drop_monitor: Convert to using devlink tracepoint

Convert drop monitor to use the recently introduced
'devlink_trap_report' tracepoint instead of having devlink call into
drop monitor.

This is both consistent with software originated drops ('kfree_skb'
tracepoint) and also allows drop monitor to be built as a module and
still report hardware originated drops.

Signed-off-by: Ido Schimmel <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Author: idosch

MAINTAINERS

@@ -12065,7 +12065,6 @@ M:	Neil Horman <[email protected]>
 L:	[email protected]
 S:	Maintained
 W:	https://fedorahosted.org/dropwatch/
-F:	include/net/drop_monitor.h
 F:	include/uapi/linux/net_dropmon.h
 F:	net/core/drop_monitor.c
 

include/net/drop_monitor.h

@@ -434,7 +434,6 @@ config NET_SOCK_MSG
 config NET_DEVLINK
 	bool
 	default n
-	imply NET_DROP_MONITOR
 
 config PAGE_POOL
 	bool

net/Kconfig

@@ -27,7 +27,6 @@
 #include <net/net_namespace.h>
 #include <net/sock.h>
 #include <net/devlink.h>
-#include <net/drop_monitor.h>
 #define CREATE_TRACE_POINTS
 #include <trace/events/devlink.h>
 
@@ -9261,24 +9260,6 @@ devlink_trap_stats_update(struct devlink_stats __percpu *trap_stats,
 	u64_stats_update_end(&stats->syncp);
 }
 
-static void
-devlink_trap_report_metadata_fill(struct net_dm_hw_metadata *hw_metadata,
-				  const struct devlink_trap_item *trap_item,
-				  struct devlink_port *in_devlink_port,
-				  const struct flow_action_cookie *fa_cookie)
-{
-	struct devlink_trap_group_item *group_item = trap_item->group_item;
-
-	hw_metadata->trap_group_name = group_item->group->name;
-	hw_metadata->trap_name = trap_item->trap->name;
-	hw_metadata->fa_cookie = fa_cookie;
-
-	spin_lock(&in_devlink_port->type_lock);
-	if (in_devlink_port->type == DEVLINK_PORT_TYPE_ETH)
-		hw_metadata->input_dev = in_devlink_port->type_dev;
-	spin_unlock(&in_devlink_port->type_lock);
-}
-
 static void
 devlink_trap_report_metadata_set(struct devlink_trap_metadata *metadata,
 				 const struct devlink_trap_item *trap_item,
@@ -9309,7 +9290,6 @@ void devlink_trap_report(struct devlink *devlink, struct sk_buff *skb,
 
 {
 	struct devlink_trap_item *trap_item = trap_ctx;
-	struct net_dm_hw_metadata hw_metadata = {};
 
 	devlink_trap_stats_update(trap_item->stats, skb->len);
 	devlink_trap_stats_update(trap_item->group_item->stats, skb->len);
@@ -9321,10 +9301,6 @@ void devlink_trap_report(struct devlink *devlink, struct sk_buff *skb,
 	if (trap_item->trap->type == DEVLINK_TRAP_TYPE_CONTROL)
 		return;
 
-	devlink_trap_report_metadata_fill(&hw_metadata, trap_item,
-					  in_devlink_port, fa_cookie);
-	net_dm_hw_report(skb, &hw_metadata);
-
 	if (trace_devlink_trap_report_enabled()) {
 		struct devlink_trap_metadata metadata = {};
 

net/core/devlink.c

@@ -26,14 +26,14 @@
 #include <linux/bitops.h>
 #include <linux/slab.h>
 #include <linux/module.h>
-#include <net/drop_monitor.h>
 #include <net/genetlink.h>
 #include <net/netevent.h>
 #include <net/flow_offload.h>
 #include <net/devlink.h>
 
 #include <trace/events/skb.h>
 #include <trace/events/napi.h>
+#include <trace/events/devlink.h>
 
 #include <asm/unaligned.h>
 
@@ -108,6 +108,13 @@ static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
 static u32 net_dm_trunc_len;
 static u32 net_dm_queue_len = 1000;
 
+struct net_dm_hw_metadata {
+	const char *trap_group_name;
+	const char *trap_name;
+	struct net_device *input_dev;
+	const struct flow_action_cookie *fa_cookie;
+};
+
 struct net_dm_alert_ops {
 	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
 				void *location);
@@ -1129,25 +1136,32 @@ static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
 	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
 };
 
-void net_dm_hw_report(struct sk_buff *skb,
-		      const struct net_dm_hw_metadata *hw_metadata)
+#if IS_ENABLED(CONFIG_NET_DEVLINK)
+static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
 {
-	rcu_read_lock();
-
-	if (!monitor_hw)
-		goto out;
+	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
+}
 
-	net_dm_alert_ops_arr[net_dm_alert_mode]->hw_probe(skb, hw_metadata);
+static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
+{
+	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
+	tracepoint_synchronize_unregister();
+}
+#else
+static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
+{
+	return -EOPNOTSUPP;
+}
 
-out:
-	rcu_read_unlock();
+static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
+{
 }
-EXPORT_SYMBOL_GPL(net_dm_hw_report);
+#endif
 
 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
 {
 	const struct net_dm_alert_ops *ops;
-	int cpu;
+	int cpu, rc;
 
 	if (monitor_hw) {
 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
@@ -1171,26 +1185,36 @@ static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
 		kfree(hw_entries);
 	}
 
+	rc = net_dm_hw_probe_register(ops);
+	if (rc) {
+		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
+		goto err_module_put;
+	}
+
 	monitor_hw = true;
 
 	return 0;
+
+err_module_put:
+	module_put(THIS_MODULE);
+	return rc;
 }
 
 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
 {
+	const struct net_dm_alert_ops *ops;
 	int cpu;
 
 	if (!monitor_hw) {
 		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
 		return;
 	}
 
+	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
+
 	monitor_hw = false;
 
-	/* After this call returns we are guaranteed that no CPU is processing
-	 * any hardware drops.
-	 */
-	synchronize_rcu();
+	net_dm_hw_probe_unregister(ops);
 
 	for_each_possible_cpu(cpu) {
 		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);