From: Prarit Bhargava <prarit@redhat.com>
Date: Thu, 19 Mar 2009 11:01:26 -0400
Subject: Revert: [x86_64] fix gettimeoday TSC overflow issue
Message-id: 20090319150125.11188.81155.sendpatchset@prarit.bos.redhat.com
O-Subject: [RHEL5 PATCH]: REVERT Fix gettimeoday() TSC overflow issue
Bugzilla: 467942
RH-Acked-by: John W. Linville <linville@redhat.com>

After including this patch two issues were seen on systems.  The first involved
the bringing up of a wireless card that required firmware, and the second was
the shutdown of xen guests.

I've requested that dzickus pull this patch.

BZ 467942.

Verified that my Sony Vaio's wireless is functioning properly after removing
this patch.

diff --git a/arch/x86_64/kernel/time.c b/arch/x86_64/kernel/time.c
index 24eb09d..2d34020 100644
--- a/arch/x86_64/kernel/time.c
+++ b/arch/x86_64/kernel/time.c
@@ -75,7 +75,6 @@ static int notsc __initdata = 0;
 #define NSEC_PER_REAL_TICK (NSEC_PER_SEC / REAL_HZ)
 
 #define NS_SCALE	10 /* 2^10, carefully chosen */
-#define NS_SCALE_22	22 /* 2^22, carefully chosen */
 #define US_SCALE	32 /* 2^32, arbitralrily chosen */
 
 unsigned int cpu_khz;					/* TSC clocks / usec, not used here */
@@ -114,7 +113,7 @@ static inline long do_gettimeoffset_tsc(void)
 	t = get_cycles_sync();
 	if (t < vxtime.last_tsc) 
 		t = vxtime.last_tsc; /* hack */
-	x = ((t - vxtime.last_tsc) * vxtime.tsc_quot) >> NS_SCALE_22;
+	x = ((t - vxtime.last_tsc) * vxtime.tsc_quot) >> NS_SCALE;
 	return x;
 }
 
@@ -370,10 +369,6 @@ void main_timer_handler(struct pt_regs *regs)
 	unsigned long tsc;
 	int delay = 0, offset = 0, lost = 0, i;
 
-	/* for re-calculate offset */
-	long last_tsc_quot = vxtime.tsc_quot;
-	unsigned long last_tsc = vxtime.last_tsc;
-
 /*
  * Here we are in the timer irq handler. We have irqs locally disabled (so we
  * don't need spin_lock_irqsave()) but we don't know if the timer_bh is running
@@ -426,8 +421,7 @@ void main_timer_handler(struct pt_regs *regs)
 #endif
 	} else {
 		offset = (((tsc - vxtime.last_tsc) *
-			   vxtime.tsc_quot) >> NS_SCALE_22) -
-			 NSEC_PER_REAL_TICK;
+			   vxtime.tsc_quot) >> NS_SCALE) - NSEC_PER_REAL_TICK;
 
 		if (offset < 0)
 			offset = 0;
@@ -441,8 +435,12 @@ void main_timer_handler(struct pt_regs *regs)
 		/* FIXME: 1000 or 1000000? */
 		monotonic_base += (tsc - vxtime.last_tsc) * 1000000 / cpu_khz;
 
-		vxtime.last_tsc = tsc - vxtime.quot * delay /
-				  (vxtime.tsc_quot >> (NS_SCALE_22 - NS_SCALE));
+		vxtime.last_tsc = tsc - vxtime.quot * delay / vxtime.tsc_quot;
+
+		if ((((tsc - vxtime.last_tsc) *
+		      vxtime.tsc_quot) >> NS_SCALE) < offset)
+			vxtime.last_tsc = tsc -
+				(((long) offset << NS_SCALE) / vxtime.tsc_quot) - 1;
 	}
 	/* SCALE: We expect tick_divider - 1 lost, ie 0 for normal behaviour */
 	if (lost > (int)tick_divider - 1)  {
@@ -474,15 +472,6 @@ void main_timer_handler(struct pt_regs *regs)
 #endif
 	}
 
-	/* re-calculate vxtime.last_tsc */
-	if (vxtime.mode != VXTIME_HPET && vxtime.mode != VXTIME_PMTMR) {
-		if (tsc > (last_tsc + cpu_khz)) {
-			vxtime.last_tsc = vxtime.last_tsc -
-			                  (tsc - last_tsc -
-					   (cpu_khz * ( lost + 1 ))) *
-			                  last_tsc_quot / vxtime.tsc_quot;
-		}
-	}
 /*
  * If we have an externally synchronized Linux clock, then update CMOS clock
  * accordingly every ~11 minutes. set_rtc_mmss() will be called in the jiffy
@@ -518,12 +507,12 @@ static unsigned int cyc2ns_scale __read_mostly;
 
 static inline void set_cyc2ns_scale(unsigned long cpu_khz)
 {
-	cyc2ns_scale = (NSEC_PER_MSEC << NS_SCALE_22) / cpu_khz;
+	cyc2ns_scale = (NSEC_PER_MSEC << NS_SCALE) / cpu_khz;
 }
 
 static inline unsigned long long cycles_2_ns(unsigned long long cyc)
 {
-	return (cyc * cyc2ns_scale) >> NS_SCALE_22;
+	return (cyc * cyc2ns_scale) >> NS_SCALE;
 }
 
 unsigned long long sched_clock(void)
@@ -729,8 +718,7 @@ static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
 
 		tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
 		if (!(freq->flags & CPUFREQ_CONST_LOOPS))
-			vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE_22) /
-					  cpu_khz;
+			vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE) / cpu_khz;
 	}
 	
 	set_cyc2ns_scale(tsc_khz_ref);
@@ -1036,7 +1024,7 @@ void __init time_init(void)
 
 	vxtime.mode = VXTIME_TSC;
 	vxtime.quot = (NSEC_PER_SEC << NS_SCALE) / vxtime_hz;
-	vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE_22) / cpu_khz;
+	vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE) / cpu_khz;
 	vxtime.last_tsc = get_cycles_sync();
 	setup_irq(0, &irq0);
 
@@ -1124,7 +1112,7 @@ void time_init_gtod(void)
 	printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n", 
 		cpu_khz / 1000, cpu_khz % 1000);
 	vxtime.quot = (NSEC_PER_SEC << NS_SCALE) / vxtime_hz;
-	vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE_22) / cpu_khz;
+	vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE) / cpu_khz;
 	vxtime.last_tsc = get_cycles_sync();
 
 	set_cyc2ns_scale(cpu_khz);
diff --git a/arch/x86_64/kernel/vsyscall.c b/arch/x86_64/kernel/vsyscall.c
index 7928f57..69719d6 100644
--- a/arch/x86_64/kernel/vsyscall.c
+++ b/arch/x86_64/kernel/vsyscall.c
@@ -53,8 +53,6 @@ int __vgetcpu_mode __section_vgetcpu_mode;
 	  ((v - fix_to_virt(VSYSCALL_FIRST_PAGE)) + __pa_symbol(&__vsyscall_0)); })
 
 #define NS_SCALE	10 /* 2^10, carefully chosen */
-#define NS_SCALE_22	22 /* 2^22, carefully chosen */
-
 
 static __always_inline void timeval_normalize(struct timeval * tv)
 {
@@ -84,7 +82,7 @@ static __always_inline void do_vgettimeofday(struct timeval * tv)
 			if (t < __vxtime.last_tsc)
 				t = __vxtime.last_tsc;
 			nsec += ((t - __vxtime.last_tsc) *
-				 __vxtime.tsc_quot) >> NS_SCALE_22;
+				 __vxtime.tsc_quot) >> NS_SCALE;
 		} else {
 			nsec += ((readl((void *)fix_to_virt(VSYSCALL_HPET) +
 					0xf0) -
diff --git a/kernel/timer.c b/kernel/timer.c
index 44ef106..0e831b5 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -1294,15 +1294,6 @@ static void update_wall_time(void)
 	xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
 	clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
 
-#ifdef CONFIG_X86_64
-#define NS_SCALE_22	22 /* 2^22, carefully chosen */
-	/* re-calculate vxtime.tsc_quot */
-	vxtime.tsc_quot = (NSEC_PER_MSEC << NS_SCALE_22) / cpu_khz *
-                          (((s64)clock->xtime_interval +
-			    (s64)clock->xtime_nsec) >>
-			   clock->shift) / NSEC_PER_MSEC;
-#endif
-
 	/* check to see if there is a new clocksource to use */
 	if (change_clocksource()) {
 		clock->error = 0;