From: Chris Lalancette <clalance@redhat.com>
Date: Tue, 20 Nov 2007 15:34:22 -0500
Subject: [misc] fix softlockup warnings/crashes
Message-id: 474344CE.5020109@redhat.com
O-Subject: [RHEL5.2 PATCH]: Fix softlockup warnings/crashes
Bugzilla: 250994

All,
     Attached is a patch to fix BZ 250994.  The original reported problem is
that after a save/restore, RHEL-5 PV guests would spit out "softlockup" warnings
on the console.  However, it has been determined in a number of other cases that
the softlockup errors can cause more serious problems, especially on dom0.  Note
that this patch only affects Xen and s390, since these are the only
architectures that support NO_IDLE_HZ in RHEL-5.  What this patch does is
basically ensure that before we go idle, we make sure that the idle time is
never greater than the next softlockup event.

This patch is upstream (and has been since June) here:
http://xenbits.xensource.com/linux-2.6.18-xen.hg?rev/61ed8662b69c

Successfully tested by me on save/restore loops, and also tested by a hardware
vendor running regression tests and a customer running Xen guests on top of GFS.

Please review and ACK.

Thanks,
Chris Lalancette

Acked-by: Jarod Wilson <jwilson@redhat.com>

Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: "Stephen C. Tweedie" <sct@redhat.com>
Acked-by: Bill Burns <bburns@redhat.com>

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 338ae7c..806ff8b 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -214,12 +214,17 @@ extern void update_process_times(int user, struct pt_regs *regs);
 extern void scheduler_tick(void);
 
 #ifdef CONFIG_DETECT_SOFTLOCKUP
+extern unsigned long softlockup_get_next_event(void);
 extern void softlockup_tick(struct pt_regs *regs);
 extern void spawn_softlockup_task(void);
 extern void touch_softlockup_watchdog(void);
 extern void touch_all_softlockup_watchdogs(void);
 extern int softlockup_thresh;
 #else
+static inline unsigned long softlockup_get_next_event(void)
+{
+	return MAX_JIFFY_OFFSET;
+}
 static inline void softlockup_tick(struct pt_regs *regs)
 {
 }
diff --git a/kernel/softlockup.c b/kernel/softlockup.c
index e35a03b..c9aad28 100644
--- a/kernel/softlockup.c
+++ b/kernel/softlockup.c
@@ -42,6 +42,19 @@ void touch_softlockup_watchdog(void)
 }
 EXPORT_SYMBOL(touch_softlockup_watchdog);
 
+unsigned long softlockup_get_next_event(void)
+{
+	int this_cpu = smp_processor_id();
+	unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
+
+	if (per_cpu(print_timestamp, this_cpu) == touch_timestamp ||
+		did_panic ||
+			!per_cpu(watchdog_task, this_cpu))
+		return MAX_JIFFY_OFFSET;
+
+	return max_t(long, 0, touch_timestamp + HZ - jiffies);
+}
+
 void touch_all_softlockup_watchdogs(void)
 {
 	int cpu;
diff --git a/kernel/timer.c b/kernel/timer.c
index a8d3320..35a40dc 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -616,7 +616,9 @@ unsigned long next_timer_interrupt(void)
 		if (hr_expires < 3)
 			return hr_expires + jiffies;
 	}
-	hr_expires += jiffies;
+	hr_expires = min_t(unsigned long,
+			   softlockup_get_next_event(),
+			   hr_expires) + jiffies;
 
 	base = __get_cpu_var(tvec_bases);
 	spin_lock(&base->lock);