Index: 2.6.12/include/linux/sched_cpustats.h
===================================================================
--- 2.6.12.orig/include/linux/sched_cpustats.h	2005-07-22 09:09:17.000000000 -0500
+++ 2.6.12/include/linux/sched_cpustats.h	2005-07-23 11:34:47.000000000 -0500
@@ -25,6 +25,7 @@
 	unsigned long long total_rt_intr_delay;
 	unsigned long long total_fork_delay;
 	unsigned long long total_sinbin;
+	unsigned long long total_latency;
 };
 
 extern DEFINE_PER_CPU(struct runq_cpustats, cpustats_runqs);
@@ -41,6 +42,7 @@
 	unsigned long long avg_cpu_per_cycle;
 	unsigned long long total_delay;
 	unsigned long long avg_delay_per_cycle;
+	unsigned long long total_latency;
 	unsigned long long total_sinbin;
 	unsigned long long avg_cycle_length;
 	unsigned long cpu_usage_rate;
@@ -48,7 +50,8 @@
 };
 
 #define CPUSTATS_WOKEN_FOR_INTR_FL (1 << 0)
-#define CPUSTATS_JUST_FORKED_FL (1 << 1)
+#define CPUSTATS_JUST_WOKEN_FL (1 << 1)
+#define CPUSTATS_JUST_FORKED_FL (1 << 2)
 
 #define INIT_CPUSTATS \
 	.cpustats = { 0, }, \
@@ -99,6 +102,7 @@
 	unsigned long long total_intr_delay;
 	unsigned long long total_rt_intr_delay;
 	unsigned long long total_fork_delay;
+	unsigned long long total_latency;
 	unsigned long long total_sinbin;
 	unsigned long long nr_switches;
 };
Index: 2.6.12/kernel/sched_cpustats.c
===================================================================
--- 2.6.12.orig/kernel/sched_cpustats.c	2005-07-22 09:09:17.000000000 -0500
+++ 2.6.12/kernel/sched_cpustats.c	2005-07-23 11:34:47.000000000 -0500
@@ -38,6 +38,7 @@
 	csrq->total_intr_delay = 0;
 	csrq->total_rt_intr_delay = 0;
 	csrq->total_fork_delay = 0;
+	csrq->total_latency = 0;
 	cpu_rq(cpu)->timestamp_last_tick = INITIAL_CPUSTATS_TIMESTAMP;
 }
 
@@ -59,6 +60,7 @@
 	TASK_CPUSTATS(p).total_cpu = 0;
 	TASK_CPUSTATS(p).total_wake_ups = 0;
 	TASK_CPUSTATS(p).intr_wake_ups = 0;
+	TASK_CPUSTATS(p).total_latency = 0;
 	TASK_CPUSTATS(p).avg_cycle_length = 0;
 	p->timestamp = now;
 	TASK_CPUSTATS(p).flags = CPUSTATS_JUST_FORKED_FL;
@@ -123,9 +125,13 @@
 		rq_stats->total_fork_delay += delta;
 		TASK_CPUSTATS(p).flags &= ~CPUSTATS_JUST_FORKED_FL;
 	}
-	if (TASK_CPUSTATS(p).flags & CPUSTATS_WOKEN_FOR_INTR_FL) {
-		rq_stats->total_intr_delay += delta;
-		TASK_CPUSTATS(p).flags &= ~CPUSTATS_WOKEN_FOR_INTR_FL;
+	if (TASK_CPUSTATS(p).flags & CPUSTATS_JUST_WOKEN_FL) {
+		rq_stats->total_latency += delta;
+		TASK_CPUSTATS(p).total_latency += delta;
+		if (TASK_CPUSTATS(p).flags & CPUSTATS_WOKEN_FOR_INTR_FL)
+			rq_stats->total_intr_delay += delta;
+		TASK_CPUSTATS(p).flags &=
+			 ~(CPUSTATS_WOKEN_FOR_INTR_FL|CPUSTATS_JUST_WOKEN_FL);
 	}
 }
 
@@ -154,6 +160,7 @@
 void update_cpustats_at_wake_up(struct task_struct *p, unsigned long long now)
 {
 	delta_sleep_cpustats(p, now);
+	TASK_CPUSTATS(p).flags |= CPUSTATS_JUST_WOKEN_FL;
 	if (in_interrupt()) {
 		TASK_CPUSTATS(p).intr_wake_ups++;
 		TASK_CPUSTATS(p).flags |= CPUSTATS_WOKEN_FOR_INTR_FL;
@@ -183,13 +190,14 @@
 	if (result)
 		return sprintf(buffer, "Data unavailable\n");
 	return sprintf(buffer,
-		"%llu (%llu) %llu (%llu) %llu (%llu) %llu %llu %llu %lu %lu @ %llu\n",
+		"%llu (%llu) %llu (%llu) %llu (%llu) %llu %llu %llu %llu %lu %lu @ %llu\n",
 		stats.total_sleep,
 		SCHED_AVG_RND(stats.avg_sleep_per_cycle),
 		stats.total_cpu,
 		SCHED_AVG_RND(stats.avg_cpu_per_cycle),
 		stats.total_delay,
 		SCHED_AVG_RND(stats.avg_delay_per_cycle),
+		stats.total_latency,
 		stats.total_sinbin,
 		stats.total_wake_ups,
 		stats.intr_wake_ups,
@@ -214,13 +222,14 @@
 			break;
 		}
 		len += sprintf(page + len,
-		    "cpu%02d %llu %llu %llu %llu %llu %llu %llu %llu @ %llu\n", i,
+		    "cpu%02d %llu %llu %llu %llu %llu %llu %llu %llu %llu @ %llu\n", i,
 		stats.total_idle,
 		stats.total_busy,
 		stats.total_delay,
 		stats.total_rt_delay,
 		stats.total_intr_delay,
 		stats.total_rt_intr_delay,
+		stats.total_latency,
 		stats.total_sinbin,
 		stats.nr_switches,
 		stats.timestamp);
@@ -230,17 +239,19 @@
 		total.total_rt_delay += stats.total_rt_delay;
 		total.total_intr_delay += stats.total_intr_delay;
 		total.total_rt_intr_delay += stats.total_rt_intr_delay;
+		total.total_latency += stats.total_latency;
 		total.total_sinbin += stats.total_sinbin;
 		total.nr_switches += stats.nr_switches;
 	}
 	if (avail)
-		len += sprintf(page + len, "total %llu %llu %llu %llu %llu %llu %llu %llu\n",
+		len += sprintf(page + len, "total %llu %llu %llu %llu %llu %llu %llu %llu %llu\n",
 			total.total_idle,
 			total.total_busy,
 			total.total_delay,
 			total.total_intr_delay,
 			total.total_rt_delay,
 			total.total_rt_intr_delay,
+			total.total_latency,
 			total.total_sinbin,
 			total.nr_switches);
 	else
Index: 2.6.12/kernel/sched_spa.c
===================================================================
--- 2.6.12.orig/kernel/sched_spa.c	2005-07-23 11:34:17.000000000 -0500
+++ 2.6.12/kernel/sched_spa.c	2005-07-25 15:23:58.000000000 -0500
@@ -522,9 +522,10 @@
 		return;
 
 	get_task_cpustats(p, &stats, &timestamp);
-	printk("SCHED_EXIT[%d] (%s) %llu %llu %llu %llu %llu %llu %lu %lu @ %llu\n",
+	printk("SCHED_EXIT[%d] (%s) %llu %llu %llu %llu %llu %llu %llu %lu %lu @ %llu\n",
 		p->pid, p->comm,
 		stats.total_sleep, stats.total_cpu, stats.total_delay,
+		stats.total_latency,
 		stats.total_sinbin, stats.total_wake_ups, stats.intr_wake_ups,
 		p->nvcsw, p->nivcsw, timestamp);
 }
@@ -1153,6 +1154,8 @@
 			stats->avg_delay_per_cycle += delta;
 			if (is_sinbinned)
 				stats->total_sinbin += delta;
+			if (stats->flags & CPUSTATS_JUST_WOKEN_FL)
+				stats->total_latency += delta;
 		} else {
 			stats->total_sleep += delta;
 			stats->avg_sleep_per_cycle += delta;
@@ -1206,6 +1209,7 @@
 	stats->total_rt_intr_delay = csrq->total_rt_intr_delay;
 	stats->total_fork_delay = csrq->total_fork_delay;
 	stats->total_sinbin = csrq->total_sinbin;
+	stats->total_latency = csrq->total_latency;
 	stats->nr_switches = rq->nr_switches;
 	spin_unlock_irq(&rq->lock);
 
@@ -1436,10 +1440,11 @@
 		total.total_rt_intr_delay += stats.total_rt_intr_delay;
 		total.total_fork_delay += stats.total_fork_delay;
 		total.total_sinbin += stats.total_sinbin;
+		total.total_latency += stats.total_latency;
 		total.nr_switches += stats.nr_switches;
 	}
 	if (avail)
-		len = sprintf(page, "%llu %llu %llu %llu %llu %llu %llu %llu %llu @ %llu\n",
+		len = sprintf(page, "%llu %llu %llu %llu %llu %llu %llu %llu %llu %llu @ %llu\n",
 			total.total_idle,
 			total.total_busy,
 			total.total_delay,
@@ -1447,6 +1452,7 @@
 			total.total_rt_delay,
 			total.total_rt_intr_delay,
 			total.total_fork_delay,
+			total.total_latency,
 			total.total_sinbin,
 			total.nr_switches,
 			timestamp);