--- linux-2.6.0-test1.virgin/kernel/sched.c.org	Sat Jul 19 09:42:12 2003
+++ linux-2.6.0-test1.virgin/kernel/sched.c	Mon Jul 21 14:29:26 2003
@@ -76,6 +76,12 @@
 #define MAX_SLEEP_AVG		(10*HZ)
 #define STARVATION_LIMIT	(10*HZ)
 #define NODE_THRESHOLD		125
+#define END_IA_PRIO		(NICE_TO_PRIO(1 - INTERACTIVE_DELTA))
+#define INTERVAL		(HZ)
+#define DURATION_PERCENT	10
+#define DURATION		(INTERVAL * DURATION_PERCENT / 100)
+#define INTERVAL_EXPIRED(rq) (time_after(jiffies, \
+	(rq)->interval_ts + ((rq)->idx ? DURATION : INTERVAL)))
 
 /*
  * If a task is 'interactive' then we reinsert it in the active
@@ -158,7 +164,7 @@
 struct runqueue {
 	spinlock_t lock;
 	unsigned long nr_running, nr_switches, expired_timestamp,
-			nr_uninterruptible;
+			nr_uninterruptible, interval_ts;
 	task_t *curr, *idle;
 	struct mm_struct *prev_mm;
 	prio_array_t *active, *expired, arrays[2];
@@ -171,6 +177,7 @@
 	struct list_head migration_queue;
 
 	atomic_t nr_iowait;
+	int idx;
 };
 
 static DEFINE_PER_CPU(struct runqueue, runqueues);
@@ -1166,6 +1173,23 @@
 			STARVATION_LIMIT * ((rq)->nr_running) + 1)))
 
 /*
+ * Must be called with the runqueue lock held.
+ */
+static void __requeue_one_expired(runqueue_t *rq)
+{
+	int idx = sched_find_first_bit(rq->expired->bitmap);
+	struct list_head *queue;
+	task_t *p;
+
+	if (idx >= MAX_PRIO)
+		return;
+	queue = rq->expired->queue + idx;
+	p = list_entry(queue->next, task_t, run_list);
+	dequeue_task(p, p->array);
+	enqueue_task(p, rq->active);
+}
+
+/*
  * This function gets called by the timer code, with HZ frequency.
  * We call it with interrupts disabled.
  *
@@ -1242,10 +1266,35 @@
 			if (!rq->expired_timestamp)
 				rq->expired_timestamp = jiffies;
 			enqueue_task(p, rq->expired);
-		} else
+		} else {
 			enqueue_task(p, rq->active);
+			if (rq->idx)
+				__requeue_one_expired(rq);
+		}
+	} else if (INTERVAL_EXPIRED(rq)) {
+		/*
+		 * If we haven't run a non-interactive task within our
+		 * interval, we take this as a hint that we may have
+		 * starvation in progress.  Trigger a queue walk-down,
+		 * and walk until either non-interactive tasks have
+		 * received DURATION ticks, or we hit the bottom.
+		 */
+		if (TASK_INTERACTIVE(p)) {
+			prio_array_t *array = p->array;
+
+			/* Requeue the hight priority expired task... */
+				__requeue_one_expired(rq);
+			/* and tell the scheduler where to start walking. */
+			rq->idx = find_next_bit(array->bitmap, MAX_PRIO, 1 + p->prio);
+			if (rq->idx >= MAX_PRIO)
+				rq->idx = 0;
+		} else
+			rq->idx = 0;
+		rq->interval_ts = jiffies;
 	}
 out_unlock:
+	if (rq->idx && TASK_INTERACTIVE(p))
+		rq->interval_ts++;
 	spin_unlock(&rq->lock);
 out:
 	rebalance_tick(rq, 0);
@@ -1312,6 +1361,8 @@
 #endif
 		next = rq->idle;
 		rq->expired_timestamp = 0;
+		rq->interval_ts = jiffies;
+		rq->idx = 0;
 		goto switch_tasks;
 	}
 
@@ -1329,6 +1380,17 @@
 	idx = sched_find_first_bit(array->bitmap);
 	queue = array->queue + idx;
 	next = list_entry(queue->next, task_t, run_list);
+	if (unlikely(rq->idx)) {
+		if(!rt_task(next)) {
+			int index = find_next_bit(array->bitmap, MAX_PRIO, rq->idx);
+			if (index < MAX_PRIO) {
+				queue = array->queue + index;
+				next = list_entry(queue->next, task_t, run_list);
+				if (index < END_IA_PRIO)
+					rq->idx = index + 1;
+			}
+		}
+	}
 
 switch_tasks:
 	prefetch(next);
@@ -2497,6 +2559,7 @@
 		rq = cpu_rq(i);
 		rq->active = rq->arrays;
 		rq->expired = rq->arrays + 1;
+		rq->interval_ts = INITIAL_JIFFIES;
 		spin_lock_init(&rq->lock);
 		INIT_LIST_HEAD(&rq->migration_queue);
 		atomic_set(&rq->nr_iowait, 0);