pkt_sched: sch_htb: Consider used jiffies in htb_do_events()
authorJarek Poplawski <jarkao2@gmail.com>
Tue, 13 Jan 2009 05:54:16 +0000 (21:54 -0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 13 Jan 2009 05:54:16 +0000 (21:54 -0800)
Next event time should consider jiffies used for recounting. Otherwise
qdisc_watchdog_schedule() triggers hrtimer immediately with the event
in the past, and may cause very high ksoftirqd cpu usage (if highres
is on).

There is also removed checking "event" for zero in htb_dequeue(): it's
always true in this place.

Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/sch_htb.c

index 5070643ce534b468a85b3df2f1666d0ae35b08ad..9ca8a26ba507621f490702b0478d8656c9483ddb 100644 (file)
@@ -685,8 +685,8 @@ static psched_time_t htb_do_events(struct htb_sched *q, int level)
                if (cl->cmode != HTB_CAN_SEND)
                        htb_add_to_wait_tree(q, cl, diff);
        }
-       /* too much load - let's continue on next jiffie */
-       return q->now + PSCHED_TICKS_PER_SEC / HZ;
+       /* too much load - let's continue on next jiffie (including above) */
+       return q->now + 2 * PSCHED_TICKS_PER_SEC / HZ;
 }
 
 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
@@ -873,7 +873,7 @@ static struct sk_buff *htb_dequeue(struct Qdisc *sch)
                } else
                        event = q->near_ev_cache[level];
 
-               if (event && next_event > event)
+               if (next_event > event)
                        next_event = event;
 
                m = ~q->row_mask[level];