[Cake] [PATCH net-next v9 7/7] sch_cake: Conditionally split GSO segments

[Toke Høiland-Jørgensen]

At lower bandwidths, the transmission time of a single GSO segment can add
an unacceptable amount of latency due to HOL blocking. Furthermore, with a
software shaper, any tuning mechanism employed by the kernel to control the
maximum size of GSO segments is thrown off by the artificial limit on
bandwidth. For this reason, we split GSO segments into their individual
packets iff the shaper is active and configured to a bandwidth <= 1 Gbps.

Signed-off-by: Toke Høiland-Jørgensen <toke at toke.dk>
---
 net/sched/sch_cake.c |   99 +++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 73 insertions(+), 26 deletions(-)

diff --git a/net/sched/sch_cake.c b/net/sched/sch_cake.c
index af25ddb333d7..1ecb2eef958e 100644
--- a/net/sched/sch_cake.c
+++ b/net/sched/sch_cake.c
@@ -82,6 +82,7 @@
 #define CAKE_QUEUES (1024)
 #define CAKE_FLOW_MASK 63
 #define CAKE_FLOW_NAT_FLAG 64
+#define CAKE_SPLIT_GSO_THRESHOLD (125000000) /* 1Gbps */
 #define US2TIME(a) (a * (u64)NSEC_PER_USEC)
 
 typedef u64 cobalt_time_t;
@@ -1479,36 +1480,73 @@ static s32 cake_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 	if (unlikely(len > b->max_skblen))
 		b->max_skblen = len;
 
-	cobalt_set_enqueue_time(skb, now);
-	get_cobalt_cb(skb)->adjusted_len = cake_overhead(q, skb);
-	flow_queue_add(flow, skb);
-
-	if (q->ack_filter)
-		ack = cake_ack_filter(q, flow);
+	if (skb_is_gso(skb) && q->rate_flags & CAKE_FLAG_SPLIT_GSO) {
+		struct sk_buff *segs, *nskb;
+		netdev_features_t features = netif_skb_features(skb);
+		unsigned int slen = 0;
+
+		segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
+		if (IS_ERR_OR_NULL(segs))
+			return qdisc_drop(skb, sch, to_free);
+
+		while (segs) {
+			nskb = segs->next;
+			segs->next = NULL;
+			qdisc_skb_cb(segs)->pkt_len = segs->len;
+			cobalt_set_enqueue_time(segs, now);
+			get_cobalt_cb(segs)->adjusted_len = cake_overhead(q,
+									  segs);
+			flow_queue_add(flow, segs);
+
+			sch->q.qlen++;
+			slen += segs->len;
+			q->buffer_used += segs->truesize;
+			b->packets++;
+			segs = nskb;
+		}
 
-	if (ack) {
-		b->ack_drops++;
-		sch->qstats.drops++;
-		b->bytes += qdisc_pkt_len(ack);
-		len -= qdisc_pkt_len(ack);
-		q->buffer_used += skb->truesize - ack->truesize;
-		if (q->rate_flags & CAKE_FLAG_INGRESS)
-			cake_advance_shaper(q, b, ack, now, true);
+		/* stats */
+		b->bytes	    += slen;
+		b->backlogs[idx]    += slen;
+		b->tin_backlog      += slen;
+		sch->qstats.backlog += slen;
+		q->avg_window_bytes += slen;
 
-		qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(ack));
-		consume_skb(ack);
+		qdisc_tree_reduce_backlog(sch, 1, len);
+		consume_skb(skb);
 	} else {
-		sch->q.qlen++;
-		q->buffer_used      += skb->truesize;
-	}
+		/* not splitting */
+		cobalt_set_enqueue_time(skb, now);
+		get_cobalt_cb(skb)->adjusted_len = cake_overhead(q, skb);
+		flow_queue_add(flow, skb);
+
+		if (q->ack_filter)
+			ack = cake_ack_filter(q, flow);
+
+		if (ack) {
+			b->ack_drops++;
+			sch->qstats.drops++;
+			b->bytes += qdisc_pkt_len(ack);
+			len -= qdisc_pkt_len(ack);
+			q->buffer_used += skb->truesize - ack->truesize;
+			if (q->rate_flags & CAKE_FLAG_INGRESS)
+				cake_advance_shaper(q, b, ack, now, true);
+
+			qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(ack));
+			consume_skb(ack);
+		} else {
+			sch->q.qlen++;
+			q->buffer_used      += skb->truesize;
+		}
 
-	/* stats */
-	b->packets++;
-	b->bytes	    += len;
-	b->backlogs[idx]    += len;
-	b->tin_backlog      += len;
-	sch->qstats.backlog += len;
-	q->avg_window_bytes += len;
+		/* stats */
+		b->packets++;
+		b->bytes	    += len;
+		b->backlogs[idx]    += len;
+		b->tin_backlog      += len;
+		sch->qstats.backlog += len;
+		q->avg_window_bytes += len;
+	}
 
 	if (q->overflow_timeout)
 		cake_heapify_up(q, b->overflow_idx[idx]);
@@ -2316,6 +2354,11 @@ static int cake_change(struct Qdisc *sch, struct nlattr *opt,
 	if (tb[TCA_CAKE_MEMORY])
 		q->buffer_config_limit = nla_get_u32(tb[TCA_CAKE_MEMORY]);
 
+	if (q->rate_bps && q->rate_bps <= CAKE_SPLIT_GSO_THRESHOLD)
+		q->rate_flags |= CAKE_FLAG_SPLIT_GSO;
+	else
+		q->rate_flags &= ~CAKE_FLAG_SPLIT_GSO;
+
 	if (q->tins) {
 		sch_tree_lock(sch);
 		cake_reconfigure(sch);
@@ -2469,6 +2512,10 @@ static int cake_dump(struct Qdisc *sch, struct sk_buff *skb)
 	if (nla_put_u32(skb, TCA_CAKE_MPU, q->rate_mpu))
 		goto nla_put_failure;
 
+	if (nla_put_u32(skb, TCA_CAKE_SPLIT_GSO,
+			!!(q->rate_flags & CAKE_FLAG_SPLIT_GSO)))
+		goto nla_put_failure;
+
 	return nla_nest_end(skb, opts);
 
 nla_put_failure: