From: Chris Lalancette <clalance@redhat.com>
Date: Thu, 23 Apr 2009 14:17:27 +0200
Subject: [x86] fix tick divider with clocksource=pit
Message-id: 49F05C57.8060806@redhat.com
O-Subject: [RHEL5.4 PATCH]: Fix tick divider with clocksource=pit
Bugzilla: 427588
RH-Acked-by: Prarit Bhargava <prarit@redhat.com>
RH-Acked-by: Rik van Riel <riel@redhat.com>
RH-Acked-by: Don Dutile <ddutile@redhat.com>
RH-Acked-by: Justin M. Forbes <jforbes@redhat.com>

All,
     The attached patch addresses BZ 427588.  Currently, if you use the tick
divider (say, divider=10), and also specify clocksource=pit, the resulting boot
will be unstable and have serious clock drift problems.
     The problem is a mis-match in the way the PIT code is calculating the
counter value.  It first reads the value out of the PIT, and then, in order to
detect PIT underflows, checks to see if this value is greater than the last read
value.  However, when using the tick divider, this check will erroneously fire,
causing time to "jump" ahead sporadically.  The fix is to change the check so
that it only fires when it is really supposed to.  A nice feature of this change
is that in the non-divided case, the check is exactly the same.

Tested by me to resolve the specific scenario mentioned in the bugzilla.
Additionally, this patch has been in the virttest kernels for some time, and has
been tested on dom0 Xen kernels, paravirt Xen kernels, fully virtualized Xen
guests, and bare-metal i386 kernels without any problems observed.

Please review and ACK.

--
Chris Lalancette

diff --git a/arch/i386/kernel/i8253.c b/arch/i386/kernel/i8253.c
index 631422e..1bb1406 100644
--- a/arch/i386/kernel/i8253.c
+++ b/arch/i386/kernel/i8253.c
@@ -86,7 +86,7 @@ static cycle_t pit_read(void)
 	 * Previous attempts to handle these cases intelligently were
 	 * buggy, so we just do the simple thing now.
 	 */
-	if (count > old_count && jifs == old_jifs) {
+	if (count > old_count && (jifs - old_jifs) < tick_divider) {
 		count = old_count;
 	}
 	old_count = count;