Split statclock into statclock and profclock, and made the method for driving

statclock based on profhz when profiling is enabled MD, since most platforms
don't use this anyway.  This removes the need for statclock_process, whose
only purpose was to subdivide profhz, and gets the profiling clock running
outside of sched_lock on platforms that implement suswintr.
Also changed the interface for starting and stopping the profiling clock to
do just that, instead of changing the rate of statclock, since they can now
be separate.

Reviewed by:	jhb, tmm
Tested on:	i386, sparc64

34 files changed, 389 insertions, 254 deletions

diff --git a/sys/alpha/alpha/clock.c b/sys/alpha/alpha/clock.c
index 2d34a99..3b113d7 100644
--- a/sys/alpha/alpha/clock.c
+++ b/sys/alpha/alpha/clock.c
@@ -446,17 +446,18 @@ handleclock(void *arg)
 	hardclock(arg);
 }
 
-/*
- * We assume newhz is either stathz or profhz, and that neither will
- * change after being set up above.  Could recalculate intervals here
- * but that would be a drag.
- */
 void
-setstatclockrate(newhz)
-	int newhz;
+cpu_startprofclock(void)
+{
+
+	/* nothing to do */
+}
+
+void
+cpu_stopprofclock(void)
 {
 
-	/* nothing we can do */
+	/* nothing to do */
 }
 
 /*
diff --git a/sys/alpha/alpha/interrupt.c b/sys/alpha/alpha/interrupt.c
index 19cd144..88eba10 100644
--- a/sys/alpha/alpha/interrupt.c
+++ b/sys/alpha/alpha/interrupt.c
@@ -472,16 +472,19 @@ alpha_clock_interrupt(struct trapframe *framep)
 #endif
 			(*platform.clockintr)(framep);
 			/* divide hz (1024) by 8 to get stathz (128) */
-			if ((++schedclk2 & 0x7) == 0)
+			if ((++schedclk2 & 0x7) == 0) {
+				if (profprocs != 0)
+					profclock((struct clockframe *)framep);
 				statclock((struct clockframe *)framep);
+			}
 #ifdef SMP
 		} else {
-			mtx_lock_spin(&sched_lock);
-			hardclock_process(curthread, TRAPF_USERMODE(framep));
-			if ((schedclk2 & 0x7) == 0)
-				statclock_process(curkse, TRAPF_PC(framep),
-				    TRAPF_USERMODE(framep));
-			mtx_unlock_spin(&sched_lock);
+			hardclock_process((struct clockframe *)framep);
+			if ((schedclk2 & 0x7) == 0) {
+				if (profprocs != 0)
+					profclock((struct clockframe *)framep);
+				statclock((struct clockframe *)framep);
+			}
 		}
 #endif
 		critical_exit();
diff --git a/sys/amd64/amd64/apic_vector.S b/sys/amd64/amd64/apic_vector.S
index f8e0321..e4b6ea5 100644
--- a/sys/amd64/amd64/apic_vector.S
+++ b/sys/amd64/amd64/apic_vector.S
@@ -365,7 +365,7 @@ Xinvlrng:
 	iret
 
 /*
- * Forward hardclock to another CPU.  Pushes a trapframe and calls
+ * Forward hardclock to another CPU.  Pushes a clockframe and calls
  * forwarded_hardclock().
  */
 	.text
@@ -389,14 +389,16 @@ Xhardclock:
 	jmp	10f
 1:
 	incl	TD_INTR_NESTING_LEVEL(%ebx)
+	pushl	$0		/* XXX convert trapframe to clockframe */
 	call	forwarded_hardclock
+	addl	$4, %esp	/* XXX convert clockframe to trapframe */
 	decl	TD_INTR_NESTING_LEVEL(%ebx)
 10:
 	MEXITCOUNT
 	jmp	doreti
 
 /*
- * Forward statclock to another CPU.  Pushes a trapframe and calls
+ * Forward statclock to another CPU.  Pushes a clockframe and calls
  * forwarded_statclock().
  */
 	.text
@@ -422,7 +424,9 @@ Xstatclock:
 	jmp	10f
 1:
 	incl	TD_INTR_NESTING_LEVEL(%ebx)
+	pushl	$0		/* XXX convert trapframe to clockframe */
 	call	forwarded_statclock
+	addl	$4, %esp	/* XXX convert clockframe to trapframe */
 	decl	TD_INTR_NESTING_LEVEL(%ebx)
 10:
 	MEXITCOUNT
diff --git a/sys/amd64/amd64/critical.c b/sys/amd64/amd64/critical.c
index de91426..7a505e1 100644
--- a/sys/amd64/amd64/critical.c
+++ b/sys/amd64/amd64/critical.c
@@ -15,6 +15,7 @@
 #include <sys/proc.h>
 #include <sys/sysctl.h>
 #include <sys/ucontext.h>
+#include <machine/clock.h>
 #include <machine/critical.h>
 
 #ifdef SMP
@@ -89,6 +90,10 @@ cpu_thread_link(struct thread *td)
 void
 i386_unpend(void)
 {
+	struct clockframe frame;
+
+	frame.cf_cs = SEL_KPL;
+	frame.cf_eip = (register_t)i386_unpend;
 	KASSERT(curthread->td_critnest == 0, ("unpend critnest != 0"));
 	KASSERT((read_eflags() & PSL_I) == 0, ("unpend interrupts enabled1"));
 	curthread->td_critnest = 1;
@@ -131,15 +136,13 @@ i386_unpend(void)
 			PCPU_SET(spending, mask & ~(1 << irq));
 			switch(irq) {
 			case 0:		/* bit 0 - hardclock */
-				mtx_lock_spin(&sched_lock);
-				hardclock_process(curthread, 0);
-				mtx_unlock_spin(&sched_lock);
+				hardclock_process(&frame);
 				break;
 			case 1:		/* bit 1 - statclock */
-				mtx_lock_spin(&sched_lock);
-				statclock_process(curthread->td_kse,
-				    (register_t)i386_unpend, 0);
-				mtx_unlock_spin(&sched_lock);
+				if (profprocs != 0)
+					profclock(&frame);
+				if (pscnt == psdiv)
+					statclock(&frame);
 				break;
 			}
 			KASSERT((read_eflags() & PSL_I) == 0,
diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 39d10d6..7960c51 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -63,6 +63,7 @@
 
 #include <machine/apic.h>
 #include <machine/atomic.h>
+#include <machine/clock.h>
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/mpapic.h>
@@ -2603,17 +2604,17 @@ ap_init(void)
  * For statclock, we send an IPI to all CPU's to have them call this
  * function.
  *
- * WARNING! unpend() will call statclock_process() directly and skip this
+ * WARNING! unpend() will call statclock() directly and skip this
  * routine.
  */
 void
-forwarded_statclock(struct trapframe frame)
+forwarded_statclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	statclock_process(curthread->td_kse, TRAPF_PC(&frame),
-	    TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	if (profprocs != 0)
+		profclock(&frame);
+	if (pscnt == psdiv)
+		statclock(&frame);
 }
 
 void
@@ -2642,12 +2643,10 @@ forward_statclock(void)
  * routine.
  */
 void
-forwarded_hardclock(struct trapframe frame)
+forwarded_hardclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	hardclock_process(curthread, TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	hardclock_process(&frame);
 }
 
 void 
diff --git a/sys/amd64/amd64/mptable.c b/sys/amd64/amd64/mptable.c
index 39d10d6..7960c51 100644
--- a/sys/amd64/amd64/mptable.c
+++ b/sys/amd64/amd64/mptable.c
@@ -63,6 +63,7 @@
 
 #include <machine/apic.h>
 #include <machine/atomic.h>
+#include <machine/clock.h>
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/mpapic.h>
@@ -2603,17 +2604,17 @@ ap_init(void)
  * For statclock, we send an IPI to all CPU's to have them call this
  * function.
  *
- * WARNING! unpend() will call statclock_process() directly and skip this
+ * WARNING! unpend() will call statclock() directly and skip this
  * routine.
  */
 void
-forwarded_statclock(struct trapframe frame)
+forwarded_statclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	statclock_process(curthread->td_kse, TRAPF_PC(&frame),
-	    TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	if (profprocs != 0)
+		profclock(&frame);
+	if (pscnt == psdiv)
+		statclock(&frame);
 }
 
 void
@@ -2642,12 +2643,10 @@ forward_statclock(void)
  * routine.
  */
 void
-forwarded_hardclock(struct trapframe frame)
+forwarded_hardclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	hardclock_process(curthread, TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	hardclock_process(&frame);
 }
 
 void 
diff --git a/sys/amd64/amd64/tsc.c b/sys/amd64/amd64/tsc.c
index 98ef6f78..8e812a1 100644
--- a/sys/amd64/amd64/tsc.c
+++ b/sys/amd64/amd64/tsc.c
@@ -130,6 +130,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #define TIMER_FREQ   1193182
@@ -380,7 +382,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1169,13 +1177,21 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
-	if (newhz == RTC_PROFRATE)
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
-	else
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
+	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = psratio;
+}
+
+void
+cpu_stopprofclock(void)
+{
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
 	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = 1;
 }
 
 static int
diff --git a/sys/amd64/include/clock.h b/sys/amd64/include/clock.h
index f6ed717..140a345 100644
--- a/sys/amd64/include/clock.h
+++ b/sys/amd64/include/clock.h
@@ -16,6 +16,8 @@
  */
 extern int	adjkerntz;
 extern int	disable_rtc_set;
+extern int	pscnt;
+extern int	psdiv;
 extern int	statclock_disable;
 extern u_int	timer_freq;
 extern int	timer0_max_count;
diff --git a/sys/amd64/include/mptable.h b/sys/amd64/include/mptable.h
index 39d10d6..7960c51 100644
--- a/sys/amd64/include/mptable.h
+++ b/sys/amd64/include/mptable.h
@@ -63,6 +63,7 @@
 
 #include <machine/apic.h>
 #include <machine/atomic.h>
+#include <machine/clock.h>
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/mpapic.h>
@@ -2603,17 +2604,17 @@ ap_init(void)
  * For statclock, we send an IPI to all CPU's to have them call this
  * function.
  *
- * WARNING! unpend() will call statclock_process() directly and skip this
+ * WARNING! unpend() will call statclock() directly and skip this
  * routine.
  */
 void
-forwarded_statclock(struct trapframe frame)
+forwarded_statclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	statclock_process(curthread->td_kse, TRAPF_PC(&frame),
-	    TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	if (profprocs != 0)
+		profclock(&frame);
+	if (pscnt == psdiv)
+		statclock(&frame);
 }
 
 void
@@ -2642,12 +2643,10 @@ forward_statclock(void)
  * routine.
  */
 void
-forwarded_hardclock(struct trapframe frame)
+forwarded_hardclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	hardclock_process(curthread, TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	hardclock_process(&frame);
 }
 
 void 
diff --git a/sys/amd64/include/smp.h b/sys/amd64/include/smp.h
index f2250ff..2f576a2 100644
--- a/sys/amd64/include/smp.h
+++ b/sys/amd64/include/smp.h
@@ -113,9 +113,9 @@ void	revoke_apic_irq(int irq);
 void	bsp_apic_configure(void);
 void	init_secondary(void);
 void	forward_statclock(void);
-void	forwarded_statclock(struct trapframe frame);
+void	forwarded_statclock(struct clockframe frame);
 void	forward_hardclock(void);
-void	forwarded_hardclock(struct trapframe frame);
+void	forwarded_hardclock(struct clockframe frame);
 void	ipi_selected(u_int cpus, u_int ipi);
 void	ipi_all(u_int ipi);
 void	ipi_all_but_self(u_int ipi);
diff --git a/sys/amd64/isa/clock.c b/sys/amd64/isa/clock.c
index 98ef6f78..8e812a1 100644
--- a/sys/amd64/isa/clock.c
+++ b/sys/amd64/isa/clock.c
@@ -130,6 +130,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #define TIMER_FREQ   1193182
@@ -380,7 +382,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1169,13 +1177,21 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
-	if (newhz == RTC_PROFRATE)
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
-	else
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
+	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = psratio;
+}
+
+void
+cpu_stopprofclock(void)
+{
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
 	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = 1;
 }
 
 static int
diff --git a/sys/i386/i386/apic_vector.s b/sys/i386/i386/apic_vector.s
index f8e0321..e4b6ea5 100644
--- a/sys/i386/i386/apic_vector.s
+++ b/sys/i386/i386/apic_vector.s
@@ -365,7 +365,7 @@ Xinvlrng:
 	iret
 
 /*
- * Forward hardclock to another CPU.  Pushes a trapframe and calls
+ * Forward hardclock to another CPU.  Pushes a clockframe and calls
  * forwarded_hardclock().
  */
 	.text
@@ -389,14 +389,16 @@ Xhardclock:
 	jmp	10f
 1:
 	incl	TD_INTR_NESTING_LEVEL(%ebx)
+	pushl	$0		/* XXX convert trapframe to clockframe */
 	call	forwarded_hardclock
+	addl	$4, %esp	/* XXX convert clockframe to trapframe */
 	decl	TD_INTR_NESTING_LEVEL(%ebx)
 10:
 	MEXITCOUNT
 	jmp	doreti
 
 /*
- * Forward statclock to another CPU.  Pushes a trapframe and calls
+ * Forward statclock to another CPU.  Pushes a clockframe and calls
  * forwarded_statclock().
  */
 	.text
@@ -422,7 +424,9 @@ Xstatclock:
 	jmp	10f
 1:
 	incl	TD_INTR_NESTING_LEVEL(%ebx)
+	pushl	$0		/* XXX convert trapframe to clockframe */
 	call	forwarded_statclock
+	addl	$4, %esp	/* XXX convert clockframe to trapframe */
 	decl	TD_INTR_NESTING_LEVEL(%ebx)
 10:
 	MEXITCOUNT
diff --git a/sys/i386/i386/critical.c b/sys/i386/i386/critical.c
index de91426..7a505e1 100644
--- a/sys/i386/i386/critical.c
+++ b/sys/i386/i386/critical.c
@@ -15,6 +15,7 @@
 #include <sys/proc.h>
 #include <sys/sysctl.h>
 #include <sys/ucontext.h>
+#include <machine/clock.h>
 #include <machine/critical.h>
 
 #ifdef SMP
@@ -89,6 +90,10 @@ cpu_thread_link(struct thread *td)
 void
 i386_unpend(void)
 {
+	struct clockframe frame;
+
+	frame.cf_cs = SEL_KPL;
+	frame.cf_eip = (register_t)i386_unpend;
 	KASSERT(curthread->td_critnest == 0, ("unpend critnest != 0"));
 	KASSERT((read_eflags() & PSL_I) == 0, ("unpend interrupts enabled1"));
 	curthread->td_critnest = 1;
@@ -131,15 +136,13 @@ i386_unpend(void)
 			PCPU_SET(spending, mask & ~(1 << irq));
 			switch(irq) {
 			case 0:		/* bit 0 - hardclock */
-				mtx_lock_spin(&sched_lock);
-				hardclock_process(curthread, 0);
-				mtx_unlock_spin(&sched_lock);
+				hardclock_process(&frame);
 				break;
 			case 1:		/* bit 1 - statclock */
-				mtx_lock_spin(&sched_lock);
-				statclock_process(curthread->td_kse,
-				    (register_t)i386_unpend, 0);
-				mtx_unlock_spin(&sched_lock);
+				if (profprocs != 0)
+					profclock(&frame);
+				if (pscnt == psdiv)
+					statclock(&frame);
 				break;
 			}
 			KASSERT((read_eflags() & PSL_I) == 0,
diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c
index 39d10d6..7960c51 100644
--- a/sys/i386/i386/mp_machdep.c
+++ b/sys/i386/i386/mp_machdep.c
@@ -63,6 +63,7 @@
 
 #include <machine/apic.h>
 #include <machine/atomic.h>
+#include <machine/clock.h>
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/mpapic.h>
@@ -2603,17 +2604,17 @@ ap_init(void)
  * For statclock, we send an IPI to all CPU's to have them call this
  * function.
  *
- * WARNING! unpend() will call statclock_process() directly and skip this
+ * WARNING! unpend() will call statclock() directly and skip this
  * routine.
  */
 void
-forwarded_statclock(struct trapframe frame)
+forwarded_statclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	statclock_process(curthread->td_kse, TRAPF_PC(&frame),
-	    TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	if (profprocs != 0)
+		profclock(&frame);
+	if (pscnt == psdiv)
+		statclock(&frame);
 }
 
 void
@@ -2642,12 +2643,10 @@ forward_statclock(void)
  * routine.
  */
 void
-forwarded_hardclock(struct trapframe frame)
+forwarded_hardclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	hardclock_process(curthread, TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	hardclock_process(&frame);
 }
 
 void 
diff --git a/sys/i386/i386/mptable.c b/sys/i386/i386/mptable.c
index 39d10d6..7960c51 100644
--- a/sys/i386/i386/mptable.c
+++ b/sys/i386/i386/mptable.c
@@ -63,6 +63,7 @@
 
 #include <machine/apic.h>
 #include <machine/atomic.h>
+#include <machine/clock.h>
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/mpapic.h>
@@ -2603,17 +2604,17 @@ ap_init(void)
  * For statclock, we send an IPI to all CPU's to have them call this
  * function.
  *
- * WARNING! unpend() will call statclock_process() directly and skip this
+ * WARNING! unpend() will call statclock() directly and skip this
  * routine.
  */
 void
-forwarded_statclock(struct trapframe frame)
+forwarded_statclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	statclock_process(curthread->td_kse, TRAPF_PC(&frame),
-	    TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	if (profprocs != 0)
+		profclock(&frame);
+	if (pscnt == psdiv)
+		statclock(&frame);
 }
 
 void
@@ -2642,12 +2643,10 @@ forward_statclock(void)
  * routine.
  */
 void
-forwarded_hardclock(struct trapframe frame)
+forwarded_hardclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	hardclock_process(curthread, TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	hardclock_process(&frame);
 }
 
 void 
diff --git a/sys/i386/i386/tsc.c b/sys/i386/i386/tsc.c
index 98ef6f78..8e812a1 100644
--- a/sys/i386/i386/tsc.c
+++ b/sys/i386/i386/tsc.c
@@ -130,6 +130,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #define TIMER_FREQ   1193182
@@ -380,7 +382,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1169,13 +1177,21 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
-	if (newhz == RTC_PROFRATE)
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
-	else
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
+	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = psratio;
+}
+
+void
+cpu_stopprofclock(void)
+{
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
 	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = 1;
 }
 
 static int
diff --git a/sys/i386/include/clock.h b/sys/i386/include/clock.h
index f6ed717..140a345 100644
--- a/sys/i386/include/clock.h
+++ b/sys/i386/include/clock.h
@@ -16,6 +16,8 @@
  */
 extern int	adjkerntz;
 extern int	disable_rtc_set;
+extern int	pscnt;
+extern int	psdiv;
 extern int	statclock_disable;
 extern u_int	timer_freq;
 extern int	timer0_max_count;
diff --git a/sys/i386/include/mptable.h b/sys/i386/include/mptable.h
index 39d10d6..7960c51 100644
--- a/sys/i386/include/mptable.h
+++ b/sys/i386/include/mptable.h
@@ -63,6 +63,7 @@
 
 #include <machine/apic.h>
 #include <machine/atomic.h>
+#include <machine/clock.h>
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/mpapic.h>
@@ -2603,17 +2604,17 @@ ap_init(void)
  * For statclock, we send an IPI to all CPU's to have them call this
  * function.
  *
- * WARNING! unpend() will call statclock_process() directly and skip this
+ * WARNING! unpend() will call statclock() directly and skip this
  * routine.
  */
 void
-forwarded_statclock(struct trapframe frame)
+forwarded_statclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	statclock_process(curthread->td_kse, TRAPF_PC(&frame),
-	    TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	if (profprocs != 0)
+		profclock(&frame);
+	if (pscnt == psdiv)
+		statclock(&frame);
 }
 
 void
@@ -2642,12 +2643,10 @@ forward_statclock(void)
  * routine.
  */
 void
-forwarded_hardclock(struct trapframe frame)
+forwarded_hardclock(struct clockframe frame)
 {
 
-	mtx_lock_spin(&sched_lock);
-	hardclock_process(curthread, TRAPF_USERMODE(&frame));
-	mtx_unlock_spin(&sched_lock);
+	hardclock_process(&frame);
 }
 
 void 
diff --git a/sys/i386/include/smp.h b/sys/i386/include/smp.h
index f2250ff..2f576a2 100644
--- a/sys/i386/include/smp.h
+++ b/sys/i386/include/smp.h
@@ -113,9 +113,9 @@ void	revoke_apic_irq(int irq);
 void	bsp_apic_configure(void);
 void	init_secondary(void);
 void	forward_statclock(void);
-void	forwarded_statclock(struct trapframe frame);
+void	forwarded_statclock(struct clockframe frame);
 void	forward_hardclock(void);
-void	forwarded_hardclock(struct trapframe frame);
+void	forwarded_hardclock(struct clockframe frame);
 void	ipi_selected(u_int cpus, u_int ipi);
 void	ipi_all(u_int ipi);
 void	ipi_all_but_self(u_int ipi);
diff --git a/sys/i386/isa/apic_vector.s b/sys/i386/isa/apic_vector.s
index f8e0321..e4b6ea5 100644
--- a/sys/i386/isa/apic_vector.s
+++ b/sys/i386/isa/apic_vector.s
@@ -365,7 +365,7 @@ Xinvlrng:
 	iret
 
 /*
- * Forward hardclock to another CPU.  Pushes a trapframe and calls
+ * Forward hardclock to another CPU.  Pushes a clockframe and calls
  * forwarded_hardclock().
  */
 	.text
@@ -389,14 +389,16 @@ Xhardclock:
 	jmp	10f
 1:
 	incl	TD_INTR_NESTING_LEVEL(%ebx)
+	pushl	$0		/* XXX convert trapframe to clockframe */
 	call	forwarded_hardclock
+	addl	$4, %esp	/* XXX convert clockframe to trapframe */
 	decl	TD_INTR_NESTING_LEVEL(%ebx)
 10:
 	MEXITCOUNT
 	jmp	doreti
 
 /*
- * Forward statclock to another CPU.  Pushes a trapframe and calls
+ * Forward statclock to another CPU.  Pushes a clockframe and calls
  * forwarded_statclock().
  */
 	.text
@@ -422,7 +424,9 @@ Xstatclock:
 	jmp	10f
 1:
 	incl	TD_INTR_NESTING_LEVEL(%ebx)
+	pushl	$0		/* XXX convert trapframe to clockframe */
 	call	forwarded_statclock
+	addl	$4, %esp	/* XXX convert clockframe to trapframe */
 	decl	TD_INTR_NESTING_LEVEL(%ebx)
 10:
 	MEXITCOUNT
diff --git a/sys/i386/isa/clock.c b/sys/i386/isa/clock.c
index 98ef6f78..8e812a1 100644
--- a/sys/i386/isa/clock.c
+++ b/sys/i386/isa/clock.c
@@ -130,6 +130,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #define TIMER_FREQ   1193182
@@ -380,7 +382,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1169,13 +1177,21 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
-	if (newhz == RTC_PROFRATE)
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
-	else
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
+	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = psratio;
+}
+
+void
+cpu_stopprofclock(void)
+{
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
 	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = 1;
 }
 
 static int
diff --git a/sys/ia64/ia64/clock.c b/sys/ia64/ia64/clock.c
index 41d8dc4..277c14b 100644
--- a/sys/ia64/ia64/clock.c
+++ b/sys/ia64/ia64/clock.c
@@ -259,12 +259,19 @@ fail:
  * change after being set up above.  Could recalculate intervals here
  * but that would be a drag.
  */
+
+void
+cpu_startprofclock(void)
+{
+
+	/* nothing to do */
+}
+
 void
-setstatclockrate(newhz)
-	int newhz;
+cpu_stopprofclock(void)
 {
 
-	/* nothing we can do */
+	/* nothing to do */
 }
 
 /*
diff --git a/sys/ia64/ia64/interrupt.c b/sys/ia64/ia64/interrupt.c
index 720dbee..bc2d324 100644
--- a/sys/ia64/ia64/interrupt.c
+++ b/sys/ia64/ia64/interrupt.c
@@ -132,16 +132,19 @@ interrupt(u_int64_t vector, struct trapframe *framep)
 #endif
 			hardclock((struct clockframe *)framep);
 			/* divide hz (1024) by 8 to get stathz (128) */
-			if ((++schedclk2 & 0x7) == 0)
+			if ((++schedclk2 & 0x7) == 0) {
+				if (profprocs != 0)
+					profclock((struct clockframe *)framep);
 				statclock((struct clockframe *)framep);
+			}
 #ifdef SMP
 		} else {
-			mtx_lock_spin(&sched_lock);
-			hardclock_process(curthread, TRAPF_USERMODE(framep));
-			if ((schedclk2 & 0x7) == 0)
-				statclock_process(curkse, TRAPF_PC(framep),
-				    TRAPF_USERMODE(framep));
-			mtx_unlock_spin(&sched_lock);
+			hardclock_process((struct clockframe *)framep);
+			if ((schedclk2 & 0x7) == 0) {
+				if (profprocs != 0)
+					profclock((struct clockframe *)framep);
+				statclock((struct clockframe *)framep);
+			}
 		}
 #endif
 		critical_exit();
diff --git a/sys/isa/atrtc.c b/sys/isa/atrtc.c
index 98ef6f78..8e812a1 100644
--- a/sys/isa/atrtc.c
+++ b/sys/isa/atrtc.c
@@ -130,6 +130,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #define TIMER_FREQ   1193182
@@ -380,7 +382,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1169,13 +1177,21 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
-	if (newhz == RTC_PROFRATE)
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
-	else
-		rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_PROF;
+	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = psratio;
+}
+
+void
+cpu_stopprofclock(void)
+{
+
+	rtc_statusa = RTCSA_DIVIDER | RTCSA_NOPROF;
 	writertc(RTC_STATUSA, rtc_statusa);
+	psdiv = pscnt = 1;
 }
 
 static int
diff --git a/sys/kern/kern_clock.c b/sys/kern/kern_clock.c
index c5e1b4a..5c6756b 100644
--- a/sys/kern/kern_clock.c
+++ b/sys/kern/kern_clock.c
@@ -122,10 +122,9 @@ long tk_rawcc;
 
 int	stathz;
 int	profhz;
-static int profprocs;
+int	profprocs;
 int	ticks;
-static int psdiv, pscnt;		/* prof => stat divider */
-int	psratio;			/* ratio: prof / stat */
+int	psratio;
 
 /*
  * Initialize clock frequencies and start both clocks running.
@@ -141,7 +140,6 @@ initclocks(dummy)
 	 * Set divisors to 1 (normal case) and let the machine-specific
 	 * code do its bit.
 	 */
-	psdiv = pscnt = 1;
 	cpu_initclocks();
 
 #ifdef DEVICE_POLLING
@@ -157,32 +155,27 @@ initclocks(dummy)
 }
 
 /*
- * Each time the real-time timer fires, this function is called on all CPUs
- * with each CPU passing in its curthread as the first argument.  If possible
- * a nice optimization in the future would be to allow the CPU receiving the
- * actual real-time timer interrupt to call this function on behalf of the
- * other CPUs rather than sending an IPI to all other CPUs so that they
- * can call this function.  Note that hardclock() calls hardclock_process()
- * for the CPU receiving the timer interrupt, so only the other CPUs in the
- * system need to call this function (or have it called on their behalf.
+ * Each time the real-time timer fires, this function is called on all CPUs.
+ * Note that hardclock() calls hardclock_process() for the boot CPU, so only
+ * the other CPUs in the system need to call this function.
  */
 void
-hardclock_process(td, user)
-	struct thread *td;
-	int user;
+hardclock_process(frame)
+	register struct clockframe *frame;
 {
 	struct pstats *pstats;
+	struct thread *td = curthread;
 	struct proc *p = td->td_proc;
 
 	/*
 	 * Run current process's virtual and profile time, as needed.
 	 */
-	mtx_assert(&sched_lock, MA_OWNED);
+	mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
 	if (p->p_flag & P_KSES) {
 		/* XXXKSE What to do? */
 	} else {
 		pstats = p->p_stats;
-		if (user &&
+		if (CLKF_USERMODE(frame) &&
 		    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
 		    itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
 			p->p_sflag |= PS_ALRMPEND;
@@ -194,6 +187,7 @@ hardclock_process(td, user)
 			td->td_kse->ke_flags |= KEF_ASTPENDING;
 		}
 	}
+	mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
 }
 
 /*
@@ -206,9 +200,7 @@ hardclock(frame)
 	int need_softclock = 0;
 
 	CTR0(KTR_CLK, "hardclock fired");
-	mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
-	hardclock_process(curthread, CLKF_USERMODE(frame));
-	mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
+	hardclock_process(frame);
 
 	tc_ticktock();
 	/*
@@ -216,8 +208,10 @@ hardclock(frame)
 	 *
 	 * XXX: this only works for UP
 	 */
-	if (stathz == 0)
+	if (stathz == 0) {
+		profclock(frame);
 		statclock(frame);
+	}
 
 #ifdef DEVICE_POLLING
 	hardclock_device_poll();	/* this is very short and quick */
@@ -312,7 +306,6 @@ void
 startprofclock(p)
 	register struct proc *p;
 {
-	int s;
 
 	/*
 	 * XXX; Right now sched_lock protects statclock(), but perhaps
@@ -322,12 +315,8 @@ startprofclock(p)
 	mtx_lock_spin(&sched_lock);
 	if ((p->p_sflag & PS_PROFIL) == 0) {
 		p->p_sflag |= PS_PROFIL;
-		if (++profprocs == 1 && stathz != 0) {
-			s = splstatclock();
-			psdiv = pscnt = psratio;
-			setstatclockrate(profhz);
-			splx(s);
-		}
+		if (++profprocs == 1)
+			cpu_startprofclock();
 	}
 	mtx_unlock_spin(&sched_lock);
 }
@@ -339,57 +328,41 @@ void
 stopprofclock(p)
 	register struct proc *p;
 {
-	int s;
 
 	mtx_lock_spin(&sched_lock);
 	if (p->p_sflag & PS_PROFIL) {
 		p->p_sflag &= ~PS_PROFIL;
-		if (--profprocs == 0 && stathz != 0) {
-			s = splstatclock();
-			psdiv = pscnt = 1;
-			setstatclockrate(stathz);
-			splx(s);
-		}
+		if (--profprocs == 0)
+			cpu_stopprofclock();
 	}
 	mtx_unlock_spin(&sched_lock);
 }
 
 /*
- * Do process and kernel statistics.  Most of the statistics are only
+ * Statistics clock.  Grab profile sample, and if divider reaches 0,
+ * do process and kernel statistics.  Most of the statistics are only
  * used by user-level statistics programs.  The main exceptions are
- * ke->ke_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu.  This function
- * should be called by all CPUs in the system for each statistics clock
- * interrupt.  See the description of hardclock_process for more detail on
- * this function's relationship to statclock.
+ * ke->ke_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu.
+ * This should be called by all active processors.
  */
 void
-statclock_process(ke, pc, user)
-	struct kse *ke;
-	register_t pc;
-	int user;
+statclock(frame)
+	register struct clockframe *frame;
 {
-#ifdef GPROF
-	struct gmonparam *g;
-	int i;
-#endif
 	struct pstats *pstats;
-	long rss;
 	struct rusage *ru;
 	struct vmspace *vm;
-	struct proc *p = ke->ke_proc;
-	struct thread *td = ke->ke_thread; /* current thread */
+	struct thread *td;
+	struct kse *ke;
+	struct proc *p;
+	long rss;
 
-	KASSERT(ke == curthread->td_kse, ("statclock_process: td != curthread"));
-	mtx_assert(&sched_lock, MA_OWNED);
-	if (user) {
-		/*
-		 * Came from user mode; CPU was in user state.
-		 * If this process is being profiled, record the tick.
-		 */
-		if (p->p_sflag & PS_PROFIL)
-			addupc_intr(ke, pc, 1);
-		if (pscnt < psdiv)
-			return;
+	td = curthread;
+	p = td->td_proc;
+
+	mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
+	ke = td->td_kse;
+	if (CLKF_USERMODE(frame)) {
 		/*
 		 * Charge the time as appropriate.
 		 */
@@ -401,21 +374,6 @@ statclock_process(ke, pc, user)
 		else
 			cp_time[CP_USER]++;
 	} else {
-#ifdef GPROF
-		/*
-		 * Kernel statistics are just like addupc_intr, only easier.
-		 */
-		g = &_gmonparam;
-		if (g->state == GMON_PROF_ON) {
-			i = pc - g->lowpc;
-			if (i < g->textsize) {
-				i /= HISTFRACTION * sizeof(*g->kcount);
-				g->kcount[i]++;
-			}
-		}
-#endif
-		if (pscnt < psdiv)
-			return;
 		/*
 		 * Came from kernel mode, so we were:
 		 * - handling an interrupt,
@@ -455,25 +413,43 @@ statclock_process(ke, pc, user)
 		if (ru->ru_maxrss < rss)
 			ru->ru_maxrss = rss;
 	}
+	mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
 }
 
-/*
- * Statistics clock.  Grab profile sample, and if divider reaches 0,
- * do process and kernel statistics.  Most of the statistics are only
- * used by user-level statistics programs.  The main exceptions are
- * ke->ke_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu.
- */
 void
-statclock(frame)
+profclock(frame)
 	register struct clockframe *frame;
 {
+	struct thread *td;
+#ifdef GPROF
+	struct gmonparam *g;
+	int i;
+#endif
 
-	CTR0(KTR_CLK, "statclock fired");
-	mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
-	if (--pscnt == 0)
-		pscnt = psdiv;
-	statclock_process(curthread->td_kse, CLKF_PC(frame), CLKF_USERMODE(frame));
-	mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
+	if (CLKF_USERMODE(frame)) {
+		/*
+		 * Came from user mode; CPU was in user state.
+		 * If this process is being profiled, record the tick.
+		 */
+		td = curthread;
+		if (td->td_proc->p_sflag & PS_PROFIL)
+			addupc_intr(td->td_kse, CLKF_PC(frame), 1);
+	}
+#ifdef GPROF
+	else {
+		/*
+		 * Kernel statistics are just like addupc_intr, only easier.
+		 */
+		g = &_gmonparam;
+		if (g->state == GMON_PROF_ON) {
+			i = CLKF_PC(frame) - g->lowpc;
+			if (i < g->textsize) {
+				i /= HISTFRACTION * sizeof(*g->kcount);
+				g->kcount[i]++;
+			}
+		}
+	}
+#endif
 }
 
 /*
diff --git a/sys/pc98/cbus/clock.c b/sys/pc98/cbus/clock.c
index 557afe8..8e2dc97 100644
--- a/sys/pc98/cbus/clock.c
+++ b/sys/pc98/cbus/clock.c
@@ -140,6 +140,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #ifdef PC98
@@ -441,7 +443,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1518,7 +1526,7 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
 #ifndef PC98
 	if (newhz == RTC_PROFRATE)
@@ -1529,6 +1537,11 @@ setstatclockrate(int newhz)
 #endif
 }
 
+void
+cpu_stopprofclock(void)
+{
+}
+
 static int
 sysctl_machdep_i8254_freq(SYSCTL_HANDLER_ARGS)
 {
diff --git a/sys/pc98/cbus/pcrtc.c b/sys/pc98/cbus/pcrtc.c
index 557afe8..8e2dc97 100644
--- a/sys/pc98/cbus/pcrtc.c
+++ b/sys/pc98/cbus/pcrtc.c
@@ -140,6 +140,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #ifdef PC98
@@ -441,7 +443,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1518,7 +1526,7 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
 #ifndef PC98
 	if (newhz == RTC_PROFRATE)
@@ -1529,6 +1537,11 @@ setstatclockrate(int newhz)
 #endif
 }
 
+void
+cpu_stopprofclock(void)
+{
+}
+
 static int
 sysctl_machdep_i8254_freq(SYSCTL_HANDLER_ARGS)
 {
diff --git a/sys/pc98/pc98/clock.c b/sys/pc98/pc98/clock.c
index 557afe8..8e2dc97 100644
--- a/sys/pc98/pc98/clock.c
+++ b/sys/pc98/pc98/clock.c
@@ -140,6 +140,8 @@ static void setup_8254_mixed_mode(void);
 int	adjkerntz;		/* local offset from GMT in seconds */
 int	clkintr_pending;
 int	disable_rtc_set;	/* disable resettodr() if != 0 */
+int	pscnt = 1;
+int	psdiv = 1;
 int	statclock_disable;
 #ifndef TIMER_FREQ
 #ifdef PC98
@@ -441,7 +443,13 @@ static void
 rtcintr(struct clockframe frame)
 {
 	while (rtcin(RTC_INTR) & RTCIR_PERIOD) {
-		statclock(&frame);
+		if (profprocs != 0) {
+			if (--pscnt == 0)
+				pscnt = psdiv;
+			profclock(&frame);
+		}
+		if (pscnt == psdiv)
+			statclock(&frame);
 #ifdef SMP
 		forward_statclock();
 #endif
@@ -1518,7 +1526,7 @@ setup_8254_mixed_mode()
 #endif
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
 {
 #ifndef PC98
 	if (newhz == RTC_PROFRATE)
@@ -1529,6 +1537,11 @@ setstatclockrate(int newhz)
 #endif
 }
 
+void
+cpu_stopprofclock(void)
+{
+}
+
 static int
 sysctl_machdep_i8254_freq(SYSCTL_HANDLER_ARGS)
 {
diff --git a/sys/powerpc/aim/clock.c b/sys/powerpc/aim/clock.c
index d1f13e2..a28d376 100644
--- a/sys/powerpc/aim/clock.c
+++ b/sys/powerpc/aim/clock.c
@@ -319,8 +319,13 @@ delay(int n)
  * Nothing to do.
  */
 void
-setstatclockrate(int arg)
+cpu_startprofclock(void)
 {
 
 	/* Do nothing */
 }
+
+void
+cpu_stopprofclock(void)
+{
+}
diff --git a/sys/powerpc/powerpc/clock.c b/sys/powerpc/powerpc/clock.c
index d1f13e2..a28d376 100644
--- a/sys/powerpc/powerpc/clock.c
+++ b/sys/powerpc/powerpc/clock.c
@@ -319,8 +319,13 @@ delay(int n)
  * Nothing to do.
  */
 void
-setstatclockrate(int arg)
+cpu_startprofclock(void)
 {
 
 	/* Do nothing */
 }
+
+void
+cpu_stopprofclock(void)
+{
+}
diff --git a/sys/sparc64/sparc64/clock.c b/sys/sparc64/sparc64/clock.c
index 1444270..37341fd 100644
--- a/sys/sparc64/sparc64/clock.c
+++ b/sys/sparc64/sparc64/clock.c
@@ -48,9 +48,13 @@ DELAY(int n)
 }
 
 void
-setstatclockrate(int newhz)
+cpu_startprofclock(void)
+{
+}
+
+void
+cpu_stopprofclock(void)
 {
-	/* TODO; */
 }
 
 int
diff --git a/sys/sparc64/sparc64/tick.c b/sys/sparc64/sparc64/tick.c
index a10ba01..f92fbb8 100644
--- a/sys/sparc64/sparc64/tick.c
+++ b/sys/sparc64/sparc64/tick.c
@@ -57,6 +57,7 @@ int tick_missed;	/* statistics */
 void
 cpu_initclocks(void)
 {
+	stathz = hz;
 	tick_start(tick_hardclock);
 }
 
@@ -64,20 +65,13 @@ static __inline void
 tick_process(struct clockframe *cf)
 {
 
-#ifdef SMP
 	if (PCPU_GET(cpuid) == 0)
 		hardclock(cf);
-	else {
-		CTR1(KTR_CLK, "tick_process: AP, cpuid=%d", PCPU_GET(cpuid));
-		mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
-		hardclock_process(curthread, CLKF_USERMODE(cf));
-		statclock_process(curthread->td_kse, CLKF_PC(cf),
-		    CLKF_USERMODE(cf));
-		mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
-	}
-#else
-	hardclock(cf);
-#endif
+	else
+		hardclock_process(cf);
+	if (profprocs != 0)
+		profclock(cf);
+	statclock(cf);
 }
 
 void
diff --git a/sys/sys/kernel.h b/sys/sys/kernel.h
index 4089d59..ba486a9 100644
--- a/sys/sys/kernel.h
+++ b/sys/sys/kernel.h
@@ -75,6 +75,7 @@ extern int hz;				/* system clock's frequency */
 extern int psratio;			/* ratio: prof / stat */
 extern int stathz;			/* statistics clock's frequency */
 extern int profhz;			/* profiling clock's frequency */
+extern int profprocs;			/* number of process's profiling */
 extern int ticks;
 extern int lbolt;			/* once a second sleep address */
 
diff --git a/sys/sys/systm.h b/sys/sys/systm.h
index a3fdcf5..363f713 100644
--- a/sys/sys/systm.h
+++ b/sys/sys/systm.h
@@ -199,14 +199,15 @@ int	suword64(void *base, int64_t word);
 void	realitexpire(void *);
 
 void	hardclock(struct clockframe *frame);
-void	hardclock_process(struct thread *td, int user);
+void	hardclock_process(struct clockframe *frame);
 void	softclock(void *);
 void	statclock(struct clockframe *frame);
-void	statclock_process(struct kse *ke, register_t pc, int user);
+void	profclock(struct clockframe *frame);
 
 void	startprofclock(struct proc *);
 void	stopprofclock(struct proc *);
-void	setstatclockrate(int hzrate);
+void	cpu_startprofclock(void);
+void	cpu_stopprofclock(void);
 
 /* flags for suser() and suser_cred() */
 #define PRISON_ROOT	1