net_sched: act: remove tcfa_qstats

tcfa_qstats is currently only used to hold drops and overlimits counters.

tcf_action_inc_drop_qstats() and tcf_action_inc_overlimit_qstats()
currently acquire a->tcfa_lock to increment these counters.

Switch to two atomic_t to get lock-free accounting.

Signed-off-by: Eric Dumazet <[e-mail]>
Reviewed-by: Jamal Hadi Salim <[e-mail]>
Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Jakub Kicinski <[e-mail]>

include/net/act_api.h

···
 	struct tcf_t			tcfa_tm;
 	struct gnet_stats_basic_sync	tcfa_bstats;
 	struct gnet_stats_basic_sync	tcfa_bstats_hw;
-	struct gnet_stats_queue		tcfa_qstats;
+
+	atomic_t			tcfa_drops;
+	atomic_t			tcfa_overlimits;
+
 	struct net_rate_estimator __rcu *tcfa_rate_est;
 	spinlock_t			tcfa_lock;
 	struct gnet_stats_basic_sync __percpu *cpu_bstats;
···
 #define tcf_action	common.tcfa_action
 #define tcf_tm		common.tcfa_tm
 #define tcf_bstats	common.tcfa_bstats
-#define tcf_qstats	common.tcfa_qstats
 #define tcf_rate_est	common.tcfa_rate_est
 #define tcf_lock	common.tcfa_lock
 
···
 		qstats_drop_inc(this_cpu_ptr(a->cpu_qstats));
 		return;
 	}
-	spin_lock(&a->tcfa_lock);
-	qstats_drop_inc(&a->tcfa_qstats);
-	spin_unlock(&a->tcfa_lock);
+	atomic_inc(&a->tcfa_drops);
 }
 
 static inline void tcf_action_inc_overlimit_qstats(struct tc_action *a)
···
 		qstats_overlimit_inc(this_cpu_ptr(a->cpu_qstats));
 		return;
 	}
-	spin_lock(&a->tcfa_lock);
-	qstats_overlimit_inc(&a->tcfa_qstats);
-	spin_unlock(&a->tcfa_lock);
+	atomic_inc(&a->tcfa_overlimits);
 }
 
 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,

net/sched/act_api.c

···
 	}
 
 	_bstats_update(&a->tcfa_bstats, bytes, packets);
-	a->tcfa_qstats.drops += drops;
+	atomic_add(drops, &a->tcfa_drops);
 	if (hw)
 		_bstats_update(&a->tcfa_bstats_hw, bytes, packets);
 }
···
 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 			  int compat_mode)
 {
-	int err = 0;
+	struct gnet_stats_queue qstats = {0};
 	struct gnet_dump d;
+	int err = 0;
 
 	if (p == NULL)
 		goto errout;
···
 	if (err < 0)
 		goto errout;
 
+	qstats.drops = atomic_read(&p->tcfa_drops);
+	qstats.overlimits = atomic_read(&p->tcfa_overlimits);
+
 	if (gnet_stats_copy_basic(&d, p->cpu_bstats,
 				  &p->tcfa_bstats, false) < 0 ||
 	    gnet_stats_copy_basic_hw(&d, p->cpu_bstats_hw,
 				     &p->tcfa_bstats_hw, false) < 0 ||
 	    gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
 	    gnet_stats_copy_queue(&d, p->cpu_qstats,
-				  &p->tcfa_qstats,
-				  p->tcfa_qstats.qlen) < 0)
+				  &qstats,
+				  qstats.qlen) < 0)
 		goto errout;
 
 	if (gnet_stats_finish_copy(&d) < 0)