Skip to contentSkip to conversation

net: imx: Provide support for L2 switch as switchdev accelerator

RFC In review Lukasz Majewski 5 participants 36 messages Jun 22, 2021 – Jun 29, 2021

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

cover net: imx: Provide support for L2 switch as switchdev accelerator

This patch series is a followup for the earlier effort [1] to bring support for L2 switch IP block on some NXP devices.

This time it augment the fec driver, so the L2 switch is treated as a HW network accelerator. This is minimal, yet functional driver, which enables bridging between imx28 ENET-MAC ports.

Links:
[1] - https://lwn.net/ml/linux-kernel/[email protected]/

Lukasz Majewski (3):
ARM: dts: imx28: Add description for L2 switch on XEA board
net: Provide switchdev driver for NXP's More Than IP L2 switch
net: imx: Adjust fec_main.c to provide support for L2 switch

 arch/arm/boot/dts/imx28-xea.dts               |  42 ++
 drivers/net/ethernet/freescale/Kconfig        |   1 +
 drivers/net/ethernet/freescale/Makefile       |   1 +
 drivers/net/ethernet/freescale/fec.h          |  36 ++
 drivers/net/ethernet/freescale/fec_main.c     | 139 +++++-
 drivers/net/ethernet/freescale/mtipsw/Kconfig |  11 +
 .../net/ethernet/freescale/mtipsw/Makefile    |   3 +
 .../net/ethernet/freescale/mtipsw/fec_mtip.c  | 438 ++++++++++++++++++
 .../net/ethernet/freescale/mtipsw/fec_mtip.h  | 213 +++++++++
 9 files changed, 860 insertions(+), 24 deletions(-)
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/Kconfig
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/Makefile
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/fec_mtip.c
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/fec_mtip.h
Signature · 1 line
2.20.1
#1 lore raw

RFC 1/3 ARM: dts: imx28: Add description for L2 switch on XEA board

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

---
 arch/arm/boot/dts/imx28-xea.dts | 42 +++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)
arch/arm/boot/dts/imx28-xea.dts +42 -0
index d649822febed..94ff801669c4 100644
@@ -23,6 +23,48 @@
 	status = "okay";
 };
 
+&mac0 {
+	pinctrl-names = "default";
+	pinctrl-0 = <&mac0_pins_a>;
+	phy-mode = "rmii";
+	phy-supply = <&reg_fec_3v3>;
+	phy-reset-gpios = <&gpio2 13 0>;
+	phy-reset-duration = <100>;
+	local-mac-address = [ 00 11 22 AA BB CC ];
+	status = "okay";
+};
+
+&mac1 {
+	pinctrl-names = "default";
+	pinctrl-0 = <&mac1_pins_a>;
+	phy-mode = "rmii";
+	phy-supply = <&reg_fec_3v3>;
+	local-mac-address = [ 00 11 22 AA BB DD ];
+	status = "okay";
+};
+
+&eth_switch {
+	compatible = "imx,mtip-l2switch";
+	reg = <0x800f8000 0x400>, <0x800fC000 0x4000>,  <0x800f8400 0x400>;
+
+	interrupts = <100>;
+	status = "okay";
+
+	ports {
+		port1@1 {
+			reg = <1>;
+			label = "eth0";
+			phy-handle = <&mac0>;
+		};
+
+		port2@2 {
+			reg = <2>;
+			label = "eth1";
+			phy-handle = <&mac1>;
+		};
+	};
+};
+
 &pinctrl {
 	pinctrl-names = "default";
 	pinctrl-0 = <&hog_pins_a &hog_pins_tiva>;
Signature · 1 line
2.20.1
#2 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Quoted 5 lines · Lukasz Majewski

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

A phy-handle points to a phy, not a MAC! Don't abuse a well known DT property like this.

Andrew

#3 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Andrew,

Quoted 9 lines · Lukasz Majewski
Quoted 5 lines

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

A phy-handle points to a phy, not a MAC! Don't abuse a well known DT property like this.

Ach.... You are right. I will change it.

Probably 'ethernet' property or 'link' will fit better?

Quoted 3 lines

Andrew

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#4 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Quoted 9 lines · Lukasz Majewski

Hi Andrew,

Quoted 7 lines · Lukasz Majewski
Quoted 5 lines

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

Quoted 6 lines
Quoted 2 lines

A phy-handle points to a phy, not a MAC! Don't abuse a well known DT property like this.

Ach.... You are right. I will change it.

Probably 'ethernet' property or 'link' will fit better?

You should first work on the overall architecture. I suspect you will end up with something more like the DSA binding, and not have the FEC nodes at all. Maybe the MDIO busses will appear under the switch?

Please don't put minimal changes to the FEC driver has your first goal. We want an architecture which is similar to other switchdev drivers. Maybe look at drivers/net/ethernet/ti/cpsw_new.c. The cpsw driver has an interesting past, it did things the wrong way for a long time, but the new switchdev driver has an architecture similar to what the FEC driver could be like.

Andrew

#5 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Andrew,

Quoted 17 lines · Lukasz Majewski
Quoted 9 lines

Hi Andrew,

Quoted 7 lines · Lukasz Majewski
Quoted 5 lines

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

Quoted 6 lines
Quoted 2 lines

A phy-handle points to a phy, not a MAC! Don't abuse a well known DT property like this.

Ach.... You are right. I will change it.

Probably 'ethernet' property or 'link' will fit better?

Quoted 7 lines

You should first work on the overall architecture. I suspect you will end up with something more like the DSA binding, and not have the FEC nodes at all. Maybe the MDIO busses will appear under the switch?

Please don't put minimal changes to the FEC driver has your first goal. We want an architecture which is similar to other switchdev drivers. Maybe look at drivers/net/ethernet/ti/cpsw_new.c.

I'm a bit confused - as I thought that with switchdev API I could just extend the current FEC driver to add bridge offload.
This patch series shows that it is doable with little changes introduced.

However, now it looks like I would need to replace FEC driver and rewrite it in a way similar to cpsw_new.c, so the switchdev could be used for both cases - with and without L2 switch offload.

This would be probably conceptually correct, but i.MX FEC driver has several issues to tackle:

- On some SoCs (vf610, imx287, etc.) the ENET-MAC ports don't have the same capabilities (eth1 is a bit special)

- Without switch we need to use DMA0 and DMA1 in the "bypass" switch mode (default). When switch is enabled we only use DMA0. The former case is best fitted with FEC driver instantiation. The latter with DSA or switchdev.

Quoted 6 lines

The cpsw
driver has an interesting past, it did things the wrong way for a long time, but the new switchdev driver has an architecture similar to what the FEC driver could be like.

Andrew

Maybe somebody from NXP can provide input to this discussion - for example to sched some light on FEC driver (near) future.

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#6 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Quoted 23 lines · Lukasz Majewski

Hi Andrew,

Quoted 21 lines · Lukasz Majewski
Quoted 9 lines

Hi Andrew,

Quoted 7 lines · Lukasz Majewski
Quoted 5 lines

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

Quoted 6 lines
Quoted 2 lines

A phy-handle points to a phy, not a MAC! Don't abuse a well known DT property like this.

Ach.... You are right. I will change it.

Probably 'ethernet' property or 'link' will fit better?

You should first work on the overall architecture. I suspect you will end up with something more like the DSA binding, and not have the FEC nodes at all. Maybe the MDIO busses will appear under the switch?

Quoted 8 lines
Quoted 3 lines

Please don't put minimal changes to the FEC driver has your first goal. We want an architecture which is similar to other switchdev drivers. Maybe look at drivers/net/ethernet/ti/cpsw_new.c.

I'm a bit confused - as I thought that with switchdev API I could just extend the current FEC driver to add bridge offload.
This patch series shows that it is doable with little changes introduced.

Regardless of how you end up implementing the switching part in the driver, one thing that you can use is the same DT binding as what DSA uses as far as representing ports of the Ethernet controller. That means that ports should ideally be embedded into an 'ethernet-ports' container node, and you describe each port individually as sub-nodes and provide, when appropriate 'phy-handle' and 'phy-mode' properties to describe how the Ethernet PHYs are connected.

Quoted 21 lines

However, now it looks like I would need to replace FEC driver and rewrite it in a way similar to cpsw_new.c, so the switchdev could be used for both cases - with and without L2 switch offload.

This would be probably conceptually correct, but i.MX FEC driver has several issues to tackle:

- On some SoCs (vf610, imx287, etc.) the ENET-MAC ports don't have the same capabilities (eth1 is a bit special)

- Without switch we need to use DMA0 and DMA1 in the "bypass" switch mode (default). When switch is enabled we only use DMA0. The former case is best fitted with FEC driver instantiation. The latter with DSA or switchdev.

Quoted 5 lines

The cpsw
driver has an interesting past, it did things the wrong way for a long time, but the new switchdev driver has an architecture similar to what the FEC driver could be like.

Quoted 4 lines
Quoted 1 line

Andrew

Maybe somebody from NXP can provide input to this discussion - for example to sched some light on FEC driver (near) future.

Seems like some folks at NXP are focusing on the STMMAC controller these days (dwmac from Synopsys), so maybe they have given up on having their own Ethernet MAC for lower end products.

Signature · 1 line
Florian
#7 lore raw

RE: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Lukasz, Florian, Andrew,

Quoted 6 lines
Quoted 2 lines

Maybe somebody from NXP can provide input to this discussion - for example to sched some light on FEC driver (near) future.

Seems like some folks at NXP are focusing on the STMMAC controller these days (dwmac from Synopsys), so maybe they have given up on having their own Ethernet MAC for lower end products.

I am very happy to take participate into this topic, but now I have no experience to DSA and i.MX28 MAC, so I may need some time to increase these knowledge, limited insight could be put to now.

Florian, Andrew could comment more and I also can learn from it :-), they are all very experienced expert.

We also want to maintain FEC driver since many SoCs implemented this IP, and as I know we would also use it for future SoCs.

Best Regards,
Joakim Zhang

#8 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Joakim,

Quoted 5 lines

Hi Lukasz, Florian, Andrew,

Quoted 3 lines
Quoted 2 lines

Maybe somebody from NXP can provide input to this discussion - for example to sched some light on FEC driver (near) future.

Quoted 7 lines
Quoted 3 lines

Seems like some folks at NXP are focusing on the STMMAC controller these days (dwmac from Synopsys), so maybe they have given up on having their own Ethernet MAC for lower end products.

I am very happy to take participate into this topic, but now I have no experience to DSA and i.MX28 MAC, so I may need some time to increase these knowledge, limited insight could be put to now.

Ok. No problem :-)

Quoted 3 lines

Florian, Andrew could comment more and I also can learn from it :-), they are all very experienced expert.

The main purpose of several RFCs for the L2 switch drivers (for DSA [1] and switchdev [2]) was to gain feedback from community as soon as possible (despite that the driver lacks some features - like VLAN, FDB, etc).

Quoted 4 lines

We also want to maintain FEC driver since many SoCs implemented this IP, and as I know we would also use it for future SoCs.

Florian, Andrew, please correct me if I'm wrong, but my impression is that upstreaming the support for L2 switch on iMX depends on FEC driver being rewritten to support switchdev?

If yes, then unfortunately, I don't have time and resources to perform that task - that is why I have asked if NXP has any plans to update the FEC (fec_main.c) driver.

Joakim, do you have any plans to re-factor the legacy FEC driver (fec_main.c) and introduce new one, which would support the switchdev?

If NXP is not planning to update the driver, then maybe it would be worth to consider adding driver from [2] to mainline? Then I could finish it and provide all required features.

Links:
[1] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-DSA-RFC_v1
[2] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-switchdev-RFC_v1

Quoted 2 lines

Best Regards,
Joakim Zhang

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#9 lore raw

RE: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Lukasz,

Quoted 53 lines

-----Original Message-----
From: Lukasz Majewski <[email protected]>
Sent: 2021年6月24日 19:21
To: Joakim Zhang <[email protected]>; Florian Fainelli <[email protected]>; Andrew Lunn <[email protected]>
Cc: David S . Miller <[email protected]>; Jakub Kicinski <[email protected]>; Madalin Bucur (OSS) <[email protected]>; Nicolas Ferre <[email protected]>; Vladimir Oltean <[email protected]>; [email protected]; Arnd Bergmann <[email protected]>; Mark Einon <[email protected]>; dl-linux-imx <[email protected]>; [email protected]
Subject: Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Joakim,

Quoted 5 lines

Hi Lukasz, Florian, Andrew,

Quoted 3 lines
Quoted 2 lines

Maybe somebody from NXP can provide input to this discussion - for example to sched some light on FEC driver (near) future.

Quoted 7 lines
Quoted 3 lines

Seems like some folks at NXP are focusing on the STMMAC controller these days (dwmac from Synopsys), so maybe they have given up on having their own Ethernet MAC for lower end products.

I am very happy to take participate into this topic, but now I have no experience to DSA and i.MX28 MAC, so I may need some time to increase these knowledge, limited insight could be put to now.

Ok. No problem :-)

Quoted 3 lines

Florian, Andrew could comment more and I also can learn from it :-), they are all very experienced expert.

The main purpose of several RFCs for the L2 switch drivers (for DSA [1] and switchdev [2]) was to gain feedback from community as soon as possible (despite that the driver lacks some features - like VLAN, FDB, etc).

Quoted 4 lines

We also want to maintain FEC driver since many SoCs implemented this IP, and as I know we would also use it for future SoCs.

Florian, Andrew, please correct me if I'm wrong, but my impression is that upstreaming the support for L2 switch on iMX depends on FEC driver being rewritten to support switchdev?

If yes, then unfortunately, I don't have time and resources to perform that task
- that is why I have asked if NXP has any plans to update the FEC (fec_main.c) driver.

Quoted 6 lines

Joakim, do you have any plans to re-factor the legacy FEC driver (fec_main.c) and introduce new one, which would support the switchdev?

If NXP is not planning to update the driver, then maybe it would be worth to consider adding driver from [2] to mainline? Then I could finish it and provide all required features.

I don't have such plan now, and have no confidence to re-factor the legacy FEC driver and introduce new one, which to support switchdev in a short time. I am not very experienced for FEC driver, since I have just maintained it for half a year. To be honest, I have no idea in my head right now, we even don't have i.MX28 boards. I'm so sorry about this, but I am also interested in it, I am finding time to increase related knowledge.

Best Regards,
Joakim Zhang

Quoted 24 lines

Links:
[1] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-u
pstream-DSA-RFC_v1
[2] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-u
pstream-switchdev-RFC_v1

Quoted 2 lines

Best Regards,
Joakim Zhang

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#10 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Joakim, Andrew,

Quoted 63 lines

Hi Lukasz,

Quoted 61 lines

-----Original Message-----
From: Lukasz Majewski <[email protected]>
Sent: 2021年6月24日 19:21
To: Joakim Zhang <[email protected]>; Florian Fainelli <[email protected]>; Andrew Lunn <[email protected]>
Cc: David S . Miller <[email protected]>; Jakub Kicinski <[email protected]>; Madalin Bucur (OSS) <[email protected]>; Nicolas Ferre <[email protected]>; Vladimir Oltean <[email protected]>; [email protected]; Arnd Bergmann <[email protected]>; Mark Einon <[email protected]>; dl-linux-imx <[email protected]>; [email protected]
Subject: Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

Hi Joakim,

Quoted 6 lines

Hi Lukasz, Florian, Andrew,

Quoted 4 lines
Quoted 3 lines

Maybe somebody from NXP can provide input to this discussion
- for example to sched some light on FEC driver (near) future.

Quoted 9 lines
Quoted 4 lines

Seems like some folks at NXP are focusing on the STMMAC controller these days (dwmac from Synopsys), so maybe they have given up on having their own Ethernet MAC for lower end products.

I am very happy to take participate into this topic, but now I have no experience to DSA and i.MX28 MAC, so I may need some time to increase these knowledge, limited insight could be put to now.

Ok. No problem :-)

Quoted 3 lines

Florian, Andrew could comment more and I also can learn from it :-), they are all very experienced expert.

The main purpose of several RFCs for the L2 switch drivers (for DSA
[1] and switchdev [2]) was to gain feedback from community as soon as possible (despite that the driver lacks some features - like VLAN, FDB, etc).

Quoted 4 lines

We also want to maintain FEC driver since many SoCs implemented this IP, and as I know we would also use it for future SoCs.

Florian, Andrew, please correct me if I'm wrong, but my impression is that upstreaming the support for L2 switch on iMX depends on FEC driver being rewritten to support switchdev?

If yes, then unfortunately, I don't have time and resources to perform that task
- that is why I have asked if NXP has any plans to update the FEC (fec_main.c) driver.

Joakim, do you have any plans to re-factor the legacy FEC driver (fec_main.c) and introduce new one, which would support the switchdev?

Quoted 7 lines
Quoted 3 lines

If NXP is not planning to update the driver, then maybe it would be worth to consider adding driver from [2] to mainline? Then I could finish it and provide all required features.

I don't have such plan now, and have no confidence to re-factor the legacy FEC driver and introduce new one, which to support switchdev in a short time.

Thanks for the clear statement, appreciated.

Quoted 2 lines

I am not very experienced for FEC driver, since I
have just maintained it for half a year.

Ok. No problem.

Quoted 2 lines

To be honest, I have no idea
in my head right now, we even don't have i.MX28 boards.

As fair as I remember there is still imx28-dev board available for purchase. You can also use vf610 based board.

Quoted 3 lines

I'm so sorry
about this, but I am also interested in it, I am finding time to increase related knowledge.

Ok.

To sum up:

- The FEC driver (legacy one) will not be rewritten anytime soon (maybe any other community member will work on this sooner...)

- Considering the above, support for L2 switch on imx28, vf610 is blocked [*]. As a result some essential functionality for still actively used SoCs is going to be maintained out of tree (for example [1][2]).

[*] - as I've stated in the other mail - what's about the situation where FEC legacy driver is not going to be excessively modified (just changes from this patch set)?

Links:

[1] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-switchdev-RFC_v1

[2] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-DSA-RFC_v1

Quoted 28 lines

Best Regards,
Joakim Zhang

Quoted 25 lines

Links:
[1] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-u
pstream-DSA-RFC_v1
[2] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-u
pstream-switchdev-RFC_v1

Quoted 2 lines

Best Regards,
Joakim Zhang

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email:
[email protected]

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#11 lore raw

Re: [RFC 1/3] ARM: dts: imx28: Add description for L2 switch on XEA board

In reply to Florian Fainelli
Quoted 35 lines · On Wed, 23 Jun 2021 17:36:59 -0700 Florian Fainelli <[email protected]>…
Quoted 25 lines · Lukasz Majewski

Hi Andrew,

Quoted 23 lines · Lukasz Majewski
Quoted 10 lines

Hi Andrew,

Quoted 8 lines · Lukasz Majewski
Quoted 6 lines

The 'eth_switch' node is now extendfed to enable support for L2 switch.

Moreover, the mac[01] nodes are defined as well and linked to the former with 'phy-handle' property.

Quoted 6 lines
Quoted 2 lines

A phy-handle points to a phy, not a MAC! Don't abuse a well known DT property like this.

Ach.... You are right. I will change it.

Probably 'ethernet' property or 'link' will fit better?

You should first work on the overall architecture. I suspect you will end up with something more like the DSA binding, and not have the FEC nodes at all. Maybe the MDIO busses will appear under the switch?

Quoted 8 lines
Quoted 3 lines

Please don't put minimal changes to the FEC driver has your first goal. We want an architecture which is similar to other switchdev drivers. Maybe look at drivers/net/ethernet/ti/cpsw_new.c.

I'm a bit confused - as I thought that with switchdev API I could just extend the current FEC driver to add bridge offload. This patch series shows that it is doable with little changes introduced.

Quoted 7 lines

Regardless of how you end up implementing the switching part in the driver, one thing that you can use is the same DT binding as what DSA uses as far as representing ports of the Ethernet controller. That means that ports should ideally be embedded into an 'ethernet-ports' container node, and you describe each port individually as sub-nodes and provide, when appropriate 'phy-handle' and 'phy-mode' properties to describe how the Ethernet PHYs are connected.

I see. Thanks for the explanation.

Quoted 22 lines
Quoted 22 lines

However, now it looks like I would need to replace FEC driver and rewrite it in a way similar to cpsw_new.c, so the switchdev could be used for both cases - with and without L2 switch offload.

This would be probably conceptually correct, but i.MX FEC driver has several issues to tackle:

- On some SoCs (vf610, imx287, etc.) the ENET-MAC ports don't have the same capabilities (eth1 is a bit special)

- Without switch we need to use DMA0 and DMA1 in the "bypass" switch mode (default). When switch is enabled we only use DMA0. The
former case is best fitted with FEC driver instantiation. The latter with DSA or switchdev.

Quoted 5 lines

The cpsw
driver has an interesting past, it did things the wrong way for a long time, but the new switchdev driver has an architecture similar to what the FEC driver could be like.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

Andrew

Maybe somebody from NXP can provide input to this discussion - for example to sched some light on FEC driver (near) future.

Seems like some folks at NXP are focusing on the STMMAC controller these days (dwmac from Synopsys), so maybe they have given up on having their own Ethernet MAC for lower end products.

For example, the imx287 SoC is still active product and is supposed to be produced for at least 15 years after release.

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#12 lore raw

RFC 2/3 net: Provide switchdev driver for NXP's More Than IP L2 switch

In reply to Lukasz Majewski

This change provides driver for More Than IP L2 switch IP block to hook into switchdev subsystem to provide bridging offloading for i.MX28 SoC (imx287 to be precise).

This code is responsible for configuring this device as L2 bridge when one decides to bridge eth[01] interfaces (so no vlan, filtering table aging supported).

This driver shall be regarded as a complementary one for NXP's FEC (fec_main.c). It reuses some code from it.

---
 drivers/net/ethernet/freescale/Kconfig        |   1 +
 drivers/net/ethernet/freescale/Makefile       |   1 +
 drivers/net/ethernet/freescale/mtipsw/Kconfig |  11 +
 .../net/ethernet/freescale/mtipsw/Makefile    |   3 +
 .../net/ethernet/freescale/mtipsw/fec_mtip.c  | 438 ++++++++++++++++++
 .../net/ethernet/freescale/mtipsw/fec_mtip.h  | 213 +++++++++
 6 files changed, 667 insertions(+)
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/Kconfig
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/Makefile
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/fec_mtip.c
 create mode 100644 drivers/net/ethernet/freescale/mtipsw/fec_mtip.h
drivers/net/ethernet/freescale/Kconfig +1 -0
index 3f9175bdce77..3cf703aa2a00 100644
@@ -102,5 +102,6 @@ config GIANFAR
 source "drivers/net/ethernet/freescale/dpaa/Kconfig"
 source "drivers/net/ethernet/freescale/dpaa2/Kconfig"
 source "drivers/net/ethernet/freescale/enetc/Kconfig"
+source "drivers/net/ethernet/freescale/mtipsw/Kconfig"
 
 endif # NET_VENDOR_FREESCALE
drivers/net/ethernet/freescale/Makefile +1 -0
index 67c436400352..12ed0c13f739 100644
@@ -27,3 +27,4 @@ obj-$(CONFIG_FSL_DPAA2_ETH) += dpaa2/
 obj-$(CONFIG_FSL_ENETC) += enetc/
 obj-$(CONFIG_FSL_ENETC_MDIO) += enetc/
 obj-$(CONFIG_FSL_ENETC_VF) += enetc/
+obj-$(CONFIG_FEC_MTIP_L2SW) += mtipsw/
drivers/net/ethernet/freescale/mtipsw/Kconfig new file +11 -0
new file mode 100644
index 000000000000..e7f40dcad0d0
@@ -0,0 +1,11 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config FEC_MTIP_L2SW
+	tristate "MoreThanIP L2 switch support to FEC driver"
+	depends on OF
+	depends on NET_SWITCHDEV
+	depends on BRIDGE
+	depends on ARCH_MXS
+	select FIXED_PHY
+	help
+	  This enables support for the MoreThan IP on i.MX SoCs (e.g. iMX28)
+	  L2 switch.
drivers/net/ethernet/freescale/mtipsw/Makefile new file +3 -0
new file mode 100644
index 000000000000..1aac85f92750
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_FEC_MTIP_L2SW) += fec_mtip.o
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +438 -0

438 lines not on this page · raw message

drivers/net/ethernet/freescale/mtipsw/fec_mtip.h new file +213 -0

213 lines not on this page · raw message

Signature · 1 line
2.20.1
#13 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 53

Quoted 7 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +7 -0
+static void write_atable(struct mtipl2sw_priv *priv, int index,
+	unsigned long write_lo, unsigned long write_hi)
+{
+	unsigned long atable_base = (unsigned long)priv->hwentry;
+
+	writel(write_lo, (volatile void *)atable_base + (index << 3));
+	writel(write_hi, (volatile void *)atable_base + (index << 3) + 4);

Using volatile is generally wrong. Why do you need it?

Quoted 80 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +80 -0
+}
+
+/*
+ * Clear complete MAC Look Up Table
+ */
+static void esw_clear_atable(struct mtipl2sw_priv *priv)
+{
+	int index;
+	for (index = 0; index < 2048; index++)
+		write_atable(priv, index, 0, 0);
+}
+
+static int mtipl2_sw_enable(struct mtipl2sw_priv *priv)
+{
+	/*
+	 * L2 switch - reset
+	 */
+	writel(MCF_ESW_MODE_SW_RST, &priv->fecp->ESW_MODE);
+	udelay(10);
+
+	/* Configure switch*/
+	writel(MCF_ESW_MODE_STATRST, &priv->fecp->ESW_MODE);
+	writel(MCF_ESW_MODE_SW_EN, &priv->fecp->ESW_MODE);
+
+	/* Management port configuration, make port 0 as management port */
+	writel(0, &priv->fecp->ESW_BMPC);
+
+	/*
+	 * Set backpressure threshold to minimize discarded frames
+	 * during due to congestion.
+	 */
+	writel(P0BC_THRESHOLD, &priv->fecp->ESW_P0BCT);
+
+	/* Set the max rx buffer size */
+	writel(L2SW_PKT_MAXBLR_SIZE, priv->hwpsw + MCF_ESW_R_BUFF_SIZE);
+	/* Enable broadcast on all ports */
+	writel(0x7, &priv->fecp->ESW_DBCR);
+
+	/* Enable multicast on all ports */
+	writel(0x7, &priv->fecp->ESW_DMCR);
+
+	esw_clear_atable(priv);
+
+	/* Clear all pending interrupts */
+	writel(0xffffffff, priv->hwpsw + FEC_IEVENT);
+
+	/* Enable interrupts we wish to service */
+	writel(FEC_MTIP_DEFAULT_IMASK, priv->hwpsw + FEC_IMASK);
+	priv->l2sw_on = true;
+
+	return 0;
+}
+
+static void mtipl2_sw_disable(struct mtipl2sw_priv *priv)
+{
+	writel(0, &priv->fecp->ESW_MODE);
+}
+
+static int mtipl2_port_enable (struct mtipl2sw_priv *priv, int port)
+{
+	u32 l2_ports_en;
+
+	pr_err("%s: PORT ENABLE %d\n", __func__, port);
+
+	/* Enable tx/rx on L2 switch ports */
+	l2_ports_en = readl(&priv->fecp->ESW_PER);
+	if (!(l2_ports_en & MCF_ESW_ENA_PORT_0))
+		l2_ports_en = MCF_ESW_ENA_PORT_0;
+
+	if (port == 0 && !(l2_ports_en & MCF_ESW_ENA_PORT_1))
+		l2_ports_en |= MCF_ESW_ENA_PORT_1;
+
+	if (port == 1 && !(l2_ports_en & MCF_ESW_ENA_PORT_2))
+		l2_ports_en |= MCF_ESW_ENA_PORT_2;
+
+	writel(l2_ports_en, &priv->fecp->ESW_PER);
+
+	/*
+	 * Check if MAC IP block is already enabled (after switch initializtion)
+	 * or if we need to enable it after mtipl2_port_disable was called.

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 143

Quoted 10 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +10 -0
+	 */
+
+	return 0;
+}
+
+static void mtipl2_port_disable (struct mtipl2sw_priv *priv, int port)
+{
+	u32 l2_ports_en;
+
+	pr_err(" %s: PORT DISABLE %d\n", __func__, port);

Please clean up debug code this this.

Quoted 13 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +13 -0
+
+	l2_ports_en = readl(&priv->fecp->ESW_PER);
+	if (port == 0)
+		l2_ports_en &= ~MCF_ESW_ENA_PORT_1;
+
+	if (port == 1)
+		l2_ports_en &= ~MCF_ESW_ENA_PORT_2;
+
+	/* Finally disable tx/rx on port 0 */
+	if (!(l2_ports_en & MCF_ESW_ENA_PORT_1) &&
+	    !(l2_ports_en & MCF_ESW_ENA_PORT_2))
+		l2_ports_en &= ~MCF_ESW_ENA_PORT_0;
+

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 166

Quoted 10 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +10 -0
+	writel(l2_ports_en, &priv->fecp->ESW_PER);
+}
+
+irqreturn_t
+mtip_l2sw_interrupt_handler(int irq, void *dev_id)
+{
+	struct mtipl2sw_priv *priv = dev_id;
+	struct fec_enet_private *fep = priv->fep[0];
+	irqreturn_t ret = IRQ_NONE;
+	u32 int_events, int_imask;

Reverse christmas tree.

Quoted 25 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +25 -0
+
+	int_events = readl(fec_hwp(fep) + FEC_IEVENT);
+	writel(int_events, fec_hwp(fep) + FEC_IEVENT);
+
+	if ((int_events & FEC_MTIP_DEFAULT_IMASK) && fep->link) {
+		ret = IRQ_HANDLED;
+
+		if (napi_schedule_prep(&fep->napi)) {
+			/* Disable RX interrupts */
+			int_imask = readl(fec_hwp(fep) + FEC_IMASK);
+			int_imask &= ~FEC_MTIP_ENET_RXF;
+			writel(int_imask, fec_hwp(fep) + FEC_IMASK);
+			__napi_schedule(&fep->napi);
+		}
+	}
+
+	return ret;
+}
+
+static int mtipl2_parse_of(struct mtipl2sw_priv *priv, struct device_node *np)
+{
+	struct device_node *port, *p, *fep_np;
+	struct platform_device *pdev;
+	struct net_device *ndev;
+	unsigned int port_num;

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 201

Quoted 10 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +10 -0
+
+	p = of_find_node_by_name(np, "ports");
+
+	for_each_available_child_of_node(p, port) {
+		if (of_property_read_u32(port, "reg", &port_num))
+			continue;
+
+		priv->n_ports = port_num;
+
+		fep_np = of_parse_phandle(port, "phy-handle", 0);

As i said, phy-handle points to a phy. It minimum, you need to call this mac-handle. But that then makes this switch driver very different to every other switch driver.

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 204

Quoted 3 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +3 -0
+		pdev = of_find_device_by_node(fep_np);
+		ndev = platform_get_drvdata(pdev);
+		priv->fep[port_num - 1] = netdev_priv(ndev);

What happens when somebody puts reg=<42>; in DT?

I would say, your basic structure needs to change, to make it more like other switchdev drivers. You need to replace the two FEC device instances with one switchdev driver. The switchdev driver will then instantiate the two netdevs for the two external MACs. You can hopefully reuse some of the FEC code for that, but i expect you are going to have to refactor the FEC driver and turn part of it into a library, which the switchdev driver can then use.

Andrew

#14 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Andrew,

Quoted 11 lines
Quoted 9 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +6 -0
+static void write_atable(struct mtipl2sw_priv *priv, int index,
+	unsigned long write_lo, unsigned long write_hi)
+{
+	unsigned long atable_base = (unsigned long)priv->hwentry;
+
+	writel(write_lo, (volatile void *)atable_base + (index <<

3));

(quoted diff) +1 -0
+	writel(write_hi, (volatile void *)atable_base + (index <<

3) + 4);

Using volatile is generally wrong. Why do you need it?

This was the code, which I took from the legacy driver. I will adjust it.

Quoted 89 lines
Quoted 89 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +25 -0
+}
+
+/*
+ * Clear complete MAC Look Up Table
+ */
+static void esw_clear_atable(struct mtipl2sw_priv *priv)
+{
+	int index;
+	for (index = 0; index < 2048; index++)
+		write_atable(priv, index, 0, 0);
+}
+
+static int mtipl2_sw_enable(struct mtipl2sw_priv *priv)
+{
+	/*
+	 * L2 switch - reset
+	 */
+	writel(MCF_ESW_MODE_SW_RST, &priv->fecp->ESW_MODE);
+	udelay(10);
+
+	/* Configure switch*/
+	writel(MCF_ESW_MODE_STATRST, &priv->fecp->ESW_MODE);
+	writel(MCF_ESW_MODE_SW_EN, &priv->fecp->ESW_MODE);
+
+	/* Management port configuration, make port 0 as

management port */

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +10 -0
+	writel(0, &priv->fecp->ESW_BMPC);
+
+	/*
+	 * Set backpressure threshold to minimize discarded frames
+	 * during due to congestion.
+	 */
+	writel(P0BC_THRESHOLD, &priv->fecp->ESW_P0BCT);
+
+	/* Set the max rx buffer size */
+	writel(L2SW_PKT_MAXBLR_SIZE, priv->hwpsw +

MCF_ESW_R_BUFF_SIZE);

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +24 -0
+	/* Enable broadcast on all ports */
+	writel(0x7, &priv->fecp->ESW_DBCR);
+
+	/* Enable multicast on all ports */
+	writel(0x7, &priv->fecp->ESW_DMCR);
+
+	esw_clear_atable(priv);
+
+	/* Clear all pending interrupts */
+	writel(0xffffffff, priv->hwpsw + FEC_IEVENT);
+
+	/* Enable interrupts we wish to service */
+	writel(FEC_MTIP_DEFAULT_IMASK, priv->hwpsw + FEC_IMASK);
+	priv->l2sw_on = true;
+
+	return 0;
+}
+
+static void mtipl2_sw_disable(struct mtipl2sw_priv *priv)
+{
+	writel(0, &priv->fecp->ESW_MODE);
+}
+
+static int mtipl2_port_enable (struct mtipl2sw_priv *priv, int

port) +{

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +19 -0
+	u32 l2_ports_en;
+
+	pr_err("%s: PORT ENABLE %d\n", __func__, port);
+
+	/* Enable tx/rx on L2 switch ports */
+	l2_ports_en = readl(&priv->fecp->ESW_PER);
+	if (!(l2_ports_en & MCF_ESW_ENA_PORT_0))
+		l2_ports_en = MCF_ESW_ENA_PORT_0;
+
+	if (port == 0 && !(l2_ports_en & MCF_ESW_ENA_PORT_1))
+		l2_ports_en |= MCF_ESW_ENA_PORT_1;
+
+	if (port == 1 && !(l2_ports_en & MCF_ESW_ENA_PORT_2))
+		l2_ports_en |= MCF_ESW_ENA_PORT_2;
+
+	writel(l2_ports_en, &priv->fecp->ESW_PER);
+
+	/*
+	 * Check if MAC IP block is already enabled (after switch

initializtion)

(quoted diff) +1 -0
+	 * or if we need to enable it after mtipl2_port_disable

was called.

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +4 -0
+	 */
+
+	return 0;
+}
Quoted 9 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 143

Quoted 6 lines
(quoted diff) +2 -0
+
+static void mtipl2_port_disable (struct mtipl2sw_priv *priv, int

port) +{

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +3 -0
+	u32 l2_ports_en;
+
+	pr_err(" %s: PORT DISABLE %d\n", __func__, port);  

Please clean up debug code this this.

Ok.

Quoted 17 lines
Quoted 17 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +17 -0
+
+	l2_ports_en = readl(&priv->fecp->ESW_PER);
+	if (port == 0)
+		l2_ports_en &= ~MCF_ESW_ENA_PORT_1;
+
+	if (port == 1)
+		l2_ports_en &= ~MCF_ESW_ENA_PORT_2;
+
+	/* Finally disable tx/rx on port 0 */
+	if (!(l2_ports_en & MCF_ESW_ENA_PORT_1) &&
+	    !(l2_ports_en & MCF_ESW_ENA_PORT_2))
+		l2_ports_en &= ~MCF_ESW_ENA_PORT_0;
+
+	writel(l2_ports_en, &priv->fecp->ESW_PER);
+}
+
+irqreturn_t
Quoted 8 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 166

Quoted 6 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +6 -0
+mtip_l2sw_interrupt_handler(int irq, void *dev_id)
+{
+	struct mtipl2sw_priv *priv = dev_id;
+	struct fec_enet_private *fep = priv->fep[0];
+	irqreturn_t ret = IRQ_NONE;
+	u32 int_events, int_imask;  

Reverse christmas tree.

Ok.

Quoted 34 lines
Quoted 34 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +10 -0
+
+	int_events = readl(fec_hwp(fep) + FEC_IEVENT);
+	writel(int_events, fec_hwp(fep) + FEC_IEVENT);
+
+	if ((int_events & FEC_MTIP_DEFAULT_IMASK) && fep->link) {
+		ret = IRQ_HANDLED;
+
+		if (napi_schedule_prep(&fep->napi)) {
+			/* Disable RX interrupts */
+			int_imask = readl(fec_hwp(fep) +

FEC_IMASK);

(quoted diff) +2 -0
+			int_imask &= ~FEC_MTIP_ENET_RXF;
+			writel(int_imask, fec_hwp(fep) +

FEC_IMASK);

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +8 -0
+			__napi_schedule(&fep->napi);
+		}
+	}
+
+	return ret;
+}
+
+static int mtipl2_parse_of(struct mtipl2sw_priv *priv, struct

device_node *np) +{

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +10 -0
+	struct device_node *port, *p, *fep_np;
+	struct platform_device *pdev;
+	struct net_device *ndev;
+	unsigned int port_num;
+
+	p = of_find_node_by_name(np, "ports");
+
+	for_each_available_child_of_node(p, port) {
+		if (of_property_read_u32(port, "reg", &port_num))
+			continue;
Quoted 8 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 201

Quoted 4 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +4 -0
+
+		priv->n_ports = port_num;
+
+		fep_np = of_parse_phandle(port, "phy-handle", 0);  

As i said, phy-handle points to a phy. It minimum, you need to call this mac-handle. But that then makes this switch driver very different to every other switch driver.

Other drivers (DSA for example) use "ethernet" or "link" properties. Maybe those can be reused?

Quoted 6 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 204

Quoted 4 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +3 -0
+		pdev = of_find_device_by_node(fep_np);
+		ndev = platform_get_drvdata(pdev);
+		priv->fep[port_num - 1] = netdev_priv(ndev);  

What happens when somebody puts reg=<42>; in DT?

I do guess that this will break the code.

However, DSA DT descriptions also rely on the exact numbering [1] (via e.g. reg property) of the ports. I've followed this paradigm.

Quoted 4 lines

I would say, your basic structure needs to change, to make it more like other switchdev drivers. You need to replace the two FEC device instances with one switchdev driver.

I've used the cpsw_new.c as the example.

Quoted 2 lines

The switchdev driver will then
instantiate the two netdevs for the two external MACs.

Then there is a question - what about eth[01], which already exists?

Shall I close them and then reuse (create as a new one?) eth0 to be connected to switch port0 (via DMA0)?

Then, I do need two net_device ports, which would only control PHY device and setup ENET-MAC for rmii, as the L2 switch will provide data for transmission. Those two ports are connected to switch's port[12] and look very similar to ports created by DSA driver (but shall not transmit and receive data).

Maybe I've overlooked something, but the rocker switchdev driver (rocker_main.c) sets netdev_ops (with .ndo_start_xmit) for ports which it creates. The prestera's prestera_sdma_xmit() (in prestera_rxtx.c) also setups the SDMA for those.

In i.MX L2 switch case - one just needs to setup DMA0 to send data to engress ports. IMHO, those ports just need to handle PHY link (up/down 10/100 change) and statistics.

Quoted 4 lines

You can
hopefully reuse some of the FEC code for that, but i expect you are going to have to refactor the FEC driver and turn part of it into a library, which the switchdev driver can then use.

To be honest - such driver for L2 switch already has been forward ported by me [2] to v4.19.

It has the L2 switch enabled after the boot and there is no way to disable it.

When you look on the code - it is a copy-paste of the FEC driver, with some necessary adjustments.

The FEC driver itself is large and used by almost _ALL_ i.MX SoCs. Turning it into library will move the already working code around. I wanted to avoid it.

The idea behind this patch series is as follows (to offload bridging to MTIP L2 switch IP block):
-------------------------
1. When bridge is created disable eth0 and eth1 (fec_close)

2. Set a flag for fec driver, so DMA descriptors registers and ones to initiate transfer are adjusted for DMA0 (eth0) device. Also L2 switch IP block has different bits positions for interrupts.

3. DMA1 - which with normal setup corresponds to eth1 - is not used.

4. FEC driver also monitors PHY changes (link up/down speed change 10/100) (for both eth0, eth1).

5. When br0 is deleted - the mtip support flag is cleared and FEC network interfaces (eth[01]) are opened again (via fec_open).

With above approach many operations are already performed - like ENET-MAC setup, buffers allocation, etc.
To make L2 switch working with this setup we need to re-map 4 registers and two interrupt bits in fec_main driver.

I'm just wondering if is it worth to refactor already working driver to library, instantiate new interfaces and re-init all the already initialized stuff ?

Quoted 2 lines

Andrew

Links:

[1] -
https://elixir.bootlin.com/linux/latest/source/arch/arm/boot/dts/armada-388-clearfog.dts#L93

[2] - https://github.com/lmajewski/linux-imx28-l2switch/commits/master

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#15 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 4 lines
Quoted 1 line

Using volatile is generally wrong. Why do you need it?

This was the code, which I took from the legacy driver. I will adjust it.

It is called 'vendor crap' for a reason.

Quoted 12 lines
Quoted 11 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 201

Quoted 7 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +7 -0
+	for_each_available_child_of_node(p, port) {
+		if (of_property_read_u32(port, "reg", &port_num))
+			continue;
+
+		priv->n_ports = port_num;
+
+		fep_np = of_parse_phandle(port, "phy-handle", 0);  

As i said, phy-handle points to a phy. It minimum, you need to call this mac-handle. But that then makes this switch driver very different to every other switch driver.

Quoted 2 lines

Other drivers (DSA for example) use "ethernet" or "link" properties. Maybe those can be reused?

Not really. They have well known meanings and they are nothing like what you are trying to do. You need a new name. Maybe 'mac-handle'?

Quoted 10 lines
Quoted 5 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 204

Quoted 3 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +3 -0
+		pdev = of_find_device_by_node(fep_np);
+		ndev = platform_get_drvdata(pdev);
+		priv->fep[port_num - 1] = netdev_priv(ndev);  

What happens when somebody puts reg=<42>; in DT?

I do guess that this will break the code.

However, DSA DT descriptions also rely on the exact numbering [1] (via e.g. reg property) of the ports. I've followed this paradigm.

DSA does a range check:

        for_each_available_child_of_node(ports, port) {
                err = of_property_read_u32(port, "reg", &reg);
                if (err)
                        goto out_put_node;

                if (reg >= ds->num_ports) {
                        dev_err(ds->dev, "port %pOF index %u exceeds num_ports (%zu)\n",
                                port, reg, ds->num_ports);
                        err = -EINVAL;
                        goto out_put_node;
                }
Quoted 10 lines
Quoted 3 lines

I would say, your basic structure needs to change, to make it more like other switchdev drivers. You need to replace the two FEC device instances with one switchdev driver.

I've used the cpsw_new.c as the example.

Quoted 2 lines

The switchdev driver will then
instantiate the two netdevs for the two external MACs.

Then there is a question - what about eth[01], which already exists?

They don't exist. cpsw_new is not used at the same time as cpsw, it replaces it. This new driver would replace the FEC driver. cpsw_new makes use of some of the code in the cpsw driver to implement two netdevs. This new FEC switch driver would do the same, make use of some of the low level code, e.g. for DMA access, MDIO bus, etc.

Quoted 2 lines

To be honest - such driver for L2 switch already has been forward ported by me [2] to v4.19.

Which is fine, you can do whatever you want in your own fork. But for mainline, we need a clean architecture. I'm not convinced your code is that clean, and how well future features can be added. Do you have support for VLANS? Adding and removing entries to the lookup tables? How will IGMP snooping work? How will STP work?

Andrew

#16 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Andrew,

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

Using volatile is generally wrong. Why do you need it?

This was the code, which I took from the legacy driver. I will adjust it.

It is called 'vendor crap' for a reason.

:-)

Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
Quoted 10 lines
(quoted diff) +2 -0
+	for_each_available_child_of_node(p, port) {
+		if (of_property_read_u32(port, "reg",

&port_num))

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +5 -0
+			continue;
+
+		priv->n_ports = port_num;
+
+		fep_np = of_parse_phandle(port, "phy-handle",

0);

As i said, phy-handle points to a phy. It minimum, you need to call this mac-handle. But that then makes this switch driver very different to every other switch driver.

Quoted 5 lines
Quoted 2 lines

Other drivers (DSA for example) use "ethernet" or "link" properties. Maybe those can be reused?

Not really. They have well known meanings and they are nothing like what you are trying to do. You need a new name. Maybe 'mac-handle'?

Ok.

Quoted 16 lines
Quoted 9 lines
Quoted 8 lines

drivers/net/ethernet/freescale/mtipsw/fec_mtip.c line 204

Quoted 6 lines
drivers/net/ethernet/freescale/mtipsw/fec_mtip.c new file +3 -0
+		pdev = of_find_device_by_node(fep_np);
+		ndev = platform_get_drvdata(pdev);
+		priv->fep[port_num - 1] = netdev_priv(ndev);

What happens when somebody puts reg=<42>; in DT?

Quoted 4 lines

I do guess that this will break the code.

However, DSA DT descriptions also rely on the exact numbering [1] (via e.g. reg property) of the ports. I've followed this paradigm.

DSA does a range check:

Quoted 12 lines
        for_each_available_child_of_node(ports, port) {
                err = of_property_read_u32(port, "reg", &reg);
                if (err)
                        goto out_put_node;

                if (reg >= ds->num_ports) {
                        dev_err(ds->dev, "port %pOF index %u exceeds
num_ports (%zu)\n", port, reg, ds->num_ports);
                        err = -EINVAL;
                        goto out_put_node;
                }

Ok.

Quoted 6 lines
Quoted 6 lines
Quoted 3 lines

I would say, your basic structure needs to change, to make it more like other switchdev drivers. You need to replace the two FEC device instances with one switchdev driver.

I've used the cpsw_new.c as the example.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

The switchdev driver will then
instantiate the two netdevs for the two external MACs.

Then there is a question - what about eth[01], which already exists?

They don't exist. cpsw_new is not used at the same time as cpsw, it replaces it. This new driver would replace the FEC driver.

I see.

Quoted 4 lines

cpsw_new
makes use of some of the code in the cpsw driver to implement two netdevs. This new FEC switch driver would do the same, make use of some of the low level code, e.g. for DMA access, MDIO bus, etc.

I'm not sure if the imx28 switch is similar to one from TI (cpsw-3g)
- it looks to me that the bypass mode for both seems to be very different. For example, on NXP when switch is disabled we need to handle two DMA[01]. When it is enabled, only one is used. The approach with two DMAs is best handled with FEC driver instantiation.

Quoted 6 lines
Quoted 3 lines

To be honest - such driver for L2 switch already has been forward ported by me [2] to v4.19.

Which is fine, you can do whatever you want in your own fork. But for mainline, we need a clean architecture.

This code is a forward port of vendor's (Freescale) old driver. It uses the _wrong_ approach, but it can (still) be used in production after some adjustments.

Quoted 2 lines

I'm not convinced your code is
that clean,

The code from [2] needs some vendor ioctl based tool (or hardcode) to configure the switch.

Quoted 3 lines

and how well future features can be added. Do you have support for VLANS? Adding and removing entries to the lookup tables? How will IGMP snooping work? How will STP work?

This can be easily added with serving netstack hooks (as it is already done with cpsw_new) in the new switchdev based version [3] (based on v5.12).

Quoted 2 lines

Andrew

Links:

[3] -
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-switchdev-RFC_v1

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#17 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 5 lines

I'm not sure if the imx28 switch is similar to one from TI (cpsw-3g)
- it looks to me that the bypass mode for both seems to be very different. For example, on NXP when switch is disabled we need to handle two DMA[01]. When it is enabled, only one is used. The approach with two DMAs is best handled with FEC driver instantiation.

I don't know if it applies to the FEC, but switches often have registers which control which egress port an ingress port can send packets to. So by default, you allow CPU to port0, CPU to port1, but block between port0 to port1. This would give you two independent interface, the switch enabled, and using one DMA. When the bridge is configured, you simply allow port0 and send/receive packets to/from port1. No change to the DMA setup, etc.

Quoted 2 lines

The code from [2] needs some vendor ioctl based tool (or hardcode) to configure the switch.

This would not be allowed. You configure switches in Linux using the existing user space tools. No vendor tools are used.

Quoted 7 lines
Quoted 3 lines

and how well future features can be added. Do you have support for VLANS? Adding and removing entries to the lookup tables? How will IGMP snooping work? How will STP work?

This can be easily added with serving netstack hooks (as it is already done with cpsw_new) in the new switchdev based version [3] (based on v5.12).

Here i'm less convinced. I expect a fully functioning switch driver is going to need switch specific versions of some of the netdev ops functions, maybe the ethtool ops as well. It is going to want to add devlink ops. By hacking around with the FEC driver in the way you are, you might get very basic switch operation working. But as we have seen with cpsw, going from very basic to a fully functioning switchdev driver required a new driver, cpsw_new. It was getting more and more difficult to add features because its structure was just wrong. We don't want to add code to the kernel which is probably a dead end.

Andrew

#18 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Andrew,

Quoted 7 lines
Quoted 6 lines

I'm not sure if the imx28 switch is similar to one from TI (cpsw-3g)
- it looks to me that the bypass mode for both seems to be very different. For example, on NXP when switch is disabled we need to handle two DMA[01]. When it is enabled, only one is used. The approach with two DMAs is best handled with FEC driver instantiation.

Quoted 7 lines

I don't know if it applies to the FEC, but switches often have registers which control which egress port an ingress port can send packets to. So by default, you allow CPU to port0, CPU to port1, but block between port0 to port1. This would give you two independent interface, the switch enabled, and using one DMA. When the bridge is configured, you simply allow port0 and send/receive packets to/from port1. No change to the DMA setup, etc.

Please correct me if I misunderstood this concept - but it seems like you refer to the use case where the switch is enabled, and by changing it's "allowed internal port's" mapping it decides if frames are passed between engress ports (port1 and port2).

	----------
DMA0 ->	|P0    P1| -> ENET-MAC (PHY control) -> eth0 (lan1)
	|L2 SW	 |
	|      P2| -> ENET-MAC (PHY control) -> eth1 (lan2)
	----------

DMA1 (not used)

We can use this approach when we keep always enabled L2 switch.

However now in FEC we use the "bypass" mode, where:
DMA0 -> ENET-MAC (FEC instance driver 1) -> eth0 DMA1 -> ENET-MAC (FEC instance driver 2) -> eth1

And the "bypass" mode is the default one.

I'm just concerned how we are going to gracefully "switch" between L2 switch and bypass configuration? In this patch series - I used the "hook" corresponding to 'ip link set eth[01] master br0' command.

In other words - how we want to manage DMA0 and DMA1 when switch is enabled and disabled (in "bypass mode").

Quoted 6 lines
Quoted 3 lines

The code from [2] needs some vendor ioctl based tool (or hardcode) to configure the switch.

This would not be allowed. You configure switches in Linux using the existing user space tools. No vendor tools are used.

Exactly - that was the rationale to bring support for L2 switch to mainline kernel.

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines

and how well future features can be added. Do you have support for VLANS? Adding and removing entries to the lookup tables? How will IGMP snooping work? How will STP work?

Quoted 7 lines
Quoted 3 lines

This can be easily added with serving netstack hooks (as it is already done with cpsw_new) in the new switchdev based version [3] (based on v5.12).

Here i'm less convinced. I expect a fully functioning switch driver is going to need switch specific versions of some of the netdev ops functions, maybe the ethtool ops as well.

Definately, the current L2 switch driver would need more work.

Quoted 2 lines

It is going to want to add
devlink ops. By hacking around with the FEC driver

I believe that I will not touch fec_main.[hc] files more than I did in the
"[RFC 3/3] net: imx: Adjust fec_main.c to provide support for L2 switch"

as the switch management (and hooks) are going to be added solely to drivers/net/ethernet/freescale/mtipsw/fec_mtip.[hc]. [*]

This would separate L2 switch driver from the current FEC driver.

Quoted 2 lines

in the way you are,
you might get very basic switch operation working.

Yes, this is the current status - only simple L2 switching works.

Quoted 3 lines

But as we have seen
with cpsw, going from very basic to a fully functioning switchdev driver required a new driver, cpsw_new.

The new driver for L2 switch has been introduced in [*]. The legacy FEC driver will also work without it.

Quoted 4 lines

It was getting more and more
difficult to add features because its structure was just wrong. We don't want to add code to the kernel which is probably a dead end.

I cannot say for sure, but all the switch/bridge related hooks can be added to [*], so fec_main will not bloat.

Quoted 1 line

Andrew

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#19 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 17 lines · Lukasz Majewski

Hi Andrew,

Quoted 7 lines
Quoted 6 lines

I'm not sure if the imx28 switch is similar to one from TI (cpsw-3g)
- it looks to me that the bypass mode for both seems to be very different. For example, on NXP when switch is disabled we need to handle two DMA[01]. When it is enabled, only one is used. The approach with two DMAs is best handled with FEC driver instantiation.

Quoted 7 lines

I don't know if it applies to the FEC, but switches often have registers which control which egress port an ingress port can send packets to. So by default, you allow CPU to port0, CPU to port1, but block between port0 to port1. This would give you two independent interface, the switch enabled, and using one DMA. When the bridge is configured, you simply allow port0 and send/receive packets to/from port1. No change to the DMA setup, etc.

Quoted 4 lines

Please correct me if I misunderstood this concept - but it seems like you refer to the use case where the switch is enabled, and by changing it's "allowed internal port's" mapping it decides if frames are passed between engress ports (port1 and port2).

Correct.

Quoted 8 lines
	----------
DMA0 ->	|P0    P1| -> ENET-MAC (PHY control) -> eth0 (lan1)
	|L2 SW	 |
	|      P2| -> ENET-MAC (PHY control) -> eth1 (lan2)
	----------

DMA1 (not used)

Quoted 7 lines

We can use this approach when we keep always enabled L2 switch.

However now in FEC we use the "bypass" mode, where:
DMA0 -> ENET-MAC (FEC instance driver 1) -> eth0 DMA1 -> ENET-MAC (FEC instance driver 2) -> eth1

And the "bypass" mode is the default one.

Which is not a problem, when you refactor the FEC into a library and a driver, plus add a new switch driver. When the FEC loads, it uses bypass mode, the switch disabled. When the new switch driver loads, it always enables the switch, but disables communication between the two ports until they both join the same bridge.

But i doubt we are actually getting anywhere. You say you don't have time to write a new driver. I'm not convinced you can hack the FEC like you are suggesting and not end up in the mess the cpsw had, before they wrote a new driver.

Andrew

#20 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Andrew,

Quoted 42 lines · Lukasz Majewski
Quoted 19 lines

Hi Andrew,

Quoted 8 lines
Quoted 7 lines

I'm not sure if the imx28 switch is similar to one from TI (cpsw-3g)
- it looks to me that the bypass mode for both seems to be very different. For example, on NXP when switch is disabled we need to handle two DMA[01]. When it is enabled, only one is used. The approach with two DMAs is best handled with FEC driver instantiation.

Quoted 8 lines

I don't know if it applies to the FEC, but switches often have registers which control which egress port an ingress port can send packets to. So by default, you allow CPU to port0, CPU to port1, but block between port0 to port1. This would give you two independent interface, the switch enabled, and using one DMA. When the bridge is configured, you simply allow port0 and send/receive packets to/from port1. No change to the DMA setup, etc.

Quoted 4 lines

Please correct me if I misunderstood this concept - but it seems like you refer to the use case where the switch is enabled, and by changing it's "allowed internal port's" mapping it decides if frames are passed between engress ports (port1 and port2).

Correct.

Quoted 14 lines
	----------
DMA0 ->	|P0    P1| -> ENET-MAC (PHY control) -> eth0 (lan1)
	|L2 SW	 |
	|      P2| -> ENET-MAC (PHY control) -> eth1 (lan2)
	----------

DMA1 (not used)

We can use this approach when we keep always enabled L2 switch.

However now in FEC we use the "bypass" mode, where:
DMA0 -> ENET-MAC (FEC instance driver 1) -> eth0 DMA1 -> ENET-MAC (FEC instance driver 2) -> eth1

Quoted 7 lines
Quoted 1 line

And the "bypass" mode is the default one.

Which is not a problem, when you refactor the FEC into a library and a driver, plus add a new switch driver. When the FEC loads, it uses bypass mode, the switch disabled. When the new switch driver loads, it always enables the switch, but disables communication between the two ports until they both join the same bridge.

Ok, the proposed idea would be to use FEC (refactored) on devices which are not equipped with the switch.

On devices, which have this IP block (like vf610, imx287) we would use the driver with switch enabled and then in switch either bridge or separate the traffic?

Quoted 3 lines

But i doubt we are actually getting anywhere. You say you don't have time to write a new driver.

Yes, I believe that this would be a very time consuming task. Joakim also pointed out that the rewrite from NXP will not happen anytime soon.

Quoted 2 lines

I'm not convinced you can hack the FEC
like you are suggesting

I do believe that I can just extend the L2 switch driver (fec_mtip.c file to be precise) to provide full blown L2 switch functionality without touching the legacy FEC more than in this patch set.

Would you consider applying this patch series then?

Quoted 2 lines

and not end up in the mess the cpsw had,
before they wrote a new driver.

I do see some conceptual differences between those two drivers.

Quoted 2 lines

Andrew

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#21 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 5 lines

I do believe that I can just extend the L2 switch driver (fec_mtip.c file to be precise) to provide full blown L2 switch functionality without touching the legacy FEC more than in this patch set.

Would you consider applying this patch series then?

What is most important is the ABI. If something is merged now, we need to ensure it does not block later refactoring to a clean new driver. The DT binding is considered ABI. So the DT binding needs to be like a traditional switchdev driver. Florian already pointed out, you can use a binding very similar to DSA. ti,cpsw-switch.yaml is another good example.

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Andrew

#22 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Andrew,

Quoted 4 lines
Quoted 4 lines

I do believe that I can just extend the L2 switch driver (fec_mtip.c file to be precise) to provide full blown L2 switch functionality without touching the legacy FEC more than in this patch set.

Quoted 8 lines
Quoted 1 line

Would you consider applying this patch series then?

What is most important is the ABI. If something is merged now, we need to ensure it does not block later refactoring to a clean new driver. The DT binding is considered ABI. So the DT binding needs to be like a traditional switchdev driver. Florian already pointed out, you can use a binding very similar to DSA. ti,cpsw-switch.yaml is another good example.

The best I could get would be:

&eth_switch {
	compatible = "imx,mtip-l2switch";
	reg = <0x800f8000 0x400>, <0x800fC000 0x4000>;

	interrupts = <100>;
	status = "okay";

	ethernet-ports {
		port1@1 {
			reg = <1>;
			label = "eth0";
			phys = <&mac0 0>;
		};

		port2@2 {
			reg = <2>;
			label = "eth1";
			phys = <&mac1 1>;
		};
	};
};

Which would abuse the "phys" properties usages - as 'mac[01]' are referring to ethernet controllers.

On TI SoCs (e.g. am33xx-l4.dtsi) phys refer to some separate driver responsible for PHY management. On NXP this is integrated with FEC driver itself.

Quoted 10 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Andrew

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#23 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 39 lines · Lukasz Majewski

Hi Andrew,

Quoted 4 lines
Quoted 4 lines

I do believe that I can just extend the L2 switch driver (fec_mtip.c file to be precise) to provide full blown L2 switch functionality without touching the legacy FEC more than in this patch set.

Quoted 8 lines
Quoted 1 line

Would you consider applying this patch series then?

What is most important is the ABI. If something is merged now, we need to ensure it does not block later refactoring to a clean new driver. The DT binding is considered ABI. So the DT binding needs to be like a traditional switchdev driver. Florian already pointed out, you can use a binding very similar to DSA. ti,cpsw-switch.yaml is another good example.

The best I could get would be:

&eth_switch {
	compatible = "imx,mtip-l2switch";
	reg = <0x800f8000 0x400>, <0x800fC000 0x4000>;

	interrupts = <100>;
	status = "okay";

	ethernet-ports {
		port1@1 {
			reg = <1>;
			label = "eth0";
			phys = <&mac0 0>;
		};

		port2@2 {
			reg = <2>;
			label = "eth1";
			phys = <&mac1 1>;
		};
	};
};
Quoted 6 lines

Which would abuse the "phys" properties usages - as 'mac[01]' are referring to ethernet controllers.

On TI SoCs (e.g. am33xx-l4.dtsi) phys refer to some separate driver responsible for PHY management. On NXP this is integrated with FEC driver itself.

If we were really honest, the binding would need to be called

port@0 {
	puppet = <&mac0>;
};
port@1 {
	puppet = <&mac1>;
};

which speaks for itself as to why accepting "puppet master" drivers is not really very compelling. I concur with the recommendation given by Andrew and Florian to refactor FEC as a multi-port single driver.

Quoted 10 lines
Quoted 10 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Andrew

I cannot stress enough how important it is for us to see STP support and consequently the ndo_start_xmit procedure for switch ports. Let me see if I understand correctly. When the switch is enabled, eth0 sends packets towards both physical switch ports, and eth1 sends packets towards none, but eth0 handles the link state of switch port 0, and eth1 handles the link state of switch port 1?

#24 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Vladimir,

Quoted 54 lines · Lukasz Majewski
Quoted 40 lines

Hi Andrew,

Quoted 5 lines
Quoted 5 lines

I do believe that I can just extend the L2 switch driver (fec_mtip.c file to be precise) to provide full blown L2 switch functionality without touching the legacy FEC more than in this patch set.

Quoted 8 lines
Quoted 1 line

Would you consider applying this patch series then?

What is most important is the ABI. If something is merged now, we need to ensure it does not block later refactoring to a clean new driver. The DT binding is considered ABI. So the DT binding needs to be like a traditional switchdev driver. Florian already pointed out, you can use a binding very similar to DSA. ti,cpsw-switch.yaml is another good example.

The best I could get would be:

&eth_switch {
	compatible = "imx,mtip-l2switch";
	reg = <0x800f8000 0x400>, <0x800fC000 0x4000>;

	interrupts = <100>;
	status = "okay";

	ethernet-ports {
		port1@1 {
			reg = <1>;
			label = "eth0";
			phys = <&mac0 0>;
		};

		port2@2 {
			reg = <2>;
			label = "eth1";
			phys = <&mac1 1>;
		};
	};
};
Quoted 6 lines

Which would abuse the "phys" properties usages - as 'mac[01]' are referring to ethernet controllers.

On TI SoCs (e.g. am33xx-l4.dtsi) phys refer to some separate driver responsible for PHY management. On NXP this is integrated with FEC driver itself.

If we were really honest, the binding would need to be called

port@0 {
	puppet = <&mac0>;
};
Quoted 7 lines
port@1 {
	puppet = <&mac1>;
};

which speaks for itself as to why accepting "puppet master" drivers is not really very compelling. I concur with the recommendation given by Andrew and Florian to refactor FEC as a multi-port single driver.

Ok.

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Quoted 4 lines
Quoted 1 line
Quoted 1 line

Andrew

I cannot stress enough how important it is for us to see STP support and consequently the ndo_start_xmit procedure for switch ports.

Ok.

Quoted 4 lines

Let me see if I understand correctly. When the switch is enabled, eth0 sends packets towards both physical switch ports, and eth1 sends packets towards none, but eth0 handles the link state of switch port 0, and eth1 handles the link state of switch port 1?

Exactly, this is how FEC driver is utilized for this switch.

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#25 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 13 lines · Lukasz Majewski
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Quoted 4 lines
Quoted 1 line
Quoted 1 line

Andrew

I cannot stress enough how important it is for us to see STP support and consequently the ndo_start_xmit procedure for switch ports.

Quoted 8 lines

Ok.

Quoted 4 lines

Let me see if I understand correctly. When the switch is enabled, eth0 sends packets towards both physical switch ports, and eth1 sends packets towards none, but eth0 handles the link state of switch port 0, and eth1 handles the link state of switch port 1?

Exactly, this is how FEC driver is utilized for this switch.

This is a much bigger problem than anything which has to do with code organization. Linux does not have any sort of support for unmanaged switches. Please try to find out if your switch is supposed to be able to be managed (run control protocols on the CPU). If not, well, I don't know what to suggest.

#26 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Vladimir,

Quoted 22 lines · Lukasz Majewski
Quoted 21 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Quoted 5 lines
Quoted 1 line
Quoted 1 line

Andrew

I cannot stress enough how important it is for us to see STP support and consequently the ndo_start_xmit procedure for switch ports.

Ok.

Quoted 4 lines

Let me see if I understand correctly. When the switch is enabled, eth0 sends packets towards both physical switch ports, and eth1 sends packets towards none, but eth0 handles the link state of switch port 0, and eth1 handles the link state of switch port 1?

Quoted 5 lines
Quoted 1 line

Exactly, this is how FEC driver is utilized for this switch.

This is a much bigger problem than anything which has to do with code organization. Linux does not have any sort of support for unmanaged switches.

My impression is similar. This switch cannot easily fit into DSA (lack of appending tags) nor to switchdev.

The latter is caused by two modes of operation:

- Bypass mode (no switch) -> DMA1 and DMA0 are used
- Switch mode -> only DMA0 is used

Moreover, from my understanding of the CPSW - looks like it uses always just a single DMA, and the switching seems to be the default operation for two ethernet ports.

The "bypass mode" from NXP's L2 switch seems to be achieved inside the CPSW switch, by configuring it to not pass packets between those ports.

Quoted 2 lines

Please try to find out if your switch is supposed to be able to be managed (run control protocols on the CPU).

It can support all the "normal" set of L2 switch features:

- VLANs, lookup table (with learning), filtering and forwarding (Multicast, Broadcast, Unicast), priority queues, IP snooping, etc.

Frames for BPDU are recognized by the switch and can be used to implement support for RSTP. However, this switch has a separate address space (not covered and accessed by FEC address).

Quoted 2 lines

If not, well, I
don't know what to suggest.

For me it looks like the NXP's L2 switch shall be treated _just_ as offloading IP block to accelerate switching (NXP already support dpaa[2] for example).

The idea with having it configured on demand, when:
ip link add name br0 type bridge; ip link set br0 up; ip link set eth0 master br0;
ip link set eth1 master br0;

Seems to be a reasonable one. In the above scenario it would work hand by hand with FEC drivers (as those would handle PHY communication setup and link up/down events).

It would be welcome if the community could come up with some rough idea how to proceed with this IP block support (especially that for example imx287 is used in many embedded devices and is going to be in active production for next 10+ years).

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#27 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Quoted 24 lines · Lukasz Majewski

Hi Vladimir,

Quoted 22 lines · Lukasz Majewski
Quoted 21 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Quoted 5 lines
Quoted 1 line
Quoted 1 line

Andrew

I cannot stress enough how important it is for us to see STP support and consequently the ndo_start_xmit procedure for switch ports.

Ok.

Quoted 4 lines

Let me see if I understand correctly. When the switch is enabled, eth0 sends packets towards both physical switch ports, and eth1 sends packets towards none, but eth0 handles the link state of switch port 0, and eth1 handles the link state of switch port 1?

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Exactly, this is how FEC driver is utilized for this switch.

This is a much bigger problem than anything which has to do with code organization. Linux does not have any sort of support for unmanaged switches.

My impression is similar. This switch cannot easily fit into DSA (lack of appending tags)

No, this is not why the switch does not fit the DSA model. DSA assumes that the master interface and the switch are two completely separate devices which manage themselves independently. Their boundary is typically at the level of a MAC-to-MAC connection, although vendors have sometimes blurred this line a bit in the case of integrated switches. But the key point is that if there are 2 external ports going to the switch, these should be managed by the switch driver. But when the switch is sandwiched between the Ethernet controller of the "DSA master" (the DMA engine of fec0) and the DSA master's MAC (still owned by fec), the separation isn't quite what DSA expects, is it? Remember that in the case of the MTIP switch, the fec driver needs to put the MACs going to the switch in promiscuous mode such that the switch behaves as a switch and actually forwards packets by MAC DA instead of dropping them. So the system is much more tightly coupled.

 +---------------------------------------------------------------------------+
 |                                                                           |
 | +--------------+        +--------------------+--------+      +------------+
 | |              |        | MTIP switch        |        |      |            |
 | |   fec 1 DMA  |---x    |                    | Port 2 |------| fec 1 MAC  |
 | |              |        |            \  /    |        |      |            |
 | +--------------+        |             \/     +--------+      +------------+
 |                         |             /\              |                   |
 | +--------------+        +--------+   /  \    +--------+      +------------+
 | |              |        |        |           |        |      |            |
 | |   fec 0 DMA  |--------| Port 0 |           | Port 1 |------| fec 0 MAC  |
 | |              |        |        |           |        |      |            |
 | +--------------+        +--------+-----------+--------+      +------------+
 |                                                                           |
 +---------------------------------------------------------------------------+

Is this DSA? I don't really think so, but you could still try to argue otherwise.

The opposite is also true. DSA supports switches that don't append tags to packets (see sja1105). This doesn't make them "less DSA", just more of a pain to work with.

Quoted 8 lines

nor to switchdev.

The latter is caused by two modes of operation:

- Bypass mode (no switch) -> DMA1 and DMA0 are used
- Switch mode -> only DMA0 is used

Quoted 6 lines

Moreover, from my understanding of the CPSW - looks like it uses always just a single DMA, and the switching seems to be the default operation for two ethernet ports.

The "bypass mode" from NXP's L2 switch seems to be achieved inside the CPSW switch, by configuring it to not pass packets between those ports.

I don't exactly see the point you're trying to make here. At the end of the day, the only thing that matters is what you expose to the user. With no way (when the switch is enabled) for a socket opened on eth0 to send/receive packets coming only from the first port, and a socket opened on eth1 to send/receive packets coming only from the second port, I think this driver attempt is a pretty far cry from what a switch driver in Linux is expected to offer, be it modeled as switchdev or DSA.

Quoted 19 lines
Quoted 2 lines

Please try to find out if your switch is supposed to be able to be managed (run control protocols on the CPU).

It can support all the "normal" set of L2 switch features:

- VLANs, lookup table (with learning), filtering and forwarding (Multicast, Broadcast, Unicast), priority queues, IP snooping, etc.

Frames for BPDU are recognized by the switch and can be used to implement support for RSTP. However, this switch has a separate address space (not covered and accessed by FEC address).

Quoted 2 lines

If not, well, I
don't know what to suggest.

For me it looks like the NXP's L2 switch shall be treated _just_ as offloading IP block to accelerate switching (NXP already support dpaa[2] for example).

Quoted 8 lines

The idea with having it configured on demand, when:
ip link add name br0 type bridge; ip link set br0 up; ip link set eth0 master br0;
ip link set eth1 master br0;

Seems to be a reasonable one. In the above scenario it would work hand by hand with FEC drivers (as those would handle PHY communication setup and link up/down events).

You seem to imply that we are suggesting something different.

Quoted 2 lines

It would be welcome if the community could come up with some rough idea how to proceed with this IP block support

Ok, so what I would do if I really cared that much about mainline support is I would refactor the FEC driver to offer its core functionality to a new multi-port driver that is able to handle the FEC DMA interfaces, the MACs and the switch. EXPORT_SYMBOL_GPL is your friend.

This driver would probe on a device tree binding with 3 "reg" values: 1 for the fec@800f0000, 1 for the fec@800f4000 and 1 for the switch@800f8000. No puppet master driver which coordinates other drivers, just a single driver that, depending on the operating state, manages all the SoC resources in a way that will offer a sane and consistent view of the Ethernet ports.

So it will have a different .ndo_start_xmit implementation depending on whether the switch is bypassed or not (if you need to send a packet on eth1 and the switch is bypassed, you send it through the DMA interface of eth1, otherwise you send it through the DMA interface of eth0 in a way in which the switch will actually route it to the eth1 physical port).

Then I would implement support for BPDU RX/TX (I haven't looked at the documentation, but I expect that what this switch offers for control traffic doesn't scale at high speeds (if it does, great, then send and receive all your packets as control packets, to have precise port identification). If it doesn't, you'll need a way to treat your data plane packets differently from the control plane packets. For the data plane, you can perhaps borrow some ideas from net/dsa/tag_8021q.c, or even from Tobias Waldekranz's proposal to just let data plane packets coming from the bridge slide into the switch with no precise control of the destination port at all, just let the switch perform FDB lookups for those packets because the switch hardware FDB is supposed to be more or less in sync with the bridge software FDB:
https://patchwork.kernel.org/project/netdevbpf/cover/[email protected]/

Quoted 2 lines

(especially that for example imx287 is used in many embedded devices and is going to be in active production for next 10+ years).

Well, I guess you have a plan then. There are still 10+ years left to enjoy the benefits of a proper driver design...

#28 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Vladimir,

Quoted 156 lines · Lukasz Majewski
Quoted 26 lines

Hi Vladimir,

Quoted 24 lines · Lukasz Majewski
Quoted 23 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines

So before considering merging your changes, i would like to see a usable binding.

I also don't remember seeing support for STP. Without that, your network has broadcast storm problems when there are loops. So i would like to see the code needed to put ports into blocking, listening, learning, and forwarding states.

Quoted 5 lines
Quoted 1 line
Quoted 1 line

Andrew

I cannot stress enough how important it is for us to see STP support and consequently the ndo_start_xmit procedure for switch ports.

Ok.

Quoted 5 lines

Let me see if I understand correctly. When the switch is enabled, eth0 sends packets towards both physical switch ports, and eth1 sends packets towards none, but eth0 handles the link state of switch port 0, and eth1 handles the link state of switch port 1?

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Exactly, this is how FEC driver is utilized for this switch.

This is a much bigger problem than anything which has to do with code organization. Linux does not have any sort of support for unmanaged switches.

My impression is similar. This switch cannot easily fit into DSA (lack of appending tags)

No, this is not why the switch does not fit the DSA model. DSA assumes that the master interface and the switch are two completely separate devices which manage themselves independently. Their boundary is typically at the level of a MAC-to-MAC connection, although vendors have sometimes blurred this line a bit in the case of integrated switches. But the key point is that if there are 2 external ports going to the switch, these should be managed by the switch driver. But when the switch is sandwiched between the Ethernet controller of the "DSA master" (the DMA engine of fec0) and the DSA master's MAC (still owned by fec), the separation isn't quite what DSA expects, is it? Remember that in the case of the MTIP switch, the fec driver needs to put the MACs going to the switch in promiscuous mode such that the switch behaves as a switch and actually forwards packets by MAC DA instead of dropping them. So the system is much more tightly coupled.

 +---------------------------------------------------------------------------+
 |
       | | +--------------+        +--------------------+--------+
  +------------+ | |              |        | MTIP switch        |
   |      |            | | |   fec 1 DMA  |---x    |
  | Port 2 |------| fec 1 MAC  | | |              |        |
  \  /    |        |      |            | | +--------------+        |
           \/     +--------+      +------------+ |
     |             /\              |                   | |
+--------------+        +--------+   /  \    +--------+
+------------+ | |              |        |        |           |
 |      |            | | |   fec 0 DMA  |--------| Port 0 |
| Port 1 |------| fec 0 MAC  | | |              |        |        |
        |        |      |            | | +--------------+
+--------+-----------+--------+      +------------+ |
                                                          |
+---------------------------------------------------------------------------+

Is this DSA? I don't really think so, but you could still try to argue otherwise.

The opposite is also true. DSA supports switches that don't append tags to packets (see sja1105). This doesn't make them "less DSA", just more of a pain to work with.

Quoted 8 lines

nor to switchdev.

The latter is caused by two modes of operation:

- Bypass mode (no switch) -> DMA1 and DMA0 are used
- Switch mode -> only DMA0 is used

Quoted 7 lines

Moreover, from my understanding of the CPSW - looks like it uses always just a single DMA, and the switching seems to be the default operation for two ethernet ports.

The "bypass mode" from NXP's L2 switch seems to be achieved inside the CPSW switch, by configuring it to not pass packets between those ports.

I don't exactly see the point you're trying to make here. At the end of the day, the only thing that matters is what you expose to the user. With no way (when the switch is enabled) for a socket opened on eth0 to send/receive packets coming only from the first port, and a socket opened on eth1 to send/receive packets coming only from the second port, I think this driver attempt is a pretty far cry from what a switch driver in Linux is expected to offer, be it modeled as switchdev or DSA.

Quoted 20 lines
Quoted 2 lines

Please try to find out if your switch is supposed to be able to be managed (run control protocols on the CPU).

It can support all the "normal" set of L2 switch features:

- VLANs, lookup table (with learning), filtering and forwarding (Multicast, Broadcast, Unicast), priority queues, IP snooping,
etc.

Frames for BPDU are recognized by the switch and can be used to implement support for RSTP. However, this switch has a separate address space (not covered and accessed by FEC address).

Quoted 2 lines

If not, well, I
don't know what to suggest.

For me it looks like the NXP's L2 switch shall be treated _just_ as offloading IP block to accelerate switching (NXP already support dpaa[2] for example).

Quoted 8 lines

The idea with having it configured on demand, when:
ip link add name br0 type bridge; ip link set br0 up; ip link set eth0 master br0;
ip link set eth1 master br0;

Seems to be a reasonable one. In the above scenario it would work hand by hand with FEC drivers (as those would handle PHY communication setup and link up/down events).

You seem to imply that we are suggesting something different.

Quoted 2 lines

It would be welcome if the community could come up with some rough idea how to proceed with this IP block support

Ok, so what I would do if I really cared that much about mainline support is I would refactor the FEC driver to offer its core functionality to a new multi-port driver that is able to handle the FEC DMA interfaces, the MACs and the switch. EXPORT_SYMBOL_GPL is your friend.

This driver would probe on a device tree binding with 3 "reg" values:
1 for the fec@800f0000, 1 for the fec@800f4000 and 1 for the switch@800f8000. No puppet master driver which coordinates other drivers, just a single driver that, depending on the operating state, manages all the SoC resources in a way that will offer a sane and consistent view of the Ethernet ports.

So it will have a different .ndo_start_xmit implementation depending on whether the switch is bypassed or not (if you need to send a packet on eth1 and the switch is bypassed, you send it through the DMA interface of eth1, otherwise you send it through the DMA interface of eth0 in a way in which the switch will actually route it to the eth1 physical port).

Quoted 14 lines

Then I would implement support for BPDU RX/TX (I haven't looked at the documentation, but I expect that what this switch offers for control traffic doesn't scale at high speeds (if it does, great, then send and receive all your packets as control packets, to have precise port identification). If it doesn't, you'll need a way to treat your data plane packets differently from the control plane packets. For the data plane, you can perhaps borrow some ideas from net/dsa/tag_8021q.c, or even from Tobias Waldekranz's proposal to just let data plane packets coming from the bridge slide into the switch with no precise control of the destination port at all, just let the switch perform FDB lookups for those packets because the switch hardware FDB is supposed to be more or less in sync with the bridge software FDB:
https://patchwork.kernel.org/project/netdevbpf/cover/[email protected]/

Thanks for sketching and sharing such detailed plan.

Quoted 5 lines
Quoted 2 lines

(especially that for example imx287 is used in many embedded devices and is going to be in active production for next 10+ years).

Well, I guess you have a plan then. There are still 10+ years left to enjoy the benefits of a proper driver design...

:-)

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#29 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

In reply to Lukasz Majewski
Quoted 24 lines

The best I could get would be:

&eth_switch {
	compatible = "imx,mtip-l2switch";
	reg = <0x800f8000 0x400>, <0x800fC000 0x4000>;

	interrupts = <100>;
	status = "okay";

	ethernet-ports {
		port1@1 {
			reg = <1>;
			label = "eth0";
			phys = <&mac0 0>;
		};

		port2@2 {
			reg = <2>;
			label = "eth1";
			phys = <&mac1 1>;
		};
	};
};
Quoted 2 lines

Which would abuse the "phys" properties usages - as 'mac[01]' are referring to ethernet controllers.

This is not how a dedicated driver would have its binding. We should not establish this as ABI.

So, sorry, but no.

Andrew

#30 lore raw

Re: [RFC 2/3] net: Provide switchdev driver for NXP's More Than IP L2 switch

Hi Andrew,

Quoted 24 lines
Quoted 24 lines

The best I could get would be:

&eth_switch {
	compatible = "imx,mtip-l2switch";
	reg = <0x800f8000 0x400>, <0x800fC000 0x4000>;

	interrupts = <100>;
	status = "okay";

	ethernet-ports {
		port1@1 {
			reg = <1>;
			label = "eth0";
			phys = <&mac0 0>;
		};

		port2@2 {
			reg = <2>;
			label = "eth1";
			phys = <&mac1 1>;
		};
	};
};
Quoted 7 lines
Quoted 2 lines

Which would abuse the "phys" properties usages - as 'mac[01]' are referring to ethernet controllers.

This is not how a dedicated driver would have its binding. We should not establish this as ABI.

So, sorry, but no.

Thanks for the clear statement about upstream requirements.

Quoted 2 lines

Andrew

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#31 lore raw

RFC 3/3 net: imx: Adjust fec_main.c to provide support for L2 switch

In reply to Lukasz Majewski

This patch provides the code for re-using generic FEC i.MX code for More Than IP L2 switch.

The trick here is to reconfigure FEC driver to use DMA0 as the one connected to switch's port 0. Port 1 and 2 are then connected to ENET-MAC.

The internal connection diagram can be found here [0].

This code has been developed on i.MX287 device, but this switch (from More ThanIP) can be also found on e.g. NXP's Vybrid VF610 SoC.

To reuse as much code as possible it was also necessary to introduce fec_hwp() wrapper, which either returns pointer to MAC or L2 switch interrupt and buffer descriptors.
Following registers shall be used with it:
FEC_{IEVENT|IMASK}, FEC_MTIP_R_DES_ACTIVE_0, FEC_MTIP_X_DES_ACTIVE_0, FEC_MTIP_R_DES_START_0, FEC_MTIP_X_DES_START_0
This driver introduces special wrappers to map on fly DMA0 to either ENET-MAC0 or L2 switch port 0.

The most intrusive changes when L2 switch is used:

- Disable RACC (Receive Data Accelerator)
When this feature is enabled the data is shifted by 16 bits and hence received packets are malformed when we try to pass data from one switch port to another.

- L2 switch shall pass the whole pkt_len packet. Current FEC code subtract 4 to remove FCS

The control of FEC clock and promisc is handled by bridge driver, so there is no need for adjustments.

This driver forward ports the legacy driver [1][2] into contemporary Linux.

Last but not least - some functions had to be exported to allow proper configuration of ethernet interfaces when bridging was enabled/disabled.

Links:
[0] - Chapter 29.3.1 in
"i.MX28 Applications Processor Reference Manual, Rev. 2, 08/2013"
[1]
From 2.6 i.MX kernel:
Repo: git://git.freescale.com/imx/linux-2.6-imx.git
Branch: imx_2.6.35_maintain
SHA1: b3912bb8a4caf3ec50909135e88af959982c43ca

[2] - https://github.com/lmajewski/linux-imx28-l2switch/commits/master

---
 drivers/net/ethernet/freescale/fec.h      |  36 ++++++
 drivers/net/ethernet/freescale/fec_main.c | 139 ++++++++++++++++++----
 2 files changed, 151 insertions(+), 24 deletions(-)
drivers/net/ethernet/freescale/fec.h +36 -0
index 0602d5d5d2ee..dc2d31321cbd 100644
@@ -29,6 +29,10 @@
  */
 #define FEC_IEVENT		0x004 /* Interrupt event reg */
 #define FEC_IMASK		0x008 /* Interrupt mask reg */
+#ifdef CONFIG_FEC_MTIP_L2SW
+#define FEC_MTIP_R_DES_ACTIVE_0	0x018 /* L2 switch Receive descriptor reg */
+#define FEC_MTIP_X_DES_ACTIVE_0	0x01C /* L2 switch Transmit descriptor reg */
+#endif
 #define FEC_R_DES_ACTIVE_0	0x010 /* Receive descriptor reg */
 #define FEC_X_DES_ACTIVE_0	0x014 /* Transmit descriptor reg */
 #define FEC_ECNTRL		0x024 /* Ethernet control reg */
@@ -59,6 +63,10 @@
 #define FEC_R_DES_START_2	0x16c /* Receive descriptor ring 2 */
 #define FEC_X_DES_START_2	0x170 /* Transmit descriptor ring 2 */
 #define FEC_R_BUFF_SIZE_2	0x174 /* Maximum receive buff ring2 size */
+#ifdef CONFIG_FEC_MTIP_L2SW
+#define FEC_MTIP_R_DES_START_0  0x0C /* L2 switch Receive descriptor ring */
+#define FEC_MTIP_X_DES_START_0	0x10 /* L2 switch Transmit descriptor ring */
+#endif
 #define FEC_R_DES_START_0	0x180 /* Receive descriptor ring */
 #define FEC_X_DES_START_0	0x184 /* Transmit descriptor ring */
 #define FEC_R_BUFF_SIZE_0	0x188 /* Maximum receive buff size */
@@ -376,6 +384,12 @@ struct bufdesc_ex {
 #define FEC_ENET_TS_AVAIL       ((uint)0x00010000)
 #define FEC_ENET_TS_TIMER       ((uint)0x00008000)
 
+#ifdef CONFIG_FEC_MTIP_L2SW
+#define FEC_MTIP_ENET_TXF ((uint)0x00000010)	/* Full frame transmitted */
+#define FEC_MTIP_ENET_RXF ((uint)0x00000004)	/* Full frame received */
+#define FEC_MTIP_DEFAULT_IMASK (FEC_MTIP_ENET_TXF | FEC_MTIP_ENET_RXF)
+#define FEC_MTIP_RX_DISABLED_IMASK (FEC_MTIP_DEFAULT_IMASK & (~FEC_MTIP_ENET_RXF))
+#endif
 #define FEC_DEFAULT_IMASK (FEC_ENET_TXF | FEC_ENET_RXF)
 #define FEC_RX_DISABLED_IMASK (FEC_DEFAULT_IMASK & (~FEC_ENET_RXF))
 
@@ -595,9 +609,31 @@ struct fec_enet_private {
 	int pps_enable;
 	unsigned int next_counter;
 
+#ifdef CONFIG_FEC_MTIP_L2SW
+	/* More Than IP L2 switch */
+	void __iomem *hwpsw;
+	struct mtipl2sw_priv *mtipl2;
+	bool mtip_l2sw;
+#endif
 	u64 ethtool_stats[];
 };
 
+/* MTIP L2 switch */
+/* Get proper base address for either L2 switch or MAC ENET */
+static inline void __iomem *fec_hwp(struct fec_enet_private *fep)
+{
+#ifdef CONFIG_FEC_MTIP_L2SW
+	if (fep->mtip_l2sw)
+		return fep->hwpsw;
+#endif
+	return fep->hwp;
+}
+
+int fec_enet_close(struct net_device *ndev);
+int fec_enet_open(struct net_device *ndev);
+const unsigned short fec_offset_des_active_rxq(struct fec_enet_private *, int);
+const unsigned short fec_offset_des_active_txq(struct fec_enet_private *, int);
+struct fec_enet_private *fec_get_priv(const struct net_device *ndev);
 void fec_ptp_init(struct platform_device *pdev, int irq_idx);
 void fec_ptp_stop(struct platform_device *pdev);
 void fec_ptp_start_cyclecounter(struct net_device *ndev);
drivers/net/ethernet/freescale/fec_main.c +115 -24

262 lines not on this page · raw message

Signature · 1 line
2.20.1
#32 lore raw

Re: [RFC 3/3] net: imx: Adjust fec_main.c to provide support for L2 switch

drivers/net/ethernet/freescale/fec.h line 35

Quoted 11 lines

index 0602d5d5d2ee..dc2d31321cbd 100644

drivers/net/ethernet/freescale/fec.h +4 -0
@@ -29,6 +29,10 @@
  */
 #define FEC_IEVENT		0x004 /* Interrupt event reg */
 #define FEC_IMASK		0x008 /* Interrupt mask reg */
+#ifdef CONFIG_FEC_MTIP_L2SW
+#define FEC_MTIP_R_DES_ACTIVE_0	0x018 /* L2 switch Receive descriptor reg */
+#define FEC_MTIP_X_DES_ACTIVE_0	0x01C /* L2 switch Transmit descriptor reg */
+#endif

Please don't scatter #ifdef everywhere.

In this case, the register exists, even if the switch it not being used, so there is no need to have it conditional.

drivers/net/ethernet/freescale/fec.h

Quoted 6 lines
drivers/net/ethernet/freescale/fec.h +3 -0
 #include <asm/cacheflush.h>
 
 #include "fec.h"
+#ifdef CONFIG_FEC_MTIP_L2SW
+#include "./mtipsw/fec_mtip.h"
+#endif

Why not just include it all the time?

Andrew

#33 lore raw

Re: [RFC 3/3] net: imx: Adjust fec_main.c to provide support for L2 switch

Hi Andrew,

Quoted 14 lines
Quoted 11 lines

index 0602d5d5d2ee..dc2d31321cbd 100644

drivers/net/ethernet/freescale/fec.h +2 -0
@@ -29,6 +29,10 @@
  */
 #define FEC_IEVENT		0x004 /* Interrupt event reg */
 #define FEC_IMASK		0x008 /* Interrupt mask reg */
+#ifdef CONFIG_FEC_MTIP_L2SW
+#define FEC_MTIP_R_DES_ACTIVE_0	0x018 /* L2 switch Receive

descriptor reg */ +#define FEC_MTIP_X_DES_ACTIVE_0 0x01C /*
L2 switch Transmit descriptor reg */ +#endif

Please don't scatter #ifdef everywhere.

Quoted 2 lines

In this case, the register exists, even if the switch it not being used, so there is no need to have it conditional.

+1

Quoted 9 lines

drivers/net/ethernet/freescale/fec.h

Quoted 7 lines

#include <asm/cacheflush.h>

#include "fec.h"

drivers/net/ethernet/freescale/fec.h +3 -0
+#ifdef CONFIG_FEC_MTIP_L2SW
+#include "./mtipsw/fec_mtip.h"
+#endif  

Why not just include it all the time?

Ok.

Quoted 2 lines

Andrew

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#34 lore raw
In reply to Lukasz Majewski
Quoted 9 lines · Lukasz Majewski

This patch series is a followup for the earlier effort [1] to bring support for L2 switch IP block on some NXP devices.

This time it augment the fec driver, so the L2 switch is treated as a HW network accelerator. This is minimal, yet functional driver, which enables bridging between imx28 ENET-MAC ports.

Links:
[1] - https://lwn.net/ml/linux-kernel/[email protected]/

On which tree are these patches supposed to apply?

Patch 1 doesn't apply on today's net-next.
git am ~/incoming/*
Applying: ARM: dts: imx28: Add description for L2 switch on XEA board
error: arch/arm/boot/dts/imx28-xea.dts: does not exist in index Patch failed at 0001 ARM: dts: imx28: Add description for L2 switch on XEA board
hint: Use 'git am --show-current-patch' to see the failed patch When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".

Patch 2 doesn't apply on today's linux-next.
git am ~/incoming/*
Applying: ARM: dts: imx28: Add description for L2 switch on XEA board
Applying: net: Provide switchdev driver for NXP's More Than IP L2 switch
error: patch failed: drivers/net/ethernet/freescale/Makefile:27
error: drivers/net/ethernet/freescale/Makefile: patch does not apply Patch failed at 0002 net: Provide switchdev driver for NXP's More Than IP L2 switch
hint: Use 'git am --show-current-patch' to see the failed patch
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".

#35 lore raw

Hi Vladimir,

Quoted 26 lines · Lukasz Majewski
Quoted 11 lines

This patch series is a followup for the earlier effort [1] to bring support for L2 switch IP block on some NXP devices.

This time it augment the fec driver, so the L2 switch is treated as a HW network accelerator. This is minimal, yet functional driver, which enables bridging between imx28 ENET-MAC ports.

Links:
[1] -
https://lwn.net/ml/linux-kernel/[email protected]/

On which tree are these patches supposed to apply?

Patch 1 doesn't apply on today's net-next.
git am ~/incoming/*
Applying: ARM: dts: imx28: Add description for L2 switch on XEA board
error: arch/arm/boot/dts/imx28-xea.dts: does not exist in index Patch failed at 0001 ARM: dts: imx28: Add description for L2 switch on XEA board hint: Use 'git am --show-current-patch' to see the failed patch When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".

Quoted 12 lines

Patch 2 doesn't apply on today's linux-next.
git am ~/incoming/*
Applying: ARM: dts: imx28: Add description for L2 switch on XEA board
Applying: net: Provide switchdev driver for NXP's More Than IP L2 switch error: patch failed: drivers/net/ethernet/freescale/Makefile:27
error: drivers/net/ethernet/freescale/Makefile: patch does not apply Patch failed at 0002 net: Provide switchdev driver for NXP's More Than IP L2 switch hint: Use 'git am --show-current-patch' to see the failed patch When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".

Please use following repos:

- RFC v1 for switchdev (sent now):
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-switchdev-RFC_v1

- RFC v1 for DSA:
https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-DSA-RFC_v1

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]