Commit f2600cf0 authored by Eric Dumazet's avatar Eric Dumazet Committed by David S. Miller
Browse files

net: sched: avoid costly atomic operation in fq_dequeue() 



Standard qdisc API to setup a timer implies an atomic operation on every
packet dequeue : qdisc_unthrottled()

It turns out this is not really needed for FQ, as FQ has no concept of
global qdisc throttling, being a qdisc handling many different flows,
some of them can be throttled, while others are not.

Fix is straightforward : add a 'bool throttle' to
qdisc_watchdog_schedule_ns(), and remove calls to qdisc_unthrottled()
in sch_fq.

Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 681d2421

include/net/pkt_sched.h

+2 −2
Original line number Original line Diff line number Diff line
@@ -65,12 +65,12 @@ struct qdisc_watchdog { 
};

};





void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);

void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);

void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires);

void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle);





static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,

static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,

					   psched_time_t expires)

					   psched_time_t expires)

{

{

	qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires));

	qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires), true);

}

}





void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);

void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);

net/sched/sch_api.c

+3 −2
Original line number Original line Diff line number Diff line
@@ -594,12 +594,13 @@ void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 
}

}

EXPORT_SYMBOL(qdisc_watchdog_init);

EXPORT_SYMBOL(qdisc_watchdog_init);





void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)

void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle)

{

{

	if (test_bit(__QDISC_STATE_DEACTIVATED,

	if (test_bit(__QDISC_STATE_DEACTIVATED,

		     &qdisc_root_sleeping(wd->qdisc)->state))

		     &qdisc_root_sleeping(wd->qdisc)->state))

		return;

		return;





	if (throttle)

		qdisc_throttled(wd->qdisc);

		qdisc_throttled(wd->qdisc);





	hrtimer_start(&wd->timer,

	hrtimer_start(&wd->timer,

net/sched/sch_fq.c

+2 −4
Original line number Original line Diff line number Diff line
@@ -377,7 +377,6 @@ static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch) 
		if (time_after(jiffies, f->age + q->flow_refill_delay))

		if (time_after(jiffies, f->age + q->flow_refill_delay))

			f->credit = max_t(u32, f->credit, q->quantum);

			f->credit = max_t(u32, f->credit, q->quantum);

		q->inactive_flows--;

		q->inactive_flows--;

		qdisc_unthrottled(sch);

	}

	}





	/* Note: this overwrites f->age */

	/* Note: this overwrites f->age */
@@ -385,7 +384,6 @@ static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch) 




	if (unlikely(f == &q->internal)) {

	if (unlikely(f == &q->internal)) {

		q->stat_internal_packets++;

		q->stat_internal_packets++;

		qdisc_unthrottled(sch);

	}

	}

	sch->q.qlen++;

	sch->q.qlen++;
@@ -433,7 +431,8 @@ begin: 
		if (!head->first) {

		if (!head->first) {

			if (q->time_next_delayed_flow != ~0ULL)

			if (q->time_next_delayed_flow != ~0ULL)

				qdisc_watchdog_schedule_ns(&q->watchdog,

				qdisc_watchdog_schedule_ns(&q->watchdog,

							   q->time_next_delayed_flow);

							   q->time_next_delayed_flow,

							   false);

			return NULL;

			return NULL;

		}

		}

	}

	}
@@ -495,7 +494,6 @@ begin: 
	}

	}

out:

out:

	qdisc_bstats_update(sch, skb);

	qdisc_bstats_update(sch, skb);

	qdisc_unthrottled(sch);

	return skb;

	return skb;

}

}

net/sched/sch_tbf.c

+2 −1
Original line number Original line Diff line number Diff line
@@ -268,7 +268,8 @@ static struct sk_buff *tbf_dequeue(struct Qdisc *sch) 
		}

		}





		qdisc_watchdog_schedule_ns(&q->watchdog,

		qdisc_watchdog_schedule_ns(&q->watchdog,

					   now + max_t(long, -toks, -ptoks));

					   now + max_t(long, -toks, -ptoks),

					   true);





		/* Maybe we have a shorter packet in the queue,

		/* Maybe we have a shorter packet in the queue,

		   which can be sent now. It sounds cool,

		   which can be sent now. It sounds cool,

CRA Git | Maintained and supported by SUSTech CRA and CCSE