[PATCH net-next 13/14] mlxsw: Add support for IFLA_OFFLOAD_XSTATS_L3_STATS

[Ido Schimmel <idosch@nvidia.com>]

From: Petr Machata <petrm@nvidia.com>

Spectrum machines support L3 stats by binding a counter to a RIF, a
hardware object representing a router interface. Recognize the netdevice
notifier events, NETDEV_OFFLOAD_XSTATS_*, to support enablement,
disablement, and reporting back to core.

As a netdevice gains a RIF, if L3 stats are enabled, install the counters,
and ping the core so that a userspace notification can be emitted.

Similarly, as a netdevice loses a RIF, push the as-yet-unreported
statistics to the core, so that they are not lost, and ping the core to
emit userspace notification.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
---
 .../net/ethernet/mellanox/mlxsw/spectrum.c    |   4 +
 .../ethernet/mellanox/mlxsw/spectrum_router.c | 291 +++++++++++++++++-
 2 files changed, 293 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum.c
index 10f32deea158..7b7b17183d10 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum.c
@@ -4829,6 +4829,10 @@ static bool mlxsw_sp_netdevice_event_is_router(unsigned long event)
 	case NETDEV_PRE_CHANGEADDR:
 	case NETDEV_CHANGEADDR:
 	case NETDEV_CHANGEMTU:
+	case NETDEV_OFFLOAD_XSTATS_ENABLE:
+	case NETDEV_OFFLOAD_XSTATS_DISABLE:
+	case NETDEV_OFFLOAD_XSTATS_REPORT_USED:
+	case NETDEV_OFFLOAD_XSTATS_REPORT_DELTA:
 		return true;
 	default:
 		return false;
diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
index 2b21fea3b37d..7ccf1c6f4643 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
@@ -225,6 +225,64 @@ int mlxsw_sp_rif_counter_value_get(struct mlxsw_sp *mlxsw_sp,
 	return 0;
 }
 
+struct mlxsw_sp_rif_counter_set_basic {
+	u64 good_unicast_packets;
+	u64 good_multicast_packets;
+	u64 good_broadcast_packets;
+	u64 good_unicast_bytes;
+	u64 good_multicast_bytes;
+	u64 good_broadcast_bytes;
+	u64 error_packets;
+	u64 discard_packets;
+	u64 error_bytes;
+	u64 discard_bytes;
+};
+
+static int
+mlxsw_sp_rif_counter_fetch_clear(struct mlxsw_sp_rif *rif,
+				 enum mlxsw_sp_rif_counter_dir dir,
+				 struct mlxsw_sp_rif_counter_set_basic *set)
+{
+	struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
+	char ricnt_pl[MLXSW_REG_RICNT_LEN];
+	unsigned int *p_counter_index;
+	int err;
+
+	if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
+		return -EINVAL;
+
+	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
+	if (!p_counter_index)
+		return -EINVAL;
+
+	mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
+			     MLXSW_REG_RICNT_OPCODE_CLEAR);
+	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
+	if (err)
+		return err;
+
+	if (!set)
+		return 0;
+
+#define MLXSW_SP_RIF_COUNTER_EXTRACT(NAME)				\
+		(set->NAME = mlxsw_reg_ricnt_ ## NAME ## _get(ricnt_pl))
+
+	MLXSW_SP_RIF_COUNTER_EXTRACT(good_unicast_packets);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(good_multicast_packets);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(good_broadcast_packets);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(good_unicast_bytes);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(good_multicast_bytes);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(good_broadcast_bytes);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(error_packets);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(discard_packets);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(error_bytes);
+	MLXSW_SP_RIF_COUNTER_EXTRACT(discard_bytes);
+
+#undef MLXSW_SP_RIF_COUNTER_EXTRACT
+
+	return 0;
+}
+
 static int mlxsw_sp_rif_counter_clear(struct mlxsw_sp *mlxsw_sp,
 				      unsigned int counter_index)
 {
@@ -242,9 +300,13 @@ int mlxsw_sp_rif_counter_alloc(struct mlxsw_sp_rif *rif,
 	unsigned int *p_counter_index;
 	int err;
 
+	if (mlxsw_sp_rif_counter_valid_get(rif, dir))
+		return 0;
+
 	p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
 	if (!p_counter_index)
 		return -EINVAL;
+
 	err = mlxsw_sp_counter_alloc(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
 				     p_counter_index);
 	if (err)
@@ -8146,6 +8208,166 @@ u16 mlxsw_sp_ipip_lb_ul_rif_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
 	return lb_rif->ul_rif_id;
 }
 
+static bool
+mlxsw_sp_router_port_l3_stats_enabled(struct mlxsw_sp_rif *rif)
+{
+	return mlxsw_sp_rif_counter_valid_get(rif,
+					      MLXSW_SP_RIF_COUNTER_EGRESS) &&
+	       mlxsw_sp_rif_counter_valid_get(rif,
+					      MLXSW_SP_RIF_COUNTER_INGRESS);
+}
+
+static int
+mlxsw_sp_router_port_l3_stats_enable(struct mlxsw_sp_rif *rif)
+{
+	int err;
+
+	err = mlxsw_sp_rif_counter_alloc(rif, MLXSW_SP_RIF_COUNTER_INGRESS);
+	if (err)
+		return err;
+
+	/* Clear stale data. */
+	err = mlxsw_sp_rif_counter_fetch_clear(rif,
+					       MLXSW_SP_RIF_COUNTER_INGRESS,
+					       NULL);
+	if (err)
+		goto err_clear_ingress;
+
+	err = mlxsw_sp_rif_counter_alloc(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
+	if (err)
+		goto err_alloc_egress;
+
+	/* Clear stale data. */
+	err = mlxsw_sp_rif_counter_fetch_clear(rif,
+					       MLXSW_SP_RIF_COUNTER_EGRESS,
+					       NULL);
+	if (err)
+		goto err_clear_egress;
+
+	return 0;
+
+err_clear_egress:
+	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
+err_alloc_egress:
+err_clear_ingress:
+	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_INGRESS);
+	return err;
+}
+
+static void
+mlxsw_sp_router_port_l3_stats_disable(struct mlxsw_sp_rif *rif)
+{
+	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_EGRESS);
+	mlxsw_sp_rif_counter_free(rif, MLXSW_SP_RIF_COUNTER_INGRESS);
+}
+
+static void
+mlxsw_sp_router_port_l3_stats_report_used(struct mlxsw_sp_rif *rif,
+					  struct netdev_notifier_offload_xstats_info *info)
+{
+	if (!mlxsw_sp_router_port_l3_stats_enabled(rif))
+		return;
+	netdev_offload_xstats_report_used(info->report_used);
+}
+
+static int
+mlxsw_sp_router_port_l3_stats_fetch(struct mlxsw_sp_rif *rif,
+				    struct rtnl_link_stats64 *p_stats)
+{
+	struct mlxsw_sp_rif_counter_set_basic ingress;
+	struct mlxsw_sp_rif_counter_set_basic egress;
+	int err;
+
+	err = mlxsw_sp_rif_counter_fetch_clear(rif,
+					       MLXSW_SP_RIF_COUNTER_INGRESS,
+					       &ingress);
+	if (err)
+		return err;
+
+	err = mlxsw_sp_rif_counter_fetch_clear(rif,
+					       MLXSW_SP_RIF_COUNTER_EGRESS,
+					       &egress);
+	if (err)
+		return err;
+
+#define MLXSW_SP_ROUTER_ALL_GOOD(SET, SFX)		\
+		((SET.good_unicast_ ## SFX) +		\
+		 (SET.good_multicast_ ## SFX) +		\
+		 (SET.good_broadcast_ ## SFX))
+
+	p_stats->rx_packets = MLXSW_SP_ROUTER_ALL_GOOD(ingress, packets);
+	p_stats->tx_packets = MLXSW_SP_ROUTER_ALL_GOOD(egress, packets);
+	p_stats->rx_bytes = MLXSW_SP_ROUTER_ALL_GOOD(ingress, bytes);
+	p_stats->tx_bytes = MLXSW_SP_ROUTER_ALL_GOOD(egress, bytes);
+	p_stats->rx_errors = ingress.error_packets;
+	p_stats->tx_errors = egress.error_packets;
+	p_stats->rx_dropped = ingress.discard_packets;
+	p_stats->tx_dropped = egress.discard_packets;
+	p_stats->multicast = ingress.good_multicast_packets +
+			     ingress.good_broadcast_packets;
+
+#undef MLXSW_SP_ROUTER_ALL_GOOD
+
+	return 0;
+}
+
+static int
+mlxsw_sp_router_port_l3_stats_report_delta(struct mlxsw_sp_rif *rif,
+					   struct netdev_notifier_offload_xstats_info *info)
+{
+	struct rtnl_link_stats64 stats = {};
+	int err;
+
+	if (!mlxsw_sp_router_port_l3_stats_enabled(rif))
+		return 0;
+
+	err = mlxsw_sp_router_port_l3_stats_fetch(rif, &stats);
+	if (err)
+		return err;
+
+	netdev_offload_xstats_report_delta(info->report_delta, &stats);
+	return 0;
+}
+
+struct mlxsw_sp_router_hwstats_notify_work {
+	struct work_struct work;
+	struct net_device *dev;
+};
+
+static void mlxsw_sp_router_hwstats_notify_work(struct work_struct *work)
+{
+	struct mlxsw_sp_router_hwstats_notify_work *hws_work =
+		container_of(work, struct mlxsw_sp_router_hwstats_notify_work,
+			     work);
+
+	rtnl_lock();
+	rtnl_offload_xstats_notify(hws_work->dev);
+	rtnl_unlock();
+	dev_put(hws_work->dev);
+	kfree(hws_work);
+}
+
+static void
+mlxsw_sp_router_hwstats_notify_schedule(struct net_device *dev)
+{
+	struct mlxsw_sp_router_hwstats_notify_work *hws_work;
+
+	/* To collect notification payload, the core ends up sending another
+	 * notifier block message, which would deadlock on the attempt to
+	 * acquire the router lock again. Just postpone the notification until
+	 * later.
+	 */
+
+	hws_work = kzalloc(sizeof(*hws_work), GFP_KERNEL);
+	if (!hws_work)
+		return;
+
+	INIT_WORK(&hws_work->work, mlxsw_sp_router_hwstats_notify_work);
+	dev_hold(dev);
+	hws_work->dev = dev;
+	mlxsw_core_schedule_work(&hws_work->work);
+}
+
 int mlxsw_sp_rif_dev_ifindex(const struct mlxsw_sp_rif *rif)
 {
 	return rif->dev->ifindex;
@@ -8156,6 +8378,16 @@ const struct net_device *mlxsw_sp_rif_dev(const struct mlxsw_sp_rif *rif)
 	return rif->dev;
 }
 
+static void mlxsw_sp_rif_push_l3_stats(struct mlxsw_sp_rif *rif)
+{
+	struct rtnl_link_stats64 stats = {};
+
+	if (!mlxsw_sp_router_port_l3_stats_fetch(rif, &stats))
+		netdev_offload_xstats_push_delta(rif->dev,
+						 NETDEV_OFFLOAD_XSTATS_TYPE_L3,
+						 &stats);
+}
+
 static struct mlxsw_sp_rif *
 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
 		    const struct mlxsw_sp_rif_params *params,
@@ -8216,10 +8448,19 @@ mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
 			goto err_mr_rif_add;
 	}
 
-	mlxsw_sp_rif_counters_alloc(rif);
+	if (netdev_offload_xstats_enabled(rif->dev,
+					  NETDEV_OFFLOAD_XSTATS_TYPE_L3)) {
+		err = mlxsw_sp_router_port_l3_stats_enable(rif);
+		if (err)
+			goto err_stats_enable;
+		mlxsw_sp_router_hwstats_notify_schedule(rif->dev);
+	} else {
+		mlxsw_sp_rif_counters_alloc(rif);
+	}
 
 	return rif;
 
+err_stats_enable:
 err_mr_rif_add:
 	for (i--; i >= 0; i--)
 		mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
@@ -8249,7 +8490,15 @@ static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif)
 	mlxsw_sp_router_rif_gone_sync(mlxsw_sp, rif);
 	vr = &mlxsw_sp->router->vrs[rif->vr_id];
 
-	mlxsw_sp_rif_counters_free(rif);
+	if (netdev_offload_xstats_enabled(rif->dev,
+					  NETDEV_OFFLOAD_XSTATS_TYPE_L3)) {
+		mlxsw_sp_rif_push_l3_stats(rif);
+		mlxsw_sp_router_port_l3_stats_disable(rif);
+		mlxsw_sp_router_hwstats_notify_schedule(rif->dev);
+	} else {
+		mlxsw_sp_rif_counters_free(rif);
+	}
+
 	for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
 		mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
 	ops->deconfigure(rif);
@@ -9126,6 +9375,35 @@ static int mlxsw_sp_router_port_pre_changeaddr_event(struct mlxsw_sp_rif *rif,
 	return -ENOBUFS;
 }
 
+static int
+mlxsw_sp_router_port_offload_xstats_cmd(struct mlxsw_sp_rif *rif,
+					unsigned long event,
+					struct netdev_notifier_offload_xstats_info *info)
+{
+	switch (info->type) {
+	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
+		break;
+	default:
+		return 0;
+	}
+
+	switch (event) {
+	case NETDEV_OFFLOAD_XSTATS_ENABLE:
+		return mlxsw_sp_router_port_l3_stats_enable(rif);
+	case NETDEV_OFFLOAD_XSTATS_DISABLE:
+		mlxsw_sp_router_port_l3_stats_disable(rif);
+		return 0;
+	case NETDEV_OFFLOAD_XSTATS_REPORT_USED:
+		mlxsw_sp_router_port_l3_stats_report_used(rif, info);
+		return 0;
+	case NETDEV_OFFLOAD_XSTATS_REPORT_DELTA:
+		return mlxsw_sp_router_port_l3_stats_report_delta(rif, info);
+	}
+
+	WARN_ON_ONCE(1);
+	return 0;
+}
+
 int mlxsw_sp_netdevice_router_port_event(struct net_device *dev,
 					 unsigned long event, void *ptr)
 {
@@ -9151,6 +9429,15 @@ int mlxsw_sp_netdevice_router_port_event(struct net_device *dev,
 	case NETDEV_PRE_CHANGEADDR:
 		err = mlxsw_sp_router_port_pre_changeaddr_event(rif, ptr);
 		break;
+	case NETDEV_OFFLOAD_XSTATS_ENABLE:
+	case NETDEV_OFFLOAD_XSTATS_DISABLE:
+	case NETDEV_OFFLOAD_XSTATS_REPORT_USED:
+	case NETDEV_OFFLOAD_XSTATS_REPORT_DELTA:
+		err = mlxsw_sp_router_port_offload_xstats_cmd(rif, event, ptr);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		break;
 	}
 
 out:
-- 
2.33.1