Index: linux-2.6.14-ck1/kernel/sched.c
===================================================================
--- linux-2.6.14-ck1.orig/kernel/sched.c	2005-10-28 20:51:30.000000000 +1000
+++ linux-2.6.14-ck1/kernel/sched.c	2005-10-28 20:51:30.000000000 +1000
@@ -122,6 +122,7 @@ struct runqueue {
 	unsigned long iso_ticks;
 	unsigned int iso_refractory;
 #ifdef CONFIG_SMP
+	unsigned long prio_bias;
 	unsigned long cpu_load[3];
 #endif
 	unsigned long long nr_switches;
@@ -563,13 +564,63 @@ static inline void enqueue_task_head(str
 	__set_bit(p->prio, rq->bitmap);
 }
 
+#ifdef CONFIG_SMP
+static inline void inc_prio_bias(runqueue_t *rq, int prio)
+{
+	rq->prio_bias += MAX_PRIO - prio;
+}
+
+static inline void dec_prio_bias(runqueue_t *rq, int prio)
+{
+	rq->prio_bias -= MAX_PRIO - prio;
+}
+
+static inline void inc_nr_running(task_t *p, runqueue_t *rq)
+{
+	rq->nr_running++;
+	if (rt_task(p)) {
+		if (p != rq->migration_thread)
+			inc_prio_bias(rq, p->prio);
+	} else
+		inc_prio_bias(rq, p->static_prio);
+}
+
+static inline void dec_nr_running(task_t *p, runqueue_t *rq)
+{
+	rq->nr_running--;
+	if (rt_task(p)) {
+		if (p != rq->migration_thread)
+			dec_prio_bias(rq, p->prio);
+	} else
+		dec_prio_bias(rq, p->static_prio);
+}
+#else
+static inline void inc_nr_running(task_t *p, runqueue_t *rq)
+{
+	rq->nr_running++;
+}
+
+static inline void dec_nr_running(task_t *p, runqueue_t *rq)
+{
+	rq->nr_running--;
+}
+
+static inline void inc_prio_bias(runqueue_t *rq, int prio)
+{
+}
+
+static inline void dec_prio_bias(runqueue_t *rq, int prio)
+{
+}
+#endif
+
 /*
  * __activate_task - move a task to the runqueue.
  */
 static void __activate_task(task_t *p, runqueue_t *rq)
 {
 	enqueue_task(p, rq);
-	rq->nr_running++;
+	inc_nr_running(p, rq);
 }
 
 /*
@@ -578,7 +629,7 @@ static void __activate_task(task_t *p, r
 static inline void __activate_idle_task(task_t *p, runqueue_t *rq)
 {
 	enqueue_task_head(p, rq);
-	rq->nr_running++;
+	inc_nr_running(p, rq);
 }
 
 /*
@@ -787,7 +838,7 @@ static void activate_task(task_t *p, run
  */
 static inline void deactivate_task(struct task_struct *p, runqueue_t *rq)
 {
-	rq->nr_running--;
+	dec_nr_running(p, rq);
 	dequeue_task(p, rq);
 }
 
@@ -923,27 +974,61 @@ void kick_process(task_t *p)
  * We want to under-estimate the load of migration sources, to
  * balance conservatively.
  */
-static inline unsigned long source_load(int cpu, int type)
+static inline unsigned long __source_load(int cpu, int type, enum idle_type idle)
 {
 	runqueue_t *rq = cpu_rq(cpu);
-	unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE;
+	unsigned long running = rq->nr_running;
+	unsigned long source_load, cpu_load = rq->cpu_load[type-1],
+		load_now = running * SCHED_LOAD_SCALE;
+
 	if (type == 0)
-		return load_now;
+		source_load = load_now;
+	else
+		source_load = min(cpu_load, load_now);
+
+	if (running > 1 || (idle == NOT_IDLE && running))
+		/*
+		 * If we are busy rebalancing the load is biased by
+		 * priority to create 'nice' support across cpus. When
+		 * idle rebalancing we should only bias the source_load if
+		 * there is more than one task running on that queue to
+		 * prevent idle rebalance from trying to pull tasks from a
+		 * queue with only one running task.
+		 */
+		source_load = source_load * rq->prio_bias / running;
+
+	return source_load;
+}
 
-	return min(rq->cpu_load[type-1], load_now);
+static inline unsigned long source_load(int cpu, int type)
+{
+	return __source_load(cpu, type, NOT_IDLE);
 }
 
 /*
  * Return a high guess at the load of a migration-target cpu
  */
-static inline unsigned long target_load(int cpu, int type)
+static inline unsigned long __target_load(int cpu, int type, enum idle_type idle)
 {
 	runqueue_t *rq = cpu_rq(cpu);
-	unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE;
+	unsigned long running = rq->nr_running;
+	unsigned long target_load, cpu_load = rq->cpu_load[type-1],
+		load_now = running * SCHED_LOAD_SCALE;
+
 	if (type == 0)
-		return load_now;
+		target_load = load_now;
+	else
+		target_load = max(cpu_load, load_now);
+
+	if (running > 1 || (idle == NOT_IDLE && running))
+		target_load = target_load * rq->prio_bias / running;
 
-	return max(rq->cpu_load[type-1], load_now);
+	return target_load;
+}
+
+static inline unsigned long target_load(int cpu, int type)
+{
+	return __target_load(cpu, type, NOT_IDLE);
 }
 
 /*
@@ -1684,9 +1769,9 @@ static inline void pull_task(runqueue_t 
 		runqueue_t *this_rq, int this_cpu)
 {
 	dequeue_task(p, src_rq);
-	src_rq->nr_running--;
+	dec_nr_running(p, src_rq);
 	set_task_cpu(p, this_cpu);
-	this_rq->nr_running++;
+	inc_nr_running(p, this_rq);
 	enqueue_task(p, this_rq);
 	p->timestamp = (p->timestamp - src_rq->timestamp_last_tick)
 				+ this_rq->timestamp_last_tick;
@@ -1840,9 +1925,9 @@ find_busiest_group(struct sched_domain *
 
 			/* Bias balancing toward cpus of our domain */
 			if (local_group)
-				load = target_load(i, load_idx);
+				load = __target_load(i, load_idx, idle);
 			else
-				load = source_load(i, load_idx);
+				load = __source_load(i, load_idx, idle);
 
 			avg_load += load;
 		}
@@ -1947,14 +2032,15 @@ out_balanced:
 /*
  * find_busiest_queue - find the busiest runqueue among the cpus in group.
  */
-static runqueue_t *find_busiest_queue(struct sched_group *group)
+static runqueue_t *find_busiest_queue(struct sched_group *group,
+	enum idle_type idle)
 {
 	unsigned long load, max_load = 0;
 	runqueue_t *busiest = NULL;
 	int i;
 
 	for_each_cpu_mask(i, group->cpumask) {
-		load = source_load(i, 0);
+		load = __source_load(i, 0, idle);
 
 		if (load > max_load) {
 			max_load = load;
@@ -1998,7 +2084,7 @@ static inline int load_balance(int this_
 		goto out_balanced;
 	}
 
-	busiest = find_busiest_queue(group);
+	busiest = find_busiest_queue(group, idle);
 	if (!busiest) {
 		schedstat_inc(sd, lb_nobusyq[idle]);
 		goto out_balanced;
@@ -2121,7 +2207,7 @@ static inline int load_balance_newidle(i
 		goto out_balanced;
 	}
 
-	busiest = find_busiest_queue(group);
+	busiest = find_busiest_queue(group, NEWLY_IDLE);
 	if (!busiest) {
 		schedstat_inc(sd, lb_nobusyq[NEWLY_IDLE]);
 		goto out_balanced;
@@ -3335,8 +3421,10 @@ void set_user_nice(task_t *p, long nice)
 		p->static_prio = NICE_TO_PRIO(nice);
 		goto out_unlock;
 	}
-	if ((queued = task_queued(p)))
+	if ((queued = task_queued(p))) {
 		dequeue_task(p, rq);
+		dec_prio_bias(rq, p->static_prio);
+	}
 
 	old_prio = p->prio;
 	new_prio = NICE_TO_PRIO(nice);
@@ -3346,6 +3434,7 @@ void set_user_nice(task_t *p, long nice)
 
 	if (queued) {
 		enqueue_task(p, rq);
+		inc_prio_bias(rq, p->static_prio);
 		/*
 		 * If the task increased its priority or is running and
 		 * lowered its priority, then reschedule its CPU: