Merge branch 'net-dsa-hsr-enable-hsr-hw-offloading-for-ksz9477'

Lukasz Majewski says:

====================
net: dsa: hsr: Enable HSR HW offloading for KSZ9477

This patch series provides support for HSR HW offloading in KSZ9477
switch IC.

To test this feature:
ip link add name hsr0 type hsr slave1 lan1 slave2 lan2 supervision 45 version 1
ip link set dev lan1 up
ip link set dev lan2 up
ip a add 192.168.0.1/24 dev hsr0
ip link set dev hsr0 up

To remove HSR network device:
ip link del hsr0

To test if one can adjust MAC address:
ip link set lan2 address 00:01:02:AA:BB:CC

It is also possible to create another HSR interface, but it will
only support HSR is software - e.g.
ip link add name hsr1 type hsr slave1 lan3 slave2 lan4 supervision 45 version 1

Test HW:
Two KSZ9477-EVB boards with HSR ports set to "Port1" and "Port2".

Performance SW used:
nuttcp -S --nofork
nuttcp -vv -T 60 -r 192.168.0.2
nuttcp -vv -T 60 -t 192.168.0.2

Code: v6.6.0-rc2+ Linux net-next repository
SHA1: 5a1b322c

Tested HSR v0 and v1
Results:
With KSZ9477 offloading support added: RX: 100 Mbps TX: 98 Mbps
With no offloading 		       RX: 63 Mbps  TX: 63 Mbps
====================

Link: https://lore.kernel.org/r/20230922133108.2090612-1-lukma@denx.deSigned-off-by: Paolo Abeni <pabeni@redhat.com>

drivers/net/dsa/microchip/ksz8795_reg.h

@@ -323,13 +323,6 @@
 	((addr) + REG_PORT_1_CTRL_0 + (port) *	\
 		(REG_PORT_2_CTRL_0 - REG_PORT_1_CTRL_0))
 
-#define REG_SW_MAC_ADDR_0		0x68
-#define REG_SW_MAC_ADDR_1		0x69
-#define REG_SW_MAC_ADDR_2		0x6A
-#define REG_SW_MAC_ADDR_3		0x6B
-#define REG_SW_MAC_ADDR_4		0x6C
-#define REG_SW_MAC_ADDR_5		0x6D
-
 #define TABLE_EXT_SELECT_S		5
 #define TABLE_EEE_V			1
 #define TABLE_ACL_V			2

drivers/net/dsa/microchip/ksz9477.c

@@ -1143,6 +1143,83 @@ int ksz9477_tc_cbs_set_cinc(struct ksz_device *dev, int port, u32 val)
 	return ksz_pwrite16(dev, port, REG_PORT_MTI_CREDIT_INCREMENT, val);
 }
 
+/* The KSZ9477 provides following HW features to accelerate
+ * HSR frames handling:
+ *
+ * 1. TX PACKET DUPLICATION FROM HOST TO SWITCH
+ * 2. RX PACKET DUPLICATION DISCARDING
+ * 3. PREVENTING PACKET LOOP IN THE RING BY SELF-ADDRESS FILTERING
+ *
+ * Only one from point 1. has the NETIF_F* flag available.
+ *
+ * Ones from point 2 and 3 are "best effort" - i.e. those will
+ * work correctly most of the time, but it may happen that some
+ * frames will not be caught - to be more specific; there is a race
+ * condition in hardware such that, when duplicate packets are received
+ * on member ports very close in time to each other, the hardware fails
+ * to detect that they are duplicates.
+ *
+ * Hence, the SW needs to handle those special cases. However, the speed
+ * up gain is considerable when above features are used.
+ *
+ * Moreover, the NETIF_F_HW_HSR_FWD feature is also enabled, as HSR frames
+ * can be forwarded in the switch fabric between HSR ports.
+ */
+#define KSZ9477_SUPPORTED_HSR_FEATURES (NETIF_F_HW_HSR_DUP | NETIF_F_HW_HSR_FWD)
+
+void ksz9477_hsr_join(struct dsa_switch *ds, int port, struct net_device *hsr)
+{
+	struct ksz_device *dev = ds->priv;
+	struct net_device *slave;
+	struct dsa_port *hsr_dp;
+	u8 data, hsr_ports = 0;
+
+	/* Program which port(s) shall support HSR */
+	ksz_rmw32(dev, REG_HSR_PORT_MAP__4, BIT(port), BIT(port));
+
+	/* Forward frames between HSR ports (i.e. bridge together HSR ports) */
+	if (dev->hsr_ports) {
+		dsa_hsr_foreach_port(hsr_dp, ds, hsr)
+			hsr_ports |= BIT(hsr_dp->index);
+
+		hsr_ports |= BIT(dsa_upstream_port(ds, port));
+		dsa_hsr_foreach_port(hsr_dp, ds, hsr)
+			ksz9477_cfg_port_member(dev, hsr_dp->index, hsr_ports);
+	}
+
+	if (!dev->hsr_ports) {
+		/* Enable discarding of received HSR frames */
+		ksz_read8(dev, REG_HSR_ALU_CTRL_0__1, &data);
+		data |= HSR_DUPLICATE_DISCARD;
+		data &= ~HSR_NODE_UNICAST;
+		ksz_write8(dev, REG_HSR_ALU_CTRL_0__1, data);
+	}
+
+	/* Enable per port self-address filtering.
+	 * The global self-address filtering has already been enabled in the
+	 * ksz9477_reset_switch() function.
+	 */
+	ksz_port_cfg(dev, port, REG_PORT_LUE_CTRL, PORT_SRC_ADDR_FILTER, true);
+
+	/* Setup HW supported features for lan HSR ports */
+	slave = dsa_to_port(ds, port)->slave;
+	slave->features |= KSZ9477_SUPPORTED_HSR_FEATURES;
+}
+
+void ksz9477_hsr_leave(struct dsa_switch *ds, int port, struct net_device *hsr)
+{
+	struct ksz_device *dev = ds->priv;
+
+	/* Clear port HSR support */
+	ksz_rmw32(dev, REG_HSR_PORT_MAP__4, BIT(port), 0);
+
+	/* Disable forwarding frames between HSR ports */
+	ksz9477_cfg_port_member(dev, port, BIT(dsa_upstream_port(ds, port)));
+
+	/* Disable per port self-address filtering */
+	ksz_port_cfg(dev, port, REG_PORT_LUE_CTRL, PORT_SRC_ADDR_FILTER, false);
+}
+
 int ksz9477_switch_init(struct ksz_device *dev)
 {
 	u8 data8;

drivers/net/dsa/microchip/ksz9477.h

@@ -56,6 +56,8 @@ int ksz9477_reset_switch(struct ksz_device *dev);
 int ksz9477_switch_init(struct ksz_device *dev);
 void ksz9477_switch_exit(struct ksz_device *dev);
 void ksz9477_port_queue_split(struct ksz_device *dev, int port);
+void ksz9477_hsr_join(struct dsa_switch *ds, int port, struct net_device *hsr);
+void ksz9477_hsr_leave(struct dsa_switch *ds, int port, struct net_device *hsr);
 
 int ksz9477_port_acl_init(struct ksz_device *dev, int port);
 void ksz9477_port_acl_free(struct ksz_device *dev, int port);

drivers/net/dsa/microchip/ksz9477_reg.h

@@ -153,13 +153,6 @@
 #define SW_DOUBLE_TAG			BIT(7)
 #define SW_RESET			BIT(1)
 
-#define REG_SW_MAC_ADDR_0		0x0302
-#define REG_SW_MAC_ADDR_1		0x0303
-#define REG_SW_MAC_ADDR_2		0x0304
-#define REG_SW_MAC_ADDR_3		0x0305
-#define REG_SW_MAC_ADDR_4		0x0306
-#define REG_SW_MAC_ADDR_5		0x0307
-
 #define REG_SW_MTU__2			0x0308
 #define REG_SW_MTU_MASK			GENMASK(13, 0)
 

drivers/net/dsa/microchip/ksz_common.c

@@ -16,6 +16,7 @@
 #include <linux/etherdevice.h>
 #include <linux/if_bridge.h>
 #include <linux/if_vlan.h>
+#include <linux/if_hsr.h>
 #include <linux/irq.h>
 #include <linux/irqdomain.h>
 #include <linux/of.h>
@@ -364,6 +365,7 @@ static const struct ksz_dev_ops lan937x_dev_ops = {
 };
 
 static const u16 ksz8795_regs[] = {
+	[REG_SW_MAC_ADDR]		= 0x68,
 	[REG_IND_CTRL_0]		= 0x6E,
 	[REG_IND_DATA_8]		= 0x70,
 	[REG_IND_DATA_CHECK]		= 0x72,
@@ -492,6 +494,7 @@ static u8 ksz8863_shifts[] = {
 };
 
 static const u16 ksz9477_regs[] = {
+	[REG_SW_MAC_ADDR]		= 0x0302,
 	[P_STP_CTRL]			= 0x0B04,
 	[S_START_CTRL]			= 0x0300,
 	[S_BROADCAST_CTRL]		= 0x0332,
@@ -3539,6 +3542,149 @@ static int ksz_setup_tc(struct dsa_switch *ds, int port,
 	}
 }
 
+static int ksz_port_set_mac_address(struct dsa_switch *ds, int port,
+				    const unsigned char *addr)
+{
+	struct dsa_port *dp = dsa_to_port(ds, port);
+
+	if (dp->hsr_dev) {
+		dev_err(ds->dev,
+			"Cannot change MAC address on port %d with active HSR offload\n",
+			port);
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+/* Program the switch's MAC address register with the MAC address of the
+ * requesting user port. This single address is used by the switch for multiple
+ * features, like HSR self-address filtering and WoL. Other user ports are
+ * allowed to share ownership of this address as long as their MAC address is
+ * the same. The user ports' MAC addresses must not change while they have
+ * ownership of the switch MAC address.
+ */
+static int ksz_switch_macaddr_get(struct dsa_switch *ds, int port,
+				  struct netlink_ext_ack *extack)
+{
+	struct net_device *slave = dsa_to_port(ds, port)->slave;
+	const unsigned char *addr = slave->dev_addr;
+	struct ksz_switch_macaddr *switch_macaddr;
+	struct ksz_device *dev = ds->priv;
+	const u16 *regs = dev->info->regs;
+	int i;
+
+	/* Make sure concurrent MAC address changes are blocked */
+	ASSERT_RTNL();
+
+	switch_macaddr = dev->switch_macaddr;
+	if (switch_macaddr) {
+		if (!ether_addr_equal(switch_macaddr->addr, addr)) {
+			NL_SET_ERR_MSG_FMT_MOD(extack,
+					       "Switch already configured for MAC address %pM",
+					       switch_macaddr->addr);
+			return -EBUSY;
+		}
+
+		refcount_inc(&switch_macaddr->refcount);
+		return 0;
+	}
+
+	switch_macaddr = kzalloc(sizeof(*switch_macaddr), GFP_KERNEL);
+	if (!switch_macaddr)
+		return -ENOMEM;
+
+	ether_addr_copy(switch_macaddr->addr, addr);
+	refcount_set(&switch_macaddr->refcount, 1);
+	dev->switch_macaddr = switch_macaddr;
+
+	/* Program the switch MAC address to hardware */
+	for (i = 0; i < ETH_ALEN; i++)
+		ksz_write8(dev, regs[REG_SW_MAC_ADDR] + i, addr[i]);
+
+	return 0;
+}
+
+static void ksz_switch_macaddr_put(struct dsa_switch *ds)
+{
+	struct ksz_switch_macaddr *switch_macaddr;
+	struct ksz_device *dev = ds->priv;
+	const u16 *regs = dev->info->regs;
+	int i;
+
+	/* Make sure concurrent MAC address changes are blocked */
+	ASSERT_RTNL();
+
+	switch_macaddr = dev->switch_macaddr;
+	if (!refcount_dec_and_test(&switch_macaddr->refcount))
+		return;
+
+	for (i = 0; i < ETH_ALEN; i++)
+		ksz_write8(dev, regs[REG_SW_MAC_ADDR] + i, 0);
+
+	dev->switch_macaddr = NULL;
+	kfree(switch_macaddr);
+}
+
+static int ksz_hsr_join(struct dsa_switch *ds, int port, struct net_device *hsr,
+			struct netlink_ext_ack *extack)
+{
+	struct ksz_device *dev = ds->priv;
+	enum hsr_version ver;
+	int ret;
+
+	ret = hsr_get_version(hsr, &ver);
+	if (ret)
+		return ret;
+
+	if (dev->chip_id != KSZ9477_CHIP_ID) {
+		NL_SET_ERR_MSG_MOD(extack, "Chip does not support HSR offload");
+		return -EOPNOTSUPP;
+	}
+
+	/* KSZ9477 can support HW offloading of only 1 HSR device */
+	if (dev->hsr_dev && hsr != dev->hsr_dev) {
+		NL_SET_ERR_MSG_MOD(extack, "Offload supported for a single HSR");
+		return -EOPNOTSUPP;
+	}
+
+	/* KSZ9477 only supports HSR v0 and v1 */
+	if (!(ver == HSR_V0 || ver == HSR_V1)) {
+		NL_SET_ERR_MSG_MOD(extack, "Only HSR v0 and v1 supported");
+		return -EOPNOTSUPP;
+	}
+
+	/* Self MAC address filtering, to avoid frames traversing
+	 * the HSR ring more than once.
+	 */
+	ret = ksz_switch_macaddr_get(ds, port, extack);
+	if (ret)
+		return ret;
+
+	ksz9477_hsr_join(ds, port, hsr);
+	dev->hsr_dev = hsr;
+	dev->hsr_ports |= BIT(port);
+
+	return 0;
+}
+
+static int ksz_hsr_leave(struct dsa_switch *ds, int port,
+			 struct net_device *hsr)
+{
+	struct ksz_device *dev = ds->priv;
+
+	WARN_ON(dev->chip_id != KSZ9477_CHIP_ID);
+
+	ksz9477_hsr_leave(ds, port, hsr);
+	dev->hsr_ports &= ~BIT(port);
+	if (!dev->hsr_ports)
+		dev->hsr_dev = NULL;
+
+	ksz_switch_macaddr_put(ds);
+
+	return 0;
+}
+
 static const struct dsa_switch_ops ksz_switch_ops = {
 	.get_tag_protocol	= ksz_get_tag_protocol,
 	.connect_tag_protocol   = ksz_connect_tag_protocol,
@@ -3558,6 +3704,9 @@ static const struct dsa_switch_ops ksz_switch_ops = {
 	.get_sset_count		= ksz_sset_count,
 	.port_bridge_join	= ksz_port_bridge_join,
 	.port_bridge_leave	= ksz_port_bridge_leave,
+	.port_hsr_join		= ksz_hsr_join,
+	.port_hsr_leave		= ksz_hsr_leave,
+	.port_set_mac_address	= ksz_port_set_mac_address,
 	.port_stp_state_set	= ksz_port_stp_state_set,
 	.port_teardown		= ksz_port_teardown,
 	.port_pre_bridge_flags	= ksz_port_pre_bridge_flags,

drivers/net/dsa/microchip/ksz_common.h

@@ -101,6 +101,11 @@ struct ksz_ptp_irq {
 	int num;
 };
 
+struct ksz_switch_macaddr {
+	unsigned char addr[ETH_ALEN];
+	refcount_t refcount;
+};
+
 struct ksz_port {
 	bool remove_tag;		/* Remove Tag flag set, for ksz8795 only */
 	bool learning;
@@ -170,6 +175,10 @@ struct ksz_device {
 	struct mutex lock_irq;		/* IRQ Access */
 	struct ksz_irq girq;
 	struct ksz_ptp_data ptp_data;
+
+	struct ksz_switch_macaddr *switch_macaddr;
+	struct net_device *hsr_dev;     /* HSR */
+	u8 hsr_ports;
 };
 
 /* List of supported models */
@@ -212,6 +221,7 @@ enum ksz_chip_id {
 };
 
 enum ksz_regs {
+	REG_SW_MAC_ADDR,
 	REG_IND_CTRL_0,
 	REG_IND_DATA_8,
 	REG_IND_DATA_CHECK,

drivers/net/dsa/xrs700x/xrs700x.c

@@ -548,7 +548,8 @@ static void xrs700x_bridge_leave(struct dsa_switch *ds, int port,
 }
 
 static int xrs700x_hsr_join(struct dsa_switch *ds, int port,
-			    struct net_device *hsr)
+			    struct net_device *hsr,
+			    struct netlink_ext_ack *extack)
 {
 	unsigned int val = XRS_HSR_CFG_HSR_PRP;
 	struct dsa_port *partner = NULL, *dp;
@@ -562,16 +563,21 @@ static int xrs700x_hsr_join(struct dsa_switch *ds, int port,
 	if (ret)
 		return ret;
 
-	/* Only ports 1 and 2 can be HSR/PRP redundant ports. */
-	if (port != 1 && port != 2)
+	if (port != 1 && port != 2) {
+		NL_SET_ERR_MSG_MOD(extack,
+				   "Only ports 1 and 2 can offload HSR/PRP");
 		return -EOPNOTSUPP;
+	}
 
-	if (ver == HSR_V1)
+	if (ver == HSR_V1) {
 		val |= XRS_HSR_CFG_HSR;
-	else if (ver == PRP_V1)
+	} else if (ver == PRP_V1) {
 		val |= XRS_HSR_CFG_PRP;
-	else
+	} else {
+		NL_SET_ERR_MSG_MOD(extack,
+				   "Only HSR v1 and PRP v1 can be offloaded");
 		return -EOPNOTSUPP;
+	}
 
 	dsa_hsr_foreach_port(dp, ds, hsr) {
 		if (dp->index != port) {

include/net/dsa.h

@@ -969,6 +969,16 @@ struct dsa_switch_ops {
 			       struct phy_device *phy);
 	void	(*port_disable)(struct dsa_switch *ds, int port);
 
+
+	/*
+	 * Notification for MAC address changes on user ports. Drivers can
+	 * currently only veto operations. They should not use the method to
+	 * program the hardware, since the operation is not rolled back in case
+	 * of other errors.
+	 */
+	int	(*port_set_mac_address)(struct dsa_switch *ds, int port,
+					const unsigned char *addr);
+
 	/*
 	 * Compatibility between device trees defining multiple CPU ports and
 	 * drivers which are not OK to use by default the numerically smallest
@@ -1198,7 +1208,8 @@ struct dsa_switch_ops {
 	 * HSR integration
 	 */
 	int	(*port_hsr_join)(struct dsa_switch *ds, int port,
-				 struct net_device *hsr);
+				 struct net_device *hsr,
+				 struct netlink_ext_ack *extack);
 	int	(*port_hsr_leave)(struct dsa_switch *ds, int port,
 				  struct net_device *hsr);
 

net/dsa/port.c

@@ -2024,7 +2024,8 @@ void dsa_shared_port_link_unregister_of(struct dsa_port *dp)
 		dsa_shared_port_setup_phy_of(dp, false);
 }
 
-int dsa_port_hsr_join(struct dsa_port *dp, struct net_device *hsr)
+int dsa_port_hsr_join(struct dsa_port *dp, struct net_device *hsr,
+		      struct netlink_ext_ack *extack)
 {
 	struct dsa_switch *ds = dp->ds;
 	int err;
@@ -2034,7 +2035,7 @@ int dsa_port_hsr_join(struct dsa_port *dp, struct net_device *hsr)
 
 	dp->hsr_dev = hsr;
 
-	err = ds->ops->port_hsr_join(ds, dp->index, hsr);
+	err = ds->ops->port_hsr_join(ds, dp->index, hsr, extack);
 	if (err)
 		dp->hsr_dev = NULL;
 

net/dsa/port.h

@@ -103,7 +103,8 @@ int dsa_port_phylink_create(struct dsa_port *dp);
 void dsa_port_phylink_destroy(struct dsa_port *dp);
 int dsa_shared_port_link_register_of(struct dsa_port *dp);
 void dsa_shared_port_link_unregister_of(struct dsa_port *dp);
-int dsa_port_hsr_join(struct dsa_port *dp, struct net_device *hsr);
+int dsa_port_hsr_join(struct dsa_port *dp, struct net_device *hsr,
+		      struct netlink_ext_ack *extack);
 void dsa_port_hsr_leave(struct dsa_port *dp, struct net_device *hsr);
 int dsa_port_tag_8021q_vlan_add(struct dsa_port *dp, u16 vid, bool broadcast);
 void dsa_port_tag_8021q_vlan_del(struct dsa_port *dp, u16 vid, bool broadcast);

net/dsa/slave.c

@@ -457,6 +457,13 @@ static int dsa_slave_set_mac_address(struct net_device *dev, void *a)
 	if (!is_valid_ether_addr(addr->sa_data))
 		return -EADDRNOTAVAIL;
 
+	if (ds->ops->port_set_mac_address) {
+		err = ds->ops->port_set_mac_address(ds, dp->index,
+						    addr->sa_data);
+		if (err)
+			return err;
+	}
+
 	/* If the port is down, the address isn't synced yet to hardware or
 	 * to the DSA master, so there is nothing to change.
 	 */
@@ -2862,7 +2869,7 @@ static int dsa_slave_changeupper(struct net_device *dev,
 		}
 	} else if (is_hsr_master(info->upper_dev)) {
 		if (info->linking) {
-			err = dsa_port_hsr_join(dp, info->upper_dev);
+			err = dsa_port_hsr_join(dp, info->upper_dev, extack);
 			if (err == -EOPNOTSUPP) {
 				NL_SET_ERR_MSG_WEAK_MOD(extack,
 							"Offloading not supported");

net/dsa/tag_ksz.c

@@ -293,6 +293,14 @@ static struct sk_buff *ksz9477_xmit(struct sk_buff *skb,
 	if (is_link_local_ether_addr(hdr->h_dest))
 		val |= KSZ9477_TAIL_TAG_OVERRIDE;
 
+	if (dev->features & NETIF_F_HW_HSR_DUP) {
+		struct net_device *hsr_dev = dp->hsr_dev;
+		struct dsa_port *other_dp;
+
+		dsa_hsr_foreach_port(other_dp, dp->ds, hsr_dev)
+			val |= BIT(other_dp->index);
+	}
+
 	*tag = cpu_to_be16(val);
 
 	return ksz_defer_xmit(dp, skb);