From: Geoff Gustafson <grgustaf@redhat.com>
Date: Fri, 28 Sep 2007 17:22:35 -0400
Subject: [x86_64] fix race conditions in setup_APIC_timer
Message-id: 20070928212235.GA8985@samurai.boston.redhat.com
O-Subject: [RHEL5.1 PATCH] fix race conditions in setup_APIC_timer (bz 251869)
Bugzilla: 251869

This patch upstream fixes race conditions in setup_APIC_timer, and fixed system hangs for the submitter.

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=b291aa7a6564e859af144e1bd14ffa463519b198

Intel has run into the same problem on a repeated reboot test, and submitted BZ 251869. A slight variant of the attached patch was appeared to fix the problem for Intel. The Intel requester would like this considered for 5.1 errata.

The upstream submitter wasn't totally sure about the patch, but it's been in Linus' tree for almost two months now. It looks sensible to me.

- Geoff

diff --git a/arch/x86_64/kernel/apic.c b/arch/x86_64/kernel/apic.c
index 32cf9f7..ca13cff 100644
--- a/arch/x86_64/kernel/apic.c
+++ b/arch/x86_64/kernel/apic.c
@@ -728,10 +728,12 @@ static void setup_APIC_timer(unsigned int clocks)
 
 	/* wait for irq slice */
  	if (vxtime.hpet_address && hpet_use_timer) {
- 		int trigger = hpet_readl(HPET_T0_CMP);
- 		while (hpet_readl(HPET_COUNTER) >= trigger)
- 			/* do nothing */ ;
- 		while (hpet_readl(HPET_COUNTER) <  trigger)
+		/*
+		 * Wait for the comparator value to change which signals that
+		 * the tick slice has expired.
+		 */
+		u32 trigger = hpet_readl(HPET_T0_CMP);
+		while (hpet_readl(HPET_T0_CMP) == trigger)
  			/* do nothing */ ;
  	} else {
 		int c1, c2;