Skip to contentSkip to conversation

: pkt_sched.h cleanup + misc changes

PATCH In review Patrick McHardy 3 participants 17 messages Mar 23, 2007

Subsystems (4)
Networking Network drivers Traffic control (tc) Netem network emulator
lore
Share & tools
lore

Patrick McHardy

author
#0 lore raw

cover : pkt_sched.h cleanup + misc changes

These patches fix an off-by-one in netem, clean up pkt_sched.h by removing most of the now unnecessary PSCHED time macros and turning the two remaining ones into inline functions, consolidate some common filter destruction code and move the TCQ_F_THROTTLED optimization from netem to qdisc_restart.

Please apply, thanks.

 include/net/pkt_sched.h   |   24 +++++++-------
 include/net/red.h         |   10 +++---
 include/net/sch_generic.h |   10 +-----
 net/sched/act_police.c    |   17 ++++------
 net/sched/sch_api.c       |   20 ++++++++++--
 net/sched/sch_atm.c       |   17 +---------
 net/sched/sch_cbq.c       |   76 ++++++++++++++++++----------------------------
 net/sched/sch_dsmark.c    |    8 ----
 net/sched/sch_generic.c   |    4 ++
 net/sched/sch_hfsc.c      |   23 +++----------
 net/sched/sch_htb.c       |   24 ++++----------
 net/sched/sch_ingress.c   |    7 ----
 net/sched/sch_netem.c     |   24 +++++---------
 net/sched/sch_prio.c      |    7 ----
 net/sched/sch_tbf.c       |    9 ++---
 15 files changed, 110 insertions(+), 170 deletions(-)

Patrick McHardy (11):
[NET_SCHED]: sch_netem: fix off-by-one in send time comparison [NET_SCHED]: kill PSCHED_AUDIT_TDIFF
[NET_SCHED]: kill PSCHED_TADD/PSCHED_TADD2 [NET_SCHED]: kill PSCHED_TLESS
[NET_SCHED]: kill PSCHED_SET_PASTPERFECT/PSCHED_IS_PASTPERFECT [NET_SCHED]: kill PSCHED_TDIFF
[NET_SCHED]: turn PSCHED_TDIFF_SAFE into inline function [NET_SCHED]: turn PSCHED_GET_TIME into inline function [NET_SCHED]: Unline tcf_destroy
[NET_SCHED]: qdisc: remove unnecessary memory barriers [NET_SCHED]: qdisc: avoid dequeue while throttled

author
#1 lore raw

PATCH 1/11 : sch_netem: fix off-by-one in send time comparison

[NET_SCHED]: sch_netem: fix off-by-one in send time comparison

netem checks PSCHED_TLESS(cb->time_to_send, now) to find out whether it is allowed to send a packet, which is equivalent to cb->time_to_send < now. Use !PSCHED_TLESS(now, cb->time_to_send) instead to properly handle cb->time_to_send == now.

Signed-off-by Patrick McHardy <[email protected]>

---
commit 9f8c07452088f432c79ac3a8d87d6adebcce57df
tree 42b214f74b8b2d5bd3065e9f63d8048beb4f3bdc
parent 3231f075945001667eafaf325abab8c992b3d1e4
author Patrick McHardy <[email protected]> Thu, 22 Mar 2007 23:57:32 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:26 +0100

 net/sched/sch_netem.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
net/sched/sch_netem.c +1 -1
index 3e1b633..bc42843 100644
@@ -286,7 +286,7 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
 		/* if more time remaining? */
 		PSCHED_GET_TIME(now);
 
-		if (PSCHED_TLESS(cb->time_to_send, now)) {
+		if (!PSCHED_TLESS(now, cb->time_to_send)) {
 			pr_debug("netem_dequeue: return skb=%p\n", skb);
 			sch->q.qlen--;
 			return skb;
#2 lore raw

Re: [NET_SCHED 01/11]: sch_netem: fix off-by-one in send time comparison

Quoted 9 lines · On Fri, 23 Mar 2007 14:35:40 +0100 (MET) Patrick McHardy <[email protected]>…

[NET_SCHED]: sch_netem: fix off-by-one in send time comparison

netem checks PSCHED_TLESS(cb->time_to_send, now) to find out whether it is allowed to send a packet, which is equivalent to cb->time_to_send < now. Use !PSCHED_TLESS(now, cb->time_to_send) instead to properly handle cb->time_to_send == now.

Signed-off-by Patrick McHardy <[email protected]>

Thanks, I saw that earlier in another spot and fixed it.

Signature · 1 line
Stephen Hemminger <[email protected]>
author
#3 lore raw

PATCH 2/11 : kill PSCHED_AUDIT_TDIFF

In reply to Patrick McHardy

[NET_SCHED]: kill PSCHED_AUDIT_TDIFF

Signed-off-by Patrick McHardy <[email protected]>

---
commit 4a4a3d59dca71f202ab063b909c84c96c8ea09a7
tree 4958bfec571a3330bd023ebe50f7b071f6dc7dd7
parent 9f8c07452088f432c79ac3a8d87d6adebcce57df
author Patrick McHardy <[email protected]> Thu, 22 Mar 2007 23:58:12 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:27 +0100

 include/net/pkt_sched.h |    1 -
 net/sched/sch_cbq.c     |    2 --
 2 files changed, 0 insertions(+), 3 deletions(-)
include/net/pkt_sched.h +0 -1
index 6555e57..276d1ad 100644
@@ -59,7 +59,6 @@ typedef long	psched_tdiff_t;
 #define PSCHED_TADD(tv, delta)		((tv) += (delta))
 #define PSCHED_SET_PASTPERFECT(t)	((t) = 0)
 #define PSCHED_IS_PASTPERFECT(t)	((t) == 0)
-#define	PSCHED_AUDIT_TDIFF(t)
 
 struct qdisc_watchdog {
 	struct hrtimer	timer;
net/sched/sch_cbq.c +0 -2
index dcd9c31..57ac6c5 100644
@@ -820,8 +820,6 @@ cbq_update(struct cbq_sched_data *q)
 			idle -= L2T(&q->link, len);
 			idle += L2T(cl, len);
 
-			PSCHED_AUDIT_TDIFF(idle);
-
 			PSCHED_TADD2(q->now, idle, cl->undertime);
 		} else {
 			/* Underlimit */
author
#4 lore raw

PATCH 3/11 : kill PSCHED_TADD/PSCHED_TADD2

In reply to Patrick McHardy

[NET_SCHED]: kill PSCHED_TADD/PSCHED_TADD2

Signed-off-by Patrick McHardy <[email protected]>

---
commit 145a1a6010c6b852ffab28c110d8911a6161aa8b
tree 84b7bf284ea3b870a9b5fd9dae3adaad9979dc26
parent 4a4a3d59dca71f202ab063b909c84c96c8ea09a7
author Patrick McHardy <[email protected]> Thu, 22 Mar 2007 23:58:42 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:27 +0100

 include/net/pkt_sched.h |    2 --
 net/sched/sch_cbq.c     |   12 ++++++------
 net/sched/sch_netem.c   |    2 +-
 3 files changed, 7 insertions(+), 9 deletions(-)
include/net/pkt_sched.h +0 -2
index 276d1ad..32cdf01 100644
@@ -55,8 +55,6 @@ typedef long	psched_tdiff_t;
 #define PSCHED_TDIFF_SAFE(tv1, tv2, bound) \
 					min_t(long long, (tv1) - (tv2), bound)
 #define PSCHED_TLESS(tv1, tv2)		((tv1) < (tv2))
-#define PSCHED_TADD2(tv, delta, tv_res) ((tv_res) = (tv) + (delta))
-#define PSCHED_TADD(tv, delta)		((tv) += (delta))
 #define PSCHED_SET_PASTPERFECT(t)	((t) = 0)
 #define PSCHED_IS_PASTPERFECT(t)	((t) == 0)
net/sched/sch_cbq.c +6 -6
index 57ac6c5..290b26b 100644
@@ -387,7 +387,7 @@ cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl)
 
 		PSCHED_GET_TIME(now);
 		incr = PSCHED_TDIFF(now, q->now_rt);
-		PSCHED_TADD2(q->now, incr, now);
+		now = q->now + incr;
 
 		do {
 			if (PSCHED_TLESS(cl->undertime, now)) {

@@ -492,7 +492,7 @@ static void cbq_ovl_classic(struct cbq_class *cl)
 			cl->avgidle = cl->minidle;
 		if (delay <= 0)
 			delay = 1;
-		PSCHED_TADD2(q->now, delay, cl->undertime);
+		cl->undertime = q->now + delay;
 
 		cl->xstats.overactions++;
 		cl->delayed = 1;

@@ -558,7 +558,7 @@ static void cbq_ovl_delay(struct cbq_class *cl)
 			delay -= (-cl->avgidle) - ((-cl->avgidle) >> cl->ewma_log);
 		if (cl->avgidle < cl->minidle)
 			cl->avgidle = cl->minidle;
-		PSCHED_TADD2(q->now, delay, cl->undertime);
+		cl->undertime = q->now + delay;
 
 		if (delay > 0) {
 			sched += delay + cl->penalty;

@@ -820,7 +820,7 @@ cbq_update(struct cbq_sched_data *q)
 			idle -= L2T(&q->link, len);
 			idle += L2T(cl, len);
 
-			PSCHED_TADD2(q->now, idle, cl->undertime);
+			cl->undertime = q->now + idle;
 		} else {
 			/* Underlimit */
 

@@ -1018,12 +1018,12 @@ cbq_dequeue(struct Qdisc *sch)
 		   cbq_time = max(real_time, work);
 		 */
 		incr2 = L2T(&q->link, q->tx_len);
-		PSCHED_TADD(q->now, incr2);
+		q->now += incr2;
 		cbq_update(q);
 		if ((incr -= incr2) < 0)
 			incr = 0;
 	}
-	PSCHED_TADD(q->now, incr);
+	q->now += incr;
 	q->now_rt = now;
 
 	for (;;) {
net/sched/sch_netem.c +1 -1
index bc42843..6044ae7 100644
@@ -218,7 +218,7 @@ static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 				  &q->delay_cor, q->delay_dist);
 
 		PSCHED_GET_TIME(now);
-		PSCHED_TADD2(now, delay, cb->time_to_send);
+		cb->time_to_send = now + delay;
 		++q->counter;
 		ret = q->qdisc->enqueue(skb, q->qdisc);
 	} else {
author
#5 lore raw

PATCH 4/11 : kill PSCHED_TLESS

In reply to Patrick McHardy

[NET_SCHED]: kill PSCHED_TLESS

Signed-off-by Patrick McHardy <[email protected]>

---
commit 4f8fc418f88c0b7ee6e726b05f27c42d8e20593c
tree c70508f2e0174aef42aaf99bf0cef4184d7ed07e
parent 145a1a6010c6b852ffab28c110d8911a6161aa8b
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 00:00:55 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:28 +0100

 include/net/pkt_sched.h |    1 -
 net/sched/sch_cbq.c     |    7 +++----
 net/sched/sch_netem.c   |    6 +++---
 3 files changed, 6 insertions(+), 8 deletions(-)
include/net/pkt_sched.h +0 -1
index 32cdf01..49325ff 100644
@@ -54,7 +54,6 @@ typedef long	psched_tdiff_t;
 #define PSCHED_TDIFF(tv1, tv2)		(long)((tv1) - (tv2))
 #define PSCHED_TDIFF_SAFE(tv1, tv2, bound) \
 					min_t(long long, (tv1) - (tv2), bound)
-#define PSCHED_TLESS(tv1, tv2)		((tv1) < (tv2))
 #define PSCHED_SET_PASTPERFECT(t)	((t) = 0)
 #define PSCHED_IS_PASTPERFECT(t)	((t) == 0)
net/sched/sch_cbq.c +3 -4
index 290b26b..9e6cdab 100644
@@ -390,7 +390,7 @@ cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl)
 		now = q->now + incr;
 
 		do {
-			if (PSCHED_TLESS(cl->undertime, now)) {
+			if (cl->undertime < now) {
 				q->toplevel = cl->level;
 				return;
 			}

@@ -845,8 +845,7 @@ cbq_under_limit(struct cbq_class *cl)
 	if (cl->tparent == NULL)
 		return cl;
 
-	if (PSCHED_IS_PASTPERFECT(cl->undertime) ||
-	    !PSCHED_TLESS(q->now, cl->undertime)) {
+	if (PSCHED_IS_PASTPERFECT(cl->undertime) || q->now >= cl->undertime) {
 		cl->delayed = 0;
 		return cl;
 	}

@@ -870,7 +869,7 @@ cbq_under_limit(struct cbq_class *cl)
 		if (cl->level > q->toplevel)
 			return NULL;
 	} while (!PSCHED_IS_PASTPERFECT(cl->undertime) &&
-		 PSCHED_TLESS(q->now, cl->undertime));
+		 q->now < cl->undertime);
 
 	cl->delayed = 0;
 	return cl;
net/sched/sch_netem.c +3 -3
index 6044ae7..5d571aa 100644
@@ -286,7 +286,7 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
 		/* if more time remaining? */
 		PSCHED_GET_TIME(now);
 
-		if (!PSCHED_TLESS(now, cb->time_to_send)) {
+		if (cb->time_to_send <= now) {
 			pr_debug("netem_dequeue: return skb=%p\n", skb);
 			sch->q.qlen--;
 			return skb;

@@ -494,7 +494,7 @@ static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
 
 	if (likely(skb_queue_len(list) < q->limit)) {
 		/* Optimize for add at tail */
-		if (likely(skb_queue_empty(list) || !PSCHED_TLESS(tnext, q->oldest))) {
+		if (likely(skb_queue_empty(list) || tnext >= q->oldest)) {
 			q->oldest = tnext;
 			return qdisc_enqueue_tail(nskb, sch);
 		}

@@ -503,7 +503,7 @@ static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
 			const struct netem_skb_cb *cb
 				= (const struct netem_skb_cb *)skb->cb;
 
-			if (!PSCHED_TLESS(tnext, cb->time_to_send))
+			if (tnext >= cb->time_to_send)
 				break;
 		}
author
#6 lore raw

PATCH 5/11 : kill PSCHED_SET_PASTPERFECT/PSCHED_IS_PASTPERFECT

In reply to Patrick McHardy

[NET_SCHED]: kill PSCHED_SET_PASTPERFECT/PSCHED_IS_PASTPERFECT

Use direct assignment and comparison instead.

Signed-off-by Patrick McHardy <[email protected]>

---
commit ec252ac5640ea38d3630cdb97333c398a75391b9
tree bce7b2c63ffb0694942484418f1adf08ed78292d
parent 4f8fc418f88c0b7ee6e726b05f27c42d8e20593c
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 00:01:32 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:29 +0100

 include/net/pkt_sched.h |    3 +--
 include/net/red.h       |    4 ++--
 net/sched/sch_cbq.c     |   17 ++++++++---------
 net/sched/sch_netem.c   |    2 +-
 4 files changed, 12 insertions(+), 14 deletions(-)
include/net/pkt_sched.h +1 -2
index 49325ff..c40147a 100644
@@ -54,8 +54,7 @@ typedef long	psched_tdiff_t;
 #define PSCHED_TDIFF(tv1, tv2)		(long)((tv1) - (tv2))
 #define PSCHED_TDIFF_SAFE(tv1, tv2, bound) \
 					min_t(long long, (tv1) - (tv2), bound)
-#define PSCHED_SET_PASTPERFECT(t)	((t) = 0)
-#define PSCHED_IS_PASTPERFECT(t)	((t) == 0)
+#define PSCHED_PASTPERFECT		0
 
 struct qdisc_watchdog {
 	struct hrtimer	timer;
include/net/red.h +2 -2
index a4eb379..d9e1149 100644
@@ -151,7 +151,7 @@ static inline void red_set_parms(struct red_parms *p,
 
 static inline int red_is_idling(struct red_parms *p)
 {
-	return !PSCHED_IS_PASTPERFECT(p->qidlestart);
+	return p->qidlestart != PSCHED_PASTPERFECT;
 }
 
 static inline void red_start_of_idle_period(struct red_parms *p)

@@ -161,7 +161,7 @@ static inline void red_start_of_idle_period(struct red_parms *p)
 
 static inline void red_end_of_idle_period(struct red_parms *p)
 {
-	PSCHED_SET_PASTPERFECT(p->qidlestart);
+	p->qidlestart = PSCHED_PASTPERFECT;
 }
 
 static inline void red_restart(struct red_parms *p)
net/sched/sch_cbq.c +8 -9
index 9e6cdab..2bb271b 100644
@@ -738,7 +738,7 @@ cbq_update_toplevel(struct cbq_sched_data *q, struct cbq_class *cl,
 	if (cl && q->toplevel >= borrowed->level) {
 		if (cl->q->q.qlen > 1) {
 			do {
-				if (PSCHED_IS_PASTPERFECT(borrowed->undertime)) {
+				if (borrowed->undertime == PSCHED_PASTPERFECT) {
 					q->toplevel = borrowed->level;
 					return;
 				}

@@ -824,7 +824,7 @@ cbq_update(struct cbq_sched_data *q)
 		} else {
 			/* Underlimit */
 
-			PSCHED_SET_PASTPERFECT(cl->undertime);
+			cl->undertime = PSCHED_PASTPERFECT;
 			if (avgidle > cl->maxidle)
 				cl->avgidle = cl->maxidle;
 			else

@@ -845,7 +845,7 @@ cbq_under_limit(struct cbq_class *cl)
 	if (cl->tparent == NULL)
 		return cl;
 
-	if (PSCHED_IS_PASTPERFECT(cl->undertime) || q->now >= cl->undertime) {
+	if (cl->undertime == PSCHED_PASTPERFECT || q->now >= cl->undertime) {
 		cl->delayed = 0;
 		return cl;
 	}

@@ -868,8 +868,7 @@ cbq_under_limit(struct cbq_class *cl)
 		}
 		if (cl->level > q->toplevel)
 			return NULL;
-	} while (!PSCHED_IS_PASTPERFECT(cl->undertime) &&
-		 q->now < cl->undertime);
+	} while (cl->undertime != PSCHED_PASTPERFECT && q->now < cl->undertime);
 
 	cl->delayed = 0;
 	return cl;

@@ -1054,11 +1053,11 @@ cbq_dequeue(struct Qdisc *sch)
 		*/
 
 		if (q->toplevel == TC_CBQ_MAXLEVEL &&
-		    PSCHED_IS_PASTPERFECT(q->link.undertime))
+		    q->link.undertime == PSCHED_PASTPERFECT)
 			break;
 
 		q->toplevel = TC_CBQ_MAXLEVEL;
-		PSCHED_SET_PASTPERFECT(q->link.undertime);
+		q->link.undertime = PSCHED_PASTPERFECT;
 	}
 
 	/* No packets in scheduler or nobody wants to give them to us :-(

@@ -1289,7 +1288,7 @@ cbq_reset(struct Qdisc* sch)
 			qdisc_reset(cl->q);
 
 			cl->next_alive = NULL;
-			PSCHED_SET_PASTPERFECT(cl->undertime);
+			cl->undertime = PSCHED_PASTPERFECT;
 			cl->avgidle = cl->maxidle;
 			cl->deficit = cl->quantum;
 			cl->cpriority = cl->priority;

@@ -1650,7 +1649,7 @@ cbq_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 	cl->xstats.avgidle = cl->avgidle;
 	cl->xstats.undertime = 0;
 
-	if (!PSCHED_IS_PASTPERFECT(cl->undertime))
+	if (cl->undertime != PSCHED_PASTPERFECT)
 		cl->xstats.undertime = PSCHED_TDIFF(cl->undertime, q->now);
 
 	if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
net/sched/sch_netem.c +1 -1
index 5d571aa..1e88301 100644
@@ -532,7 +532,7 @@ static int tfifo_init(struct Qdisc *sch, struct rtattr *opt)
 	} else
 		q->limit = max_t(u32, sch->dev->tx_queue_len, 1);
 
-	PSCHED_SET_PASTPERFECT(q->oldest);
+	q->oldest = PSCHED_PASTPERFECT;
 	return 0;
 }
author
#7 lore raw

PATCH 6/11 : kill PSCHED_TDIFF

In reply to Patrick McHardy

[NET_SCHED]: kill PSCHED_TDIFF

Signed-off-by Patrick McHardy <[email protected]>

---
commit d72d57707edf96c31e62da0841faf59c011dcd92
tree 8b6192c94e025fb8b6e1be3b02526d4792bd4fa1
parent ec252ac5640ea38d3630cdb97333c398a75391b9
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 00:01:47 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:29 +0100

 include/net/pkt_sched.h |    1 -
 net/sched/sch_cbq.c     |   14 +++++++-------
 2 files changed, 7 insertions(+), 8 deletions(-)
include/net/pkt_sched.h +0 -1
index c40147a..1639737 100644
@@ -51,7 +51,6 @@ typedef long	psched_tdiff_t;
 #define PSCHED_GET_TIME(stamp) \
 	((stamp) = PSCHED_NS2US(ktime_to_ns(ktime_get())))
 
-#define PSCHED_TDIFF(tv1, tv2)		(long)((tv1) - (tv2))
 #define PSCHED_TDIFF_SAFE(tv1, tv2, bound) \
 					min_t(long long, (tv1) - (tv2), bound)
 #define PSCHED_PASTPERFECT		0
net/sched/sch_cbq.c +7 -7
index 2bb271b..f9e8403 100644
@@ -386,7 +386,7 @@ cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl)
 		psched_tdiff_t incr;
 
 		PSCHED_GET_TIME(now);
-		incr = PSCHED_TDIFF(now, q->now_rt);
+		incr = now - q->now_rt;
 		now = q->now + incr;
 
 		do {

@@ -474,7 +474,7 @@ cbq_requeue(struct sk_buff *skb, struct Qdisc *sch)
 static void cbq_ovl_classic(struct cbq_class *cl)
 {
 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
-	psched_tdiff_t delay = PSCHED_TDIFF(cl->undertime, q->now);
+	psched_tdiff_t delay = cl->undertime - q->now;
 
 	if (!cl->delayed) {
 		delay += cl->offtime;

@@ -509,7 +509,7 @@ static void cbq_ovl_classic(struct cbq_class *cl)
 		psched_tdiff_t base_delay = q->wd_expires;
 
 		for (b = cl->borrow; b; b = b->borrow) {
-			delay = PSCHED_TDIFF(b->undertime, q->now);
+			delay = b->undertime - q->now;
 			if (delay < base_delay) {
 				if (delay <= 0)
 					delay = 1;

@@ -547,7 +547,7 @@ static void cbq_ovl_rclassic(struct cbq_class *cl)
 static void cbq_ovl_delay(struct cbq_class *cl)
 {
 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
-	psched_tdiff_t delay = PSCHED_TDIFF(cl->undertime, q->now);
+	psched_tdiff_t delay = cl->undertime - q->now;
 
 	if (!cl->delayed) {
 		psched_time_t sched = q->now;

@@ -776,7 +776,7 @@ cbq_update(struct cbq_sched_data *q)
 			 idle = (now - last) - last_pktlen/rate
 		 */
 
-		idle = PSCHED_TDIFF(q->now, cl->last);
+		idle = q->now - cl->last;
 		if ((unsigned long)idle > 128*1024*1024) {
 			avgidle = cl->maxidle;
 		} else {

@@ -1004,7 +1004,7 @@ cbq_dequeue(struct Qdisc *sch)
 	psched_tdiff_t incr;
 
 	PSCHED_GET_TIME(now);
-	incr = PSCHED_TDIFF(now, q->now_rt);
+	incr = now - q->now_rt;
 
 	if (q->tx_class) {
 		psched_tdiff_t incr2;

@@ -1650,7 +1650,7 @@ cbq_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 	cl->xstats.undertime = 0;
 
 	if (cl->undertime != PSCHED_PASTPERFECT)
-		cl->xstats.undertime = PSCHED_TDIFF(cl->undertime, q->now);
+		cl->xstats.undertime = cl->undertime - q->now;
 
 	if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
 #ifdef CONFIG_NET_ESTIMATOR
author
#8 lore raw

PATCH 7/11 : turn PSCHED_TDIFF_SAFE into inline function

In reply to Patrick McHardy

[NET_SCHED]: turn PSCHED_TDIFF_SAFE into inline function

Also rename to psched_tdiff_bounded.

Signed-off-by Patrick McHardy <[email protected]>

---
commit c86b236046f7de4094ceb2b2cb069c32969ee36c
tree 27c99a0d619bcabf384838adeae3c0469472b86b
parent d72d57707edf96c31e62da0841faf59c011dcd92
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 00:01:59 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:30 +0100

 include/net/pkt_sched.h |    8 ++++++--
 include/net/red.h       |    2 +-
 net/sched/act_police.c  |    8 ++++----
 net/sched/sch_htb.c     |    4 ++--
 net/sched/sch_tbf.c     |    2 +-
 5 files changed, 14 insertions(+), 10 deletions(-)
include/net/pkt_sched.h +6 -2
index 1639737..e6b1da0 100644
@@ -51,10 +51,14 @@ typedef long	psched_tdiff_t;
 #define PSCHED_GET_TIME(stamp) \
 	((stamp) = PSCHED_NS2US(ktime_to_ns(ktime_get())))
 
-#define PSCHED_TDIFF_SAFE(tv1, tv2, bound) \
-					min_t(long long, (tv1) - (tv2), bound)
 #define PSCHED_PASTPERFECT		0
 
+static inline psched_tdiff_t
+psched_tdiff_bounded(psched_time_t tv1, psched_time_t tv2, psched_time_t bound)
+{
+	return min(tv1 - tv2, bound);
+}
+
 struct qdisc_watchdog {
 	struct hrtimer	timer;
 	struct Qdisc	*qdisc;
include/net/red.h +1 -1
index d9e1149..0bc1691 100644
@@ -178,7 +178,7 @@ static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
 	int  shift;
 
 	PSCHED_GET_TIME(now);
-	us_idle = PSCHED_TDIFF_SAFE(now, p->qidlestart, p->Scell_max);
+	us_idle = psched_tdiff_bounded(now, p->qidlestart, p->Scell_max);
 
 	/*
 	 * The problem: ideally, average length queue recalcultion should
net/sched/act_police.c +4 -4
index 0a5679e..65d60a3 100644
@@ -298,8 +298,8 @@ static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
 
 		PSCHED_GET_TIME(now);
 
-		toks = PSCHED_TDIFF_SAFE(now, police->tcfp_t_c,
-					 police->tcfp_burst);
+		toks = psched_tdiff_bounded(now, police->tcfp_t_c,
+					    police->tcfp_burst);
 		if (police->tcfp_P_tab) {
 			ptoks = toks + police->tcfp_ptoks;
 			if (ptoks > (long)L2T_P(police, police->tcfp_mtu))

@@ -544,8 +544,8 @@ int tcf_police(struct sk_buff *skb, struct tcf_police *police)
 		}
 
 		PSCHED_GET_TIME(now);
-		toks = PSCHED_TDIFF_SAFE(now, police->tcfp_t_c,
-					 police->tcfp_burst);
+		toks = psched_tdiff_bounded(now, police->tcfp_t_c,
+					    police->tcfp_burst);
 		if (police->tcfp_P_tab) {
 			ptoks = toks + police->tcfp_ptoks;
 			if (ptoks > (long)L2T_P(police, police->tcfp_mtu))
net/sched/sch_htb.c +2 -2
index d265ac4..f629ce2 100644
@@ -729,7 +729,7 @@ static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
 	cl->T = toks
 
 	while (cl) {
-		diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer);
+		diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
 		if (cl->level >= level) {
 			if (cl->level == level)
 				cl->xstats.lends++;

@@ -789,7 +789,7 @@ static psched_time_t htb_do_events(struct htb_sched *q, int level)
 			return cl->pq_key;
 
 		htb_safe_rb_erase(p, q->wait_pq + level);
-		diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer);
+		diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
 		htb_change_class_mode(q, cl, &diff);
 		if (cl->cmode != HTB_CAN_SEND)
 			htb_add_to_wait_tree(q, cl, diff);
net/sched/sch_tbf.c +1 -1
index 626ce96..da9f40e 100644
@@ -201,7 +201,7 @@ static struct sk_buff *tbf_dequeue(struct Qdisc* sch)
 
 		PSCHED_GET_TIME(now);
 
-		toks = PSCHED_TDIFF_SAFE(now, q->t_c, q->buffer);
+		toks = psched_tdiff_bounded(now, q->t_c, q->buffer);
 
 		if (q->P_tab) {
 			ptoks = toks + q->ptokens;
author
#9 lore raw

PATCH 8/11 : turn PSCHED_GET_TIME into inline function

In reply to Patrick McHardy

[NET_SCHED]: turn PSCHED_GET_TIME into inline function

Signed-off-by Patrick McHardy <[email protected]>

---
commit 231788aa3b9eef85b72ecac2e33441bd842ce3f4
tree f302e509ec32a86bc9a6c3712d188fc91455a213
parent c86b236046f7de4094ceb2b2cb069c32969ee36c
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 00:02:12 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:30 +0100

 include/net/pkt_sched.h |    8 +++++---
 include/net/red.h       |    4 ++--
 net/sched/act_police.c  |    9 ++++-----
 net/sched/sch_cbq.c     |   10 +++++-----
 net/sched/sch_hfsc.c    |   10 ++++------
 net/sched/sch_htb.c     |    6 +++---
 net/sched/sch_netem.c   |    8 +++-----
 net/sched/sch_tbf.c     |    7 +++----
 8 files changed, 29 insertions(+), 33 deletions(-)
include/net/pkt_sched.h +5 -3
index e6b1da0..b2cc9a8 100644
@@ -48,11 +48,13 @@ typedef long	psched_tdiff_t;
 #define PSCHED_NS2US(x)			((x) >> 10)
 
 #define PSCHED_TICKS_PER_SEC		PSCHED_NS2US(NSEC_PER_SEC)
-#define PSCHED_GET_TIME(stamp) \
-	((stamp) = PSCHED_NS2US(ktime_to_ns(ktime_get())))
-
 #define PSCHED_PASTPERFECT		0
 
+static inline psched_time_t psched_get_time(void)
+{
+	return PSCHED_NS2US(ktime_to_ns(ktime_get()));
+}
+
 static inline psched_tdiff_t
 psched_tdiff_bounded(psched_time_t tv1, psched_time_t tv2, psched_time_t bound)
 {
include/net/red.h +2 -2
index 0bc1691..3cf31d4 100644
@@ -156,7 +156,7 @@ static inline int red_is_idling(struct red_parms *p)
 
 static inline void red_start_of_idle_period(struct red_parms *p)
 {
-	PSCHED_GET_TIME(p->qidlestart);
+	p->qidlestart = psched_get_time();
 }
 
 static inline void red_end_of_idle_period(struct red_parms *p)

@@ -177,7 +177,7 @@ static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
 	long us_idle;
 	int  shift;
 
-	PSCHED_GET_TIME(now);
+	now = psched_get_time();
 	us_idle = psched_tdiff_bounded(now, p->qidlestart, p->Scell_max);
 
 	/*
net/sched/act_police.c +4 -5
index 65d60a3..616f465 100644
@@ -241,7 +241,7 @@ override:
 	if (ret != ACT_P_CREATED)
 		return ret;
 
-	PSCHED_GET_TIME(police->tcfp_t_c);
+	police->tcfp_t_c = psched_get_time();
 	police->tcf_index = parm->index ? parm->index :
 		tcf_hash_new_index(&police_idx_gen, &police_hash_info);
 	h = tcf_hash(police->tcf_index, POL_TAB_MASK);

@@ -296,8 +296,7 @@ static int tcf_act_police(struct sk_buff *skb, struct tc_action *a,
 			return police->tcfp_result;
 		}
 
-		PSCHED_GET_TIME(now);
-
+		now = psched_get_time();
 		toks = psched_tdiff_bounded(now, police->tcfp_t_c,
 					    police->tcfp_burst);
 		if (police->tcfp_P_tab) {

@@ -495,7 +494,7 @@ struct tcf_police *tcf_police_locate(struct rtattr *rta, struct rtattr *est)
 	}
 	if (police->tcfp_P_tab)
 		police->tcfp_ptoks = L2T_P(police, police->tcfp_mtu);
-	PSCHED_GET_TIME(police->tcfp_t_c);
+	police->tcfp_t_c = psched_get_time();
 	police->tcf_index = parm->index ? parm->index :
 		tcf_police_new_index();
 	police->tcf_action = parm->action;

@@ -543,7 +542,7 @@ int tcf_police(struct sk_buff *skb, struct tcf_police *police)
 			return police->tcfp_result;
 		}
 
-		PSCHED_GET_TIME(now);
+		now = psched_get_time();
 		toks = psched_tdiff_bounded(now, police->tcfp_t_c,
 					    police->tcfp_burst);
 		if (police->tcfp_P_tab) {
net/sched/sch_cbq.c +5 -5
index f9e8403..414a97c 100644
@@ -385,7 +385,7 @@ cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl)
 		psched_time_t now;
 		psched_tdiff_t incr;
 
-		PSCHED_GET_TIME(now);
+		now = psched_get_time();
 		incr = now - q->now_rt;
 		now = q->now + incr;
 

@@ -654,7 +654,7 @@ static enum hrtimer_restart cbq_undelay(struct hrtimer *timer)
 	psched_tdiff_t delay = 0;
 	unsigned pmask;
 
-	PSCHED_GET_TIME(now);
+	now = psched_get_time();
 
 	pmask = q->pmask;
 	q->pmask = 0;

@@ -1003,7 +1003,7 @@ cbq_dequeue(struct Qdisc *sch)
 	psched_time_t now;
 	psched_tdiff_t incr;
 
-	PSCHED_GET_TIME(now);
+	now = psched_get_time();
 	incr = now - q->now_rt;
 
 	if (q->tx_class) {

@@ -1277,7 +1277,7 @@ cbq_reset(struct Qdisc* sch)
 	qdisc_watchdog_cancel(&q->watchdog);
 	hrtimer_cancel(&q->delay_timer);
 	q->toplevel = TC_CBQ_MAXLEVEL;
-	PSCHED_GET_TIME(q->now);
+	q->now = psched_get_time();
 	q->now_rt = q->now;
 
 	for (prio = 0; prio <= TC_CBQ_MAXPRIO; prio++)

@@ -1448,7 +1448,7 @@ static int cbq_init(struct Qdisc *sch, struct rtattr *opt)
 	hrtimer_init(&q->delay_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 	q->delay_timer.function = cbq_undelay;
 	q->toplevel = TC_CBQ_MAXLEVEL;
-	PSCHED_GET_TIME(q->now);
+	q->now = psched_get_time();
 	q->now_rt = q->now;
 
 	cbq_link_class(&q->link);
net/sched/sch_hfsc.c +4 -6
index 79126e5..0913f71 100644
@@ -629,9 +629,7 @@ rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
 static void
 init_ed(struct hfsc_class *cl, unsigned int next_len)
 {
-	u64 cur_time;
-
-	PSCHED_GET_TIME(cur_time);
+	u64 cur_time = psched_get_time();
 
 	/* update the deadline curve */
 	rtsc_min(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);

@@ -754,7 +752,7 @@ init_vf(struct hfsc_class *cl, unsigned int len)
 			if (cl->cl_flags & HFSC_USC) {
 				/* class has upper limit curve */
 				if (cur_time == 0)
-					PSCHED_GET_TIME(cur_time);
+					cur_time = psched_get_time();
 
 				/* update the ulimit curve */
 				rtsc_min(&cl->cl_ulimit, &cl->cl_usc, cur_time,

@@ -1038,7 +1036,7 @@ hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
 			if (cl->cl_parent == NULL && parentid != TC_H_ROOT)
 				return -EINVAL;
 		}
-		PSCHED_GET_TIME(cur_time);
+		cur_time = psched_get_time();
 
 		sch_tree_lock(sch);
 		if (rsc != NULL)

@@ -1637,7 +1635,7 @@ hfsc_dequeue(struct Qdisc *sch)
 	if ((skb = __skb_dequeue(&q->requeue)))
 		goto out;
 
-	PSCHED_GET_TIME(cur_time);
+	cur_time = psched_get_time();
 
 	/*
 	 * if there are eligible classes, use real-time criteria.
net/sched/sch_htb.c +3 -3
index f629ce2..00f4c4f 100644
@@ -965,7 +965,7 @@ static struct sk_buff *htb_dequeue(struct Qdisc *sch)
 
 	if (!sch->q.qlen)
 		goto fin;
-	PSCHED_GET_TIME(q->now);
+	q->now = psched_get_time();
 
 	next_event = q->now + 5 * PSCHED_TICKS_PER_SEC;
 	q->nwc_hit = 0;

@@ -1274,7 +1274,7 @@ static void htb_parent_to_leaf(struct htb_class *cl, struct Qdisc *new_q)
 	parent->un.leaf.prio = parent->prio;
 	parent->tokens = parent->buffer;
 	parent->ctokens = parent->cbuffer;
-	PSCHED_GET_TIME(parent->t_c);
+	parent->t_c = psched_get_time();
 	parent->cmode = HTB_CAN_SEND;
 }
 

@@ -1471,7 +1471,7 @@ static int htb_change_class(struct Qdisc *sch, u32 classid,
 		cl->tokens = hopt->buffer;
 		cl->ctokens = hopt->cbuffer;
 		cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC;	/* 1min */
-		PSCHED_GET_TIME(cl->t_c);
+		cl->t_c = psched_get_time();
 		cl->cmode = HTB_CAN_SEND;
 
 		/* attach to the hash list and parent's family */
net/sched/sch_netem.c +3 -5
index 1e88301..5d9d8bc 100644
@@ -217,7 +217,7 @@ static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		delay = tabledist(q->latency, q->jitter,
 				  &q->delay_cor, q->delay_dist);
 
-		PSCHED_GET_TIME(now);
+		now = psched_get_time();
 		cb->time_to_send = now + delay;
 		++q->counter;
 		ret = q->qdisc->enqueue(skb, q->qdisc);

@@ -226,7 +226,7 @@ static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		 * Do re-ordering by putting one out of N packets at the front
 		 * of the queue.
 		 */
-		PSCHED_GET_TIME(cb->time_to_send);
+		cb->time_to_send = psched_get_time();
 		q->counter = 0;
 		ret = q->qdisc->ops->requeue(skb, q->qdisc);
 	}

@@ -281,11 +281,9 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
 	if (skb) {
 		const struct netem_skb_cb *cb
 			= (const struct netem_skb_cb *)skb->cb;
-		psched_time_t now;
+		psched_time_t now = psched_get_time();
 
 		/* if more time remaining? */
-		PSCHED_GET_TIME(now);
-
 		if (cb->time_to_send <= now) {
 			pr_debug("netem_dequeue: return skb=%p\n", skb);
 			sch->q.qlen--;
net/sched/sch_tbf.c +3 -4
index da9f40e..5386295 100644
@@ -199,8 +199,7 @@ static struct sk_buff *tbf_dequeue(struct Qdisc* sch)
 		long ptoks = 0;
 		unsigned int len = skb->len;
 
-		PSCHED_GET_TIME(now);
-
+		now = psched_get_time();
 		toks = psched_tdiff_bounded(now, q->t_c, q->buffer);
 
 		if (q->P_tab) {

@@ -254,7 +253,7 @@ static void tbf_reset(struct Qdisc* sch)
 
 	qdisc_reset(q->qdisc);
 	sch->q.qlen = 0;
-	PSCHED_GET_TIME(q->t_c);
+	q->t_c = psched_get_time();
 	q->tokens = q->buffer;
 	q->ptokens = q->mtu;
 	qdisc_watchdog_cancel(&q->watchdog);

@@ -364,7 +363,7 @@ static int tbf_init(struct Qdisc* sch, struct rtattr *opt)
 	if (opt == NULL)
 		return -EINVAL;
 
-	PSCHED_GET_TIME(q->t_c);
+	q->t_c = psched_get_time();
 	qdisc_watchdog_init(&q->watchdog, sch);
 	q->qdisc = &noop_qdisc;
author
#10 lore raw

PATCH 9/11 : Unline tcf_destroy

In reply to Patrick McHardy

[NET_SCHED]: Unline tcf_destroy

Uninline tcf_destroy and add a helper function to destroy an entire filter chain.

Signed-off-by Patrick McHardy <[email protected]>

---
commit 8da4bcec7e54c8344c8fd77c72a61f24ce12cfc3
tree 7f36f4af8e9413637fb9b65501f281fd8a915da3
parent 231788aa3b9eef85b72ecac2e33441bd842ce3f4
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:31 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:31 +0100

 include/net/sch_generic.h |   10 ++--------
 net/sched/sch_api.c       |   18 ++++++++++++++++++
 net/sched/sch_atm.c       |   17 ++---------------
 net/sched/sch_cbq.c       |   14 ++------------
 net/sched/sch_dsmark.c    |    8 +-------
 net/sched/sch_hfsc.c      |   13 +------------
 net/sched/sch_htb.c       |   14 ++------------
 net/sched/sch_ingress.c   |    7 +------
 net/sched/sch_prio.c      |    7 +------
 9 files changed, 30 insertions(+), 78 deletions(-)
include/net/sch_generic.h +2 -8
index a3f4ddd..1b8e351 100644
@@ -177,14 +177,8 @@ extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
 extern struct Qdisc *qdisc_alloc(struct net_device *dev, struct Qdisc_ops *ops);
 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
 				       struct Qdisc_ops *ops, u32 parentid);
-
-static inline void
-tcf_destroy(struct tcf_proto *tp)
-{
-	tp->ops->destroy(tp);
-	module_put(tp->ops->owner);
-	kfree(tp);
-}
+extern void tcf_destroy(struct tcf_proto *tp);
+extern void tcf_destroy_chain(struct tcf_proto *fl);
 
 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
 				       struct sk_buff_head *list)
net/sched/sch_api.c +18 -0
index 5873250..5b5bce0 100644
@@ -1220,6 +1220,24 @@ reclassify:
 	return -1;
 }
 
+void tcf_destroy(struct tcf_proto *tp)
+{
+	tp->ops->destroy(tp);
+	module_put(tp->ops->owner);
+	kfree(tp);
+}
+
+void tcf_destroy_chain(struct tcf_proto *fl)
+{
+	struct tcf_proto *tp;
+
+	while ((tp = fl) != NULL) {
+		fl = tp->next;
+		tcf_destroy(tp);
+	}
+}
+EXPORT_SYMBOL(tcf_destroy_chain);
+
 #ifdef CONFIG_PROC_FS
 static int psched_show(struct seq_file *seq, void *v)
 {
net/sched/sch_atm.c +2 -15
index 0cc3c9b..be7d299 100644
@@ -158,19 +158,6 @@ static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
 	return atm_tc_get(sch,classid);
 }
 
-
-static void destroy_filters(struct atm_flow_data *flow)
-{
-	struct tcf_proto *filter;
-
-	while ((filter = flow->filter_list)) {
-		DPRINTK("destroy_filters: destroying filter %p\n",filter);
-		flow->filter_list = filter->next;
-		tcf_destroy(filter);
-	}
-}
-
-
 /*
  * atm_tc_put handles all destructions, including the ones that are explicitly
  * requested (atm_tc_destroy, etc.). The assumption here is that we never drop

@@ -195,7 +182,7 @@ static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
 	*prev = flow->next;
 	DPRINTK("atm_tc_put: qdisc %p\n",flow->q);
 	qdisc_destroy(flow->q);
-	destroy_filters(flow);
+	tcf_destroy_chain(flow->filter_list);
 	if (flow->sock) {
 		DPRINTK("atm_tc_put: f_count %d\n",
 		    file_count(flow->sock->file));

@@ -611,7 +598,7 @@ static void atm_tc_destroy(struct Qdisc *sch)
 	DPRINTK("atm_tc_destroy(sch %p,[qdisc %p])\n",sch,p);
 	/* races ? */
 	while ((flow = p->flows)) {
-		destroy_filters(flow);
+		tcf_destroy_chain(flow->filter_list);
 		if (flow->ref > 1)
 			printk(KERN_ERR "atm_destroy: %p->ref = %d\n",flow,
 			    flow->ref);
net/sched/sch_cbq.c +2 -12
index 414a97c..a294542 100644
@@ -1717,23 +1717,13 @@ static unsigned long cbq_get(struct Qdisc *sch, u32 classid)
 	return 0;
 }
 
-static void cbq_destroy_filters(struct cbq_class *cl)
-{
-	struct tcf_proto *tp;
-
-	while ((tp = cl->filter_list) != NULL) {
-		cl->filter_list = tp->next;
-		tcf_destroy(tp);
-	}
-}
-
 static void cbq_destroy_class(struct Qdisc *sch, struct cbq_class *cl)
 {
 	struct cbq_sched_data *q = qdisc_priv(sch);
 
 	BUG_TRAP(!cl->filters);
 
-	cbq_destroy_filters(cl);
+	tcf_destroy_chain(cl->filter_list);
 	qdisc_destroy(cl->q);
 	qdisc_put_rtab(cl->R_tab);
 #ifdef CONFIG_NET_ESTIMATOR

@@ -1760,7 +1750,7 @@ cbq_destroy(struct Qdisc* sch)
 	 */
 	for (h = 0; h < 16; h++)
 		for (cl = q->classes[h]; cl; cl = cl->next)
-			cbq_destroy_filters(cl);
+			tcf_destroy_chain(cl->filter_list);
 
 	for (h = 0; h < 16; h++) {
 		struct cbq_class *next;
net/sched/sch_dsmark.c +1 -7
index 2c857af..e38e0d0 100644
@@ -412,16 +412,10 @@ static void dsmark_reset(struct Qdisc *sch)
 static void dsmark_destroy(struct Qdisc *sch)
 {
 	struct dsmark_qdisc_data *p = PRIV(sch);
-	struct tcf_proto *tp;
 
 	DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
 
-	while (p->filter_list) {
-		tp = p->filter_list;
-		p->filter_list = tp->next;
-		tcf_destroy(tp);
-	}
-
+	tcf_destroy_chain(p->filter_list);
 	qdisc_destroy(p->q);
 	kfree(p->mask);
 }
net/sched/sch_hfsc.c +1 -12
index 0913f71..22cec11 100644
@@ -1122,22 +1122,11 @@ hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
 }
 
 static void
-hfsc_destroy_filters(struct tcf_proto **fl)
-{
-	struct tcf_proto *tp;
-
-	while ((tp = *fl) != NULL) {
-		*fl = tp->next;
-		tcf_destroy(tp);
-	}
-}
-
-static void
 hfsc_destroy_class(struct Qdisc *sch, struct hfsc_class *cl)
 {
 	struct hfsc_sched *q = qdisc_priv(sch);
 
-	hfsc_destroy_filters(&cl->filter_list);
+	tcf_destroy_chain(cl->filter_list);
 	qdisc_destroy(cl->qdisc);
 #ifdef CONFIG_NET_ESTIMATOR
 	gen_kill_estimator(&cl->bstats, &cl->rate_est);
net/sched/sch_htb.c +2 -12
index 00f4c4f..71db121 100644
@@ -1236,16 +1236,6 @@ static unsigned long htb_get(struct Qdisc *sch, u32 classid)
 	return (unsigned long)cl;
 }
 
-static void htb_destroy_filters(struct tcf_proto **fl)
-{
-	struct tcf_proto *tp;
-
-	while ((tp = *fl) != NULL) {
-		*fl = tp->next;
-		tcf_destroy(tp);
-	}
-}
-
 static inline int htb_parent_last_child(struct htb_class *cl)
 {
 	if (!cl->parent)

@@ -1289,7 +1279,7 @@ static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
 	qdisc_put_rtab(cl->rate);
 	qdisc_put_rtab(cl->ceil);
 
-	htb_destroy_filters(&cl->filter_list);
+	tcf_destroy_chain(cl->filter_list);
 
 	while (!list_empty(&cl->children))
 		htb_destroy_class(sch, list_entry(cl->children.next,

@@ -1321,7 +1311,7 @@ static void htb_destroy(struct Qdisc *sch)
 	   and surprisingly it worked in 2.4. But it must precede it
 	   because filter need its target class alive to be able to call
 	   unbind_filter on it (without Oops). */
-	htb_destroy_filters(&q->filter_list);
+	tcf_destroy_chain(q->filter_list);
 
 	while (!list_empty(&q->root))
 		htb_destroy_class(sch, list_entry(q->root.next,
net/sched/sch_ingress.c +1 -6
index f63d5c6..1fb60ab 100644
@@ -346,14 +346,9 @@ static void ingress_reset(struct Qdisc *sch)
 static void ingress_destroy(struct Qdisc *sch)
 {
 	struct ingress_qdisc_data *p = PRIV(sch);
-	struct tcf_proto *tp;
 
 	DPRINTK("ingress_destroy(sch %p,[qdisc %p])\n", sch, p);
-	while (p->filter_list) {
-		tp = p->filter_list;
-		p->filter_list = tp->next;
-		tcf_destroy(tp);
-	}
+	tcf_destroy_chain(p->filter_list);
 #if 0
 /* for future use */
 	qdisc_destroy(p->q);
net/sched/sch_prio.c +1 -6
index f139963..5cfe60b 100644
@@ -189,13 +189,8 @@ prio_destroy(struct Qdisc* sch)
 {
 	int prio;
 	struct prio_sched_data *q = qdisc_priv(sch);
-	struct tcf_proto *tp;
-
-	while ((tp = q->filter_list) != NULL) {
-		q->filter_list = tp->next;
-		tcf_destroy(tp);
-	}
 
+	tcf_destroy_chain(q->filter_list);
 	for (prio=0; prio<q->bands; prio++)
 		qdisc_destroy(q->queues[prio]);
 }
author
#11 lore raw

PATCH 10/11 : qdisc: remove unnecessary memory barriers

In reply to Patrick McHardy

[NET_SCHED]: qdisc: remove unnecessary memory barriers

We're holding dev->queue_lock in qdisc_watchdog_schedule and qdisc_watchdog_cancel, no need for the barriers.

Signed-off-by Patrick McHardy <[email protected]>

---
commit e2459acd7dee06fb4d5e980f26c23d31db0e5de1
tree 47287fb285e9d920392863121e88e69e45779cbb
parent 8da4bcec7e54c8344c8fd77c72a61f24ce12cfc3
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:31 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:31 +0100

 net/sched/sch_api.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)
net/sched/sch_api.c +0 -2
index 5b5bce0..2e863bd 100644
@@ -322,7 +322,6 @@ void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
 	ktime_t time;
 
 	wd->qdisc->flags |= TCQ_F_THROTTLED;
-	smp_wmb();
 	time = ktime_set(0, 0);
 	time = ktime_add_ns(time, PSCHED_US2NS(expires));
 	hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);

@@ -333,7 +332,6 @@ void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
 {
 	hrtimer_cancel(&wd->timer);
 	wd->qdisc->flags &= ~TCQ_F_THROTTLED;
-	smp_wmb();
 }
 EXPORT_SYMBOL(qdisc_watchdog_cancel);
author
#12 lore raw

PATCH 11/11 : qdisc: avoid dequeue while throttled

In reply to Patrick McHardy

[NET_SCHED]: qdisc: avoid dequeue while throttled

Avoid dequeueing while the device is throttled.

Signed-off-by Patrick McHardy <[email protected]>

---
commit 69f6039ded0bdaf93e3fc954140caae1e7fc9927
tree 9894ce3afc9f7e5cadb26269827ba4d4bca393fc
parent e2459acd7dee06fb4d5e980f26c23d31db0e5de1
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:32 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 10:31:32 +0100

 net/sched/sch_generic.c |    4 ++++
 net/sched/sch_netem.c   |    4 ----
 2 files changed, 4 insertions(+), 4 deletions(-)
net/sched/sch_generic.c +4 -0
index 52eb343..39c5312 100644
@@ -93,6 +93,10 @@ static inline int qdisc_restart(struct net_device *dev)
 	struct Qdisc *q = dev->qdisc;
 	struct sk_buff *skb;
 
+	smp_rmb();
+	if (q->flags & TCQ_F_THROTTLED)
+		return q->q.qlen;
+
 	/* Dequeue packet */
 	if (((skb = dev->gso_skb)) || ((skb = q->dequeue(q)))) {
 		unsigned nolock = (dev->features & NETIF_F_LLTX);
net/sched/sch_netem.c +0 -4
index 5d9d8bc..4c7a8d8 100644
@@ -273,10 +273,6 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
 	struct netem_sched_data *q = qdisc_priv(sch);
 	struct sk_buff *skb;
 
-	smp_mb();
-	if (sch->flags & TCQ_F_THROTTLED)
-		return NULL;
-
 	skb = q->qdisc->dequeue(q->qdisc);
 	if (skb) {
 		const struct netem_skb_cb *cb
author
#13 lore raw

Re: [NET_SCHED 11/11]: qdisc: avoid dequeue while throttled

Quoted 1 line · Patrick McHardy

[NET_SCHED]: qdisc: avoid dequeue while throttled

It just occured to me that this doesn't work properly with qdiscs that have multiple classes since they all don't properly maintain the TCQ_F_THROTTLED flag. They set it on dequeue when no active class is willing to give out packets, but when enqueueing to a non-active class (thereby activating it) it is still set even though we don't know if that class could be dequeued.

So this updated patch unsets the TCQ_F_THROTTLED flag whenever we activate a class. Additionally it removes the unsetting of TCQ_F_THROTTLED on successful dequeue since we're now guaranteed that it was not set before.

[NET_SCHED]: qdisc: avoid dequeue while throttled

Avoid dequeueing while the device is throttled.

Signed-off-by Patrick McHardy <[email protected]>

---
commit 073456c84a46736a3aa1ae4cc9d953a9e97b327c
tree 805a29224001180c88a429e65812b97a489c427a
parent e2459acd7dee06fb4d5e980f26c23d31db0e5de1
author Patrick McHardy <[email protected]> Fri, 23 Mar 2007 15:37:51 +0100 committer Patrick McHardy <[email protected]> Fri, 23 Mar 2007 15:37:51 +0100

 net/sched/sch_cbq.c     |    5 +++--
 net/sched/sch_generic.c |    4 ++++
 net/sched/sch_hfsc.c    |    5 +++--
 net/sched/sch_htb.c     |    6 ++++--
 net/sched/sch_netem.c   |    4 ----
 net/sched/sch_tbf.c     |    1 -
 6 files changed, 14 insertions(+), 11 deletions(-)
net/sched/sch_cbq.c +3 -2
index a294542..151f8e3 100644
@@ -424,8 +424,10 @@ cbq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		sch->bstats.packets++;
 		sch->bstats.bytes+=len;
 		cbq_mark_toplevel(q, cl);
-		if (!cl->next_alive)
+		if (!cl->next_alive) {
 			cbq_activate_class(cl);
+			sch->flags &= ~TCQ_F_THROTTLED;
+		}
 		return ret;
 	}
 

@@ -1030,7 +1032,6 @@ cbq_dequeue(struct Qdisc *sch)
 		skb = cbq_dequeue_1(sch);
 		if (skb) {
 			sch->q.qlen--;
-			sch->flags &= ~TCQ_F_THROTTLED;
 			return skb;
 		}
net/sched/sch_generic.c +4 -0
index 52eb343..39c5312 100644
@@ -93,6 +93,10 @@ static inline int qdisc_restart(struct net_device *dev)
 	struct Qdisc *q = dev->qdisc;
 	struct sk_buff *skb;
 
+	smp_rmb();
+	if (q->flags & TCQ_F_THROTTLED)
+		return q->q.qlen;
+
 	/* Dequeue packet */
 	if (((skb = dev->gso_skb)) || ((skb = q->dequeue(q)))) {
 		unsigned nolock = (dev->features & NETIF_F_LLTX);
net/sched/sch_hfsc.c +3 -2
index 22cec11..c6da436 100644
@@ -1597,8 +1597,10 @@ hfsc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		return err;
 	}
 
-	if (cl->qdisc->q.qlen == 1)
+	if (cl->qdisc->q.qlen == 1) {
 		set_active(cl, len);
+		sch->flags &= ~TCQ_F_THROTTLED;
+	}
 
 	cl->bstats.packets++;
 	cl->bstats.bytes += len;

@@ -1672,7 +1674,6 @@ hfsc_dequeue(struct Qdisc *sch)
 	}
 
  out:
-	sch->flags &= ~TCQ_F_THROTTLED;
 	sch->q.qlen--;
 
 	return skb;
net/sched/sch_htb.c +4 -2
index 71db121..1387b7b 100644
@@ -615,6 +615,8 @@ static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		/* enqueue to helper queue */
 		if (q->direct_queue.qlen < q->direct_qlen) {
 			__skb_queue_tail(&q->direct_queue, skb);
+			if (q->direct_queue.qlen == 1)
+				sch->flags &= ~TCQ_F_THROTTLED;
 			q->direct_pkts++;
 		} else {
 			kfree_skb(skb);

@@ -637,6 +639,8 @@ static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 		cl->bstats.packets++;
 		cl->bstats.bytes += skb->len;
 		htb_activate(q, cl);
+		if (cl->un.leaf.q->q.qlen == 1)
+			sch->flags &= ~TCQ_F_THROTTLED;
 	}
 
 	sch->q.qlen++;

@@ -958,7 +962,6 @@ static struct sk_buff *htb_dequeue(struct Qdisc *sch)
 	/* try to dequeue direct packets as high prio (!) to minimize cpu work */
 	skb = __skb_dequeue(&q->direct_queue);
 	if (skb != NULL) {
-		sch->flags &= ~TCQ_F_THROTTLED;
 		sch->q.qlen--;
 		return skb;
 	}

@@ -991,7 +994,6 @@ static struct sk_buff *htb_dequeue(struct Qdisc *sch)
 			skb = htb_dequeue_tree(q, prio, level);
 			if (likely(skb != NULL)) {
 				sch->q.qlen--;
-				sch->flags &= ~TCQ_F_THROTTLED;
 				goto fin;
 			}
 		}
net/sched/sch_netem.c +0 -4
index 5d9d8bc..4c7a8d8 100644
@@ -273,10 +273,6 @@ static struct sk_buff *netem_dequeue(struct Qdisc *sch)
 	struct netem_sched_data *q = qdisc_priv(sch);
 	struct sk_buff *skb;
 
-	smp_mb();
-	if (sch->flags & TCQ_F_THROTTLED)
-		return NULL;
-
 	skb = q->qdisc->dequeue(q->qdisc);
 	if (skb) {
 		const struct netem_skb_cb *cb
net/sched/sch_tbf.c +0 -1
index 5386295..ed7e581 100644
@@ -218,7 +218,6 @@ static struct sk_buff *tbf_dequeue(struct Qdisc* sch)
 			q->tokens = toks;
 			q->ptokens = ptoks;
 			sch->q.qlen--;
-			sch->flags &= ~TCQ_F_THROTTLED;
 			return skb;
 		}
author
#14 lore raw

Re: [NET_SCHED 11/11]: qdisc: avoid dequeue while throttled

Quoted 10 lines · Patrick McHardy
Quoted 1 line

[NET_SCHED]: qdisc: avoid dequeue while throttled

It just occured to me that this doesn't work properly with qdiscs that have multiple classes since they all don't properly maintain the TCQ_F_THROTTLED flag. They set it on dequeue when no active class is willing to give out packets, but when enqueueing to a non-active class (thereby activating it) it is still set even though we don't know if that class could be dequeued.

Quoted 4 lines

So this updated patch unsets the TCQ_F_THROTTLED flag whenever we activate a class. Additionally it removes the unsetting of TCQ_F_THROTTLED on successful dequeue since we're now guaranteed that it was not set before.

I found another case that doesn't work properly, so let me retract this patch until I've properly thought this through.

maintainer
#15 lore raw

Re: [NET_SCHED 11/11]: qdisc: avoid dequeue while throttled

From: Patrick McHardy <[email protected]>
Date: Fri, 23 Mar 2007 15:57:08 +0100

Quoted 2 lines

I found another case that doesn't work properly, so let me retract this patch until I've properly thought this through.

Ok, I'll apply the rest.

maintainer
#16 lore raw
In reply to Patrick McHardy

From: Patrick McHardy <[email protected]>
Date: Fri, 23 Mar 2007 14:35:38 +0100 (MET)

Quoted 6 lines

These patches fix an off-by-one in netem, clean up pkt_sched.h by removing most of the now unnecessary PSCHED time macros and turning the two remaining ones into inline functions, consolidate some common filter destruction code and move the TCQ_F_THROTTLED optimization from netem to qdisc_restart.

Please apply, thanks.

Patches 1-10 applied.