diff -urpN numaq-2.5.45/drivers/block/ll_rw_blk.c iowait-2.5.45/drivers/block/ll_rw_blk.c
--- numaq-2.5.45/drivers/block/ll_rw_blk.c	2002-11-01 00:53:40.000000000 -0800
+++ iowait-2.5.45/drivers/block/ll_rw_blk.c	2002-11-01 01:29:00.000000000 -0800
@@ -56,7 +56,6 @@ static int queue_nr_requests;
 static int batch_requests;
 
 unsigned long blk_max_low_pfn, blk_max_pfn;
-atomic_t nr_iowait_tasks = ATOMIC_INIT(0);
 int blk_nohighio = 0;
 
 static struct congestion_state {
@@ -115,27 +114,6 @@ static void set_queue_congested(request_
 		atomic_inc(&congestion_states[rw].nr_congested_queues);
 }
 
-/*
- * This task is about to go to sleep on IO.  Increment nr_iowait_tasks so
- * that process accounting knows that this is a task in IO wait state.
- *
- * But don't do that if it is a deliberate, throttling IO wait (this task
- * has set its backing_dev_info: the queue against which it should throttle)
- */
-void io_schedule(void)
-{
-	atomic_inc(&nr_iowait_tasks);
-	schedule();
-	atomic_dec(&nr_iowait_tasks);
-}
-
-void io_schedule_timeout(long timeout)
-{
-	atomic_inc(&nr_iowait_tasks);
-	schedule_timeout(timeout);
-	atomic_dec(&nr_iowait_tasks);
-}
-
 /**
  * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
  * @dev:	device
diff -urpN numaq-2.5.45/fs/proc/proc_misc.c iowait-2.5.45/fs/proc/proc_misc.c
--- numaq-2.5.45/fs/proc/proc_misc.c	2002-11-01 00:53:40.000000000 -0800
+++ iowait-2.5.45/fs/proc/proc_misc.c	2002-11-01 01:28:45.000000000 -0800
@@ -424,7 +424,7 @@ static int kstat_read_proc(char *page, c
 		xtime.tv_sec - jif / HZ,
 		total_forks,
 		nr_running(),
-		atomic_read(&nr_iowait_tasks));
+		nr_iowait());
 
 	return proc_calc_metrics(page, start, off, count, eof, len);
 }
diff -urpN numaq-2.5.45/include/linux/blkdev.h iowait-2.5.45/include/linux/blkdev.h
--- numaq-2.5.45/include/linux/blkdev.h	2002-11-01 00:53:40.000000000 -0800
+++ iowait-2.5.45/include/linux/blkdev.h	2002-11-01 01:40:57.000000000 -0800
@@ -467,9 +467,4 @@ static inline void put_dev_sector(Sector
 #endif 
  
 
-
-extern atomic_t nr_iowait_tasks;
-void io_schedule(void);
-void io_schedule_timeout(long timeout);
-
 #endif
diff -urpN numaq-2.5.45/include/linux/sched.h iowait-2.5.45/include/linux/sched.h
--- numaq-2.5.45/include/linux/sched.h	2002-11-01 00:53:41.000000000 -0800
+++ iowait-2.5.45/include/linux/sched.h	2002-11-01 01:41:12.000000000 -0800
@@ -89,6 +89,7 @@ extern int nr_threads;
 extern int last_pid;
 extern unsigned long nr_running(void);
 extern unsigned long nr_uninterruptible(void);
+extern unsigned long nr_iowait(void);
 
 #include <linux/time.h>
 #include <linux/param.h>
@@ -161,6 +162,8 @@ extern unsigned long cache_decay_ticks;
 #define	MAX_SCHEDULE_TIMEOUT	LONG_MAX
 extern signed long FASTCALL(schedule_timeout(signed long timeout));
 asmlinkage void schedule(void);
+void io_schedule(void);
+void io_schedule_timeout(long timeout);
 
 struct namespace;
 
diff -urpN numaq-2.5.45/kernel/sched.c iowait-2.5.45/kernel/sched.c
--- numaq-2.5.45/kernel/sched.c	2002-11-01 00:53:41.000000000 -0800
+++ iowait-2.5.45/kernel/sched.c	2002-11-01 01:36:14.000000000 -0800
@@ -157,6 +157,7 @@ struct runqueue {
 	task_t *migration_thread;
 	struct list_head migration_queue;
 
+	atomic_t nr_iowait;
 } ____cacheline_aligned;
 
 static struct runqueue runqueues[NR_CPUS] __cacheline_aligned;
@@ -573,6 +574,16 @@ unsigned long nr_context_switches(void)
 	return sum;
 }
 
+unsigned long nr_iowait(void)
+{
+	unsigned long i, sum = 0;
+
+	for (i = 0; i < NR_CPUS; ++i)
+		sum += atomic_read(&cpu_rq(i)->nr_iowait);
+
+	return sum;
+}
+
 /*
  * double_rq_lock - safely lock two runqueues
  *
@@ -875,7 +886,7 @@ void scheduler_tick(int user_ticks, int 
 		/* note: this timer irq context must be accounted for as well */
 		if (irq_count() - HARDIRQ_OFFSET >= SOFTIRQ_OFFSET)
 			kstat_cpu(cpu).cpustat.system += sys_ticks;
-		else if (atomic_read(&nr_iowait_tasks) > 0)
+		else if (atomic_read(&rq->nr_iowait) > 0)
 			kstat_cpu(cpu).cpustat.iowait += sys_ticks;
 		else
 			kstat_cpu(cpu).cpustat.idle += sys_ticks;
@@ -1712,6 +1723,35 @@ void yield(void)
 	sys_sched_yield();
 }
 
+/*
+ * This task is about to go to sleep on IO.  Increment rq->nr_iowait so
+ * that process accounting knows that this is a task in IO wait state.
+ *
+ * But don't do that if it is a deliberate, throttling IO wait (this task
+ * has set its backing_dev_info: the queue against which it should throttle)
+ */
+void io_schedule(void)
+{
+	struct runqueue *rq;
+	preempt_disable();
+	rq = this_rq();
+	atomic_inc(&rq->nr_iowait);
+	schedule();
+	atomic_dec(&rq->nr_iowait);
+	preempt_enable();
+}
+
+void io_schedule_timeout(long timeout)
+{
+	struct runqueue *rq;
+	preempt_disable();
+	rq = this_rq();
+	atomic_inc(&rq->nr_iowait);
+	schedule_timeout(timeout);
+	atomic_dec(&rq->nr_iowait);
+	preempt_enable();
+}
+
 /**
  * sys_sched_get_priority_max - return maximum RT priority.
  * @policy: scheduling class.
@@ -2160,6 +2200,7 @@ void __init sched_init(void)
 		rq->expired = rq->arrays + 1;
 		spin_lock_init(&rq->lock);
 		INIT_LIST_HEAD(&rq->migration_queue);
+		atomic_set(&rq->nr_iowait, 0);
 
 		for (j = 0; j < 2; j++) {
 			array = rq->arrays + j;