net/sched: sch_cake: share shaper state across sub-instances of cake_mq

This commit adds shared shaper state across the cake instances beneath a
cake_mq qdisc. It works by periodically tracking the number of active
instances, and scaling the configured rate by the number of active
queues.

The scan is lockless and simply reads the qlen and the last_active state
variable of each of the instances configured beneath the parent cake_mq
instance. Locking is not required since the values are only updated by
the owning instance, and eventual consistency is sufficient for the
purpose of estimating the number of active queues.

The interval for scanning the number of active queues is set to 200 us.
We found this to be a good tradeoff between overhead and response time.
For a detailed analysis of this aspect see the Netdevconf talk:

https://netdevconf.info/0x19/docs/netdev-0x19-paper16-talk-paper.pdf

Reviewed-by: Jamal Hadi Salim <[e-mail]>
Signed-off-by: Jonas Köppeler <[e-mail]>
Signed-off-by: Toke Høiland-Jørgensen <[e-mail]>
Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Paolo Abeni <[e-mail]>

Documentation/netlink/specs/tc.yaml

···
       -
         name: blue-timer-us
         type: s32
+      -
+        name: active-queues
+        type: u32
   -
     name: cake-tin-stats-attrs
     name-prefix: tca-cake-tin-stats-

include/uapi/linux/pkt_sched.h

···
 	TCA_CAKE_STATS_DROP_NEXT_US,
 	TCA_CAKE_STATS_P_DROP,
 	TCA_CAKE_STATS_BLUE_TIMER_US,
+	TCA_CAKE_STATS_ACTIVE_QUEUES,
 	__TCA_CAKE_STATS_MAX
 };
 #define TCA_CAKE_STATS_MAX (__TCA_CAKE_STATS_MAX - 1)

net/sched/sch_cake.c

···
 	u64		rate_bps;
 	u64		interval;
 	u64		target;
+	u64		sync_time;
 	u32		buffer_config_limit;
 	u32		fwmark_mask;
 	u16		fwmark_shft;
···
 	u16		max_adjlen;
 	u16		min_netlen;
 	u16		min_adjlen;
+
+	/* mq sync state */
+	u64		last_checked_active;
+	u64		last_active;
+	u32		active_queues;
 };
 
 enum {
···
 	1239850263, 1191209601, 1147878294, 1108955788
 };
 
+static void cake_set_rate(struct cake_tin_data *b, u64 rate, u32 mtu,
+			  u64 target_ns, u64 rtt_est_ns);
 /* http://en.wikipedia.org/wiki/Methods_of_computing_square_roots
  * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2)
  *
···
 	u64 delay;
 	u32 len;
 
+	if (q->config->is_shared && now - q->last_checked_active >= q->config->sync_time) {
+		struct net_device *dev = qdisc_dev(sch);
+		struct cake_sched_data *other_priv;
+		u64 new_rate = q->config->rate_bps;
+		u64 other_qlen, other_last_active;
+		struct Qdisc *other_sch;
+		u32 num_active_qs = 1;
+		unsigned int ntx;
+
+		for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
+			other_sch = rcu_dereference(netdev_get_tx_queue(dev, ntx)->qdisc_sleeping);
+			other_priv = qdisc_priv(other_sch);
+
+			if (other_priv == q)
+				continue;
+
+			other_qlen = READ_ONCE(other_sch->q.qlen);
+			other_last_active = READ_ONCE(other_priv->last_active);
+
+			if (other_qlen || other_last_active > q->last_checked_active)
+				num_active_qs++;
+		}
+
+		if (num_active_qs > 1)
+			new_rate = div64_u64(q->config->rate_bps, num_active_qs);
+
+		/* mtu = 0 is used to only update the rate and not mess with cobalt params */
+		cake_set_rate(b, new_rate, 0, 0, 0);
+		q->last_checked_active = now;
+		q->active_queues = num_active_qs;
+		q->rate_ns = b->tin_rate_ns;
+		q->rate_shft = b->tin_rate_shft;
+	}
+
 begin:
 	if (!sch->q.qlen)
 		return NULL;
···
 
 	b->tin_ecn_mark += !!flow->cvars.ecn_marked;
 	qdisc_bstats_update(sch, skb);
+	WRITE_ONCE(q->last_active, now);
 
 	/* collect delay stats */
 	delay = ktime_to_ns(ktime_sub(now, cobalt_get_enqueue_time(skb)));
···
 	b->tin_rate_bps  = rate;
 	b->tin_rate_ns   = rate_ns;
 	b->tin_rate_shft = rate_shft;
+
+	if (mtu == 0)
+		return;
 
 	byte_target_ns = (byte_target * rate_ns) >> rate_shft;
 
···
 			       */
 	q->rate_flags |= CAKE_FLAG_SPLIT_GSO;
 	q->is_shared = is_shared;
+	q->sync_time = 200 * NSEC_PER_USEC;
 }
 
 static int cake_init(struct Qdisc *sch, struct nlattr *opt,
···
 	qd->avg_peak_bandwidth = q->rate_bps;
 	qd->min_netlen = ~0;
 	qd->min_adjlen = ~0;
+	qd->active_queues = 0;
+	qd->last_checked_active = 0;
+
 	return 0;
 err:
 	kvfree(qd->config);
···
 	PUT_STAT_U32(MAX_ADJLEN, q->max_adjlen);
 	PUT_STAT_U32(MIN_NETLEN, q->min_netlen);
 	PUT_STAT_U32(MIN_ADJLEN, q->min_adjlen);
+	PUT_STAT_U32(ACTIVE_QUEUES, q->active_queues);
 
 #undef PUT_STAT_U32
 #undef PUT_STAT_U64