Index: sys/x86/isa/clock.c
===================================================================
--- sys/x86/isa/clock.c	(revision 219697)
+++ sys/x86/isa/clock.c	(working copy)
@@ -62,6 +62,7 @@ __FBSDID("$FreeBSD$");
 
 #include <machine/clock.h>
 #include <machine/cpu.h>
+#include <machine/cputypes.h>
 #include <machine/intr_machdep.h>
 #include <machine/ppireg.h>
 #include <machine/timerreg.h>
@@ -245,40 +246,85 @@ getit(void)
 	return ((high << 8) | low);
 }
 
-static __inline void
-delay_tsc(int n)
+#ifdef __i386__
+static __inline uint64_t
+atomic_fetch_quad_i386(uint64_t *p)
 {
-	uint64_t start, end, now;
+	uint64_t v;
 
-	sched_pin();
-	start = rdtsc();
-	end = start + (tsc_freq * n) / 1000000;
-	do {
-		cpu_spinwait();
-		now = rdtsc();
-	} while (now < end || (now > start && end < start));
-	sched_unpin();
+	/* i486 does not support SMP. */
+	__asm __volatile(
+	"	pushfl ;			"
+	"	cli ;				"
+	"	movl (%1), %%eax ;		"
+	"	movl 4(%1), %%edx ;		"
+	"	popfl"
+	: "=&A" (v) : "r" (p));
+	return (v);
 }
 
-static __inline void
-delay_timecounter(struct timecounter *tc, int n)
+static __inline uint64_t
+atomic_fetch_quad_i586(uint64_t *p)
 {
-	uint64_t end, now;
+	uint64_t v;
+
+	__asm __volatile(
+	"	movl %%ebx,%%eax ;		"
+	"	movl %%ecx,%%edx ;		"
+	"	" MPLOCKED " cmpxchg8b (%1)"
+	: "=&A" (v) : "r" (p) : "cc");
+	return (v);
+}
+
+static __inline uint64_t
+_fetch_frequency(uint64_t *p)
+{
+
+	if (cpu_class == CPUCLASS_486)
+		return (atomic_fetch_quad_i386(p));
+	return (atomic_fetch_quad_i586(p));
+}
+#else
+#define	_fetch_frequency(p)	(*(p))
+#endif
+
+static __inline int
+_delay(int n)
+{
+	struct timecounter *tc;
+	uint64_t end, freq, now;
 	u_int last, mask, u;
+	int use_tsc;
 
-	mask = tc->tc_counter_mask;
-	last = tc->tc_get_timecount(tc) & mask;
-	end = tc->tc_frequency * n / 1000000;
+	tc = timecounter;
+	freq = _fetch_frequency(&tsc_freq);
+	use_tsc = tsc_is_invariant && freq != 0;
+	if (use_tsc) {
+		mask = ~0u;
+		sched_pin();
+		last = rdtsc();
+	} else {
+		if (tc->tc_quality <= 0)
+			return (0);
+		freq = _fetch_frequency(&tc->tc_frequency);
+		mask = tc->tc_counter_mask;
+		last = tc->tc_get_timecount(tc);
+	}
+	last &= mask;
+	end = freq * n / 1000000;
 	now = 0;
 	do {
 		cpu_spinwait();
-		u = tc->tc_get_timecount(tc) & mask;
+		u = (use_tsc ? rdtsc() : tc->tc_get_timecount(tc)) & mask;
 		if (u < last)
 			now += mask - last + u + 1;
 		else
 			now += u - last;
 		last = u;
 	} while (now < end);
+	if (use_tsc)
+		sched_unpin();
+	return (1);
 }
 
 /*
@@ -289,7 +335,6 @@ getit(void)
 void
 DELAY(int n)
 {
-	struct timecounter *tc;
 	int delta, prev_tick, tick, ticks_left;
 
 #ifdef DELAYDEBUG
@@ -298,15 +343,8 @@ DELAY(int n)
 	static int state = 0;
 #endif
 
-	if (tsc_freq != 0) {
-		delay_tsc(n);
+	if (_delay(n))
 		return;
-	}
-	tc = timecounter;
-	if (tc->tc_quality > 0) {
-		delay_timecounter(tc, n);
-		return;
-	}
 #ifdef DELAYDEBUG
 	if (state == 0) {
 		state = 1;