Skip to contentSkip to conversation

Add MV88E6xxx DSA driver and use on gwventana

PATCH Superseded Tim Harvey 3 participants 15 messages Oct 10, 2022 – Oct 26, 2022

Subsystems (4)

v6 of 9 Superseded Latest is v9 →

lore
Share & tools
lore

Tim Harvey

author
#0 lore raw

v6 cover Add MV88E6xxx DSA driver and use on gwventana

This series adds a DSA driver for the MV88E6xxx based on drivers/net/phy/mv88e61xx and uses it in the gwventana_gw5904_defconfig.

The hope is that the other three boards that use the MV88E61xx driver can move to this as well eventually so that we can remove the non-dm driver and the 4 Kconfig options it requires.

The MV88E6xxx has an MDIO interface thus DM_MDIO must be used so support for a UCLASS_MDIO driver is added to the fec_mxc ethernet driver in a way that allows a fallback to the previous non DM_MDIO case as there are many boards out there using this driver that define DM_MDIO but do not have the required dt props for a DM_MDIO driver which would cause a regression.

Additionally some other patches are here suggested by Vladimir:
- ensure MDIO children are scanned on post-bind is needed
- sanity check DSA driver required ops are present
- allow DSA drivers to not require xmit/recv functions
- remove unecessary xmit/recv functions from ksz9477 driver

v6:
- update commit messages
- squash accidently change to mv88e6xxx driver into previous patch
- remove unused dt label for cpu port
- removed unnecessary semicolon
- update commit msg (s/MV88E61xx/MV88E6xxx)
- remove GbE from commit msg and Kconfig desc
- squash accidental change from patch 7 into patch 6
- added error print on failure to read switch id
- mv88e6xxx_probe:
- check for switch enabled
- remove unused variable enabled
- remove unnecessary call to dsa_set_tagging
- port_probe:
- new function to configure phy features by calling phy_config
- port_enable:
- only enable energy-detect sensing on phy ports
- add phy/cmode verification mistmatch warning
- remove mv88e6xxx_fixed_port_setup()
- always force link up for fixed ports
- always set SERDES mode regardless of cpu port
- remove unnecessary setting of CPUDEST
- port_disable:
- remove pointless error check
- removed unnecessary priv data for mdio controller
- fix indentation
- change variable name for clarity

v5:
- fix typo in defconfig update
- added support for MV88E6320
- added Fabio's rb tag

v4:
- use standard Linux internal MDIO dt structure
- use PHY_FIXED_ID define
- rename to mv88e6xxx
- sort includes alphabetically
- remove dsa term from function names
- reduce indentation level and remove unecessary code in of probe_mdio function
- rename pdev to mdev to represent mdio device

v3:
- fix mdios node in dt
- add Vladimir's rb tag's

v2:
- added Ramon's rb tag's to first two patches
- add patches for dsa-uclass to sanity check ops and make xmit/recv optional
- fec: fix fallback for non conforming DM_MDIO dts
- mv88e61xx:
- rebase on v2022.07-rc2 (use ofnode_get_phy_node)
- remove unused commented out fields from struct
- remove unused PORT_MASK macro
- remove phy from priv struct name
- refactor code from original drivers/net/phy/mv88e61xx with suggestions from review to consolidate some functions into mv88e61xx_dsa_port_enable
- remove unecessary skiping of disabling of CPU port
- remove unecessary dev_set_parent_priv
- remove unnecessary static init flag
- replace debug with a dev_warn if switch device-id unsupported
- remove unnecessary xmit/recv functions as we rely on the fact that only a single port is active instead of mangling packets

Tested on a Gateworks GW5904 which has a Marvell 88E6176 switch hanging off the IMX6 FEC.

Best Regards,

Tim

Tim Harvey (8):
net: mdio-uclass: scan for dm mdio children on post-bind
net: dsa: move cpu port probe to dsa_post_probe
net: dsa: ensure dsa driver has proper ops
net: dsa: allow rcv() and xmit() to be optional
net: ksz9477: remove unnecessary xmit and recv functions
net: fec: add support for DM_MDIO
net: add MV88E61xx DSA driver
board: gw_ventana: enable MV88E61XX DSA support

 arch/arm/dts/imx6qdl-gw5904.dtsi        |  32 +-
 board/gateworks/gw_ventana/gw_ventana.c |  50 +-
 configs/gwventana_gw5904_defconfig      |   7 +-
 drivers/net/Kconfig                     |   7 +
 drivers/net/Makefile                    |   1 +
 drivers/net/fec_mxc.c                   |  90 ++-
 drivers/net/ksz9477.c                   |  23 -
 drivers/net/mv88e6xxx.c                 | 832 ++++++++++++++++++++++++
 net/dsa-uclass.c                        |  55 +-
 net/mdio-uclass.c                       |   4 +
 10 files changed, 1029 insertions(+), 72 deletions(-)
 create mode 100644 drivers/net/mv88e6xxx.c
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH v6 1/8 net: mdio-uclass: scan for dm mdio children on post-bind

If a DM_MDIO driver is used we need to scan the subnodes as well.

---
v6:
- no changes
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- no changes
v2:
- added Ramon's rb tag

---
 net/mdio-uclass.c | 4 ++++
 1 file changed, 4 insertions(+)
net/mdio-uclass.c +4 -0
index 4401492ca015..d80037d0ac71 100644
@@ -49,7 +49,11 @@ static int dm_mdio_post_bind(struct udevice *dev)
 		return -EINVAL;
 	}
 
+#if CONFIG_IS_ENABLED(OF_REAL)
+	return dm_scan_fdt_dev(dev);
+#else
 	return 0;
+#endif
 }
 
 int dm_mdio_read(struct udevice *mdio_dev, int addr, int devad, int reg)
Signature · 1 line
2.25.1
author
#2 lore raw

PATCH v6 2/8 net: dsa: move cpu port probe to dsa_post_probe

In reply to Tim Harvey

In order to ensure that a DSA driver probe gets called before dsa_ops->port_probe move the port_probe of the cpu_port to a post-probe function.

---
v6:
- removed unnecessary semicolon
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- added Vladimir's rb tag
v2:
- added Ramon's rb tag

---
 net/dsa-uclass.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
net/dsa-uclass.c +10 -1
index 5b7046432ff3..5759cedcbec3 100644
@@ -466,7 +466,6 @@ static int dsa_pre_probe(struct udevice *dev)
 {
 	struct dsa_pdata *pdata = dev_get_uclass_plat(dev);
 	struct dsa_priv *priv = dev_get_uclass_priv(dev);
-	struct dsa_ops *ops = dsa_get_ops(dev);
 	int err;
 
 	priv->num_ports = pdata->num_ports;
@@ -482,6 +481,15 @@ static int dsa_pre_probe(struct udevice *dev)
 	if (err)
 		return err;
 
+	return 0;
+}
+
+static int dsa_post_probe(struct udevice *dev)
+{
+	struct dsa_priv *priv = dev_get_uclass_priv(dev);
+	struct dsa_ops *ops = dsa_get_ops(dev);
+	int err;
+
 	/* Simulate a probing event for the CPU port */
 	if (ops->port_probe) {
 		err = ops->port_probe(dev, priv->cpu_port,
@@ -498,6 +506,7 @@ UCLASS_DRIVER(dsa) = {
 	.name = "dsa",
 	.post_bind = dsa_post_bind,
 	.pre_probe = dsa_pre_probe,
+	.post_probe = dsa_post_probe,
 	.per_device_auto = sizeof(struct dsa_priv),
 	.per_device_plat_auto = sizeof(struct dsa_pdata),
 	.per_child_plat_auto = sizeof(struct dsa_port_pdata),
Signature · 1 line
2.25.1
author
#3 lore raw

PATCH v6 3/8 net: dsa: ensure dsa driver has proper ops

In reply to Tim Harvey

Add a function to sanity check a dsa driver having proper ops.

---
v6:
- no changes
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- added Vladimir's rb tag
v2: new patch

---
 net/dsa-uclass.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
net/dsa-uclass.c +17 -0
index 5759cedcbec3..211a991cdd0d 100644
@@ -342,6 +342,19 @@ U_BOOT_DRIVER(dsa_port) = {
 	.plat_auto = sizeof(struct eth_pdata),
 };
 
+static int dsa_sanitize_ops(struct udevice *dev)
+{
+	struct dsa_ops *ops = dsa_get_ops(dev);
+
+	if ((!ops->xmit || !ops->rcv) &&
+	    (!ops->port_enable && !ops->port_disable)) {
+		dev_err(dev, "Packets cannot be steered to ports\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 /*
  * This function mostly deals with pulling information out of the device tree
  * into the pdata structure.
@@ -358,6 +371,10 @@ static int dsa_post_bind(struct udevice *dev)
 	if (!ofnode_valid(node))
 		return -ENODEV;
 
+	err = dsa_sanitize_ops(dev);
+	if (err)
+		return err;
+
 	pdata->master_node = ofnode_null();
 
 	node = ofnode_find_subnode(node, "ports");
Signature · 1 line
2.25.1
author
#4 lore raw

PATCH v6 4/8 net: dsa: allow rcv() and xmit() to be optional

In reply to Tim Harvey

Allow rcv() and xmit() dsa driver ops to be optional in case a driver does not care to mangle a packet as in U-Boot only one network port is enabled at a time and thus no packet mangling is necessary.

---
v6:
- no changes
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- added Vladimir's rb tag
v2: new patch

---
 net/dsa-uclass.c | 27 ++++++++++++++++++---------
 1 file changed, 18 insertions(+), 9 deletions(-)
net/dsa-uclass.c +18 -9
index 211a991cdd0d..eba8347ae21a 100644
@@ -131,16 +131,14 @@ static void dsa_port_stop(struct udevice *pdev)
  * We copy the frame to a stack buffer where we have reserved headroom and
  * tailroom space.  Headroom and tailroom are set to 0.
  */
-static int dsa_port_send(struct udevice *pdev, void *packet, int length)
+static int dsa_port_mangle_packet(struct udevice *pdev, void *packet, int length)
 {
+	struct dsa_port_pdata *port_pdata = dev_get_parent_plat(pdev);
 	struct udevice *dev = dev_get_parent(pdev);
 	struct dsa_priv *priv = dev_get_uclass_priv(dev);
 	int head = priv->headroom, tail = priv->tailroom;
-	struct udevice *master = dsa_get_master(dev);
 	struct dsa_ops *ops = dsa_get_ops(dev);
 	uchar dsa_packet_tmp[PKTSIZE_ALIGN];
-	struct dsa_port_pdata *port_pdata;
-	int err;
 
 	if (length + head + tail > PKTSIZE_ALIGN)
 		return -EINVAL;
@@ -152,10 +150,21 @@ static int dsa_port_send(struct udevice *pdev, void *packet, int length)
 	/* copy back to preserve original buffer alignment */
 	memcpy(packet, dsa_packet_tmp, length);
 
-	port_pdata = dev_get_parent_plat(pdev);
-	err = ops->xmit(dev, port_pdata->index, packet, length);
-	if (err)
-		return err;
+	return ops->xmit(dev, port_pdata->index, packet, length);
+}
+
+static int dsa_port_send(struct udevice *pdev, void *packet, int length)
+{
+	struct udevice *dev = dev_get_parent(pdev);
+	struct udevice *master = dsa_get_master(dev);
+	struct dsa_ops *ops = dsa_get_ops(dev);
+	int err;
+
+	if (ops->xmit) {
+		err = dsa_port_mangle_packet(pdev, packet, length);
+		if (err)
+			return err;
+	}
 
 	return eth_get_ops(master)->send(master, packet, length);
 }
@@ -172,7 +181,7 @@ static int dsa_port_recv(struct udevice *pdev, int flags, uchar **packetp)
 	int length, port_index, err;
 
 	length = eth_get_ops(master)->recv(master, flags, packetp);
-	if (length <= 0)
+	if (length <= 0 || !ops->rcv)
 		return length;
 
 	/*
Signature · 1 line
2.25.1
#5 lore raw

Re: [PATCH v6 4/8] net: dsa: allow rcv() and xmit() to be optional

Hi Tim,

Quoted 19 lines · Tim Harvey

Allow rcv() and xmit() dsa driver ops to be optional in case a driver does not care to mangle a packet as in U-Boot only one network port is enabled at a time and thus no packet mangling is necessary.

---
v6:
- no changes
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- added Vladimir's rb tag
v2: new patch

Quoted 3 lines
---
 net/dsa-uclass.c | 27 ++++++++++++++++++---------
 1 file changed, 18 insertions(+), 9 deletions(-)

Should the tests be updated for this?

Regards,
Simon

#6 lore raw

Re: [PATCH v6 4/8] net: dsa: allow rcv() and xmit() to be optional

Hi Simon,

Quoted 9 lines · Simon Glass

Hi Tim,

Quoted 4 lines · Tim Harvey

Allow rcv() and xmit() dsa driver ops to be optional in case a driver does not care to mangle a packet as in U-Boot only one network port is enabled at a time and thus no packet mangling is necessary.

Should the tests be updated for this?

Patch 3/8 serves as a runtime validation for this change. I don't believe that adding something more to test/dm/dsa.c is warranted, and I don't see exactly what can be added, either.

author
#7 lore raw

PATCH v6 5/8 net: ksz9477: remove unnecessary xmit and recv functions

In reply to Tim Harvey

Remove the unnecessary xmit and recv functions.

---
v6:
- no changes
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- added Vladimir's rb tag
v2: new patch

---
 drivers/net/ksz9477.c | 23 -----------------------
 1 file changed, 23 deletions(-)
drivers/net/ksz9477.c +0 -23
index ed8f1895cb12..fb5c76c600be 100644
@@ -62,7 +62,6 @@
 
 struct ksz_dsa_priv {
 	struct udevice *dev;
-	int active_port;
 };
 
 static inline int ksz_read8(struct udevice *dev, u32 reg, u8 *val)
@@ -382,9 +381,6 @@ static int ksz_port_enable(struct udevice *dev, int port, struct phy_device *phy
 	data8 |= SW_START;
 	ksz_write8(priv->dev, REG_SW_OPERATION, data8);
 
-	/* keep track of current enabled non-cpu port */
-	priv->active_port = port;
-
 	return 0;
 }
 
@@ -413,28 +409,9 @@ static void ksz_port_disable(struct udevice *dev, int port, struct phy_device *p
 	 */
 }
 
-static int ksz_xmit(struct udevice *dev, int port, void *packet, int length)
-{
-	dev_dbg(dev, "%s P%d %d\n", __func__, port + 1, length);
-
-	return 0;
-}
-
-static int ksz_recv(struct udevice *dev, int *port, void *packet, int length)
-{
-	struct ksz_dsa_priv *priv = dev_get_priv(dev);
-
-	dev_dbg(dev, "%s P%d %d\n", __func__, priv->active_port + 1, length);
-	*port = priv->active_port;
-
-	return 0;
-};
-
 static const struct dsa_ops ksz_dsa_ops = {
 	.port_enable = ksz_port_enable,
 	.port_disable = ksz_port_disable,
-	.xmit = ksz_xmit,
-	.rcv = ksz_recv,
 };
 
 static int ksz_probe_mdio(struct udevice *dev)
Signature · 1 line
2.25.1
author
#8 lore raw

PATCH v6 6/8 net: fec: add support for DM_MDIO

In reply to Tim Harvey

Add support for DM_MDIO by registering a UCLASS_MDIO driver and attempting to use it. This is necessary if wanting to use a DSA driver for example hanging off of the FEC MAC.

Care is taken to fallback to non DM_MDIO mii bus as several boards define DM_MDIO without having the proper device-tree configuration necessary such as an mdio subnode, a phy-mode prop, and either a valid phy-handle prop or fixed-phy subnode which will cause dm_eth_phy_connect() to fail.

---
v6:
- no changes
v5:
- added Fabio's rb tag
v4:
- no changes
v3:
- no changes
v2:
- fix fallback mechanism for legacy dt's that do not have phy-handle and mdio subnode

---
 drivers/net/fec_mxc.c | 90 ++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 88 insertions(+), 2 deletions(-)
drivers/net/fec_mxc.c +88 -2
index bbc4434ddb9e..7057472710fa 100644
@@ -30,6 +30,8 @@
 #include <asm/arch/imx-regs.h>
 #include <asm/mach-imx/sys_proto.h>
 #include <asm-generic/gpio.h>
+#include <dm/device_compat.h>
+#include <dm/lists.h>
 
 #include "fec_mxc.h"
 #include <eth_phy.h>
@@ -1026,6 +1028,81 @@ struct mii_dev *fec_get_miibus(ulong base_addr, int dev_id)
 	return bus;
 }
 
+#ifdef CONFIG_DM_MDIO
+struct dm_fec_mdio_priv {
+	struct ethernet_regs *regs;
+};
+
+static int dm_fec_mdio_read(struct udevice *dev, int addr, int devad, int reg)
+{
+	struct dm_fec_mdio_priv *priv = dev_get_priv(dev);
+
+	return fec_mdio_read(priv->regs, addr, reg);
+}
+
+static int dm_fec_mdio_write(struct udevice *dev, int addr, int devad, int reg, u16 data)
+{
+	struct dm_fec_mdio_priv *priv = dev_get_priv(dev);
+
+	return fec_mdio_write(priv->regs, addr, reg, data);
+}
+
+static const struct mdio_ops dm_fec_mdio_ops = {
+	.read = dm_fec_mdio_read,
+	.write = dm_fec_mdio_write,
+};
+
+static int dm_fec_mdio_probe(struct udevice *dev)
+{
+	struct dm_fec_mdio_priv *priv = dev_get_priv(dev);
+
+	priv->regs = (struct ethernet_regs *)ofnode_get_addr(dev_ofnode(dev->parent));
+
+	return 0;
+}
+
+U_BOOT_DRIVER(fec_mdio) = {
+	.name		= "fec_mdio",
+	.id		= UCLASS_MDIO,
+	.probe		= dm_fec_mdio_probe,
+	.ops		= &dm_fec_mdio_ops,
+	.priv_auto	= sizeof(struct dm_fec_mdio_priv),
+};
+
+static int dm_fec_bind_mdio(struct udevice *dev)
+{
+	struct udevice *mdiodev;
+	const char *name;
+	ofnode mdio;
+	int ret = -ENODEV;
+
+	/* for a UCLASS_MDIO driver we need to bind and probe manually
+	 * for an internal MDIO bus that has no dt compatible of its own
+	 */
+	ofnode_for_each_subnode(mdio, dev_ofnode(dev)) {
+		name = ofnode_get_name(mdio);
+
+		if (strcmp(name, "mdio"))
+			continue;
+
+		ret = device_bind_driver_to_node(dev, "fec_mdio",
+						 name, mdio, &mdiodev);
+		if (ret) {
+			printf("%s bind %s failed: %d\n", __func__, name, ret);
+			break;
+		}
+
+		/* need to probe it as there is no compatible to do so */
+		ret = uclass_get_device_by_ofnode(UCLASS_MDIO, mdio, &mdiodev);
+		if (!ret)
+			return 0;
+		printf("%s probe %s failed: %d\n", __func__, name, ret);
+	}
+
+	return ret;
+}
+#endif
+
 static int fecmxc_read_rom_hwaddr(struct udevice *dev)
 {
 	struct fec_priv *priv = dev_get_priv(dev);
@@ -1089,7 +1166,7 @@ static int device_get_phy_addr(struct fec_priv *priv, struct udevice *dev)
 
 static int fec_phy_init(struct fec_priv *priv, struct udevice *dev)
 {
-	struct phy_device *phydev;
+	struct phy_device *phydev = NULL;
 	int addr;
 
 	addr = device_get_phy_addr(priv, dev);
@@ -1097,7 +1174,10 @@ static int fec_phy_init(struct fec_priv *priv, struct udevice *dev)
 	addr = CONFIG_FEC_MXC_PHYADDR;
 #endif
 
-	phydev = phy_connect(priv->bus, addr, dev, priv->interface);
+	if (IS_ENABLED(CONFIG_DM_MDIO))
+		phydev = dm_eth_phy_connect(dev);
+	if (!phydev)
+		phydev = phy_connect(priv->bus, addr, dev, priv->interface);
 	if (!phydev)
 		return -ENODEV;
 
@@ -1228,6 +1308,12 @@ static int fecmxc_probe(struct udevice *dev)
 
 	priv->dev_id = dev_seq(dev);
 
+	if (IS_ENABLED(CONFIG_DM_MDIO)) {
+		ret = dm_fec_bind_mdio(dev);
+		if (ret && ret != -ENODEV)
+			return ret;
+	}
+
 #ifdef CONFIG_DM_ETH_PHY
 	bus = eth_phy_get_mdio_bus(dev);
 #endif
Signature · 1 line
2.25.1
author
#9 lore raw

PATCH v6 7/8 net: add MV88E61xx DSA driver

In reply to Tim Harvey

Add a DSA driver for the MV88E6xxx compatible Ethernet switches.

---
v6:
- update commit msg (s/MV88E61xx/MV88E6xxx)
- remove GbE from commit msg and Kconfig desc
- squash accidental change from patch 7 into patch 6
- added error print on failure to read switch id
- mv88e6xxx_probe:
- check for switch enabled
- remove unused variable enabled
- remove unnecessary call to dsa_set_tagging
- port_probe:
- new function to configure phy features by calling phy_config
- port_enable:
- only enable energy-detect sensing on phy ports
- add phy/cmode verification mistmatch warning
- remove mv88e6xxx_fixed_port_setup()
- always force link up for fixed ports
- always set SERDES mode regardless of cpu port
- remove unnecessary setting of CPUDEST
- port_disable:
- remove pointless error check
- removed unnecessary priv data for mdio controller
- fix indentation
- change variable name for clarity
v5:
- added support for MV88E6320
- added Fabio's rb tag
v4:
- rename to mv88e6xxx
- sort includes alphabetically
- remove dsa term from function names
- reduce indentation level and remove unecessary code in of probe_mdio function
- rename pdev to mdev to represent mdio device
v3:
- Added Vladimir's rb tag
v2:
- rebase on v2022.07-rc1 (use ofnode_get_phy_node)
- remove unused commented out fields from struct
- remove unused PORT_MASK macro
- remove phy from priv struct name
- refactor code from original drivers/net/phy/mv88e61xx with suggestions from review to consolidate some functions
into mv88e61xx_dsa_port_enable
- remove unecessary skiping of disabling of CPU port
- remove unecessary dev_set_parent_priv
- remove unnecessary static init flag
- replace debug with a dev_warn if switch device-id unsupported
- remove unnecessary xmit/recv functions as we rely on the fact that only a single port is active instead of mangling packets

---
 drivers/net/Kconfig     |   7 +
 drivers/net/Makefile    |   1 +
 drivers/net/mv88e6xxx.c | 832 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 840 insertions(+)
 create mode 100644 drivers/net/mv88e6xxx.c
drivers/net/Kconfig +7 -0
index 6bbbadc5eef3..67f23917f42c 100644
@@ -438,6 +438,13 @@ config KSZ9477
 	  This driver implements a DSA switch driver for the KSZ9477 family
 	  of GbE switches using the I2C interface.
 
+config MV88E6XXX
+	bool "Marvell MV88E6xxx Ethernet switch DSA driver"
+	depends on DM_DSA && DM_MDIO
+	help
+	  This driver implements a DSA switch driver for the MV88E6xxx family
+	  of Ethernet switches using the MDIO interface
+
 config MVGBE
 	bool "Marvell Orion5x/Kirkwood network interface support"
 	depends on ARCH_KIRKWOOD || ARCH_ORION5X
drivers/net/Makefile +1 -0
index 96b7678e9882..f05fd58e3fa7 100644
@@ -59,6 +59,7 @@ obj-$(CONFIG_MEDIATEK_ETH) += mtk_eth.o
 obj-$(CONFIG_MPC8XX_FEC) += mpc8xx_fec.o
 obj-$(CONFIG_MT7620_ETH) += mt7620-eth.o
 obj-$(CONFIG_MT7628_ETH) += mt7628-eth.o
+obj-$(CONFIG_MV88E6XXX) += mv88e6xxx.o
 obj-$(CONFIG_MVGBE) += mvgbe.o
 obj-$(CONFIG_MVMDIO) += mvmdio.o
 obj-$(CONFIG_MVNETA) += mvneta.o
drivers/net/mv88e6xxx.c new file +832 -0

832 lines not on this page · raw message

Signature · 1 line
2.25.1
#10 lore raw

Re: [PATCH v6 7/8] net: add MV88E61xx DSA driver

Quoted 10 lines · Tim Harvey

Add a DSA driver for the MV88E6xxx compatible Ethernet switches.

---
v6:
- update commit msg (s/MV88E61xx/MV88E6xxx)

really?

Quoted 7 lines

- remove GbE from commit msg and Kconfig desc
- squash accidental change from patch 7 into patch 6
- added error print on failure to read switch id
- mv88e6xxx_probe:
- check for switch enabled
- remove unused variable enabled
- remove unnecessary call to dsa_set_tagging

Quoted 8 lines

- port_probe:
- new function to configure phy features by calling phy_config
- port_enable:
- only enable energy-detect sensing on phy ports
- add phy/cmode verification mistmatch warning
- remove mv88e6xxx_fixed_port_setup()
- always force link up for fixed ports
- always set SERDES mode regardless of cpu port

really?

drivers/net/mv88e6xxx.c line 200

Quoted 13 lines
drivers/net/mv88e6xxx.c new file +13 -0
+struct mv88e6xxx_priv {
+	int smi_addr;
+	int id;
+	int port_count;		/* Number of switch ports */
+	int port_reg_base;	/* Base of the switch port registers */
+	u8 global1;	/* Offset of Switch Global 1 registers */
+	u8 global2;	/* Offset of Switch Global 2 registers */
+	u8 phy_ctrl1_en_det_shift; /* 'EDet' bit field offset */
+	u8 phy_ctrl1_en_det_width; /* Width of 'EDet' bit field */
+	u8 phy_ctrl1_en_det_ctrl;  /* 'EDet' control value */
+};
+
+static inline int smi_cmd(int cmd, int addr, int reg)

If the convention in U-Boot is the same as in Linux (which it well may be, except for the fact that more things escape review), then you should avoid manually defining functions as inline in C files. The compiler will decide whether to inline or not regardless of what you specify here.

Although I wonder if these couldn't be in fact expressed more clearly using plain, boring old logical | and << operators as part of a macro.

Quoted 59 lines
drivers/net/mv88e6xxx.c new file +59 -0
+{
+	cmd = bitfield_replace(cmd, SMI_CMD_ADDR_SHIFT, SMI_CMD_ADDR_WIDTH,
+			       addr);
+	cmd = bitfield_replace(cmd, SMI_CMD_REG_SHIFT, SMI_CMD_REG_WIDTH, reg);
+	return cmd;
+}
+
+static inline int smi_cmd_read(int addr, int reg)
+{
+	return smi_cmd(SMI_CMD_READ, addr, reg);
+}
+
+static inline int smi_cmd_write(int addr, int reg)
+{
+	return smi_cmd(SMI_CMD_WRITE, addr, reg);
+}
+
+/* Wait for the current SMI indirect command to complete */
+static int mv88e6xxx_smi_wait(struct udevice *dev, int smi_addr)
+{
+	int val;
+	u32 timeout = 100;
+
+	do {
+		val = dm_mdio_read(dev->parent, smi_addr, MDIO_DEVAD_NONE, SMI_CMD_REG);
+		if (val >= 0 && (val & SMI_BUSY) == 0)
+			return 0;
+
+		mdelay(1);
+	} while (--timeout);
+
+	dev_err(dev, "SMI busy timeout\n");
+	return -ETIMEDOUT;
+}
+
+static int mv88e6xxx_serdes_init(struct udevice *dev)
+{
+	int val;
+
+	val = mv88e6xxx_set_page(dev, DEVADDR_SERDES, PHY_PAGE_SERDES);
+	if (val < 0)
+		return val;
+
+	/* Power up serdes module */
+	val = mv88e6xxx_phy_read(dev, DEVADDR_SERDES, MII_BMCR);
+	if (val < 0)
+		return val;
+	val &= ~(BMCR_PDOWN);
+	val = mv88e6xxx_phy_write(dev, DEVADDR_SERDES, MII_BMCR, val);
+	if (val < 0)
+		return val;
+
+	return 0;
+}
+
+static int mv88e6xxx_port_enable(struct udevice *dev, int port, struct phy_device *phy)
+{
+	struct dsa_pdata *dsa_pdata = dev_get_uclass_plat(dev);
+	struct mv88e6xxx_priv *priv = dev_get_priv(dev);

drivers/net/mv88e6xxx.c

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+	int val, ret;
+
+	dev_dbg(dev, "%s P%d phy:0x%08x %s\n", __func__, port,
+		phy->phy_id, phy_string_for_interface(phy->interface));
+
+	/*
+	 * Enable energy-detect sensing on PHY, used to determine when a PHY
+	 * port is physically connected
+	 */
+	if (port != dsa_pdata->cpu_port) {

Can you test based on phy->interface == PHY_INTERFACE_MODE_INTERNAL? Support may come later for cascade ports, and those are ports without an internal PHY which are not CPU ports.

Quoted 14 lines
drivers/net/mv88e6xxx.c new file +14 -0
+		val = mv88e6xxx_phy_read(dev, port, PHY_REG_CTRL1);
+		if (val < 0)
+			return val;
+		val = bitfield_replace(val, priv->phy_ctrl1_en_det_shift,
+				       priv->phy_ctrl1_en_det_width,
+				       priv->phy_ctrl1_en_det_ctrl);
+		val = mv88e6xxx_phy_write(dev, port, PHY_REG_CTRL1, val);
+		if (val < 0)
+			return val;
+	}
+
+	/* enable port */
+	val = mv88e6xxx_port_read(dev, port, PORT_REG_CTRL);
+	if (val < 0)

drivers/net/mv88e6xxx.c line 654

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+		return val;
+	val = bitfield_replace(val, PORT_REG_CTRL_PSTATE_SHIFT,
+			       PORT_REG_CTRL_PSTATE_WIDTH,
+			       PORT_REG_CTRL_PSTATE_FORWARD);
+	val = mv88e6xxx_port_write(dev, port, PORT_REG_CTRL, val);
+	if (val < 0)
+		return val;
+
+	/* configure RGMII delays for fixed link */
+	if (phy->phy_id == PHY_FIXED_ID) {

I believe this code should really be executed only for RGMII ports. In other words, move it inside a function that gets called inside the switch (phy->interface) block you added below.

drivers/net/mv88e6xxx.c line 660

Quoted 6 lines
drivers/net/mv88e6xxx.c new file +6 -0
+		/* Physical Control register: Table 62 */
+		val = mv88e6xxx_port_read(dev, port, PORT_REG_PHYS_CTRL);
+
+		/* RGMII delays */
+		val &= ~(PORT_REG_PHYS_CTRL_RGMII_DELAY_RXCLK ||
+			 PORT_REG_PHYS_CTRL_RGMII_DELAY_TXCLK);

I find the style a bit chaotic when in other places you use bitfield_replace() but not everywhere.

Quoted 7 lines
drivers/net/mv88e6xxx.c new file +7 -0
+		if (phy->interface == PHY_INTERFACE_MODE_RGMII_ID ||
+		    phy->interface == PHY_INTERFACE_MODE_RGMII_RXID)
+			val |= PORT_REG_PHYS_CTRL_RGMII_DELAY_RXCLK;
+		if (phy->interface == PHY_INTERFACE_MODE_RGMII_ID ||
+		    phy->interface == PHY_INTERFACE_MODE_RGMII_TXID)
+			val |= PORT_REG_PHYS_CTRL_RGMII_DELAY_TXCLK;
+		/* Force Link */

drivers/net/mv88e6xxx.c line 677

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+		val |= PORT_REG_PHYS_CTRL_LINK_VALUE |
+		       PORT_REG_PHYS_CTRL_LINK_FORCE;
+
+		ret = mv88e6xxx_port_write(dev, port, PORT_REG_PHYS_CTRL, val);
+		if (ret < 0)
+			return ret;
+	}
+
+	if (mv88e6xxx_6352_family(dev)) {
+		val = mv88e6xxx_get_cmode(dev, dsa_pdata->cpu_port);

What gets handled by dsa_ops :: port_enable() should be a function of "int port", whereas dsa_pdata->cpu_port is an invariant of that.

Otherwise said, you check for the cmode of the CPU port for as many times as ports there are, and never for the other ports. Probably not what you intend.

drivers/net/mv88e6xxx.c line 684

Quoted 7 lines
drivers/net/mv88e6xxx.c new file +7 -0
+		if (val < 0)
+			return val;
+		/* initialize serdes */
+		if (val == PORT_REG_STATUS_CMODE_100BASE_X ||
+		    val == PORT_REG_STATUS_CMODE_1000BASE_X ||
+		    val == PORT_REG_STATUS_CMODE_SGMII) {
+			ret = mv88e6xxx_serdes_init(dev);

Again, you call mv88e6xxx_serdes_init() 10 times on the CPU port. The point of getting the C_Mode per port was to avoid directly checking for dsa_pdata->cpu_port.

drivers/net/mv88e6xxx.c line 689

Quoted 5 lines
drivers/net/mv88e6xxx.c new file +5 -0
+			if (ret < 0)
+				return ret;
+		}
+		/* validate interface type */
+		if (phy->phy_id == PHY_FIXED_ID) {

I think you mean here to validate the C_Mode only for ports not connected to internal PHYs. But SERDES/RGMII ports may also have PHYs, albeit external. You want to validate the C_Mode for them too. So this check must go. Please check for PHY_INTERFACE_MODE_INTERNAL, and avoid comparing it to the C_Mode in that case.

Quoted 13 lines
drivers/net/mv88e6xxx.c new file +13 -0
+			switch (phy->interface) {
+			case PHY_INTERFACE_MODE_RGMII:
+			case PHY_INTERFACE_MODE_RGMII_RXID:
+			case PHY_INTERFACE_MODE_RGMII_TXID:
+			case PHY_INTERFACE_MODE_RGMII_ID:
+				if (val != PORT_REG_STATUS_CMODE_RGMII)
+					goto mismatch;
+				break;
+			case PHY_INTERFACE_MODE_1000BASEX:
+				if (val != PORT_REG_STATUS_CMODE_1000BASE_X)
+					goto mismatch;
+				break;
+mismatch:

drivers/net/mv88e6xxx.c line 712

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+			default:
+				dev_err(dev, "Mismatched PHY mode %s on port %d!\n",
+					phy_string_for_interface(phy->interface), port);
+				break;
+			}
+		}
+	}
+
+	return phy_startup(phy);
+}

Overall it looks better than v5, but I believe that it could use a bit more attention to detail.

Also, I don't work with Marvell hardware, so I'm not all that familiar with it. I understand this only supports the SERDES as seen on the 6176 (where there is a single SERDES port), but I would appreciate if someone could take a second look at the hardware configuration in the physical interface area.

author
#11 lore raw

Re: [PATCH v6 7/8] net: add MV88E61xx DSA driver

Quoted 13 lines · Vladimir Oltean
Quoted 10 lines · Tim Harvey

Add a DSA driver for the MV88E6xxx compatible Ethernet switches.

---
v6:
- update commit msg (s/MV88E61xx/MV88E6xxx)

Quoted 1 line

really?

will fix in v7 - I updated the long msg and missed updating the short msg

Quoted 17 lines
Quoted 16 lines

- remove GbE from commit msg and Kconfig desc
- squash accidental change from patch 7 into patch 6
- added error print on failure to read switch id
- mv88e6xxx_probe:
- check for switch enabled
- remove unused variable enabled
- remove unnecessary call to dsa_set_tagging
- port_probe:
- new function to configure phy features by calling phy_config
- port_enable:
- only enable energy-detect sensing on phy ports
- add phy/cmode verification mistmatch warning
- remove mv88e6xxx_fixed_port_setup()
- always force link up for fixed ports
- always set SERDES mode regardless of cpu port

Quoted 1 line

really?

The SERDES configuration on the 88E6352/88E6240/88E6176/88E6172 (which I have documentation for) is not in port specific registers so I think I should move that to the probe function for the switch.

Quoted 20 lines
Quoted 4 lines
drivers/net/mv88e6xxx.c new file +3 -0
+struct mv88e6xxx_priv {
+     int smi_addr;
+     int id;

drivers/net/mv88e6xxx.c line 200

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+     int port_count;         /* Number of switch ports */
+     int port_reg_base;      /* Base of the switch port registers */
+     u8 global1;     /* Offset of Switch Global 1 registers */
+     u8 global2;     /* Offset of Switch Global 2 registers */
+     u8 phy_ctrl1_en_det_shift; /* 'EDet' bit field offset */
+     u8 phy_ctrl1_en_det_width; /* Width of 'EDet' bit field */
+     u8 phy_ctrl1_en_det_ctrl;  /* 'EDet' control value */
+};
+
+static inline int smi_cmd(int cmd, int addr, int reg)

If the convention in U-Boot is the same as in Linux (which it well may be, except for the fact that more things escape review), then you should avoid manually defining functions as inline in C files. The compiler will decide whether to inline or not regardless of what you specify here.

Quoted 2 lines

Although I wonder if these couldn't be in fact expressed more clearly using plain, boring old logical | and << operators as part of a macro.

The bitfield_replace came from the original drivers/net/phy/mv88e61xx.c but I also find it easier to read logical | and << operators so I'll change to that throughout for v7

Quoted 66 lines
Quoted 66 lines
drivers/net/mv88e6xxx.c new file +65 -0
+{
+     cmd = bitfield_replace(cmd, SMI_CMD_ADDR_SHIFT, SMI_CMD_ADDR_WIDTH,
+                            addr);
+     cmd = bitfield_replace(cmd, SMI_CMD_REG_SHIFT, SMI_CMD_REG_WIDTH, reg);
+     return cmd;
+}
+
+static inline int smi_cmd_read(int addr, int reg)
+{
+     return smi_cmd(SMI_CMD_READ, addr, reg);
+}
+
+static inline int smi_cmd_write(int addr, int reg)
+{
+     return smi_cmd(SMI_CMD_WRITE, addr, reg);
+}
+
+/* Wait for the current SMI indirect command to complete */
+static int mv88e6xxx_smi_wait(struct udevice *dev, int smi_addr)
+{
+     int val;
+     u32 timeout = 100;
+
+     do {
+             val = dm_mdio_read(dev->parent, smi_addr, MDIO_DEVAD_NONE, SMI_CMD_REG);
+             if (val >= 0 && (val & SMI_BUSY) == 0)
+                     return 0;
+
+             mdelay(1);
+     } while (--timeout);
+
+     dev_err(dev, "SMI busy timeout\n");
+     return -ETIMEDOUT;
+}
+
+static int mv88e6xxx_serdes_init(struct udevice *dev)
+{
+     int val;
+
+     val = mv88e6xxx_set_page(dev, DEVADDR_SERDES, PHY_PAGE_SERDES);
+     if (val < 0)
+             return val;
+
+     /* Power up serdes module */
+     val = mv88e6xxx_phy_read(dev, DEVADDR_SERDES, MII_BMCR);
+     if (val < 0)
+             return val;
+     val &= ~(BMCR_PDOWN);
+     val = mv88e6xxx_phy_write(dev, DEVADDR_SERDES, MII_BMCR, val);
+     if (val < 0)
+             return val;
+
+     return 0;
+}
+
+static int mv88e6xxx_port_enable(struct udevice *dev, int port, struct phy_device *phy)
+{
+     struct dsa_pdata *dsa_pdata = dev_get_uclass_plat(dev);
+     struct mv88e6xxx_priv *priv = dev_get_priv(dev);
+     int val, ret;
+
+     dev_dbg(dev, "%s P%d phy:0x%08x %s\n", __func__, port,
+             phy->phy_id, phy_string_for_interface(phy->interface));
+
+     /*
Quoted 8 lines

drivers/net/mv88e6xxx.c line 630

Quoted 4 lines
drivers/net/mv88e6xxx.c new file +4 -0
+      * Enable energy-detect sensing on PHY, used to determine when a PHY
+      * port is physically connected
+      */
+     if (port != dsa_pdata->cpu_port) {

Can you test based on phy->interface == PHY_INTERFACE_MODE_INTERNAL? Support may come later for cascade ports, and those are ports without an internal PHY which are not CPU ports.

In my config there are no ports with interface == PHY_INTERFACE_MODE_INTERNAL so I'm not sure what you mean here. What situations does that come up in?

Quoted 21 lines
Quoted 21 lines
drivers/net/mv88e6xxx.c new file +20 -0
+             val = mv88e6xxx_phy_read(dev, port, PHY_REG_CTRL1);
+             if (val < 0)
+                     return val;
+             val = bitfield_replace(val, priv->phy_ctrl1_en_det_shift,
+                                    priv->phy_ctrl1_en_det_width,
+                                    priv->phy_ctrl1_en_det_ctrl);
+             val = mv88e6xxx_phy_write(dev, port, PHY_REG_CTRL1, val);
+             if (val < 0)
+                     return val;
+     }
+
+     /* enable port */
+     val = mv88e6xxx_port_read(dev, port, PORT_REG_CTRL);
+     if (val < 0)
+             return val;
+     val = bitfield_replace(val, PORT_REG_CTRL_PSTATE_SHIFT,
+                            PORT_REG_CTRL_PSTATE_WIDTH,
+                            PORT_REG_CTRL_PSTATE_FORWARD);
+     val = mv88e6xxx_port_write(dev, port, PORT_REG_CTRL, val);
+     if (val < 0)
Quoted 8 lines

drivers/net/mv88e6xxx.c line 654

Quoted 4 lines
drivers/net/mv88e6xxx.c new file +4 -0
+             return val;
+
+     /* configure RGMII delays for fixed link */
+     if (phy->phy_id == PHY_FIXED_ID) {

I believe this code should really be executed only for RGMII ports. In other words, move it inside a function that gets called inside the switch (phy->interface) block you added below.

It makes sense to only do this for RGMII modes but I'll need to take it out of the if (mv88e61xx_6352_famiily()) as it looks like other switches that could be supported by this driver support RGMII interfaces.

Quoted 10 lines

drivers/net/mv88e6xxx.c line 660

Quoted 7 lines
drivers/net/mv88e6xxx.c new file +6 -0
+             /* Physical Control register: Table 62 */
+             val = mv88e6xxx_port_read(dev, port, PORT_REG_PHYS_CTRL);
+
+             /* RGMII delays */
+             val &= ~(PORT_REG_PHYS_CTRL_RGMII_DELAY_RXCLK ||
+                      PORT_REG_PHYS_CTRL_RGMII_DELAY_TXCLK);

I find the style a bit chaotic when in other places you use bitfield_replace() but not everywhere.

I'll get rid of bitfield_replace and leave these as is

Quoted 22 lines
Quoted 8 lines
drivers/net/mv88e6xxx.c new file +7 -0
+             if (phy->interface == PHY_INTERFACE_MODE_RGMII_ID ||
+                 phy->interface == PHY_INTERFACE_MODE_RGMII_RXID)
+                     val |= PORT_REG_PHYS_CTRL_RGMII_DELAY_RXCLK;
+             if (phy->interface == PHY_INTERFACE_MODE_RGMII_ID ||
+                 phy->interface == PHY_INTERFACE_MODE_RGMII_TXID)
+                     val |= PORT_REG_PHYS_CTRL_RGMII_DELAY_TXCLK;
+             /* Force Link */

drivers/net/mv88e6xxx.c line 677

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+             val |= PORT_REG_PHYS_CTRL_LINK_VALUE |
+                    PORT_REG_PHYS_CTRL_LINK_FORCE;
+
+             ret = mv88e6xxx_port_write(dev, port, PORT_REG_PHYS_CTRL, val);
+             if (ret < 0)
+                     return ret;
+     }
+
+     if (mv88e6xxx_6352_family(dev)) {
+             val = mv88e6xxx_get_cmode(dev, dsa_pdata->cpu_port);

What gets handled by dsa_ops :: port_enable() should be a function of "int port", whereas dsa_pdata->cpu_port is an invariant of that.

Quoted 3 lines

Otherwise said, you check for the cmode of the CPU port for as many times as ports there are, and never for the other ports. Probably not what you intend.

The original mv88e61xx driver initializes SERDES in mv88e61xx_set_cpu_port() called from mv88e61xx_switch_init() called from mv88e61xx_phy_config() for the cpu port PHY, only does this for the 6172/6176/6240/6352, and the registers are not port specific:

         /* If CPU is connected to serdes, initialize serdes */
        if (mv88e61xx_6352_family(phydev)) {
                val = mv88e61xx_get_cmode(phydev, CONFIG_MV88E61XX_CPU_PORT);
                if (val < 0)
                        return val;
                if (val == PORT_REG_STATUS_CMODE_100BASE_X ||
                    val == PORT_REG_STATUS_CMODE_1000BASE_X ||
                    val == PORT_REG_STATUS_CMODE_SGMII) {
                        val = mv88e61xx_serdes_init(phydev);
                        if (val < 0)
                                return val;
                }
        }

static int mv88e61xx_serdes_init(struct phy_device *phydev)

{
        int val;

        val = mv88e61xx_set_page(phydev, DEVADDR_SERDES, PHY_PAGE_SERDES);
        if (val < 0)
                return val;

        /* Power up serdes module */
        val = mv88e61xx_phy_read(phydev, DEVADDR_SERDES, MII_BMCR);
        if (val < 0)
                return val;
        val &= ~(BMCR_PDOWN);
        val = mv88e61xx_phy_write(phydev, DEVADDR_SERDES, MII_BMCR, val);
        if (val < 0)
                return val;

        return 0;
}

So I think the proper thing to do is move this SERDES init into my mv88e6xxx_probe function.

Quoted 19 lines

drivers/net/mv88e6xxx.c line 684

Quoted 8 lines
drivers/net/mv88e6xxx.c new file +7 -0
+             if (val < 0)
+                     return val;
+             /* initialize serdes */
+             if (val == PORT_REG_STATUS_CMODE_100BASE_X ||
+                 val == PORT_REG_STATUS_CMODE_1000BASE_X ||
+                 val == PORT_REG_STATUS_CMODE_SGMII) {
+                     ret = mv88e6xxx_serdes_init(dev);

Again, you call mv88e6xxx_serdes_init() 10 times on the CPU port. The point of getting the C_Mode per port was to avoid directly checking for dsa_pdata->cpu_port.

drivers/net/mv88e6xxx.c line 689

Quoted 5 lines
drivers/net/mv88e6xxx.c new file +5 -0
+                     if (ret < 0)
+                             return ret;
+             }
+             /* validate interface type */
+             if (phy->phy_id == PHY_FIXED_ID) {
Quoted 5 lines

I think you mean here to validate the C_Mode only for ports not connected to internal PHYs. But SERDES/RGMII ports may also have PHYs, albeit external. You want to validate the C_Mode for them too. So this check must go. Please check for PHY_INTERFACE_MODE_INTERNAL, and avoid comparing it to the C_Mode in that case.

I think my flaw here was that I checked c_mode for the cpu port above as you pointed out:

val = mv88e6xxx_get_cmode(dev, dsa_pdata->cpu_port);

Changing that to mv88e6xxx_get_cmode(dev, port) should resolve that but again I'm not sure where I would be seeing an interface mode of PHY_INTERFACE_MODE_INTERNAL

Quoted 28 lines
Quoted 14 lines
drivers/net/mv88e6xxx.c new file +13 -0
+                     switch (phy->interface) {
+                     case PHY_INTERFACE_MODE_RGMII:
+                     case PHY_INTERFACE_MODE_RGMII_RXID:
+                     case PHY_INTERFACE_MODE_RGMII_TXID:
+                     case PHY_INTERFACE_MODE_RGMII_ID:
+                             if (val != PORT_REG_STATUS_CMODE_RGMII)
+                                     goto mismatch;
+                             break;
+                     case PHY_INTERFACE_MODE_1000BASEX:
+                             if (val != PORT_REG_STATUS_CMODE_1000BASE_X)
+                                     goto mismatch;
+                             break;
+mismatch:

drivers/net/mv88e6xxx.c line 712

Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+                     default:
+                             dev_err(dev, "Mismatched PHY mode %s on port %d!\n",
+                                     phy_string_for_interface(phy->interface), port);
+                             break;
+                     }
+             }
+     }
+
+     return phy_startup(phy);
+}

Overall it looks better than v5, but I believe that it could use a bit more attention to detail.

Quoted 5 lines

Also, I don't work with Marvell hardware, so I'm not all that familiar with it. I understand this only supports the SERDES as seen on the 6176 (where there is a single SERDES port), but I would appreciate if someone could take a second look at the hardware configuration in the physical interface area.

The only hardware I have access to for testing has a MV88E6176 and the only register documentionat I have is from the '88E6352/88E6240/88E6176/88E6172 Functional Specification' which is what the driver considers the '6352 family' of Gigabit Ethernet switches. I did manage to find a Marvell 2020 Product Selector Guide
[1] that does give an outline of the various switches between pages 15 and 17 but without register definitions for the other switches all I have to go off of is the original u-boot driver and perhaps the Linux driver.

Best Regards,

Tim
[1] https://jp.marvell.com/content/dam/marvell/en/psg/marvell_psg.pdf

#12 lore raw

Re: [PATCH v6 7/8] net: add MV88E61xx DSA driver

Sorry for the late reply.

Quoted 3 lines · Tim Harvey

The SERDES configuration on the 88E6352/88E6240/88E6176/88E6172 (which I have documentation for) is not in port specific registers so I think I should move that to the probe function for the switch.

It's not in port specific registers for this switch family, because there's a single SERDES port. What you could do is you could do the SERDES initialization once per (this) switch in the per-port procedure, if the current port is a SERDES port according to the C_MODE. This would have to be done anyway for other switches. Anyway, what you did in v7 works too, you can leave it like that and others can refactor if needed (I think; I didn't look in too much detail yet).

Quoted 19 lines
Quoted 10 lines
Quoted 10 lines
drivers/net/mv88e6xxx.c new file +10 -0
+static int mv88e6xxx_port_enable(struct udevice *dev, int port, struct phy_device *phy)
+{
+     struct dsa_pdata *dsa_pdata = dev_get_uclass_plat(dev);
+     struct mv88e6xxx_priv *priv = dev_get_priv(dev);
+     int val, ret;
+
+     dev_dbg(dev, "%s P%d phy:0x%08x %s\n", __func__, port,
+             phy->phy_id, phy_string_for_interface(phy->interface));
+
+     /*
Quoted 8 lines

drivers/net/mv88e6xxx.c line 630

Quoted 4 lines
drivers/net/mv88e6xxx.c new file +4 -0
+      * Enable energy-detect sensing on PHY, used to determine when a PHY
+      * port is physically connected
+      */
+     if (port != dsa_pdata->cpu_port) {

Can you test based on phy->interface == PHY_INTERFACE_MODE_INTERNAL? Support may come later for cascade ports, and those are ports without an internal PHY which are not CPU ports.

Quoted 3 lines

In my config there are no ports with interface == PHY_INTERFACE_MODE_INTERNAL so I'm not sure what you mean here. What situations does that come up in?

Ugh. Again the nasty Linux dt-binding legacy for this switch....

So normally, an Ethernet port has a phy-handle/fixed-link, and a phy-mode. For a port connected to an internal PHY, you should have phy-handle = <&phy> (which you do), and phy-mode = "internal" (which you don't). Linux supports skipping both the phy-handle and the phy-mode.

Sorry for not realizing this in time. This is again a thing so that U-Boot can know that you're operating on a port with an internal PHY, rather than hardcoding this based on the assumption that the SERDES port is the CPU port, and everything else has internal PHY (which assumes RGMII does not exist).

author
#13 lore raw

PATCH v6 8/8 board: gw_ventana: enable MV88E6XXX DSA support

In reply to Tim Harvey

Add MV88E6XXX DSA support:
- update dt to provide internal MDIO bus and port handles. U-Boot requires a more restrictive subset of the dt bindings required by Linux for the sake of simplifying code
- update defconfig to remove old driver and enable new one
- replace mv88e61xx_hw_reset weak override with board_phy_config support for register configuration that is outside the scope of the DSA driver

---
v6:
- update commit message
- squash accidently change to mv88e6xxx driver into previous patch
- remove unused label for cpu port
v5:
- fix typo in defconfig s/MV88E61XX/MV88E6XXX
- added Fabio's rb tag
v4:
- use standard Linux internal MDIO dt structure
- use PHY_FIXED_ID define
v3:
- move mdio's mdio@0 subnode
v2: no changes

---
 arch/arm/dts/imx6qdl-gw5904.dtsi        | 32 +++++++++++++++-
 board/gateworks/gw_ventana/gw_ventana.c | 50 +++++++++----------------
 configs/gwventana_gw5904_defconfig      |  7 ++--
 3 files changed, 52 insertions(+), 37 deletions(-)
arch/arm/dts/imx6qdl-gw5904.dtsi +31 -1
index 612b6e068e28..c0790183c013 100644
@@ -212,6 +212,27 @@
 			compatible = "marvell,mv88e6085";
 			reg = <0>;
 
+			mdio {
+				#address-cells = <1>;
+				#size-cells = <0>;
+
+				sw_phy0: ethernet-phy@0 {
+					reg = <0x0>;
+				};
+
+				sw_phy1: ethernet-phy@1 {
+					reg = <0x1>;
+				};
+
+				sw_phy2: ethernet-phy@2 {
+					reg = <0x2>;
+				};
+
+				sw_phy3: ethernet-phy@3 {
+					reg = <0x3>;
+				};
+			};
+
 			ports {
 				#address-cells = <1>;
 				#size-cells = <0>;
@@ -219,27 +240,36 @@
 				port@0 {
 					reg = <0>;
 					label = "lan4";
+					phy-handle = <&sw_phy0>;
 				};
 
 				port@1 {
 					reg = <1>;
 					label = "lan3";
+					phy-handle = <&sw_phy1>;
 				};
 
 				port@2 {
 					reg = <2>;
 					label = "lan2";
+					phy-handle = <&sw_phy2>;
 				};
 
 				port@3 {
 					reg = <3>;
 					label = "lan1";
+					phy-handle = <&sw_phy3>;
 				};
 
 				port@5 {
 					reg = <5>;
-					label = "cpu";
 					ethernet = <&fec>;
+					phy-mode = "rgmii-id";
+
+					fixed-link {
+						speed = <1000>;
+						full-duplex;
+					};
 				};
 			};
 		};
board/gateworks/gw_ventana/gw_ventana.c +18 -32

62 lines not on this page · raw message

configs/gwventana_gw5904_defconfig +3 -4
index fb5870fa5803..50115b21e263 100644
@@ -109,13 +109,12 @@ CONFIG_SUPPORT_EMMC_BOOT=y
 CONFIG_FSL_USDHC=y
 CONFIG_MTD=y
 CONFIG_PHYLIB=y
-CONFIG_MV88E61XX_SWITCH=y
-CONFIG_MV88E61XX_CPU_PORT=5
-CONFIG_MV88E61XX_PHY_PORTS=0xf
-CONFIG_MV88E61XX_FIXED_PORTS=0x0
+CONFIG_PHY_FIXED=y
 CONFIG_DM_MDIO=y
+CONFIG_DM_DSA=y
 CONFIG_E1000=y
 CONFIG_FEC_MXC=y
+CONFIG_MV88E6XXX=y
 CONFIG_MII=y
 CONFIG_PCI=y
 CONFIG_PCIE_IMX=y
Signature · 1 line
2.25.1
Reviewed-by on v6 8/8
#14 lore raw

Re: [PATCH v6 8/8] board: gw_ventana: enable MV88E6XXX DSA support

Quoted 11 lines · Tim Harvey

Add MV88E6XXX DSA support:
- update dt to provide internal MDIO bus and port handles. U-Boot requires a more restrictive subset of the dt bindings required by Linux for the sake of simplifying code
- update defconfig to remove old driver and enable new one
- replace mv88e61xx_hw_reset weak override with board_phy_config support for register configuration that is outside the scope of the DSA driver

---