From: Paolo Abeni <pabeni@redhat.com>
Date: Wed, 10 Apr 2019 14:32:40 +0200
Subject: net: sched: when clearing NOLOCK, clear TCQ_F_CPUSTATS, too
Patch-mainline: v5.2-rc1
Git-commit: 8a53e616de294873fec1a75ddb77ecb3d225cee0
References: bsc#1183405

Since stats updating is always consistent with TCQ_F_CPUSTATS flag,
we can disable it at qdisc creation time flipping such bit.

In my experiments, if the NOLOCK flag is cleared, per CPU stats
accounting does not give any measurable performance gain, but it
waste some memory.

Let's clear TCQ_F_CPUSTATS together with NOLOCK, when enslaving
a NOLOCK qdisc to 'lock' one.

Use stats update helper inside pfifo_fast, to cope correctly with
TCQ_F_CPUSTATS flag change.

As a side effect, q.qlen value for any child qdiscs is always
consistent for all lock classfull qdiscs.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Michal Kubecek <mkubecek@suse.cz>

---
 include/net/sch_generic.h | 26 ++++++++++++++++++++++++++
 net/sched/sch_api.c       | 15 ++++++++++++++-
 net/sched/sch_generic.c   | 10 ++--------
 3 files changed, 42 insertions(+), 9 deletions(-)

--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -1011,6 +1011,32 @@ static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
 	return skb;
 }
 
+static inline void qdisc_update_stats_at_dequeue(struct Qdisc *sch,
+						 struct sk_buff *skb)
+{
+	if (qdisc_is_percpu_stats(sch)) {
+		qdisc_qstats_cpu_backlog_dec(sch, skb);
+		qdisc_bstats_cpu_update(sch, skb);
+		qdisc_qstats_atomic_qlen_dec(sch);
+	} else {
+		qdisc_qstats_backlog_dec(sch, skb);
+		qdisc_bstats_update(sch, skb);
+		sch->q.qlen--;
+	}
+}
+
+static inline void qdisc_update_stats_at_enqueue(struct Qdisc *sch,
+						 unsigned int pkt_len)
+{
+	if (qdisc_is_percpu_stats(sch)) {
+		qdisc_qstats_atomic_qlen_inc(sch);
+		this_cpu_add(sch->cpu_qstats->backlog, pkt_len);
+	} else {
+		sch->qstats.backlog += pkt_len;
+		sch->q.qlen++;
+	}
+}
+
 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
 {
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -941,6 +941,19 @@ static void notify_and_destroy(struct net *net, struct sk_buff *skb,
 		qdisc_put(old);
 }
 
+static void qdisc_clear_nolock(struct Qdisc *sch)
+{
+	sch->flags &= ~TCQ_F_NOLOCK;
+	if (!(sch->flags & TCQ_F_CPUSTATS))
+		return;
+
+	free_percpu(sch->cpu_bstats);
+	free_percpu(sch->cpu_qstats);
+	sch->cpu_bstats = NULL;
+	sch->cpu_qstats = NULL;
+	sch->flags &= ~TCQ_F_CPUSTATS;
+}
+
 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
  * to device "dev".
  *
@@ -1016,7 +1029,7 @@ static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
 		/* Only support running class lockless if parent is lockless */
 		if (new && (new->flags & TCQ_F_NOLOCK) &&
 		    parent && !(parent->flags & TCQ_F_NOLOCK))
-			new->flags &= ~TCQ_F_NOLOCK;
+			qdisc_clear_nolock(new);
 
 		err = -EOPNOTSUPP;
 		if (cops && cops->graft) {
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -631,11 +631,7 @@ static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
 	if (unlikely(err))
 		return qdisc_drop_cpu(skb, qdisc, to_free);
 
-	qdisc_qstats_atomic_qlen_inc(qdisc);
-	/* Note: skb can not be used after skb_array_produce(),
-	 * so we better not use qdisc_qstats_cpu_backlog_inc()
-	 */
-	this_cpu_add(qdisc->cpu_qstats->backlog, pkt_len);
+	qdisc_update_stats_at_enqueue(qdisc, pkt_len);
 	return NET_XMIT_SUCCESS;
 }
 
@@ -654,9 +650,7 @@ static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
 		skb = __skb_array_consume(q);
 	}
 	if (likely(skb)) {
-		qdisc_qstats_cpu_backlog_dec(qdisc, skb);
-		qdisc_bstats_cpu_update(qdisc, skb);
-		qdisc_qstats_atomic_qlen_dec(qdisc);
+		qdisc_update_stats_at_dequeue(qdisc, skb);
 	} else {
 		qdisc->empty = true;
 	}