Back out per-CPU kernel profiling, it shouldn't modify a public header

at this moment.

21 files changed, 128 insertions, 278 deletions

diff --git a/lib/libc/gmon/gmon.c b/lib/libc/gmon/gmon.c
index 13c7a9a4847..240c53c2570 100644
--- a/lib/libc/gmon/gmon.c
+++ b/lib/libc/gmon/gmon.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gmon.c,v 1.21 2013/02/12 07:31:13 mpi Exp $ */
+/*	$OpenBSD: gmon.c,v 1.22 2013/02/12 08:06:22 mpi Exp $ */
 /*-
  * Copyright (c) 1983, 1992, 1993
  *	The Regents of the University of California.  All rights reserved.
@@ -47,9 +47,6 @@ static int	s_scale;
 /* see profil(2) where this is describe (incorrectly) */
 #define		SCALE_1_TO_1	0x10000L
 
-#define ROUNDDOWN(x,y)(((x)/(y))*(y))
-#define ROUNDUP(x,y)((((x)+(y)-1)/(y))*(y))
-
 #define ERR(s) write(STDERR_FILENO, s, sizeof(s))
 
 void	moncontrol(int);
diff --git a/lib/libc/gmon/mcount.c b/lib/libc/gmon/mcount.c
index f18b1b52089..e7619c7599b 100644
--- a/lib/libc/gmon/mcount.c
+++ b/lib/libc/gmon/mcount.c
@@ -1,5 +1,4 @@
-/*	$OpenBSD: mcount.c,v 1.12 2013/02/11 17:17:03 mpi Exp $ */
-
+/*	$OpenBSD: mcount.c,v 1.13 2013/02/12 08:06:22 mpi Exp $ */
 /*-
  * Copyright (c) 1983, 1992, 1993
  *	The Regents of the University of California.  All rights reserved.
@@ -42,6 +41,10 @@
  * _mcount updates data structures that represent traversals of the
  * program's call graph edges.  frompc and selfpc are the return
  * address and function address that represents the given call graph edge.
+ *
+ * Note: the original BSD code used the same variable (frompcindex) for
+ * both frompcindex and frompc.  Any reasonable, modern compiler will
+ * perform this optimization.
  */
 _MCOUNT_DECL(u_long frompc, u_long selfpc) __used;
 /* _mcount; may be static, inline, etc */
@@ -53,21 +56,9 @@ _MCOUNT_DECL(u_long frompc, u_long selfpc)
 	long toindex;
 #ifdef _KERNEL
 	int s;
+#endif
 
-	/*
-	 * Do not profile execution if memory for the current CPU
-	 * desciptor and profiling buffers has not yet been allocated
-	 * or if the CPU we are running on has not yet set its trap
-	 * handler.
-	 */
-	if (gmoninit == 0)
-		return;
-
-	if ((p = curcpu()->ci_gmon) == NULL)
-		return;
-#else
 	p = &_gmonparam;
-#endif
 	/*
 	 * check that we are profiling
 	 * and that we aren't recursively invoked.
@@ -180,8 +171,10 @@ overflow:
 	return;
 }
 
+#ifndef lint
 /*
  * Actual definition of mcount function.  Defined in <machine/profile.h>,
  * which is included by <sys/gmon.h>.
  */
 MCOUNT
+#endif
diff --git a/sys/arch/alpha/include/cpu.h b/sys/arch/alpha/include/cpu.h
index 523dcfbcdea..6ce34749606 100644
--- a/sys/arch/alpha/include/cpu.h
+++ b/sys/arch/alpha/include/cpu.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: cpu.h,v 1.47 2013/02/11 17:05:25 mpi Exp $ */
+/* $OpenBSD: cpu.h,v 1.48 2013/02/12 08:06:22 mpi Exp $ */
 /* $NetBSD: cpu.h,v 1.45 2000/08/21 02:03:12 thorpej Exp $ */
 
 /*-
@@ -203,9 +203,6 @@ struct cpu_info {
 	u_long ci_ipis;			/* interprocessor interrupts pending */
 #endif
 	u_int32_t ci_randseed;
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 #define	CPUF_PRIMARY	0x01		/* CPU is primary CPU */
diff --git a/sys/arch/amd64/include/cpu.h b/sys/arch/amd64/include/cpu.h
index 9baaf42dc8c..61e9889d71f 100644
--- a/sys/arch/amd64/include/cpu.h
+++ b/sys/arch/amd64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.77 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.78 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 2003/04/26 18:39:39 fvdl Exp $	*/
 
 /*-
@@ -128,9 +128,6 @@ struct cpu_info {
 
 	struct ksensordev	ci_sensordev;
 	struct ksensor		ci_sensor;
-#ifdef GPROF
-	struct gmonparam	*ci_gmon;
-#endif
 };
 
 #define CPUF_BSP	0x0001		/* CPU is the original BSP */
diff --git a/sys/arch/arm/include/cpu.h b/sys/arch/arm/include/cpu.h
index c82ce63009d..7e1ca74122c 100644
--- a/sys/arch/arm/include/cpu.h
+++ b/sys/arch/arm/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.32 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.33 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.34 2003/06/23 11:01:08 martin Exp $	*/
 
 /*
@@ -198,9 +198,6 @@ struct cpu_info {
 
 	uint32_t ci_cpl;
 	uint32_t ci_ipending;
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 #ifndef MULTIPROCESSOR
diff --git a/sys/arch/hppa/include/cpu.h b/sys/arch/hppa/include/cpu.h
index 203ec4e3675..b477fc6b15a 100644
--- a/sys/arch/hppa/include/cpu.h
+++ b/sys/arch/hppa/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.82 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.83 2013/02/12 08:06:22 mpi Exp $	*/
 
 /*
  * Copyright (c) 2000-2004 Michael Shalayeff
@@ -103,9 +103,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int		ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 } __attribute__((__aligned__(64)));
 
 #define		CPUF_RUNNING	0x0001		/* CPU is running. */
diff --git a/sys/arch/hppa64/include/cpu.h b/sys/arch/hppa64/include/cpu.h
index 211a3ded9ab..509a8a21b14 100644
--- a/sys/arch/hppa64/include/cpu.h
+++ b/sys/arch/hppa64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.29 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.30 2013/02/12 08:06:22 mpi Exp $	*/
 
 /*
  * Copyright (c) 2005 Michael Shalayeff
@@ -109,9 +109,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int		ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 struct cpu_info *curcpu(void);
diff --git a/sys/arch/i386/include/cpu.h b/sys/arch/i386/include/cpu.h
index d68a669ac4e..2fe58c1d2a5 100644
--- a/sys/arch/i386/include/cpu.h
+++ b/sys/arch/i386/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.126 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.127 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.35 1996/05/05 19:29:26 christos Exp $	*/
 
 /*-
@@ -148,9 +148,6 @@ struct cpu_info {
 
 	struct ksensordev	ci_sensordev;
 	struct ksensor		ci_sensor;
-#ifdef GPROF
-	struct gmonparam	*ci_gmon;
-#endif
 };
 
 /*
diff --git a/sys/arch/m68k/include/cpu.h b/sys/arch/m68k/include/cpu.h
index 723e39c1693..bac5eedc662 100644
--- a/sys/arch/m68k/include/cpu.h
+++ b/sys/arch/m68k/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.27 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.28 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.3 1997/02/02 06:56:57 thorpej Exp $	*/
 
 /*
@@ -59,9 +59,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 extern struct cpu_info cpu_info_store;
diff --git a/sys/arch/m88k/include/cpu.h b/sys/arch/m88k/include/cpu.h
index 4da10bede57..f5055a5463e 100644
--- a/sys/arch/m88k/include/cpu.h
+++ b/sys/arch/m88k/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.57 2013/02/11 17:05:25 mpi Exp $ */
+/*	$OpenBSD: cpu.h,v 1.58 2013/02/12 08:06:22 mpi Exp $ */
 /*
  * Copyright (c) 1996 Nivas Madhur
  * Copyright (c) 1992, 1993
@@ -172,9 +172,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 extern cpuid_t master_cpu;
diff --git a/sys/arch/mips64/include/cpu.h b/sys/arch/mips64/include/cpu.h
index 1bb6fabc04b..c7ba15de925 100644
--- a/sys/arch/mips64/include/cpu.h
+++ b/sys/arch/mips64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.92 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.93 2013/02/12 08:06:22 mpi Exp $	*/
 
 /*-
  * Copyright (c) 1992, 1993
@@ -198,9 +198,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 #define	CPUF_PRIMARY	0x01		/* CPU is primary CPU */
diff --git a/sys/arch/powerpc/include/cpu.h b/sys/arch/powerpc/include/cpu.h
index 7715195bde4..a9561d6e820 100644
--- a/sys/arch/powerpc/include/cpu.h
+++ b/sys/arch/powerpc/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.49 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.50 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 1996/09/30 16:34:21 ws Exp $	*/
 
 /*
@@ -85,9 +85,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 static __inline struct cpu_info *
diff --git a/sys/arch/sh/include/cpu.h b/sys/arch/sh/include/cpu.h
index ce6a69da2b6..e313d701d22 100644
--- a/sys/arch/sh/include/cpu.h
+++ b/sys/arch/sh/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.23 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.24 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.41 2006/01/21 04:24:12 uwe Exp $	*/
 
 /*-
@@ -65,9 +65,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 extern struct cpu_info cpu_info_store;
diff --git a/sys/arch/sparc64/include/cpu.h b/sys/arch/sparc64/include/cpu.h
index b9e6115098a..7b5e4fee0a0 100644
--- a/sys/arch/sparc64/include/cpu.h
+++ b/sys/arch/sparc64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.80 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.81 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.28 2001/06/14 22:56:58 thorpej Exp $ */
 
 /*
@@ -156,9 +156,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 #define CPUF_RUNNING	0x0001		/* CPU is running */
diff --git a/sys/arch/vax/include/cpu.h b/sys/arch/vax/include/cpu.h
index f8f6ba281c2..509556a3358 100644
--- a/sys/arch/vax/include/cpu.h
+++ b/sys/arch/vax/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.43 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.44 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: cpu.h,v 1.41 1999/10/21 20:01:36 ragge Exp $	*/
 
 /*
@@ -55,9 +55,6 @@ struct cpu_info {
 #ifdef DIAGNOSTIC
 	int	ci_mutex_level;
 #endif
-#ifdef GPROF
-	struct gmonparam *ci_gmon;
-#endif
 };
 
 extern struct cpu_info cpu_info_store;
diff --git a/sys/kern/kern_clock.c b/sys/kern/kern_clock.c
index 57220881f03..d7d51646c2c 100644
--- a/sys/kern/kern_clock.c
+++ b/sys/kern/kern_clock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_clock.c,v 1.77 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: kern_clock.c,v 1.78 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: kern_clock.c,v 1.34 1996/06/09 04:51:03 briggs Exp $	*/
 
 /*-
@@ -421,8 +421,8 @@ statclock(struct clockframe *frame)
 		/*
 		 * Kernel statistics are just like addupc_intr, only easier.
 		 */
-		g = ci->ci_gmon;
-		if (g != NULL && g->state == GMON_PROF_ON) {
+		g = &_gmonparam;
+		if (g->state == GMON_PROF_ON) {
 			i = CLKF_PC(frame) - g->lowpc;
 			if (i < g->textsize) {
 				i /= HISTFRACTION * sizeof(*g->kcount);
diff --git a/sys/kern/subr_prof.c b/sys/kern/subr_prof.c
index a7eff46dc2f..3786cb11df8 100644
--- a/sys/kern/subr_prof.c
+++ b/sys/kern/subr_prof.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: subr_prof.c,v 1.22 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: subr_prof.c,v 1.23 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: subr_prof.c,v 1.12 1996/04/22 01:38:50 christos Exp $	*/
 
 /*-
@@ -49,75 +49,49 @@
 #include <uvm/uvm_extern.h>
 
 /*
- * Protect CPUs from executing profiling while they are not yet in a
- * sane state.
+ * Froms is actually a bunch of unsigned shorts indexing tos
  */
-int gmoninit = 0;
+struct gmonparam _gmonparam = { GMON_PROF_OFF };
 
 extern char etext[];
 
-#define ROUNDDOWN(x,y)	(((x)/(y))*(y))
-#define ROUNDUP(x,y)	((((x)+(y)-1)/(y))*(y))
 
 void
 kmstartup(void)
 {
-	CPU_INFO_ITERATOR cii;
-	struct cpu_info *ci;
-	struct gmonparam *p;
-	u_long lowpc, highpc, textsize;
-	u_long kcountsize, fromssize, tossize;
-	long tolimit;
 	char *cp;
+	struct gmonparam *p = &_gmonparam;
 	int size;
 
 	/*
 	 * Round lowpc and highpc to multiples of the density we're using
 	 * so the rest of the scaling (here and in gprof) stays in ints.
 	 */
-	lowpc = ROUNDDOWN(KERNBASE, HISTFRACTION * sizeof(HISTCOUNTER));
-	highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER));
-	textsize = highpc - lowpc;
+	p->lowpc = ROUNDDOWN(KERNBASE, HISTFRACTION * sizeof(HISTCOUNTER));
+	p->highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER));
+	p->textsize = p->highpc - p->lowpc;
 	printf("Profiling kernel, textsize=%ld [%lx..%lx]\n",
-	    textsize, lowpc, highpc);
-	kcountsize = textsize / HISTFRACTION;
-	fromssize = textsize / HASHFRACTION;
-	tolimit = textsize * ARCDENSITY / 100;
-	if (tolimit < MINARCS)
-		tolimit = MINARCS;
-	else if (tolimit > MAXARCS)
-		tolimit = MAXARCS;
-	tossize = tolimit * sizeof(struct tostruct);
-	size = sizeof(*p) + kcountsize + fromssize + tossize;
-
-	/* Allocate and initialize one profiling buffer per CPU. */
-	CPU_INFO_FOREACH(cii, ci) {
-		cp = km_alloc(round_page(size), &kv_any, &kp_zero, &kd_nowait);
-		if (cp == NULL) {
-			printf("No memory for profiling.\n");
-			return;
-		}
-
-		p = (struct gmonparam *)cp;
-		cp += sizeof(*p);
-		p->tos = (struct tostruct *)cp;
-		cp += tossize;
-		p->kcount = (u_short *)cp;
-		cp += kcountsize;
-		p->froms = (u_short *)cp;
-
-		p->state = GMON_PROF_OFF;
-		p->lowpc = lowpc;
-		p->highpc = highpc;
-		p->textsize = textsize;
-		p->hashfraction = HASHFRACTION;
-		p->kcountsize = kcountsize;
-		p->fromssize = fromssize;
-		p->tolimit = tolimit;
-		p->tossize = tossize;
-
-		ci->ci_gmon = p;
+	       p->textsize, p->lowpc, p->highpc);
+	p->kcountsize = p->textsize / HISTFRACTION;
+	p->hashfraction = HASHFRACTION;
+	p->fromssize = p->textsize / HASHFRACTION;
+	p->tolimit = p->textsize * ARCDENSITY / 100;
+	if (p->tolimit < MINARCS)
+		p->tolimit = MINARCS;
+	else if (p->tolimit > MAXARCS)
+		p->tolimit = MAXARCS;
+	p->tossize = p->tolimit * sizeof(struct tostruct);
+	size = p->kcountsize + p->fromssize + p->tossize;
+	cp = (char *)uvm_km_zalloc(kernel_map, round_page(size));
+	if (cp == 0) {
+		printf("No memory for profiling.\n");
+		return;
 	}
+	p->tos = (struct tostruct *)cp;
+	cp += p->tossize;
+	p->kcount = (u_short *)cp;
+	cp += p->kcountsize;
+	p->froms = (u_short *)cp;
 }
 
 /*
@@ -127,32 +101,14 @@ int
 sysctl_doprof(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
     size_t newlen)
 {
-	CPU_INFO_ITERATOR cii;
-	struct cpu_info *ci;
-	struct gmonparam *gp = NULL;
-	int error, cpuid, op;
+	struct gmonparam *gp = &_gmonparam;
+	int error;
 
-	/* all sysctl names at this level are name and field */
-	if (namelen != 2)
+	/* all sysctl names at this level are terminal */
+	if (namelen != 1)
 		return (ENOTDIR);		/* overloaded */
 
-	op = name[0];
-	cpuid = name[1];
-
-	CPU_INFO_FOREACH(cii, ci) {
-		if (cpuid == CPU_INFO_UNIT(ci)) {
-			gp = ci->ci_gmon;
-			break;
-		}
-	}
-
-	if (gp == NULL)
-		return (EOPNOTSUPP);
-
-	/* Assume that if we're here it is safe to execute profiling. */
-	gmoninit = 1;
-
-	switch (op) {
+	switch (name[0]) {
 	case GPROF_STATE:
 		error = sysctl_int(oldp, oldlenp, newp, newlen, &gp->state);
 		if (error)
diff --git a/sys/lib/libkern/mcount.c b/sys/lib/libkern/mcount.c
index bcb5c1286bd..c66314cefe0 100644
--- a/sys/lib/libkern/mcount.c
+++ b/sys/lib/libkern/mcount.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mcount.c,v 1.11 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: mcount.c,v 1.12 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: mcount.c,v 1.3.6.1 1996/06/12 04:23:01 cgd Exp $	*/
 
 /*-
@@ -43,10 +43,13 @@
  * _mcount updates data structures that represent traversals of the
  * program's call graph edges.  frompc and selfpc are the return
  * address and function address that represents the given call graph edge.
+ * 
+ * Note: the original BSD code used the same variable (frompcindex) for
+ * both frompcindex and frompc.  Any reasonable, modern compiler will
+ * perform this optimization.
  */
 _MCOUNT_DECL(u_long frompc, u_long selfpc) __used;
-/* _mcount; may be static, inline, etc */
-_MCOUNT_DECL(u_long frompc, u_long selfpc)
+_MCOUNT_DECL(u_long frompc, u_long selfpc)	/* _mcount; may be static, inline, etc */
 {
 	u_short *frompcindex;
 	struct tostruct *top, *prevtop;
@@ -54,21 +57,9 @@ _MCOUNT_DECL(u_long frompc, u_long selfpc)
 	long toindex;
 #ifdef _KERNEL
 	int s;
+#endif
 
-	/*
-	 * Do not profile execution if memory for the current CPU
-	 * desciptor and profiling buffers has not yet been allocated
-	 * or if the CPU we are running on has not yet set its trap
-	 * handler.
-	 */
-	if (gmoninit == 0)
-		return;
-
-	if ((p = curcpu()->ci_gmon) == NULL)
-		return;
-#else
 	p = &_gmonparam;
-#endif
 	/*
 	 * check that we are profiling
 	 * and that we aren't recursively invoked.
@@ -165,6 +156,7 @@ _MCOUNT_DECL(u_long frompc, u_long selfpc)
 			*frompcindex = toindex;
 			goto done;
 		}
+		
 	}
 done:
 #ifdef _KERNEL
diff --git a/sys/sys/gmon.h b/sys/sys/gmon.h
index 0eaa36490d4..192b7c7ebbc 100644
--- a/sys/sys/gmon.h
+++ b/sys/sys/gmon.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gmon.h,v 1.5 2013/02/11 17:05:25 mpi Exp $	*/
+/*	$OpenBSD: gmon.h,v 1.6 2013/02/12 08:06:22 mpi Exp $	*/
 /*	$NetBSD: gmon.h,v 1.5 1996/04/09 20:55:30 cgd Exp $	*/
 
 /*-
@@ -115,6 +115,12 @@ struct rawarc {
 };
 
 /*
+ * general rounding functions.
+ */
+#define ROUNDDOWN(x,y)	(((x)/(y))*(y))
+#define ROUNDUP(x,y)	((((x)+(y)-1)/(y))*(y))
+
+/*
  * The profiling data structures are housed in this structure.
  */
 struct gmonparam {
@@ -131,19 +137,15 @@ struct gmonparam {
 	u_long		textsize;
 	u_long		hashfraction;
 };
-#ifdef _KERNEL
-extern int gmoninit;		/* Is the kernel ready for beeing profiled? */
-#else
 extern struct gmonparam _gmonparam;
-#endif
 
 /*
  * Possible states of profiling.
  */
-#define	GMON_PROF_OFF	0
+#define	GMON_PROF_ON	0
 #define	GMON_PROF_BUSY	1
 #define	GMON_PROF_ERROR	2
-#define	GMON_PROF_ON	3
+#define	GMON_PROF_OFF	3
 
 /*
  * Sysctl definitions for extracting profiling information from the kernel.
diff --git a/usr.sbin/kgmon/kgmon.8 b/usr.sbin/kgmon/kgmon.8
index 5518a1aabb3..6f12062f5c5 100644
--- a/usr.sbin/kgmon/kgmon.8
+++ b/usr.sbin/kgmon/kgmon.8
@@ -1,4 +1,4 @@
-.\" $OpenBSD: kgmon.8,v 1.11 2013/02/11 17:32:33 jmc Exp $
+.\" $OpenBSD: kgmon.8,v 1.12 2013/02/12 08:06:22 mpi Exp $
 .\" Copyright (c) 1983, 1991, 1993
 .\"	The Regents of the University of California.  All rights reserved.
 .\"
@@ -27,9 +27,9 @@
 .\" SUCH DAMAGE.
 .\"
 .\"     from: @(#)kgmon.8	8.1 (Berkeley) 6/6/93
-.\"	$Id: kgmon.8,v 1.11 2013/02/11 17:32:33 jmc Exp $
+.\"	$Id: kgmon.8,v 1.12 2013/02/12 08:06:22 mpi Exp $
 .\"
-.Dd $Mdocdate: February 11 2013 $
+.Dd $Mdocdate: February 12 2013 $
 .Dt KGMON 8
 .Os
 .Sh NAME
@@ -38,7 +38,6 @@
 .Sh SYNOPSIS
 .Nm kgmon
 .Op Fl bhpr
-.Op Fl c Ar cpuid
 .Op Fl M Ar core
 .Op Fl N Ar system
 .Sh DESCRIPTION
@@ -46,7 +45,7 @@
 is a tool used when profiling the operating system.
 When no arguments are supplied,
 .Nm kgmon
-indicates the state of per-CPU operating system profilings as
+indicates the state of operating system profiling as
 .Dq running ,
 .Dq off ,
 or
@@ -118,8 +117,10 @@ Users with only read permission on
 .Pa /dev/kmem
 cannot change the state
 of profiling collection.
-They can get profile files with the warning that the data may be inconsistent
-if profiling is in progress.
+They can get a
+.Pa gmon.out
+file with the warning that the data may be
+inconsistent if profiling is in progress.
 .Sh SEE ALSO
 .Xr gprof 1 ,
 .Xr config 8
diff --git a/usr.sbin/kgmon/kgmon.c b/usr.sbin/kgmon/kgmon.c
index 887a8c84368..82b9700f98f 100644
--- a/usr.sbin/kgmon/kgmon.c
+++ b/usr.sbin/kgmon/kgmon.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kgmon.c,v 1.16 2013/02/11 17:08:53 mpi Exp $	*/
+/*	$OpenBSD: kgmon.c,v 1.17 2013/02/12 08:06:22 mpi Exp $	*/
 
 /*
  * Copyright (c) 1983, 1992, 1993
@@ -58,32 +58,28 @@ struct kvmvars {
 	struct gmonparam gpm;
 };
 
-extern char *__progname;
-
-int	bflag, cflag, hflag, kflag, rflag, pflag;
+int	bflag, hflag, kflag, rflag, pflag;
 int	debug = 0;
-void	kgmon(char *, char *, struct kvmvars *, int);
-void	setprof(struct kvmvars *, int, int);
-void	dumpstate(struct kvmvars *, int);
-void	reset(struct kvmvars *, int);
+void	setprof(struct kvmvars *, int);
+void	dumpstate(struct kvmvars *);
+void	reset(struct kvmvars *);
 void	kern_readonly(int);
-int	getprof(struct kvmvars *, int);
+int	getprof(struct kvmvars *);
 int	getprofhz(struct kvmvars *);
-int	openfiles(char *, char *, struct kvmvars *, int);
-int	getncpu(void);
+int	openfiles(char *, char *, struct kvmvars *);
 
 int
 main(int argc, char **argv)
 {
-	int ch, err, ncpu, cpuid = -1;
+	extern char *__progname;
+	int ch, mode, disp, accessmode;
 	struct kvmvars kvmvars;
 	char *sys, *kmemf;
-	const char *p;
 
 	seteuid(getuid());
 	kmemf = NULL;
 	sys = NULL;
-	while ((ch = getopt(argc, argv, "M:N:bc:hpr")) != -1) {
+	while ((ch = getopt(argc, argv, "M:N:bhpr")) != -1) {
 		switch((char)ch) {
 
 		case 'M':
@@ -99,13 +95,6 @@ main(int argc, char **argv)
 			bflag = 1;
 			break;
 
-		case 'c':
-			cflag = 1;
-			cpuid = strtonum(optarg, 0, 1024, &p);
-			if (p)
-				errx(1, "illegal CPU id %s: %s", optarg, p);
-			break;
-
 		case 'h':
 			hflag = 1;
 			break;
@@ -119,8 +108,9 @@ main(int argc, char **argv)
 			break;
 
 		default:
-			fprintf(stderr, "usage: %s [-bhpr] "
-			    "[-c cpuid] [-M core] [-N system]\n", __progname);
+			fprintf(stderr,
+			    "usage: %s [-bhpr] [-M core] [-N system]\n",
+			    __progname);
 			exit(1);
 		}
 	}
@@ -137,25 +127,8 @@ main(int argc, char **argv)
 		}
 	}
 #endif
-
-	if (cflag) {
-		kgmon(sys, kmemf, &kvmvars, cpuid);
-	} else {
-		ncpu = getncpu();
-		for (cpuid = 0; cpuid < ncpu; cpuid++)
-			kgmon(sys, kmemf, &kvmvars, cpuid);
-	}
-
-	return (0);
-}
-
-void
-kgmon(char *sys, char *kmemf, struct kvmvars *kvp, int cpuid)
-{
-	int mode, disp, accessmode;
-
-	accessmode = openfiles(sys, kmemf, kvp, cpuid);
-	mode = getprof(kvp, cpuid);
+	accessmode = openfiles(sys, kmemf, &kvmvars);
+	mode = getprof(&kvmvars);
 	if (hflag)
 		disp = GMON_PROF_OFF;
 	else if (bflag)
@@ -163,22 +136,23 @@ kgmon(char *sys, char *kmemf, struct kvmvars *kvp, int cpuid)
 	else
 		disp = mode;
 	if (pflag)
-		dumpstate(kvp, cpuid);
+		dumpstate(&kvmvars);
 	if (rflag)
-		reset(kvp, cpuid);
+		reset(&kvmvars);
 	if (accessmode == O_RDWR)
-		setprof(kvp, cpuid, disp);
-	printf("%s: kernel profiling is %s for cpu %d.\n", __progname,
-	    disp == GMON_PROF_OFF ? "off" : "running", cpuid);
+		setprof(&kvmvars, disp);
+	printf("%s: kernel profiling is %s.\n", __progname,
+	    disp == GMON_PROF_OFF ? "off" : "running");
+	return (0);
 }
 
 /*
  * Check that profiling is enabled and open any ncessary files.
  */
 int
-openfiles(char *sys, char *kmemf, struct kvmvars *kvp, int cpuid)
+openfiles(char *sys, char *kmemf, struct kvmvars *kvp)
 {
-	int mib[4], state, openmode;
+	int mib[3], state, openmode;
 	size_t size;
 	char errbuf[_POSIX2_LINE_MAX];
 
@@ -186,15 +160,14 @@ openfiles(char *sys, char *kmemf, struct kvmvars *kvp, int cpuid)
 		mib[0] = CTL_KERN;
 		mib[1] = KERN_PROF;
 		mib[2] = GPROF_STATE;
-		mib[3] = cpuid;
 		size = sizeof state;
-		if (sysctl(mib, 4, &state, &size, NULL, 0) < 0)
+		if (sysctl(mib, 3, &state, &size, NULL, 0) < 0)
 			errx(20, "profiling not defined in kernel.");
 		if (!(bflag || hflag || rflag ||
 		    (pflag && state == GMON_PROF_ON)))
 			return (O_RDONLY);
 		(void)seteuid(0);
-		if (sysctl(mib, 4, NULL, NULL, &state, size) >= 0)
+		if (sysctl(mib, 3, NULL, NULL, &state, size) >= 0)
 			return (O_RDWR);
 		(void)seteuid(getuid());
 		kern_readonly(state);
@@ -243,9 +216,9 @@ kern_readonly(int mode)
  * Get the state of kernel profiling.
  */
 int
-getprof(struct kvmvars *kvp, int cpuid)
+getprof(struct kvmvars *kvp)
 {
-	int mib[4];
+	int mib[3];
 	size_t size;
 
 	if (kflag) {
@@ -255,9 +228,8 @@ getprof(struct kvmvars *kvp, int cpuid)
 		mib[0] = CTL_KERN;
 		mib[1] = KERN_PROF;
 		mib[2] = GPROF_GMONPARAM;
-		mib[3] = cpuid;
 		size = sizeof kvp->gpm;
-		if (sysctl(mib, 4, &kvp->gpm, &size, NULL, 0) < 0)
+		if (sysctl(mib, 3, &kvp->gpm, &size, NULL, 0) < 0)
 			size = 0;
 	}
 	if (size != sizeof kvp->gpm)
@@ -270,10 +242,10 @@ getprof(struct kvmvars *kvp, int cpuid)
  * Enable or disable kernel profiling according to the state variable.
  */
 void
-setprof(struct kvmvars *kvp, int cpuid, int state)
+setprof(struct kvmvars *kvp, int state)
 {
 	struct gmonparam *p = (struct gmonparam *)nl[N_GMONPARAM].n_value;
-	int mib[4], oldstate;
+	int mib[3], oldstate;
 	size_t sz;
 
 	sz = sizeof(state);
@@ -281,13 +253,12 @@ setprof(struct kvmvars *kvp, int cpuid, int state)
 		mib[0] = CTL_KERN;
 		mib[1] = KERN_PROF;
 		mib[2] = GPROF_STATE;
-		mib[3] = cpuid;
-		if (sysctl(mib, 4, &oldstate, &sz, NULL, 0) < 0)
+		if (sysctl(mib, 3, &oldstate, &sz, NULL, 0) < 0)
 			goto bad;
 		if (oldstate == state)
 			return;
 		(void)seteuid(0);
-		if (sysctl(mib, 4, NULL, NULL, &state, sz) >= 0) {
+		if (sysctl(mib, 3, NULL, NULL, &state, sz) >= 0) {
 			(void)seteuid(getuid());
 			return;
 		}
@@ -304,25 +275,22 @@ bad:
  * Build the gmon.out file.
  */
 void
-dumpstate(struct kvmvars *kvp, int cpuid)
+dumpstate(struct kvmvars *kvp)
 {
 	FILE *fp;
 	struct rawarc rawarc;
 	struct tostruct *tos;
 	u_long frompc;
 	u_short *froms, *tickbuf;
-	int mib[4];
+	int mib[3];
 	size_t i;
 	struct gmonhdr h;
 	int fromindex, endfrom, toindex;
-	char buf[16];
 
-	snprintf(buf, sizeof(buf), "gmon-%02d.out", cpuid);
-
-	setprof(kvp, cpuid, GMON_PROF_OFF);
-	fp = fopen(buf, "w");
+	setprof(kvp, GMON_PROF_OFF);
+	fp = fopen("gmon.out", "w");
 	if (fp == 0) {
-		perror(buf);
+		perror("gmon.out");
 		return;
 	}
 
@@ -349,9 +317,8 @@ dumpstate(struct kvmvars *kvp, int cpuid)
 		    kvp->gpm.kcountsize);
 	} else {
 		mib[2] = GPROF_COUNT;
-		mib[3] = cpuid;
 		i = kvp->gpm.kcountsize;
-		if (sysctl(mib, 4, tickbuf, &i, NULL, 0) < 0)
+		if (sysctl(mib, 3, tickbuf, &i, NULL, 0) < 0)
 			i = 0;
 	}
 	if (i != kvp->gpm.kcountsize)
@@ -372,9 +339,8 @@ dumpstate(struct kvmvars *kvp, int cpuid)
 		    kvp->gpm.fromssize);
 	} else {
 		mib[2] = GPROF_FROMS;
-		mib[3] = cpuid;
 		i = kvp->gpm.fromssize;
-		if (sysctl(mib, 4, froms, &i, NULL, 0) < 0)
+		if (sysctl(mib, 3, froms, &i, NULL, 0) < 0)
 			i = 0;
 	}
 	if (i != kvp->gpm.fromssize)
@@ -388,9 +354,8 @@ dumpstate(struct kvmvars *kvp, int cpuid)
 		    kvp->gpm.tossize);
 	} else {
 		mib[2] = GPROF_TOS;
-		mib[3] = cpuid;
 		i = kvp->gpm.tossize;
-		if (sysctl(mib, 4, tos, &i, NULL, 0) < 0)
+		if (sysctl(mib, 3, tos, &i, NULL, 0) < 0)
 			i = 0;
 	}
 	if (i != kvp->gpm.tossize)
@@ -450,13 +415,13 @@ getprofhz(struct kvmvars *kvp)
  * Reset the kernel profiling date structures.
  */
 void
-reset(struct kvmvars *kvp, int cpuid)
+reset(struct kvmvars *kvp)
 {
 	char *zbuf;
 	u_long biggest;
-	int mib[4];
+	int mib[3];
 
-	setprof(kvp, cpuid, GMON_PROF_OFF);
+	setprof(kvp, GMON_PROF_OFF);
 
 	biggest = kvp->gpm.kcountsize;
 	if (kvp->gpm.fromssize > biggest)
@@ -482,31 +447,14 @@ reset(struct kvmvars *kvp, int cpuid)
 	mib[0] = CTL_KERN;
 	mib[1] = KERN_PROF;
 	mib[2] = GPROF_COUNT;
-	mib[3] = cpuid;
-	if (sysctl(mib, 4, NULL, NULL, zbuf, kvp->gpm.kcountsize) < 0)
+	if (sysctl(mib, 3, NULL, NULL, zbuf, kvp->gpm.kcountsize) < 0)
 		err(13, "tickbuf zero");
 	mib[2] = GPROF_FROMS;
-	if (sysctl(mib, 4, NULL, NULL, zbuf, kvp->gpm.fromssize) < 0)
+	if (sysctl(mib, 3, NULL, NULL, zbuf, kvp->gpm.fromssize) < 0)
 		err(14, "froms zero");
 	mib[2] = GPROF_TOS;
-	if (sysctl(mib, 4, NULL, NULL, zbuf, kvp->gpm.tossize) < 0)
+	if (sysctl(mib, 3, NULL, NULL, zbuf, kvp->gpm.tossize) < 0)
 		err(15, "tos zero");
 	(void)seteuid(getuid());
 	free(zbuf);
 }
-
-int
-getncpu(void)
-{
-	int mib[2] = { CTL_HW, HW_NCPU };
-	size_t size;
-	int ncpu;
-
-	size = sizeof(ncpu);
-	if (sysctl(mib, 2, &ncpu, &size, NULL, 0) < 0) {
-		warnx("cannot read hw.ncpu");
-		return (1);
-	}
-
-	return (ncpu);
-}