduce a new `master-slave` string property in the ethernet-phy > > binding to specify the link role for Single Pair Ethernet > > (1000/100/10Base-T1) PHYs. This property supports the values > > `forced-master` and `forced-slave`, which allow the PHY to operate in a > > predefined role, necessary when hardware strap pins are unavailable or > > wrongly set. > > > > Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> > > --- > > changes v2: > > - use string property instead of multiple flags > > --- > > .../devicetree/bindings/net/ethernet-phy.yaml | 14 ++++++++++++++ > > 1 file changed, 14 insertions(+) > > > > diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > index d9b62741a2259..025e59f6be6f3 100644 > > --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > @@ -158,6 +158,20 @@ properties: > > Mark the corresponding energy efficient ethernet mode as > > broken and request the ethernet to stop advertising it. > > > > + master-slave: > > Outdated terminology and kind of vague what it is for... > > The usual transformation to 'controller-device' would not make much > sense though. I think a better name would be "spe-link-role" or > "spe-link-mode". This applies to more than Single Pair Ethernet. This property could also be used for 2 and 4 pair cables. So spe-link-mode would be wrong. Also: https://grouper.ieee.org/groups/802/3/dc/comments/P8023_D2p0_comments_final_by_cls.pdf On 3 December 2020, the IEEE SA Standard Board passed the following resolution. (See <https://standards.ieee.org/about/sasb/resolutions.html>.) "IEEE standards (including recommended practices and guides) shall be written in such a way as to unambiguously communicate the technical necessities, preferences, and options of the standard to best enable market adoption, conformity assessment, interoperability, and other technical aspirations of the developing standards committee. IEEE standards should be written in such a way as to avoid non-inclusive and insensitive terminology (see IEEE Policy 9.27) and other deprecated terminology (see clause 10 of the IEEE SA Style Manual) except when required by safety, legal, regulatory, and other similar considerations. Terms such as master/slave, blacklist, and whitelist should be avoided." In IEEE Std 802.3, 1000BASE-T, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, and MultiGBASE-T PHYs use the terms "master" and "slave" to indicate whether the clock is derived from an external source or from the received signal. In these cases, the terms appear in the text, figures, state names, variable names, register/bit names, etc. A direct substitution of terms will create disconnects between the standard and the documentation for devices in the field (e.g., the register interface) and also risks the introduction of technical errors. Note that "master" and "slave" are also occasionally used to describe the relationship between an ONT and an ONU for EPON and between a CNT and a CNU for EPoC. The approach that other IEEE standards are taking to address this issue have been considered. For example, IEEE P1588g proposes to define "optional alternative suitable and inclusive terminology" but not replace the original terms. (See <https://development.standards.ieee.org/myproject-web/public/view.html#pardetail/8858>.) It is understood that an annex to the IEEE 1588 standard has been proposed that defines the inclusive terminology. It is also understood that the inclusive terminology has been chosen to be "leader" and "follower". The IEEE P802.1ASdr project proposes to align to the IEEE P1588g inclusive terminology. (See <https://development.standards.ieee.org/myprojectweb/public/view.html#pardetail/9009>.) Based on this, it seems reasonable to include an annex that defines optional alternative inclusive terminology and, for consistency, to use the terms "leader" and "follower" as the inclusive terminology The 2022 revision of 802.3 still has master/slave when describing the registers, but it does have Annex K as described above saying "leader" and "follower" are optional substitutions. The Linux code has not changed, and the uAPI has not changed. It seems like the best compromise would be to allow both 'force-master' and 'force-leader', as well as 'force-slave' and 'force-follower', and a reference to 802.3 Annex K. As to you comment about it being unclear what it means i would suggest a reference to 802.3 section 1.4.389: 1.4.389 master Physical Layer device (PHY): Within IEEE 802.3, in a 100BASE-T2, 1000BASE-T, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, or any MultiGBASE-T link containing a pair of PHYs, the PHY that uses an external clock for generating its clock signals to determine the timing of transmitter and receiver operations. It also uses the master transmit scrambler generator polynomial for side-stream scrambling. Master and slave PHY status is determined during the Auto-Negotiation process that takes place prior to establishing the transmission link, or in the case of a PHY where Auto-Negotiation is optional and not used, master and slave PHY status Andrew
On Mon, Sep 9, 2024 at 12:00 PM Andrew Lunn <andrew@lunn.ch> wrote: > > On Mon, Sep 09, 2024 at 11:20:09AM -0500, Rob Herring wrote: > > On Mon, Sep 09, 2024 at 02:43:40PM +0200, Oleksij Rempel wrote: > > > Introduce a new `master-slave` string property in the ethernet-phy > > > binding to specify the link role for Single Pair Ethernet > > > (1000/100/10Base-T1) PHYs. This property supports the values > > > `forced-master` and `forced-slave`, which allow the PHY to operate in a > > > predefined role, necessary when hardware strap pins are unavailable or > > > wrongly set. > > > > > > Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> > > > --- > > > changes v2: > > > - use string property instead of multiple flags > > > --- > > > .../devicetree/bindings/net/ethernet-phy.yaml | 14 ++++++++++++++ > > > 1 file changed, 14 insertions(+) > > > > > > diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > > index d9b62741a2259..025e59f6be6f3 100644 > > > --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > > +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml > > > @@ -158,6 +158,20 @@ properties: > > > Mark the corresponding energy efficient ethernet mode as > > > broken and request the ethernet to stop advertising it. > > > > > > + master-slave: > > > > Outdated terminology and kind of vague what it is for... > > > > The usual transformation to 'controller-device' would not make much > > sense though. I think a better name would be "spe-link-role" or > > "spe-link-mode". > > This applies to more than Single Pair Ethernet. This property could > also be used for 2 and 4 pair cables. So spe-link-mode would be wrong. I kind of figured that... Propose something that's not just duplicating possible values. > > Also: > > https://grouper.ieee.org/groups/802/3/dc/comments/P8023_D2p0_comments_final_by_cls.pdf > > On 3 December 2020, the IEEE SA Standard Board passed the following resolution. (See > <https://standards.ieee.org/about/sasb/resolutions.html>.) > > "IEEE standards (including recommended practices and guides) shall > be written in such a way as to unambiguously communicate the > technical necessities, preferences, and options of the standard to > best enable market adoption, conformity assessment, > interoperability, and other technical aspirations of the developing > standards committee. IEEE standards should be written in such a way > as to avoid non-inclusive and insensitive terminology (see IEEE > Policy 9.27) and other deprecated terminology (see clause 10 of the > IEEE SA Style Manual) except when required by safety, legal, > regulatory, and other similar considerations. Terms such as > master/slave, blacklist, and whitelist should be avoided." > > In IEEE Std 802.3, 1000BASE-T, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, > and MultiGBASE-T PHYs use the terms "master" and "slave" to indicate > whether the clock is derived from an external source or from the > received signal. In these cases, the terms appear in the text, > figures, state names, variable names, register/bit names, etc. A > direct substitution of terms will create disconnects between the > standard and the documentation for devices in the field (e.g., the > register interface) and also risks the introduction of technical > errors. Note that "master" and "slave" are also occasionally used to > describe the relationship between an ONT and an ONU for EPON and > between a CNT and a CNU for EPoC. > > The approach that other IEEE standards are taking to address this > issue have been considered. For example, IEEE P1588g proposes to > define "optional alternative suitable and inclusive terminology" but > not replace the original terms. (See > <https://development.standards.ieee.org/myproject-web/public/view.html#pardetail/8858>.) > It is understood that an annex to the IEEE 1588 standard has been > proposed that defines the inclusive terminology. It is also > understood that the inclusive terminology has been chosen to be > "leader" and "follower". > > The IEEE P802.1ASdr project proposes to align to the IEEE P1588g > inclusive terminology. (See > <https://development.standards.ieee.org/myprojectweb/public/view.html#pardetail/9009>.) > Based on this, it seems reasonable to include an annex that defines > optional alternative inclusive terminology and, for consistency, to > use the terms "leader" and "follower" as the inclusive terminology > > The 2022 revision of 802.3 still has master/slave when describing the > registers, but it does have Annex K as described above saying "leader" > and "follower" are optional substitutions. > > The Linux code has not changed, and the uAPI has not changed. It seems > like the best compromise would be to allow both 'force-master' and > 'force-leader', as well as 'force-slave' and 'force-follower', and a > reference to 802.3 Annex K. It seems silly to maintain both forever. I'd rather have one or the other than both. > As to you comment about it being unclear what it means i would suggest > a reference to 802.3 section 1.4.389: > > 1.4.389 master Physical Layer device (PHY): Within IEEE 802.3, in a > 100BASE-T2, 1000BASE-T, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, or any > MultiGBASE-T link containing a pair of PHYs, the PHY that uses an > external clock for generating its clock signals to determine the > timing of transmitter and receiver operations. It also uses the > master transmit scrambler generator polynomial for side-stream > scrambling. Master and slave PHY status is determined during the > Auto-Negotiation process that takes place prior to establishing the > transmission link, or in the case of a PHY where Auto-Negotiation is > optional and not used, master and slave PHY status phy-status? Shrug. Another thought. Is it possible that h/w strapping disables auto-neg, but you actually want to override that and force auto-neg? Rob
On Tue, Sep 10, 2024 at 11:54:04AM -0500, Rob Herring wrote: > On Mon, Sep 9, 2024 at 12:00 PM Andrew Lunn <andrew@lunn.ch> wrote: .... > > The 2022 revision of 802.3 still has master/slave when describing the > > registers, but it does have Annex K as described above saying "leader" > > and "follower" are optional substitutions. > > > > The Linux code has not changed, and the uAPI has not changed. It seems > > like the best compromise would be to allow both 'force-master' and > > 'force-leader', as well as 'force-slave' and 'force-follower', and a > > reference to 802.3 Annex K. > > It seems silly to maintain both forever. I'd rather have one or the > other than both. I'll accept what ever will be decided. Even if we will decide to use words 'leader' and 'follower' for the properties, we still need to document that they correspond to 'master' and 'slave' in the IEEE specification and in the kernel UAPI. I can imagine a mdi-link-clock-role = 'leader' or 'follower' > > As to you comment about it being unclear what it means i would suggest > > a reference to 802.3 section 1.4.389: > > > > 1.4.389 master Physical Layer device (PHY): Within IEEE 802.3, in a > > 100BASE-T2, 1000BASE-T, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, or any > > MultiGBASE-T link containing a pair of PHYs, the PHY that uses an > > external clock for generating its clock signals to determine the > > timing of transmitter and receiver operations. It also uses the > > master transmit scrambler generator polynomial for side-stream > > scrambling. Master and slave PHY status is determined during the > > Auto-Negotiation process that takes place prior to establishing the > > transmission link, or in the case of a PHY where Auto-Negotiation is > > optional and not used, master and slave PHY status > > phy-status? Shrug. In my understanding, the 'status' is result not actual configuration request. > Another thought. Is it possible that h/w strapping disables auto-neg, > but you actually want to override that and force auto-neg? Yes. If I would need it, I would recommend to have a different override property, something like: autoneg-default = 'on' or 'off' Regards, Oleksij
> It seems silly to maintain both forever. I'd rather have one or the > other than both. It currently seems like 802.3 is going to keep with master/slave in the body of the text. And they don't even have to deal with breaking backwards compatibility. So i suggest we keep with master/slave, but comment that an annex of the standard proposes alternative names of leader/follower. But don't actually accept them. > > > As to you comment about it being unclear what it means i would suggest > > a reference to 802.3 section 1.4.389: > > > > 1.4.389 master Physical Layer device (PHY): Within IEEE 802.3, in a > > 100BASE-T2, 1000BASE-T, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, or any > > MultiGBASE-T link containing a pair of PHYs, the PHY that uses an > > external clock for generating its clock signals to determine the > > timing of transmitter and receiver operations. It also uses the > > master transmit scrambler generator polynomial for side-stream > > scrambling. Master and slave PHY status is determined during the > > Auto-Negotiation process that takes place prior to establishing the > > transmission link, or in the case of a PHY where Auto-Negotiation is > > optional and not used, master and slave PHY status > > phy-status? Shrug. phy-status is too generic. Maybe 'timing-role' ? > > Another thought. Is it possible that h/w strapping disables auto-neg, > but you actually want to override that and force auto-neg? Autoneg can be used for a bunch of parameters. In automotive situations, it is generally disabled and those parameters are forced. In more tradition settings those parameters are negotiated. However, even with autoneg enabled, you can force each individual parameter, rather than negotiate it. So we would need a DT parameter about autoneg in general. And then a DT parameter about 'timing-role', where force-master/force-slave means don't negotiate, and prefer-master/prefer-slave means do negotiate with the given preference. Andrew
diff --git a/Documentation/devicetree/bindings/net/ethernet-phy.yaml b/Documentation/devicetree/bindings/net/ethernet-phy.yaml index d9b62741a2259..025e59f6be6f3 100644 --- a/Documentation/devicetree/bindings/net/ethernet-phy.yaml +++ b/Documentation/devicetree/bindings/net/ethernet-phy.yaml @@ -158,6 +158,20 @@ properties: Mark the corresponding energy efficient ethernet mode as broken and request the ethernet to stop advertising it. + master-slave: + $ref: /schemas/types.yaml#/definitions/string + enum: + - forced-master + - forced-slave + description: | + Specifies the predefined link role for the PHY in Single Pair Ethernet + (1000/100/10Base-T1). This property is required for setups where the link + role must be assigned by the device tree due to limitations in using + hardware strap pins. + + - 'forced-master': The PHY is forced to operate as a master. + - 'forced-slave': The PHY is forced to operate as a slave. + pses: $ref: /schemas/types.yaml#/definitions/phandle-array maxItems: 1
Introduce a new `master-slave` string property in the ethernet-phy binding to specify the link role for Single Pair Ethernet (1000/100/10Base-T1) PHYs. This property supports the values `forced-master` and `forced-slave`, which allow the PHY to operate in a predefined role, necessary when hardware strap pins are unavailable or wrongly set. Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de> --- changes v2: - use string property instead of multiple flags --- .../devicetree/bindings/net/ethernet-phy.yaml | 14 ++++++++++++++ 1 file changed, 14 insertions(+)