net_sched: sch_fq: remove q->ktime_cache
authorEric Dumazet <edumazet@google.com>
Mon, 2 Oct 2023 13:17:35 +0000 (13:17 +0000)
committerPaolo Abeni <pabeni@redhat.com>
Thu, 5 Oct 2023 11:27:20 +0000 (13:27 +0200)
Now that both enqueue() and dequeue() need to use ktime_get_ns(),
there is no point wasting 8 bytes in struct fq_sched_data.

This makes room for future fields. ;)

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Dave Taht <dave.taht@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net/sched/sch_fq.c

index d35419db7b9407cedc8f48eaab783f41b366b2a2..818ac786379d4de1812e7be7d883adcbef4fc737 100644 (file)
@@ -106,7 +106,6 @@ struct fq_sched_data {
 
        struct rb_root  delayed;        /* for rate limited flows */
        u64             time_next_delayed_flow;
-       u64             ktime_cache;    /* copy of last ktime_get_ns() */
        unsigned long   unthrottle_latency_ns;
 
        struct fq_flow  internal;       /* for non classified or high prio packets */
@@ -282,12 +281,13 @@ static void fq_gc(struct fq_sched_data *q,
  *
  * FQ can not use generic TCQ_F_CAN_BYPASS infrastructure.
  */
-static bool fq_fastpath_check(const struct Qdisc *sch, struct sk_buff *skb)
+static bool fq_fastpath_check(const struct Qdisc *sch, struct sk_buff *skb,
+                             u64 now)
 {
        const struct fq_sched_data *q = qdisc_priv(sch);
        const struct sock *sk;
 
-       if (fq_skb_cb(skb)->time_to_send > q->ktime_cache)
+       if (fq_skb_cb(skb)->time_to_send > now)
                return false;
 
        if (sch->q.qlen != 0) {
@@ -317,7 +317,8 @@ static bool fq_fastpath_check(const struct Qdisc *sch, struct sk_buff *skb)
        return true;
 }
 
-static struct fq_flow *fq_classify(struct Qdisc *sch, struct sk_buff *skb)
+static struct fq_flow *fq_classify(struct Qdisc *sch, struct sk_buff *skb,
+                                  u64 now)
 {
        struct fq_sched_data *q = qdisc_priv(sch);
        struct rb_node **p, *parent;
@@ -360,7 +361,7 @@ static struct fq_flow *fq_classify(struct Qdisc *sch, struct sk_buff *skb)
                sk = (struct sock *)((hash << 1) | 1UL);
        }
 
-       if (fq_fastpath_check(sch, skb)) {
+       if (fq_fastpath_check(sch, skb, now)) {
                q->internal.stat_fastpath_packets++;
                return &q->internal;
        }
@@ -497,9 +498,9 @@ static void flow_queue_add(struct fq_flow *flow, struct sk_buff *skb)
 }
 
 static bool fq_packet_beyond_horizon(const struct sk_buff *skb,
-                                   const struct fq_sched_data *q)
+                                    const struct fq_sched_data *q, u64 now)
 {
-       return unlikely((s64)skb->tstamp > (s64)(q->ktime_cache + q->horizon));
+       return unlikely((s64)skb->tstamp > (s64)(now + q->horizon));
 }
 
 static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch,
@@ -507,27 +508,28 @@ static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 {
        struct fq_sched_data *q = qdisc_priv(sch);
        struct fq_flow *f;
+       u64 now;
 
        if (unlikely(sch->q.qlen >= sch->limit))
                return qdisc_drop(skb, sch, to_free);
 
-       q->ktime_cache = ktime_get_ns();
+       now = ktime_get_ns();
        if (!skb->tstamp) {
-               fq_skb_cb(skb)->time_to_send = q->ktime_cache;
+               fq_skb_cb(skb)->time_to_send = now;
        } else {
                /* Check if packet timestamp is too far in the future. */
-               if (fq_packet_beyond_horizon(skb, q)) {
+               if (fq_packet_beyond_horizon(skb, q, now)) {
                        if (q->horizon_drop) {
                                        q->stat_horizon_drops++;
                                        return qdisc_drop(skb, sch, to_free);
                        }
                        q->stat_horizon_caps++;
-                       skb->tstamp = q->ktime_cache + q->horizon;
+                       skb->tstamp = now + q->horizon;
                }
                fq_skb_cb(skb)->time_to_send = skb->tstamp;
        }
 
-       f = fq_classify(sch, skb);
+       f = fq_classify(sch, skb, now);
 
        if (f != &q->internal) {
                if (unlikely(f->qlen >= q->flow_plimit)) {
@@ -602,7 +604,7 @@ static struct sk_buff *fq_dequeue(struct Qdisc *sch)
                goto out;
        }
 
-       q->ktime_cache = now = ktime_get_ns();
+       now = ktime_get_ns();
        fq_check_throttled(q, now);
 begin:
        head = &q->new_flows;