Merge branch 'lockless-qdisc-opts'

Yunsheng Lin says:

====================
Some optimization for lockless qdisc

Patch 1: remove unnecessary seqcount operation.
Patch 2: implement TCQ_F_CAN_BYPASS.
Patch 3: remove qdisc->empty.

Performance data for pktgen in queue_xmit mode + dummy netdev
with pfifo_fast:

 threads    unpatched           patched             delta
    1       2.60Mpps            3.21Mpps             +23%
    2       3.84Mpps            5.56Mpps             +44%
    4       5.52Mpps            5.58Mpps             +1%
    8       2.77Mpps            2.76Mpps             -0.3%
   16       2.24Mpps            2.23Mpps             -0.4%

Performance for IP forward testing: 1.05Mpps increases to
1.16Mpps, about 10% improvement.

V3: Add 'Acked-by' from Jakub and 'Tested-by' from Vladimir,
    and resend based on latest net-next.
V2: Adjust the comment and commit log according to discussion
    in V1.
V1: Drop RFC tag, add nolock_qdisc_is_empty() and do the qdisc
    empty checking without the protection of qdisc->seqlock to
    aviod doing unnecessary spin_trylock() for contention case.
RFC v4: Use STATE_MISSED and STATE_DRAINING to indicate non-empty
        qdisc, and add patch 1 and 3.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 16 insertions, 7 deletions

diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index e9c0afc8becc..d9ac60ffe927 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -52,6 +52,8 @@ static void qdisc_maybe_clear_missed(struct Qdisc *q,
 	 */
 	if (!netif_xmit_frozen_or_stopped(txq))
 		set_bit(__QDISC_STATE_MISSED, &q->state);
+	else
+		set_bit(__QDISC_STATE_DRAINING, &q->state);
 }
 
 /* Main transmission queue. */
@@ -164,9 +166,13 @@ static inline void dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
 
 		skb = next;
 	}
-	if (lock)
+
+	if (lock) {
 		spin_unlock(lock);
-	__netif_schedule(q);
+		set_bit(__QDISC_STATE_MISSED, &q->state);
+	} else {
+		__netif_schedule(q);
+	}
 }
 
 static void try_bulk_dequeue_skb(struct Qdisc *q,
@@ -409,7 +415,11 @@ void __qdisc_run(struct Qdisc *q)
 	while (qdisc_restart(q, &packets)) {
 		quota -= packets;
 		if (quota <= 0) {
-			__netif_schedule(q);
+			if (q->flags & TCQ_F_NOLOCK)
+				set_bit(__QDISC_STATE_MISSED, &q->state);
+			else
+				__netif_schedule(q);
+
 			break;
 		}
 	}
@@ -698,13 +708,14 @@ retry:
 	if (likely(skb)) {
 		qdisc_update_stats_at_dequeue(qdisc, skb);
 	} else if (need_retry &&
-		   test_bit(__QDISC_STATE_MISSED, &qdisc->state)) {
+		   READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY) {
 		/* Delay clearing the STATE_MISSED here to reduce
 		 * the overhead of the second spin_trylock() in
 		 * qdisc_run_begin() and __netif_schedule() calling
 		 * in qdisc_run_end().
 		 */
 		clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
+		clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
 
 		/* Make sure dequeuing happens after clearing
 		 * STATE_MISSED.
@@ -714,8 +725,6 @@ retry:
 		need_retry = false;
 
 		goto retry;
-	} else {
-		WRITE_ONCE(qdisc->empty, true);
 	}
 
 	return skb;
@@ -916,7 +925,6 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 	sch->enqueue = ops->enqueue;
 	sch->dequeue = ops->dequeue;
 	sch->dev_queue = dev_queue;
-	sch->empty = true;
 	dev_hold(dev);
 	refcount_set(&sch->refcnt, 1);
 
@@ -1222,6 +1230,7 @@ static void dev_reset_queue(struct net_device *dev,
 	spin_unlock_bh(qdisc_lock(qdisc));
 	if (nolock) {
 		clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
+		clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
 		spin_unlock_bh(&qdisc->seqlock);
 	}
 }