net: dsa: mv88e6xxx: Link aggregation support

Support offloading of LAGs to hardware. LAGs may be attached to a
bridge in which case VLANs, multicast groups, etc. are also offloaded
as usual.

Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 295 insertions, 1 deletions

diff --git a/drivers/net/dsa/mv88e6xxx/chip.c b/drivers/net/dsa/mv88e6xxx/chip.c
index 4aa7d0a8f197..dcb1726b68cc 100644
--- a/drivers/net/dsa/mv88e6xxx/chip.c
+++ b/drivers/net/dsa/mv88e6xxx/chip.c
@@ -1396,15 +1396,32 @@ static int mv88e6xxx_mac_setup(struct mv88e6xxx_chip *chip)
 
 static int mv88e6xxx_pvt_map(struct mv88e6xxx_chip *chip, int dev, int port)
 {
+	struct dsa_switch_tree *dst = chip->ds->dst;
+	struct dsa_switch *ds;
+	struct dsa_port *dp;
 	u16 pvlan = 0;
 
 	if (!mv88e6xxx_has_pvt(chip))
 		return 0;
 
 	/* Skip the local source device, which uses in-chip port VLAN */
-	if (dev != chip->ds->index)
+	if (dev != chip->ds->index) {
 		pvlan = mv88e6xxx_port_vlan(chip, dev, port);
 
+		ds = dsa_switch_find(dst->index, dev);
+		dp = ds ? dsa_to_port(ds, port) : NULL;
+		if (dp && dp->lag_dev) {
+			/* As the PVT is used to limit flooding of
+			 * FORWARD frames, which use the LAG ID as the
+			 * source port, we must translate dev/port to
+			 * the special "LAG device" in the PVT, using
+			 * the LAG ID as the port number.
+			 */
+			dev = MV88E6XXX_G2_PVT_ADRR_DEV_TRUNK;
+			port = dsa_lag_id(dst, dp->lag_dev);
+		}
+	}
+
 	return mv88e6xxx_g2_pvt_write(chip, dev, port, pvlan);
 }
 
@@ -5364,6 +5381,271 @@ static int mv88e6xxx_port_egress_floods(struct dsa_switch *ds, int port,
 	return err;
 }
 
+static bool mv88e6xxx_lag_can_offload(struct dsa_switch *ds,
+				      struct net_device *lag,
+				      struct netdev_lag_upper_info *info)
+{
+	struct dsa_port *dp;
+	int id, members = 0;
+
+	id = dsa_lag_id(ds->dst, lag);
+	if (id < 0 || id >= ds->num_lag_ids)
+		return false;
+
+	dsa_lag_foreach_port(dp, ds->dst, lag)
+		/* Includes the port joining the LAG */
+		members++;
+
+	if (members > 8)
+		return false;
+
+	/* We could potentially relax this to include active
+	 * backup in the future.
+	 */
+	if (info->tx_type != NETDEV_LAG_TX_TYPE_HASH)
+		return false;
+
+	/* Ideally we would also validate that the hash type matches
+	 * the hardware. Alas, this is always set to unknown on team
+	 * interfaces.
+	 */
+	return true;
+}
+
+static int mv88e6xxx_lag_sync_map(struct dsa_switch *ds, struct net_device *lag)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	struct dsa_port *dp;
+	u16 map = 0;
+	int id;
+
+	id = dsa_lag_id(ds->dst, lag);
+
+	/* Build the map of all ports to distribute flows destined for
+	 * this LAG. This can be either a local user port, or a DSA
+	 * port if the LAG port is on a remote chip.
+	 */
+	dsa_lag_foreach_port(dp, ds->dst, lag)
+		map |= BIT(dsa_towards_port(ds, dp->ds->index, dp->index));
+
+	return mv88e6xxx_g2_trunk_mapping_write(chip, id, map);
+}
+
+static const u8 mv88e6xxx_lag_mask_table[8][8] = {
+	/* Row number corresponds to the number of active members in a
+	 * LAG. Each column states which of the eight hash buckets are
+	 * mapped to the column:th port in the LAG.
+	 *
+	 * Example: In a LAG with three active ports, the second port
+	 * ([2][1]) would be selected for traffic mapped to buckets
+	 * 3,4,5 (0x38).
+	 */
+	{ 0xff,    0,    0,    0,    0,    0,    0,    0 },
+	{ 0x0f, 0xf0,    0,    0,    0,    0,    0,    0 },
+	{ 0x07, 0x38, 0xc0,    0,    0,    0,    0,    0 },
+	{ 0x03, 0x0c, 0x30, 0xc0,    0,    0,    0,    0 },
+	{ 0x03, 0x0c, 0x30, 0x40, 0x80,    0,    0,    0 },
+	{ 0x03, 0x0c, 0x10, 0x20, 0x40, 0x80,    0,    0 },
+	{ 0x03, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,    0 },
+	{ 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80 },
+};
+
+static void mv88e6xxx_lag_set_port_mask(u16 *mask, int port,
+					int num_tx, int nth)
+{
+	u8 active = 0;
+	int i;
+
+	num_tx = num_tx <= 8 ? num_tx : 8;
+	if (nth < num_tx)
+		active = mv88e6xxx_lag_mask_table[num_tx - 1][nth];
+
+	for (i = 0; i < 8; i++) {
+		if (BIT(i) & active)
+			mask[i] |= BIT(port);
+	}
+}
+
+static int mv88e6xxx_lag_sync_masks(struct dsa_switch *ds)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	unsigned int id, num_tx;
+	struct net_device *lag;
+	struct dsa_port *dp;
+	int i, err, nth;
+	u16 mask[8];
+	u16 ivec;
+
+	/* Assume no port is a member of any LAG. */
+	ivec = BIT(mv88e6xxx_num_ports(chip)) - 1;
+
+	/* Disable all masks for ports that _are_ members of a LAG. */
+	list_for_each_entry(dp, &ds->dst->ports, list) {
+		if (!dp->lag_dev || dp->ds != ds)
+			continue;
+
+		ivec &= ~BIT(dp->index);
+	}
+
+	for (i = 0; i < 8; i++)
+		mask[i] = ivec;
+
+	/* Enable the correct subset of masks for all LAG ports that
+	 * are in the Tx set.
+	 */
+	dsa_lags_foreach_id(id, ds->dst) {
+		lag = dsa_lag_dev(ds->dst, id);
+		if (!lag)
+			continue;
+
+		num_tx = 0;
+		dsa_lag_foreach_port(dp, ds->dst, lag) {
+			if (dp->lag_tx_enabled)
+				num_tx++;
+		}
+
+		if (!num_tx)
+			continue;
+
+		nth = 0;
+		dsa_lag_foreach_port(dp, ds->dst, lag) {
+			if (!dp->lag_tx_enabled)
+				continue;
+
+			if (dp->ds == ds)
+				mv88e6xxx_lag_set_port_mask(mask, dp->index,
+							    num_tx, nth);
+
+			nth++;
+		}
+	}
+
+	for (i = 0; i < 8; i++) {
+		err = mv88e6xxx_g2_trunk_mask_write(chip, i, true, mask[i]);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static int mv88e6xxx_lag_sync_masks_map(struct dsa_switch *ds,
+					struct net_device *lag)
+{
+	int err;
+
+	err = mv88e6xxx_lag_sync_masks(ds);
+
+	if (!err)
+		err = mv88e6xxx_lag_sync_map(ds, lag);
+
+	return err;
+}
+
+static int mv88e6xxx_port_lag_change(struct dsa_switch *ds, int port)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	int err;
+
+	mv88e6xxx_reg_lock(chip);
+	err = mv88e6xxx_lag_sync_masks(ds);
+	mv88e6xxx_reg_unlock(chip);
+	return err;
+}
+
+static int mv88e6xxx_port_lag_join(struct dsa_switch *ds, int port,
+				   struct net_device *lag,
+				   struct netdev_lag_upper_info *info)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	int err, id;
+
+	if (!mv88e6xxx_lag_can_offload(ds, lag, info))
+		return -EOPNOTSUPP;
+
+	id = dsa_lag_id(ds->dst, lag);
+
+	mv88e6xxx_reg_lock(chip);
+
+	err = mv88e6xxx_port_set_trunk(chip, port, true, id);
+	if (err)
+		goto err_unlock;
+
+	err = mv88e6xxx_lag_sync_masks_map(ds, lag);
+	if (err)
+		goto err_clear_trunk;
+
+	mv88e6xxx_reg_unlock(chip);
+	return 0;
+
+err_clear_trunk:
+	mv88e6xxx_port_set_trunk(chip, port, false, 0);
+err_unlock:
+	mv88e6xxx_reg_unlock(chip);
+	return err;
+}
+
+static int mv88e6xxx_port_lag_leave(struct dsa_switch *ds, int port,
+				    struct net_device *lag)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	int err_sync, err_trunk;
+
+	mv88e6xxx_reg_lock(chip);
+	err_sync = mv88e6xxx_lag_sync_masks_map(ds, lag);
+	err_trunk = mv88e6xxx_port_set_trunk(chip, port, false, 0);
+	mv88e6xxx_reg_unlock(chip);
+	return err_sync ? : err_trunk;
+}
+
+static int mv88e6xxx_crosschip_lag_change(struct dsa_switch *ds, int sw_index,
+					  int port)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	int err;
+
+	mv88e6xxx_reg_lock(chip);
+	err = mv88e6xxx_lag_sync_masks(ds);
+	mv88e6xxx_reg_unlock(chip);
+	return err;
+}
+
+static int mv88e6xxx_crosschip_lag_join(struct dsa_switch *ds, int sw_index,
+					int port, struct net_device *lag,
+					struct netdev_lag_upper_info *info)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	int err;
+
+	if (!mv88e6xxx_lag_can_offload(ds, lag, info))
+		return -EOPNOTSUPP;
+
+	mv88e6xxx_reg_lock(chip);
+
+	err = mv88e6xxx_lag_sync_masks_map(ds, lag);
+	if (err)
+		goto unlock;
+
+	err = mv88e6xxx_pvt_map(chip, sw_index, port);
+
+unlock:
+	mv88e6xxx_reg_unlock(chip);
+	return err;
+}
+
+static int mv88e6xxx_crosschip_lag_leave(struct dsa_switch *ds, int sw_index,
+					 int port, struct net_device *lag)
+{
+	struct mv88e6xxx_chip *chip = ds->priv;
+	int err_sync, err_pvt;
+
+	mv88e6xxx_reg_lock(chip);
+	err_sync = mv88e6xxx_lag_sync_masks_map(ds, lag);
+	err_pvt = mv88e6xxx_pvt_map(chip, sw_index, port);
+	mv88e6xxx_reg_unlock(chip);
+	return err_sync ? : err_pvt;
+}
+
 static const struct dsa_switch_ops mv88e6xxx_switch_ops = {
 	.get_tag_protocol	= mv88e6xxx_get_tag_protocol,
 	.setup			= mv88e6xxx_setup,
@@ -5416,6 +5698,12 @@ static const struct dsa_switch_ops mv88e6xxx_switch_ops = {
 	.devlink_param_get	= mv88e6xxx_devlink_param_get,
 	.devlink_param_set	= mv88e6xxx_devlink_param_set,
 	.devlink_info_get	= mv88e6xxx_devlink_info_get,
+	.port_lag_change	= mv88e6xxx_port_lag_change,
+	.port_lag_join		= mv88e6xxx_port_lag_join,
+	.port_lag_leave		= mv88e6xxx_port_lag_leave,
+	.crosschip_lag_change	= mv88e6xxx_crosschip_lag_change,
+	.crosschip_lag_join	= mv88e6xxx_crosschip_lag_join,
+	.crosschip_lag_leave	= mv88e6xxx_crosschip_lag_leave,
 };
 
 static int mv88e6xxx_register_switch(struct mv88e6xxx_chip *chip)
@@ -5435,6 +5723,12 @@ static int mv88e6xxx_register_switch(struct mv88e6xxx_chip *chip)
 	ds->ageing_time_min = chip->info->age_time_coeff;
 	ds->ageing_time_max = chip->info->age_time_coeff * U8_MAX;
 
+	/* Some chips support up to 32, but that requires enabling the
+	 * 5-bit port mode, which we do not support. 640k^W16 ought to
+	 * be enough for anyone.
+	 */
+	ds->num_lag_ids = 16;
+
 	dev_set_drvdata(dev, ds);
 
 	return dsa_register_switch(ds);