debranch SMP, have fun

144 files changed, 10647 insertions, 1668 deletions

diff --git a/sys/arch/alpha/alpha/cpu.c b/sys/arch/alpha/alpha/cpu.c
index a8045be3246..6d5c8a9d347 100644
--- a/sys/arch/alpha/alpha/cpu.c
+++ b/sys/arch/alpha/alpha/cpu.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: cpu.c,v 1.18 2004/06/08 20:13:21 miod Exp $ */
+/* $OpenBSD: cpu.c,v 1.19 2004/06/13 21:49:11 niklas Exp $ */
 /* $NetBSD: cpu.c,v 1.44 2000/05/23 05:12:53 thorpej Exp $ */
 
 /*-
@@ -565,7 +565,8 @@ cpu_iccb_send(cpu_id, msg)
 	strlcpy(pcsp->pcs_iccb.iccb_rxbuf, msg,
 	    sizeof pcsp->pcs_iccb.iccb_rxbuf);
 	pcsp->pcs_iccb.iccb_rxlen = strlen(msg);
-	atomic_setbits_ulong(&hwrpb->rpb_rxrdy, cpumask);
+	/* XXX cast to __volatile */
+	atomic_setbits_ulong((__volatile u_long *)&hwrpb->rpb_rxrdy, cpumask);
 
 	/* Wait for the message to be received. */
 	for (timeout = 10000; timeout != 0; timeout--) {
diff --git a/sys/arch/alpha/alpha/genassym.cf b/sys/arch/alpha/alpha/genassym.cf
index 27d0faa330b..2a8802256f3 100644
--- a/sys/arch/alpha/alpha/genassym.cf
+++ b/sys/arch/alpha/alpha/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.5 2003/06/02 23:27:43 millert Exp $
+#	$OpenBSD: genassym.cf,v 1.6 2004/06/13 21:49:11 niklas Exp $
 
 # Copyright (c) 1994, 1995 Gordon W. Ross
 # Copyright (c) 1993 Adam Glass
@@ -120,6 +120,8 @@ struct	prochd
 member	ph_link
 member	ph_rlink
 
+export	SONPROC
+
 # offsets needed by cpu_switch() to switch mappings.
 define	VM_MAP_PMAP		offsetof(struct vmspace, vm_map.pmap) 
 
diff --git a/sys/arch/alpha/alpha/locore.s b/sys/arch/alpha/alpha/locore.s
index d30e1f646f1..48849130727 100644
--- a/sys/arch/alpha/alpha/locore.s
+++ b/sys/arch/alpha/alpha/locore.s
@@ -1,4 +1,4 @@
-/* $OpenBSD: locore.s,v 1.21 2003/10/18 20:14:40 jmc Exp $ */
+/* $OpenBSD: locore.s,v 1.22 2004/06/13 21:49:11 niklas Exp $ */
 /* $NetBSD: locore.s,v 1.94 2001/04/26 03:10:44 ross Exp $ */
 
 /*-
@@ -891,7 +891,6 @@ cpu_switch_queuescan:
 	 *
 	 * Note: GET_CPUINFO clobbers v0, t0, t8...t11.
 	 */
-#if 0
 #ifdef __alpha_bwx__
 	ldiq	t0, SONPROC			/* p->p_stat = SONPROC */
 	stb	t0, P_STAT(s2)
@@ -904,7 +903,6 @@ cpu_switch_queuescan:
 	or	t0, t1, t0
 	stq_u	t0, 0(t3)
 #endif /* __alpha_bwx__ */
-#endif
 
 	GET_CPUINFO
 	/* p->p_cpu initialized in fork1() for single-processor */
diff --git a/sys/arch/alpha/alpha/pmap.c b/sys/arch/alpha/alpha/pmap.c
index d8860e84b38..484eb661c93 100644
--- a/sys/arch/alpha/alpha/pmap.c
+++ b/sys/arch/alpha/alpha/pmap.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: pmap.c,v 1.41 2004/06/08 20:13:21 miod Exp $ */
+/* $OpenBSD: pmap.c,v 1.42 2004/06/13 21:49:11 niklas Exp $ */
 /* $NetBSD: pmap.c,v 1.154 2000/12/07 22:18:55 thorpej Exp $ */
 
 /*-
@@ -388,11 +388,6 @@ struct lock pmap_main_lock;
 struct simplelock pmap_all_pmaps_slock;
 struct simplelock pmap_growkernel_slock;
 
-#ifdef __OpenBSD__
-#define spinlockinit(lock, name, flags)  lockinit(lock, 0, name, 0, flags)
-#define spinlockmgr(lock, flags, slock) lockmgr(lock, flags, slock, curproc)
-#endif
-
 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
 #define	PMAP_MAP_TO_HEAD_LOCK() \
 	spinlockmgr(&pmap_main_lock, LK_SHARED, NULL)
diff --git a/sys/arch/alpha/include/cpu.h b/sys/arch/alpha/include/cpu.h
index 7bfc81c4624..c793d217035 100644
--- a/sys/arch/alpha/include/cpu.h
+++ b/sys/arch/alpha/include/cpu.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: cpu.h,v 1.24 2004/06/08 20:13:23 miod Exp $ */
+/* $OpenBSD: cpu.h,v 1.25 2004/06/13 21:49:12 niklas Exp $ */
 /* $NetBSD: cpu.h,v 1.45 2000/08/21 02:03:12 thorpej Exp $ */
 
 /*-
@@ -105,6 +105,7 @@ typedef union alpha_t_float {
 #ifdef _KERNEL
 
 #include <machine/bus.h>
+#include <sys/device.h>
 
 struct pcb;
 struct proc;
@@ -177,6 +178,7 @@ struct mchkinfo {
 };
 
 struct cpu_info {
+	struct device *ci_dev;		/* pointer to our device */
 	/*
 	 * Public members.
 	 */
@@ -185,12 +187,13 @@ struct cpu_info {
 	u_long ci_simple_locks;		/* # of simple locks held */
 #endif
 	struct proc *ci_curproc;	/* current owner of the processor */
+	struct simplelock ci_slock;	/* lock on this data structure */
+	cpuid_t ci_cpuid;		/* our CPU ID */
 
 	/*
 	 * Private members.
 	 */
 	struct mchkinfo ci_mcinfo;	/* machine check info */
-	cpuid_t ci_cpuid;		/* our CPU ID */
 	struct proc *ci_fpcurproc;	/* current owner of the FPU */
 	paddr_t ci_curpcb;		/* PA of current HW PCB */
 	struct pcb *ci_idle_pcb;	/* our idle PCB */
@@ -265,15 +268,20 @@ struct clockframe {
 /*
  * Preempt the current process if in interrupt from user mode,
  * or after the current trap/syscall if in system mode.
- *
- * XXXSMP
- * need_resched() needs to take a cpu_info *.
  */
-#define	need_resched()							\
+#ifdef MULTIPROCESSOR
+#define	need_resched(ci)						\
+do {									\
+	ci->ci_want_resched = 1;					\
+	aston(curcpu());						\
+} while (/*CONSTCOND*/0)
+#else
+#define	need_resched(ci)						\
 do {									\
 	curcpu()->ci_want_resched = 1;					\
 	aston(curcpu());						\
 } while (/*CONSTCOND*/0)
+#endif
 
 /*
  * Give a profiling tick to the current process when the user profiling
diff --git a/sys/arch/alpha/include/intr.h b/sys/arch/alpha/include/intr.h
index a5813c6e1bf..25fdcba079d 100644
--- a/sys/arch/alpha/include/intr.h
+++ b/sys/arch/alpha/include/intr.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: intr.h,v 1.16 2004/04/16 04:52:05 pvalchev Exp $ */
+/* $OpenBSD: intr.h,v 1.17 2004/06/13 21:49:12 niklas Exp $ */
 /* $NetBSD: intr.h,v 1.26 2000/06/03 20:47:41 thorpej Exp $ */
 
 /*-
@@ -159,6 +159,7 @@ extern	ipifunc_t ipifuncs[ALPHA_NIPIS];
 
 void	alpha_send_ipi(unsigned long, unsigned long);
 void	alpha_broadcast_ipi(unsigned long);
+void	alpha_multicast_ipi(unsigned long, unsigned long);
 
 /*
  * Alpha shared-interrupt-line common code.
diff --git a/sys/arch/alpha/include/pmap.h b/sys/arch/alpha/include/pmap.h
index e61bba6497d..1a5c4784915 100644
--- a/sys/arch/alpha/include/pmap.h
+++ b/sys/arch/alpha/include/pmap.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: pmap.h,v 1.17 2004/06/09 20:17:23 tedu Exp $ */
+/* $OpenBSD: pmap.h,v 1.18 2004/06/13 21:49:12 niklas Exp $ */
 /* $NetBSD: pmap.h,v 1.37 2000/11/19 03:16:35 thorpej Exp $ */
 
 /*-
@@ -178,7 +178,7 @@ struct pv_head {
 
 #if defined(MULTIPROCESSOR)
 void	pmap_tlb_shootdown(pmap_t, vaddr_t, pt_entry_t);
-void	pmap_do_tlb_shootdown(void);
+void	pmap_do_tlb_shootdown(struct cpu_info *, struct trapframe *);
 void	pmap_tlb_shootdown_q_drain(u_long, boolean_t);
 #define	PMAP_TLB_SHOOTDOWN(pm, va, pte)					\
 	pmap_tlb_shootdown((pm), (va), (pte))
diff --git a/sys/arch/amd64/amd64/genassym.cf b/sys/arch/amd64/amd64/genassym.cf
index de6ab55ca81..a56769705b4 100644
--- a/sys/arch/amd64/amd64/genassym.cf
+++ b/sys/arch/amd64/amd64/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.3 2004/05/13 20:20:24 sturm Exp $
+#	$OpenBSD: genassym.cf,v 1.4 2004/06/13 21:49:12 niklas Exp $
 
 #	Written by Artur Grabowski art@openbsd.org, Public Domain
 
@@ -17,6 +17,7 @@ include <machine/vmparam.h>
 include <machine/intr.h>
 
 export	SRUN
+export	SONPROC
 
 export	L4_SLOT_KERNBASE
 define	L3_SLOT_KERNBASE	pl3_pi(KERNBASE)
diff --git a/sys/arch/amd64/amd64/locore.S b/sys/arch/amd64/amd64/locore.S
index a61e1719e61..732eea0a0df 100644
--- a/sys/arch/amd64/amd64/locore.S
+++ b/sys/arch/amd64/amd64/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.10 2004/05/13 20:20:24 sturm Exp $	*/
+/*	$OpenBSD: locore.S,v 1.11 2004/06/13 21:49:12 niklas Exp $	*/
 /*	$NetBSD: locore.S,v 1.2 2003/04/26 19:34:45 fvdl Exp $	*/
 
 /*
@@ -848,9 +848,7 @@ switch_resume:
 	movq	%rax,P_BACK(%r12)
 
 	/* Record new proc. */
-#ifdef MULTIPROCESSOR
-	movb	$SONPROC,P_STAT(%r12)	# l->l_stat = SONPROC
-#endif
+	movb	$SONPROC,P_STAT(%r12)	# p->p_stat = SONPROC
 	SET_CURPROC(%r12,%rcx)
 
 	sti
diff --git a/sys/arch/amd64/include/cpu.h b/sys/arch/amd64/include/cpu.h
index cc688f39aa6..9caf05b561d 100644
--- a/sys/arch/amd64/include/cpu.h
+++ b/sys/arch/amd64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.6 2004/03/09 23:05:13 deraadt Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.7 2004/06/13 21:49:13 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 2003/04/26 18:39:39 fvdl Exp $	*/
 
 /*-
@@ -191,25 +191,15 @@ extern struct cpu_info cpu_info_primary;
  * or after the current trap/syscall if in system mode.
  */
 
-#ifdef MULTIPROCESSOR
 #define need_resched(ci)						\
 do {									\
-	struct cpu_info *__ci = (ci);					\
+	struct cpu_info *__ci = curcpu();					\
 	__ci->ci_want_resched = 1;					\
 	if (__ci->ci_curproc != NULL)					\
 		aston(__ci->ci_curproc);				\
 } while (/*CONSTCOND*/0)
-#else
-#define need_resched()							\
-do {									\
-	struct cpu_info *__ci = curcpu();				\
-	__ci->ci_want_resched = 1;					\
-	if (__ci->ci_curproc != NULL)					\
-		aston(__ci->ci_curproc);				\
-} while (/*CONSTCOND*/0)
-#endif
 
-#endif
+#endif	/* MULTIPROCESSOR */
 
 #define aston(p)	((p)->p_md.md_astpending = 1)
 
diff --git a/sys/arch/amd64/include/param.h b/sys/arch/amd64/include/param.h
index ea59b9e5273..09e070357a6 100644
--- a/sys/arch/amd64/include/param.h
+++ b/sys/arch/amd64/include/param.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: param.h,v 1.3 2004/04/19 22:55:48 deraadt Exp $	*/
+/*	$OpenBSD: param.h,v 1.4 2004/06/13 21:49:13 niklas Exp $	*/
 
 /*-
  * Copyright (c) 1990 The Regents of the University of California.
@@ -162,9 +162,3 @@
 #define btop(x)				x86_btop(x)
 #define ptob(x)				x86_ptob(x)
 #define round_pdr(x)			x86_round_pdr(x)
-
-/* XXX - oh, the horror.. Just for now. */
-#define KERNEL_PROC_LOCK(p)
-#define KERNEL_PROC_UNLOCK(p)
-#define KERNEL_LOCK(i)
-#define KERNEL_UNLOCK()
diff --git a/sys/arch/amd64/isa/isa_machdep.c b/sys/arch/amd64/isa/isa_machdep.c
index f86a49a00ed..2b3e249e7ae 100644
--- a/sys/arch/amd64/isa/isa_machdep.c
+++ b/sys/arch/amd64/isa/isa_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: isa_machdep.c,v 1.1 2004/01/28 01:39:39 mickey Exp $	*/
+/*	$OpenBSD: isa_machdep.c,v 1.2 2004/06/13 21:49:13 niklas Exp $	*/
 /*	$NetBSD: isa_machdep.c,v 1.22 1997/06/12 23:57:32 thorpej Exp $	*/
 
 #define ISA_DMA_STATS
@@ -132,6 +132,7 @@
 #include <machine/intr.h>
 #include <machine/pio.h>
 #include <machine/cpufunc.h>
+#include <machine/i8259.h>
 
 #include <dev/isa/isareg.h>
 #include <dev/isa/isavar.h>
@@ -139,7 +140,6 @@
 #include <dev/isa/isadmavar.h>
 #endif
 #include <i386/isa/isa_machdep.h>
-#include <i386/isa/icu.h>
 
 #include "isadma.h"
 
diff --git a/sys/arch/arm/arm/cpuswitch.S b/sys/arch/arm/arm/cpuswitch.S
index e7f673a8a45..813bb1c813f 100644
--- a/sys/arch/arm/arm/cpuswitch.S
+++ b/sys/arch/arm/arm/cpuswitch.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpuswitch.S,v 1.2 2004/02/01 06:10:33 drahn Exp $	*/
+/*	$OpenBSD: cpuswitch.S,v 1.3 2004/06/13 21:49:13 niklas Exp $	*/
 /*	$NetBSD: cpuswitch.S,v 1.41 2003/11/15 08:44:18 scw Exp $	*/
 
 /*
@@ -537,11 +537,9 @@ ENTRY(cpu_switch)
 	/* l->l_cpu initialized in fork1() for single-processor */
 #endif
 
-#if 0
 	/* Process is now on a processor. */
-	mov	r0, #LSONPROC			/* l->l_stat = LSONPROC */
-	str	r0, [r6, #(P_STAT)]
-#endif
+	mov	r0, #SONPROC			/* p->p_stat = SONPROC */
+	strb	r0, [r6, #(P_STAT)]
 
 	/* We have a new curproc now so make a note it */
 	ldr	r7, .Lcurproc
diff --git a/sys/arch/arm/arm/genassym.cf b/sys/arch/arm/arm/genassym.cf
index 6803c524393..83733107879 100644
--- a/sys/arch/arm/arm/genassym.cf
+++ b/sys/arch/arm/arm/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.2 2004/02/14 15:34:31 miod Exp $
+#	$OpenBSD: genassym.cf,v 1.3 2004/06/13 21:49:13 niklas Exp $
 #	$NetBSD: genassym.cf,v 1.27 2003/11/04 10:33:16 dsl Exp$
 
 # Copyright (c) 1982, 1990 The Regents of the University of California.
@@ -83,6 +83,8 @@ export	PAGE_SHIFT
 export	P_TRACED
 export	P_PROFIL
 
+export	SONPROC
+
 struct	proc
 member	p_forw
 member	p_back
diff --git a/sys/arch/hp300/include/cpu.h b/sys/arch/hp300/include/cpu.h
index b304d633999..183dbb8ded8 100644
--- a/sys/arch/hp300/include/cpu.h
+++ b/sys/arch/hp300/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.23 2004/05/20 09:20:41 kettenis Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.24 2004/06/13 21:49:13 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.28 1998/02/13 07:41:51 scottr Exp $	*/
 
 /*
@@ -91,7 +91,7 @@ struct clockframe {
  * or after the current trap/syscall if in system mode.
  */
 extern int want_resched;	/* resched() was called */
-#define	need_resched()	{ want_resched++; aston(); }
+#define	need_resched(ci)	{ want_resched++; aston(); }
 
 /*
  * Give a profiling tick to the current process when the user profiling
diff --git a/sys/arch/hppa/hppa/genassym.cf b/sys/arch/hppa/hppa/genassym.cf
index ee718b10e25..c9f494d4e20 100644
--- a/sys/arch/hppa/hppa/genassym.cf
+++ b/sys/arch/hppa/hppa/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.25 2004/06/08 22:00:25 mickey Exp $
+#	$OpenBSD: genassym.cf,v 1.26 2004/06/13 21:49:14 niklas Exp $
 
 #
 # Copyright (c) 1982, 1990, 1993
@@ -142,6 +142,7 @@ member	P_MD_FLAGS		p_md.md_flags
 member	P_MD_REGS		p_md.md_regs
 
 export	SRUN
+export	SONPROC
 
 struct	pcb
 member	pcb_fpregs
diff --git a/sys/arch/hppa/hppa/locore.S b/sys/arch/hppa/hppa/locore.S
index 883c4d3eab8..218c95974b2 100644
--- a/sys/arch/hppa/hppa/locore.S
+++ b/sys/arch/hppa/hppa/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.133 2004/06/08 22:00:25 mickey Exp $	*/
+/*	$OpenBSD: locore.S,v 1.134 2004/06/13 21:49:14 niklas Exp $	*/
 
 /*
  * Copyright (c) 1998-2004 Michael Shalayeff
@@ -2824,6 +2824,8 @@ link_ok
 	copy    arg1, t2
 kstack_ok
 #endif
+	ldi	SONPROC, t1
+	stb	t1, P_STAT(arg1)
 	/* Skip context switch if same process. */
 	comb,=,n arg1, arg2, switch_return
 
diff --git a/sys/arch/hppa/include/cpu.h b/sys/arch/hppa/include/cpu.h
index eff31c9d750..5eb720ab0e9 100644
--- a/sys/arch/hppa/include/cpu.h
+++ b/sys/arch/hppa/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.42 2004/04/08 17:10:18 mickey Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.43 2004/06/13 21:49:14 niklas Exp $	*/
 
 /*
  * Copyright (c) 2000-2004 Michael Shalayeff
@@ -124,7 +124,7 @@ extern int cpu_hvers;
 #define	CLKF_SYSCALL(framep)	((framep)->tf_flags & TFF_SYS)
 
 #define	signotify(p)		(setsoftast())
-#define	need_resched()		(want_resched = 1, setsoftast())
+#define	need_resched(ci)	(want_resched = 1, setsoftast())
 #define	need_proftick(p)	((p)->p_flag |= P_OWEUPC, setsoftast())
 
 #ifndef _LOCORE
diff --git a/sys/arch/i386/compile/.cvsignore b/sys/arch/i386/compile/.cvsignore
index bf45b1f61e5..07993a81a78 100644
--- a/sys/arch/i386/compile/.cvsignore
+++ b/sys/arch/i386/compile/.cvsignore
@@ -1,5 +1,6 @@
 DISKLESS
 GENERIC
+GENERIC.MP
 RAMDISK
 RAMDISKB
 RAMDISKC
diff --git a/sys/arch/i386/conf/GENERIC b/sys/arch/i386/conf/GENERIC
index 7e6a393b230..19025bed6e7 100644
--- a/sys/arch/i386/conf/GENERIC
+++ b/sys/arch/i386/conf/GENERIC
@@ -1,4 +1,4 @@
-#	$OpenBSD: GENERIC,v 1.370 2004/06/06 04:50:57 pvalchev Exp $
+#	$OpenBSD: GENERIC,v 1.371 2004/06/13 21:49:15 niklas Exp $
 #
 #	GENERIC -- everything that's currently supported
 #
@@ -39,7 +39,8 @@ config		bsd	swap generic
 
 mainbus0 at root
 
-bios0	at mainbus0
+cpu0	at mainbus? apid ?
+bios0	at mainbus0 apid ?
 apm0	at bios0 flags 0x0000	# flags 0x0101 to force protocol version 1.1
 pcibios0 at bios0 flags 0x0000	# use 0x30 for a total verbose
 
diff --git a/sys/arch/i386/conf/GENERIC.MP b/sys/arch/i386/conf/GENERIC.MP
new file mode 100644
index 00000000000..4bebe33f96d
--- /dev/null
+++ b/sys/arch/i386/conf/GENERIC.MP
@@ -0,0 +1,12 @@
+#	$OpenBSD: GENERIC.MP,v 1.2 2004/06/13 21:49:15 niklas Exp $
+#
+#	GENERIC.MP - sample multiprocessor kernel
+#
+
+include "arch/i386/conf/GENERIC"
+
+option		MULTIPROCESSOR	# Multiple processor support
+option		CPU
+
+cpu*		at mainbus? apid ?
+ioapic* 	at mainbus? apid ?
diff --git a/sys/arch/i386/conf/RAMDISK b/sys/arch/i386/conf/RAMDISK
index b7f528ea3d3..c377ae6ce8f 100644
--- a/sys/arch/i386/conf/RAMDISK
+++ b/sys/arch/i386/conf/RAMDISK
@@ -1,4 +1,4 @@
-#	$OpenBSD: RAMDISK,v 1.129 2004/06/03 18:22:21 grange Exp $
+#	$OpenBSD: RAMDISK,v 1.130 2004/06/13 21:49:15 niklas Exp $
 
 machine		i386		# architecture, used by config; REQUIRED
 
@@ -42,7 +42,8 @@ config		bsd	root on rd0a swap on rd0b and wd0b and sd0b
 
 mainbus0 at root
 
-bios0	at mainbus0
+cpu0	at mainbus? apid ?
+bios0	at mainbus0 apid ?
 apm0	at bios0 flags 0x0000	# flags 0x0101 to force protocol version 1.1
 #option	APMDEBUG
 pcibios0 at bios0 flags 0x0000
diff --git a/sys/arch/i386/conf/RAMDISKB b/sys/arch/i386/conf/RAMDISKB
index c7a0d727224..53184d25203 100644
--- a/sys/arch/i386/conf/RAMDISKB
+++ b/sys/arch/i386/conf/RAMDISKB
@@ -1,4 +1,4 @@
-#	$OpenBSD: RAMDISKB,v 1.70 2004/06/03 18:22:21 grange Exp $
+#	$OpenBSD: RAMDISKB,v 1.71 2004/06/13 21:49:15 niklas Exp $
 
 machine		i386		# architecture, used by config; REQUIRED
 
@@ -42,7 +42,8 @@ config		bsd	root on rd0a swap on rd0b and wd0b and sd0b
 
 mainbus0 at root
 
-bios0	at mainbus0
+cpu0	at mainbus? apid ?
+bios0	at mainbus0 apid ?
 apm0	at bios0 flags 0x0000	# flags 0x0101 to force protocol version 1.1
 #option	APMDEBUG
 pcibios0 at bios0 flags 0x0000
diff --git a/sys/arch/i386/conf/RAMDISKC b/sys/arch/i386/conf/RAMDISKC
index 76e9db57d6f..81764b62557 100644
--- a/sys/arch/i386/conf/RAMDISKC
+++ b/sys/arch/i386/conf/RAMDISKC
@@ -1,4 +1,4 @@
-#	$OpenBSD: RAMDISKC,v 1.39 2004/06/03 18:22:21 grange Exp $
+#	$OpenBSD: RAMDISKC,v 1.40 2004/06/13 21:49:15 niklas Exp $
 
 machine		i386		# architecture, used by config; REQUIRED
 
@@ -42,7 +42,8 @@ config		bsd	root on rd0a swap on rd0b and wd0b and sd0b
 
 mainbus0 at root
 
-bios0	at mainbus0
+cpu0	at mainbus? apid ?
+bios0	at mainbus0 apid ?
 apm0	at bios0 flags 0x0000	# flags 0x0101 to force protocol version 1.1
 #option	APMDEBUG
 pcibios0 at bios0 flags 0x0000
diff --git a/sys/arch/i386/conf/RAMDISK_CD b/sys/arch/i386/conf/RAMDISK_CD
index 830c763eabc..1ea5cf46878 100644
--- a/sys/arch/i386/conf/RAMDISK_CD
+++ b/sys/arch/i386/conf/RAMDISK_CD
@@ -1,4 +1,4 @@
-#	$OpenBSD: RAMDISK_CD,v 1.75 2004/06/06 04:50:57 pvalchev Exp $
+#	$OpenBSD: RAMDISK_CD,v 1.76 2004/06/13 21:49:15 niklas Exp $
 
 machine		i386		# architecture, used by config; REQUIRED
 
@@ -43,7 +43,8 @@ config		bsd	root on rd0a swap on rd0b and wd0b and sd0b
 
 mainbus0 at root
 
-bios0	at mainbus0
+cpu0	at mainbus? apid ?
+bios0	at mainbus0 apid ?
 apm0	at bios0 flags 0x0000	# flags 0x0101 to force protocol version 1.1
 #option	APMDEBUG
 pcibios0 at bios0 flags 0x0000
diff --git a/sys/arch/i386/conf/files.i386 b/sys/arch/i386/conf/files.i386
index 02b5d8ee2c5..a4cc18bd5e9 100644
--- a/sys/arch/i386/conf/files.i386
+++ b/sys/arch/i386/conf/files.i386
@@ -1,4 +1,4 @@
-#	$OpenBSD: files.i386,v 1.122 2004/06/05 15:06:22 grange Exp $
+#	$OpenBSD: files.i386,v 1.123 2004/06/13 21:49:15 niklas Exp $
 #
 # new style config file for i386 architecture
 #
@@ -44,6 +44,10 @@ file	arch/i386/i386/dkcsum.c		bios
 file	lib/libz/adler32.c		!ppp_deflate & !ipsec & !crypto
 file	dev/cons.c
 file	dev/cninit.c
+file	arch/i386/i386/mptramp.s	multiprocessor
+file	arch/i386/i386/lock_machdep.c	multiprocessor
+file	arch/i386/i386/ipifuncs.c	multiprocessor
+file	arch/i386/i386/db_mp.c		multiprocessor
 file	arch/i386/i386/wscons_machdep.c	wsdisplay
 major	{vnd = 14}
 
@@ -78,7 +82,7 @@ include "../../../dev/ata/files.ata"
 # System bus types
 #
 
-define mainbus { }
+define mainbus { apid = -1 }
 device	mainbus: isabus, eisabus, pcibus, mainbus
 attach	mainbus at root
 file	arch/i386/i386/mainbus.c	mainbus
@@ -272,6 +276,28 @@ attach	apm at bios
 file	arch/i386/i386/apm.c		apm needs-count
 file	arch/i386/i386/apmcall.S	apm
 
+# Intel SMP specification 1.4
+define	mpbios
+file	arch/i386/i386/mpbios.c		mpbios needs-flag
+
+# CPUS
+define	cpu { apid = -1 }
+device	cpu
+attach	cpu at mainbus
+file	arch/i386/i386/cpu.c		cpu
+
+# Common APIC support routines
+file	arch/i386/i386/apic.c		ioapic | lapic
+
+# Local APIC (required for multiprocessor)
+define	lapic
+file	arch/i386/i386/lapic.c		lapic needs-flag
+
+# I/O APICs
+device	ioapic: mpbios, lapic
+attach	ioapic at mainbus
+file	arch/i386/i386/ioapic.c		ioapic needs-flag
+
 device	pcibios
 attach	pcibios at bios
 file	arch/i386/pci/pcibios.c		pcibios needs-count
diff --git a/sys/arch/i386/eisa/eisa_machdep.c b/sys/arch/i386/eisa/eisa_machdep.c
index 178950e9e86..e086091fd75 100644
--- a/sys/arch/i386/eisa/eisa_machdep.c
+++ b/sys/arch/i386/eisa/eisa_machdep.c
@@ -1,5 +1,5 @@
-/*	$OpenBSD: eisa_machdep.c,v 1.7 2004/06/09 20:13:10 deraadt Exp $	*/
-/*	$NetBSD: eisa_machdep.c,v 1.6 1997/06/06 23:12:52 thorpej Exp $	*/
+/*	$OpenBSD: eisa_machdep.c,v 1.8 2004/06/13 21:49:15 niklas Exp $	*/
+/*	$NetBSD: eisa_machdep.c,v 1.10.22.2 2000/06/25 19:36:58 sommerfeld Exp $	*/
 
 /*-
  * Copyright (c) 1997 The NetBSD Foundation, Inc.
@@ -81,8 +81,8 @@
 
 #define _I386_BUS_DMA_PRIVATE
 #include <machine/bus.h>
+#include <machine/i8259.h>
 
-#include <i386/isa/icu.h>
 #include <dev/isa/isavar.h>
 #include <dev/eisa/eisavar.h>
 
@@ -133,6 +133,9 @@ eisa_intr_map(ec, irq, ihp)
 	u_int irq;
 	eisa_intr_handle_t *ihp;
 {
+#if NIOAPIC > 0
+	struct mp_intr_map *mip;
+#endif
 
 	if (irq >= ICU_LEN) {
 		printf("eisa_intr_map: bad IRQ %d\n", irq);
@@ -144,6 +147,26 @@ eisa_intr_map(ec, irq, ihp)
 		irq = 9;
 	}
 
+#if NIOAPIC > 0
+	if (mp_busses != NULL) {
+		/*
+		 * Assumes 1:1 mapping between PCI bus numbers and
+		 * the numbers given by the MP bios.
+		 * XXX Is this a valid assumption?
+		 */
+		
+		for (mip = mp_busses[bus].mb_intrs; mip != NULL;
+		    mip = mip->next) {
+			if (mip->bus_pin == irq) {
+				*ihp = mip->ioapic_ih | irq;
+				return (0);
+			}
+		}
+		if (mip == NULL)
+			printf("eisa_intr_map: no MP mapping found\n");
+	}
+#endif
+
 	*ihp = irq;
 	return (0);
 }
@@ -155,9 +178,17 @@ eisa_intr_string(ec, ih)
 {
 	static char irqstr[8];		/* 4 + 2 + NUL + sanity */
 
-	if (ih == 0 || ih >= ICU_LEN || ih == 2)
+	if (ih == 0 || (ih & 0xff) >= ICU_LEN || ih == 2)
 		panic("eisa_intr_string: bogus handle 0x%x", ih);
 
+#if NIOAPIC > 0
+	if (ih & APIC_INT_VIA_APIC) {
+		sprintf(irqstr, "apic %d int %d (irq %d)",
+		    APIC_IRQ_APIC(ih), APIC_IRQ_PIN(ih), ih & 0xff);
+		return (irqstr);
+	}
+#endif
+
 	snprintf(irqstr, sizeof irqstr, "irq %d", ih);
 	return (irqstr);
 	
@@ -171,7 +202,14 @@ eisa_intr_establish(ec, ih, type, level, func, arg, what)
 	void *arg;
 	char *what;
 {
-
+#if NIOAPIC > 0
+	if (ih != -1) {
+		if (ih != -1 && (ih & APIC_INT_VIA_APIC)) {
+			return (apic_intr_establish(ih, type, level, func, arg,
+			    what));
+		}
+	}
+#endif
 	if (ih == 0 || ih >= ICU_LEN || ih == 2)
 		panic("eisa_intr_establish: bogus handle 0x%x", ih);
 
@@ -183,6 +221,5 @@ eisa_intr_disestablish(ec, cookie)
 	eisa_chipset_tag_t ec;
 	void *cookie;
 {
-
 	return (isa_intr_disestablish(NULL, cookie));
 }
diff --git a/sys/arch/i386/eisa/eisa_machdep.h b/sys/arch/i386/eisa/eisa_machdep.h
index a313d805fae..fccdd343245 100644
--- a/sys/arch/i386/eisa/eisa_machdep.h
+++ b/sys/arch/i386/eisa/eisa_machdep.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: eisa_machdep.h,v 1.5 2002/03/14 03:15:53 millert Exp $	*/
+/*	$OpenBSD: eisa_machdep.h,v 1.6 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: eisa_machdep.h,v 1.4 1997/06/06 23:12:52 thorpej Exp $	*/
 
 /*
@@ -45,6 +45,9 @@
 
 extern struct i386_bus_dma_tag eisa_bus_dma_tag;
 
+#define ELCR0	0x4d0		/* eisa irq 0-7 */
+#define ELCR1	0x4d1		/* eisa irq 8-15 */
+
 /*
  * Types provided to machine-independent EISA code.
  */
diff --git a/sys/arch/i386/i386/apic.c b/sys/arch/i386/i386/apic.c
new file mode 100644
index 00000000000..6ed294770f3
--- /dev/null
+++ b/sys/arch/i386/i386/apic.c
@@ -0,0 +1,79 @@
+/*	$OpenBSD: apic.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/* $NetBSD: apic.c,v 1.1.2.2 2000/02/21 18:51:00 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/types.h>
+#include <sys/systm.h>
+
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+#include <machine/apicvar.h>
+
+
+const char redirlofmt[] = "\177\20"
+	"f\0\10vector\0"
+	"f\10\3delmode\0"
+	"b\13logical\0"
+	"b\14pending\0"
+	"b\15actlo\0"
+	"b\16irrpending\0"
+	"b\17level\0"
+	"b\20masked\0"
+	"f\22\1dest\0" "=\1self" "=\2all" "=\3all-others";
+
+const char redirhifmt[] = "\177\20"
+	"f\30\10target\0";
+
+void
+apic_format_redir(where1, where2, idx, redirhi, redirlo)
+	char *where1;
+	char *where2;
+	int idx;
+	u_int32_t redirhi;
+	u_int32_t redirlo;
+{
+	printf("%s: %s%d 0x%x", where1, where2, idx, redirlo);
+
+	if ((redirlo & LAPIC_DEST_MASK) == 0)
+		printf(" 0x%x", redirhi);
+
+	printf("\n");
+}
+
diff --git a/sys/arch/i386/i386/apicvec.s b/sys/arch/i386/i386/apicvec.s
new file mode 100644
index 00000000000..c2fadf429be
--- /dev/null
+++ b/sys/arch/i386/i386/apicvec.s
@@ -0,0 +1,250 @@
+/* $OpenBSD: apicvec.s,v 1.2 2004/06/13 21:49:15 niklas Exp $ */	
+/* $NetBSD: apicvec.s,v 1.1.2.2 2000/02/21 21:54:01 sommerfeld Exp $ */	
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+	
+#include <machine/i82093reg.h>
+#include <machine/i82489reg.h>	
+
+#ifdef __ELF__
+#define XINTR(vec) Xintr/**/vec
+#else
+#define XINTR(vec) _Xintr/**/vec
+#endif
+
+#ifdef MULTIPROCESSOR
+	.globl	XINTR(ipi)
+XINTR(ipi):
+	pushl	$0		
+	pushl	$T_ASTFLT
+	INTRENTRY		
+	MAKE_FRAME		
+	pushl	CPL
+	movl	_C_LABEL(lapic_ppr),%eax
+	movl	%eax,CPL
+	ioapic_asm_ack()
+        sti			/* safe to take interrupts.. */
+	call	_C_LABEL(i386_ipi_handler)
+	jmp	_C_LABEL(Xdoreti)
+#endif
+	
+	/*
+	 * Interrupt from the local APIC timer.
+	 */
+	.globl	XINTR(ltimer)
+XINTR(ltimer):			
+	pushl	$0		
+	pushl	$T_ASTFLT
+	INTRENTRY		
+	MAKE_FRAME		
+	pushl	CPL
+	movl	_C_LABEL(lapic_ppr),%eax
+	movl	%eax,CPL
+	ioapic_asm_ack()
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
+	movl	%esp,%eax
+	pushl	%eax
+	call	_C_LABEL(lapic_clockintr)
+	addl	$4,%esp		
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintunlock)
+#endif
+	jmp	_C_LABEL(Xdoreti)
+
+	.globl	XINTR(softclock), XINTR(softnet), XINTR(softtty)
+XINTR(softclock):
+	pushl	$0		
+	pushl	$T_ASTFLT
+	INTRENTRY		
+	MAKE_FRAME		
+	pushl	CPL
+	movl	$IPL_SOFTCLOCK,CPL
+	andl	$~(1<<SIR_CLOCK),_C_LABEL(ipending)
+	ioapic_asm_ack()
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
+	call	_C_LABEL(softclock)
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintunlock)
+#endif
+	jmp	_C_LABEL(Xdoreti)
+	
+#define DONETISR(s, c) \
+	.globl  _C_LABEL(c)	;\
+	testl	$(1 << s),%edi	;\
+	jz	1f		;\
+	call	_C_LABEL(c)	;\
+1:
+
+XINTR(softnet):
+	pushl	$0		
+	pushl	$T_ASTFLT
+	INTRENTRY		
+	MAKE_FRAME		
+	pushl	CPL
+	movl	$IPL_SOFTNET,CPL
+	andl	$~(1<<SIR_NET),_C_LABEL(ipending)
+	ioapic_asm_ack()
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
+	xorl	%edi,%edi
+	xchgl	_C_LABEL(netisr),%edi
+#include <net/netisr_dispatch.h>
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintunlock)
+#endif
+	jmp	_C_LABEL(Xdoreti)
+#undef DONETISR
+
+XINTR(softtty):	
+	pushl	$0		
+	pushl	$T_ASTFLT
+	INTRENTRY		
+	MAKE_FRAME		
+	pushl	CPL
+	movl	$IPL_SOFTTTY,CPL
+	andl	$~(1<<SIR_TTY),_C_LABEL(ipending)
+	ioapic_asm_ack()
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
+	call	_C_LABEL(comsoft)
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintunlock)
+#endif
+	jmp	_C_LABEL(Xdoreti)
+
+#if NIOAPIC > 0
+
+#define voidop(num)
+
+	/*
+	 * I/O APIC interrupt.
+	 * We sort out which one is which based on the value of 
+	 * the processor priority register.
+	 *
+	 * XXX no stray interrupt mangling stuff..
+	 * XXX use cmove when appropriate.
+	 */
+	
+#define APICINTR(name, num, early_ack, late_ack, mask, unmask, level_mask) \
+_C_LABEL(Xintr_/**/name/**/num):					\
+	pushl	$0							;\
+	pushl	$T_ASTFLT						;\
+	INTRENTRY							;\
+	MAKE_FRAME							;\
+	pushl	CPL							;\
+	movl	_C_LABEL(lapic_ppr),%eax				;\
+	movl	%eax,CPL						;\
+	mask(num)			/* mask it in hardware */	;\
+	early_ack(num)			/* and allow other intrs */	;\
+	incl	MY_COUNT+V_INTR		/* statistical info */		;\
+	sti								;\
+	orl	$num,%eax						;\
+	incl	_C_LABEL(apic_intrcount)(,%eax,4)			;\
+	movl	_C_LABEL(apic_intrhand)(,%eax,4),%ebx /* chain head */	;\
+	testl	%ebx,%ebx						;\
+	jz	8f			/* oops, no handlers.. */	;\
+7:									 \
+	LOCK_KERNEL							;\
+	movl	IH_ARG(%ebx),%eax	/* get handler arg */		;\
+	testl	%eax,%eax						;\
+	jnz	6f							;\
+	movl	%esp,%eax		/* 0 means frame pointer */	;\
+6:									 \
+	pushl	%eax							;\
+	call	*IH_FUN(%ebx)		/* call it */			;\
+	addl	$4,%esp			/* toss the arg */		;\
+	UNLOCK_KERNEL							;\
+	incl	IH_COUNT(%ebx)		/* count the intrs */		;\
+	movl	IH_NEXT(%ebx),%ebx	/* next handler in chain */	;\
+	testl	%ebx,%ebx						;\
+	jnz	7b							;\
+8:									 \
+	unmask(num)			/* unmask it in hardware */	;\
+	late_ack(num)							;\
+	jmp	_C_LABEL(Xdoreti)
+	
+APICINTR(ioapic,0, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,1, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,2, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,3, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,4, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,5, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,6, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,7, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,8, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,9, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,10, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,11, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,12, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,13, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,14, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+APICINTR(ioapic,15, voidop, ioapic_asm_ack, voidop, voidop, voidop)
+
+	.globl	_C_LABEL(Xintr_ioapic0),_C_LABEL(Xintr_ioapic1)
+	.globl	_C_LABEL(Xintr_ioapic2),_C_LABEL(Xintr_ioapic3)
+	.globl	_C_LABEL(Xintr_ioapic4),_C_LABEL(Xintr_ioapic5)
+	.globl	_C_LABEL(Xintr_ioapic6),_C_LABEL(Xintr_ioapic7)
+	.globl	_C_LABEL(Xintr_ioapic8),_C_LABEL(Xintr_ioapic9)
+	.globl	_C_LABEL(Xintr_ioapic10),_C_LABEL(Xintr_ioapic11)
+	.globl	_C_LABEL(Xintr_ioapic12),_C_LABEL(Xintr_ioapic13)
+	.globl	_C_LABEL(Xintr_ioapic14),_C_LABEL(Xintr_ioapic15)
+	.globl _C_LABEL(apichandler)
+
+_C_LABEL(apichandler):	
+	.long	_C_LABEL(Xintr_ioapic0),_C_LABEL(Xintr_ioapic1)
+	.long	_C_LABEL(Xintr_ioapic2),_C_LABEL(Xintr_ioapic3)
+	.long	_C_LABEL(Xintr_ioapic4),_C_LABEL(Xintr_ioapic5)
+	.long	_C_LABEL(Xintr_ioapic6),_C_LABEL(Xintr_ioapic7)
+	.long	_C_LABEL(Xintr_ioapic8),_C_LABEL(Xintr_ioapic9)
+	.long	_C_LABEL(Xintr_ioapic10),_C_LABEL(Xintr_ioapic11)
+	.long	_C_LABEL(Xintr_ioapic12),_C_LABEL(Xintr_ioapic13)
+	.long	_C_LABEL(Xintr_ioapic14),_C_LABEL(Xintr_ioapic15)
+
+#endif
+	
diff --git a/sys/arch/i386/i386/apm.c b/sys/arch/i386/i386/apm.c
index 1ade5dfb400..f996be58a37 100644
--- a/sys/arch/i386/i386/apm.c
+++ b/sys/arch/i386/i386/apm.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: apm.c,v 1.62 2004/05/27 08:19:59 tedu Exp $	*/
+/*	$OpenBSD: apm.c,v 1.63 2004/06/13 21:49:15 niklas Exp $	*/
 
 /*-
  * Copyright (c) 1998-2001 Michael Shalayeff. All rights reserved.
@@ -777,7 +777,6 @@ apmattach(parent, self, aux)
 	struct device *parent, *self;
 	void *aux;
 {
-	extern union descriptor *dynamic_gdt;
 	struct bios_attach_args *ba = aux;
 	bios_apminfo_t *ap = ba->bios_apmp;
 	struct apm_softc *sc = (void *)self;
@@ -843,12 +842,12 @@ apmattach(parent, self, aux)
 		else
 			ch16 += ap->apm_code16_base - cbase;
 
-		setsegment(&dynamic_gdt[GAPM32CODE_SEL].sd, (void *)ch32,
-			   ap->apm_code_len, SDT_MEMERA, SEL_KPL, 1, 0);
-		setsegment(&dynamic_gdt[GAPM16CODE_SEL].sd, (void *)ch16,
-			   ap->apm_code16_len, SDT_MEMERA, SEL_KPL, 0, 0);
-		setsegment(&dynamic_gdt[GAPMDATA_SEL].sd, (void *)dh,
-			   ap->apm_data_len, SDT_MEMRWA, SEL_KPL, 1, 0);
+		setgdt(GAPM32CODE_SEL, (void *)ch32, ap->apm_code_len,
+		    SDT_MEMERA, SEL_KPL, 1, 0);
+		setgdt(GAPM16CODE_SEL, (void *)ch16, ap->apm_code16_len,
+		    SDT_MEMERA, SEL_KPL, 0, 0);
+		setgdt(GAPMDATA_SEL, (void *)dh, ap->apm_data_len, SDT_MEMRWA,
+		    SEL_KPL, 1, 0);
 		DPRINTF((": flags %x code 32:%x/%x[%x] 16:%x/%x[%x] "
 		       "data %x/%x/%x ep %x (%x:%x)\n%s", apm_flags,
 		    ap->apm_code32_base, ch32, ap->apm_code_len,
@@ -890,9 +889,9 @@ apmattach(parent, self, aux)
 		} else
 			kthread_create_deferred(apm_thread_create, sc);
 	} else {
-		dynamic_gdt[GAPM32CODE_SEL] = dynamic_gdt[GNULL_SEL];
-		dynamic_gdt[GAPM16CODE_SEL] = dynamic_gdt[GNULL_SEL];
-		dynamic_gdt[GAPMDATA_SEL] = dynamic_gdt[GNULL_SEL];
+		setgdt(GAPM32CODE_SEL, NULL, 0, 0, 0, 0, 0);
+		setgdt(GAPM16CODE_SEL, NULL, 0, 0, 0, 0, 0);
+		setgdt(GAPMDATA_SEL, NULL, 0, 0, 0, 0, 0);
 	}
 }
 
@@ -901,6 +900,15 @@ apm_thread_create(v)
 	void *v;
 {
 	struct apm_softc *sc = v;
+
+#ifdef MULTIPROCESSOR
+	if (ncpus > 1) {
+		apm_disconnect(sc);
+		apm_dobusy = apm_doidle = 0;
+		return;
+	}
+#endif
+
 	if (kthread_create(apm_thread, sc, &sc->sc_thread,
 	    "%s", sc->sc_dev.dv_xname)) {
 		apm_disconnect(sc);
diff --git a/sys/arch/i386/i386/autoconf.c b/sys/arch/i386/i386/autoconf.c
index f0a5b2f1de0..95a3059984b 100644
--- a/sys/arch/i386/i386/autoconf.c
+++ b/sys/arch/i386/i386/autoconf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: autoconf.c,v 1.52 2003/10/15 03:56:21 david Exp $	*/
+/*	$OpenBSD: autoconf.c,v 1.53 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: autoconf.c,v 1.20 1996/05/03 19:41:56 christos Exp $	*/
 
 /*-
@@ -61,6 +61,12 @@
 
 #include <dev/cons.h>
 
+#include "ioapic.h"
+
+#if NIOAPIC > 0
+#include <machine/i82093var.h>
+#endif
+
 int findblkmajor(struct device *dv);
 char *findblkname(int);
 
@@ -109,6 +115,14 @@ cpu_configure()
 	printf("biomask %x netmask %x ttymask %x\n", (u_short)IMASK(IPL_BIO),
 	    (u_short)IMASK(IPL_NET), (u_short)IMASK(IPL_TTY));
 
+#if NIOAPIC > 0
+	ioapic_enable();
+#endif
+
+#ifdef MULTIPROCESSOR
+	/* propagate TSS and LDT configuration to the idle pcb's. */
+	cpu_init_idle_pcbs();
+#endif
 	spl0();
 
 	/*
diff --git a/sys/arch/i386/i386/bios.c b/sys/arch/i386/i386/bios.c
index 74691de83d9..a9576ccd41f 100644
--- a/sys/arch/i386/i386/bios.c
+++ b/sys/arch/i386/i386/bios.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: bios.c,v 1.54 2004/01/29 01:36:13 tom Exp $	*/
+/*	$OpenBSD: bios.c,v 1.55 2004/06/13 21:49:15 niklas Exp $	*/
 
 /*
  * Copyright (c) 1997-2001 Michael Shalayeff
@@ -85,9 +85,12 @@ bios_apminfo_t *apm;
 bios_pciinfo_t *bios_pciinfo;
 #endif
 bios_diskinfo_t *bios_diskinfo;
-bios_memmap_t	*bios_memmap;
+bios_memmap_t  *bios_memmap;
 u_int32_t	bios_cksumlen;
 struct bios32_entry bios32_entry;
+#ifdef MULTIPROCESSOR
+void	       *bios_smpinfo;
+#endif
 
 bios_diskinfo_t *bios_getdiskinfo(dev_t);
 
@@ -324,6 +327,12 @@ bios_getopt()
 				cnset(cdp->consdev);
 			}
 			break;
+#ifdef MULTIPROCESSOR
+		case BOOTARG_SMPINFO:
+			bios_smpinfo = q->ba_arg;
+			printf(" smpinfo %p", bios_smpinfo);
+			break;
+#endif
 
 		default:
 #ifdef BIOS_DEBUG
@@ -356,9 +365,6 @@ bios32_service(service, e, ei)
 	bios32_entry_t e;
 	bios32_entry_info_t ei;
 {
-	extern union descriptor *dynamic_gdt;
-	extern int gdt_get_slot(void);
-
 	u_long pa, endpa;
 	vaddr_t va, sva;
 	u_int32_t base, count, off, ent;
@@ -388,8 +394,7 @@ bios32_service(service, e, ei)
 		return (0);
 
 	slot = gdt_get_slot();
-	setsegment(&dynamic_gdt[slot].sd, (caddr_t)va, BIOS32_END,
-	    SDT_MEMERA, SEL_KPL, 1, 0);
+	setgdt(slot, (caddr_t)va, BIOS32_END, SDT_MEMERA, SEL_KPL, 1, 0);
 
 	for (pa = i386_trunc_page(BIOS32_START),
 	     va += i386_trunc_page(BIOS32_START);
diff --git a/sys/arch/i386/i386/cpu.c b/sys/arch/i386/i386/cpu.c
new file mode 100644
index 00000000000..c447da4e533
--- /dev/null
+++ b/sys/arch/i386/i386/cpu.c
@@ -0,0 +1,606 @@
+/*	$OpenBSD: cpu.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/* $NetBSD: cpu.c,v 1.1.2.7 2000/06/26 02:04:05 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Copyright (c) 1999 Stefan Grefen
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed by the NetBSD
+ *      Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include "lapic.h"
+#include "ioapic.h"
+
+#include <sys/param.h>
+#include <sys/proc.h>
+#include <sys/user.h>
+#include <sys/systm.h>
+#include <sys/device.h>
+
+#include <uvm/uvm_extern.h>
+
+#include <machine/cpu.h>
+#include <machine/cpufunc.h>
+#include <machine/cpuvar.h>
+#include <machine/pmap.h>
+#include <machine/vmparam.h>
+#include <machine/mpbiosvar.h>
+#include <machine/npx.h>
+#include <machine/pcb.h>
+#include <machine/specialreg.h>
+#include <machine/segments.h>
+#include <machine/gdt.h>
+#include <machine/pio.h>
+
+#if NLAPIC > 0
+#include <machine/apicvar.h>
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+#endif
+
+#if NIOAPIC > 0
+#include <machine/i82093var.h>
+#endif
+
+#include <dev/ic/mc146818reg.h>
+#include <i386/isa/nvram.h>
+#include <dev/isa/isareg.h>
+
+int     cpu_match(struct device *, void *, void *);
+void    cpu_attach(struct device *, struct device *, void *);
+
+#ifdef MULTIPROCESSOR
+int mp_cpu_start(struct cpu_info *); 
+void mp_cpu_start_cleanup(struct cpu_info *);
+struct cpu_functions mp_cpu_funcs =
+    { mp_cpu_start, NULL, mp_cpu_start_cleanup };
+#endif
+
+/*
+ * Statically-allocated CPU info for the primary CPU (or the only
+ * CPU, on uniprocessors).  The CPU info list is initialized to
+ * point at it.
+ */
+struct cpu_info cpu_info_primary;
+struct cpu_info *cpu_info_list = &cpu_info_primary;
+
+void	cpu_init_tss(struct i386tss *, void *, void *);
+void	cpu_set_tss_gates(struct cpu_info *);
+
+#ifdef MULTIPROCESSOR
+/*
+ * Array of CPU info structures.  Must be statically-allocated because
+ * curproc, etc. are used early.
+ */
+
+struct cpu_info *cpu_info[I386_MAXPROCS] = { &cpu_info_primary };
+
+void   	cpu_hatch(void *);
+void   	cpu_boot_secondary(struct cpu_info *);
+void	cpu_copy_trampoline(void);
+
+/*
+ * Runs once per boot once multiprocessor goo has been detected and
+ * the local APIC has been mapped.
+ * Called from mpbios_scan();
+ */
+void
+cpu_init_first()
+{
+	int cpunum = cpu_number();
+
+	if (cpunum != 0) {
+		cpu_info[0] = NULL;
+		cpu_info[cpunum] = &cpu_info_primary;
+	}
+
+	cpu_copy_trampoline();
+}
+#endif
+
+struct cfattach cpu_ca = {
+	sizeof(struct cpu_info), cpu_match, cpu_attach
+};
+
+struct cfdriver cpu_cd = {
+	NULL, "cpu", DV_DULL /* XXX DV_CPU */
+};
+
+int
+cpu_match(parent, matchv, aux)
+    struct device *parent;
+    void *matchv;
+    void *aux;
+{
+  	struct cfdata *match = (struct cfdata *)matchv;
+	struct cpu_attach_args *caa = (struct cpu_attach_args *)aux;
+
+	if (strcmp(caa->caa_name, match->cf_driver->cd_name) == 0)
+		return (1);
+	return (0);
+}
+
+void
+cpu_attach(parent, self, aux)
+	struct device *parent, *self;
+	void *aux;
+{
+	struct cpu_info *ci = (struct cpu_info *)self;
+	struct cpu_attach_args *caa = (struct cpu_attach_args *)aux;
+
+#ifdef MULTIPROCESSOR
+	int cpunum = caa->cpu_number;
+	vaddr_t kstack;
+	struct pcb *pcb;
+
+	if (caa->cpu_role != CPU_ROLE_AP) {
+		if (cpunum != cpu_number()) {
+			panic("%s: running cpu is at apic %d"
+			    " instead of at expected %d\n",
+			    self->dv_xname, cpu_number(), cpunum);
+		}
+
+		ci = &cpu_info_primary;
+		bcopy(self, &ci->ci_dev, sizeof *self);
+
+		/* special-case boot CPU */			    /* XXX */
+		if (cpu_info[cpunum] == &cpu_info_primary) {	    /* XXX */
+			cpu_info[cpunum] = NULL; 		    /* XXX */
+		}				 		    /* XXX */
+	}
+	if (cpu_info[cpunum] != NULL)
+		panic("cpu at apic id %d already attached?", cpunum);
+
+	cpu_info[cpunum] = ci;
+#endif
+
+	ci->ci_self = ci;
+	ci->ci_apicid = caa->cpu_number;
+#ifdef MULTIPROCESSOR
+	ci->ci_cpuid = ci->ci_apicid;
+#else
+	ci->ci_cpuid = 0;	/* False for APs, so what, they're not used */
+#endif
+	ci->ci_signature = caa->cpu_signature;
+	ci->ci_feature_flags = caa->feature_flags;
+	ci->ci_func = caa->cpu_func;
+
+#ifdef MULTIPROCESSOR
+	/*
+	 * Allocate UPAGES contiguous pages for the idle PCB and stack.
+	 */
+
+	kstack = uvm_km_alloc(kernel_map, USPACE);
+	if (kstack == 0) {
+		if (cpunum == 0) { /* XXX */
+			panic("cpu_attach: unable to allocate idle stack for"
+			    " primary");
+		}
+		printf("%s: unable to allocate idle stack\n",
+		    ci->ci_dev.dv_xname);
+		return;
+	}
+	pcb = ci->ci_idle_pcb = (struct pcb *)kstack;
+	memset(pcb, 0, USPACE);
+
+	pcb->pcb_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
+	pcb->pcb_tss.tss_esp0 = kstack + USPACE - 16 -
+	    sizeof (struct trapframe);
+	pcb->pcb_tss.tss_esp = kstack + USPACE - 16 -
+	    sizeof (struct trapframe);
+	pcb->pcb_pmap = pmap_kernel();
+	pcb->pcb_cr3 = vtophys(pcb->pcb_pmap->pm_pdir);
+	/* pcb->pcb_cr3 = pcb->pcb_pmap->pm_pdir - KERNBASE; XXX ??? */
+#endif
+
+	/* further PCB init done later. */
+
+/* XXXSMP: must be shared with UP */
+#ifdef MULTIPROCESSOR
+	printf(": ");
+
+	switch (caa->cpu_role) {
+	case CPU_ROLE_SP:
+		printf("(uniprocessor)\n");
+		ci->ci_flags |= CPUF_PRESENT | CPUF_SP | CPUF_PRIMARY;
+		identifycpu(ci);
+		cpu_init(ci);
+		break;
+
+	case CPU_ROLE_BP:
+		printf("apid %d (", caa->cpu_number);
+		printf("boot processor");
+		ci->ci_flags |= CPUF_PRESENT | CPUF_BSP | CPUF_PRIMARY;
+		printf(")\n");
+		identifycpu(ci);
+		cpu_init(ci);
+
+#if NLAPIC > 0
+		/*
+		 * Enable local apic
+		 */
+		lapic_enable();
+		lapic_calibrate_timer(ci);
+#endif
+#if NIOAPIC > 0
+		ioapic_bsp_id = caa->cpu_number;
+#endif
+		break;
+
+	case CPU_ROLE_AP:
+		/*
+		 * report on an AP
+		 */
+		printf("apid %d (application processor)\n", caa->cpu_number);
+
+#ifdef MULTIPROCESSOR
+		gdt_alloc_cpu(ci);
+		ci->ci_flags |= CPUF_PRESENT | CPUF_AP;
+		identifycpu(ci);
+		ci->ci_next = cpu_info_list->ci_next;
+		cpu_info_list->ci_next = ci;
+		ncpus++;
+#else
+		printf("%s: not started\n", ci->ci_dev.dv_xname);
+#endif
+		break;
+
+	default:
+		panic("unknown processor type??\n");
+	}
+#else	/* MULTIPROCESSOR */
+	printf("\n");
+#endif	/* !MULTIPROCESSOR */
+
+#ifdef MULTIPROCESSOR
+	if (mp_verbose) {
+		printf("%s: kstack at 0x%lx for %d bytes\n",
+		    ci->ci_dev.dv_xname, kstack, USPACE);
+		printf("%s: idle pcb at %p, idle sp at 0x%x\n",
+		    ci->ci_dev.dv_xname, pcb, pcb->pcb_esp);
+	}
+#endif
+}
+
+/*
+ * Initialize the processor appropriately.
+ */
+
+void
+cpu_init(ci)
+	struct cpu_info *ci;
+{
+	/* configure the CPU if needed */
+	if (ci->cpu_setup != NULL)
+		(*ci->cpu_setup)(NULL, 0, 0);
+
+#if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
+	/*
+	 * On a 486 or above, enable ring 0 write protection.
+	 */
+	if (ci->cpu_class >= CPUCLASS_486)
+		lcr0(rcr0() | CR0_WP);
+#endif
+	if (cpu_feature & CPUID_PGE)
+		lcr4(rcr4() | CR4_PGE);	/* enable global TLB caching */
+
+	ci->ci_flags |= CPUF_RUNNING;
+}
+
+
+#ifdef MULTIPROCESSOR
+
+void
+cpu_boot_secondary_processors()
+{
+	struct cpu_info *ci;
+	u_long i;
+
+	for (i = 0; i < I386_MAXPROCS; i++) {
+		ci = cpu_info[i];
+		if (ci == NULL)
+			continue;
+		if (ci->ci_idle_pcb == NULL)
+			continue;
+		if ((ci->ci_flags & CPUF_PRESENT) == 0)
+			continue;
+		if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
+			continue;
+		cpu_boot_secondary(ci);
+	}
+}
+
+void
+cpu_init_idle_pcbs()
+{
+	struct cpu_info *ci;
+	u_long i;
+
+	for (i=0; i < I386_MAXPROCS; i++) {
+		ci = cpu_info[i];
+		if (ci == NULL)
+			continue;
+		if (ci->ci_idle_pcb == NULL)
+			continue;
+		if ((ci->ci_flags & CPUF_PRESENT) == 0)
+			continue;
+		i386_init_pcb_tss_ldt(ci);
+	}
+}
+
+void
+cpu_boot_secondary (ci)
+	struct cpu_info *ci;
+{
+	struct pcb *pcb;
+	int i;
+	struct pmap *kpm = pmap_kernel();
+	extern u_int32_t mp_pdirpa;
+
+	printf("%s: starting", ci->ci_dev.dv_xname);
+
+	/* XXX move elsewhere, not per CPU. */
+	mp_pdirpa = vtophys(kpm->pm_pdir);
+
+	pcb = ci->ci_idle_pcb;
+
+	if (mp_verbose)
+		printf(", init idle stack ptr is 0x%x", pcb->pcb_esp);
+
+	printf("\n");
+
+	CPU_STARTUP(ci);
+
+	/*
+	 * wait for it to become ready
+	 */
+	for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
+		delay(10);
+	}
+	if (!(ci->ci_flags & CPUF_RUNNING)) {
+		printf("cpu failed to become ready\n");
+		Debugger();
+	}
+
+	CPU_START_CLEANUP(ci);
+}
+
+/*
+ * The CPU ends up here when its ready to run
+ * XXX should share some of this with init386 in machdep.c
+ * for now it jumps into an infinite loop.
+ */
+void
+cpu_hatch(void *v)
+{
+	struct cpu_info *ci = (struct cpu_info *)v;
+	int s;
+
+	cpu_init_idt();
+	lapic_enable();
+	lapic_initclocks();
+	lapic_set_lvt();
+	gdt_init_cpu(ci);
+	npxinit(ci);
+
+	lldt(GSEL(GLDT_SEL, SEL_KPL));
+
+	cpu_init(ci);
+
+	s = splhigh();		/* XXX prevent softints from running here.. */
+	lapic_tpr = 0;
+	enable_intr();
+	printf("%s: CPU %ld running\n", ci->ci_dev.dv_xname, ci->ci_cpuid);
+	splx(s);
+}
+
+void
+cpu_copy_trampoline()
+{
+	/*
+	 * Copy boot code.
+	 */
+	extern u_char cpu_spinup_trampoline[];
+	extern u_char cpu_spinup_trampoline_end[];
+
+	pmap_kenter_pa((vaddr_t)MP_TRAMPOLINE,	/* virtual */
+	    (paddr_t)MP_TRAMPOLINE,		/* physical */
+	    VM_PROT_ALL);			/* protection */
+	bcopy(cpu_spinup_trampoline, (caddr_t)MP_TRAMPOLINE,
+	    cpu_spinup_trampoline_end - cpu_spinup_trampoline);
+}
+
+#endif
+
+#ifdef notyet
+void
+cpu_init_tss(struct i386tss *tss, void *stack, void *func)
+{
+	memset(tss, 0, sizeof *tss);
+	tss->tss_esp0 = tss->tss_esp = (int)((char *)stack + USPACE - 16);
+	tss->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
+	tss->__tss_cs = GSEL(GCODE_SEL, SEL_KPL);
+	tss->tss_fs = GSEL(GCPU_SEL, SEL_KPL);
+	tss->tss_gs = tss->__tss_es = tss->__tss_ds =
+	    tss->__tss_ss = GSEL(GDATA_SEL, SEL_KPL);
+	tss->tss_cr3 = pmap_kernel()->pm_pdirpa;
+	tss->tss_esp = (int)((char *)stack + USPACE - 16);
+	tss->tss_ldt = GSEL(GLDT_SEL, SEL_KPL);
+	tss->__tss_eflags = PSL_MBO | PSL_NT;	/* XXX not needed? */
+	tss->__tss_eip = (int)func;
+}
+
+/* XXX */
+#define IDTVEC(name)	__CONCAT(X, name)
+typedef void (vector)(void);
+extern vector IDTVEC(tss_trap08);
+#ifdef DDB
+extern vector Xintrddbipi;
+extern int ddb_vec;
+#endif
+
+void
+cpu_set_tss_gates(struct cpu_info *ci)
+{
+	struct segment_descriptor sd;
+
+	ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE);
+	cpu_init_tss(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
+	    IDTVEC(tss_trap08));
+	setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
+	    SDT_SYS386TSS, SEL_KPL, 0, 0);
+	ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
+	setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
+	    GSEL(GTRAPTSS_SEL, SEL_KPL));
+
+#if defined(DDB) && defined(MULTIPROCESSOR)
+	/*
+	 * Set up seperate handler for the DDB IPI, so that it doesn't
+	 * stomp on a possibly corrupted stack.
+	 *
+	 * XXX overwriting the gate set in db_machine_init.
+	 * Should rearrange the code so that it's set only once.
+	 */
+	ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE);
+	cpu_init_tss(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
+	    Xintrddbipi);
+
+	setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
+	    SDT_SYS386TSS, SEL_KPL, 0, 0);
+	ci->ci_gdt[GIPITSS_SEL].sd = sd;
+
+	setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
+	    GSEL(GIPITSS_SEL, SEL_KPL));
+#endif
+}
+#endif
+
+#ifdef MULTIPROCESSOR
+int
+mp_cpu_start(struct cpu_info *ci)
+{
+#if NLAPIC > 0
+	int error;
+#endif
+	unsigned short dwordptr[2];
+
+	/*
+	 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
+	 */
+
+	outb(IO_RTC, NVRAM_RESET);
+	outb(IO_RTC+1, NVRAM_RESET_JUMP);
+
+	/*
+	 * "and the warm reset vector (DWORD based at 40:67) to point
+	 * to the AP startup code ..."
+	 */
+
+	dwordptr[0] = 0;
+	dwordptr[1] = MP_TRAMPOLINE >> 4;
+
+	pmap_kenter_pa(0, 0, VM_PROT_READ|VM_PROT_WRITE);
+	memcpy((u_int8_t *)0x467, dwordptr, 4);
+	pmap_kremove(0, PAGE_SIZE);
+
+#if NLAPIC > 0
+	/*
+	 * ... prior to executing the following sequence:"
+	 */
+
+	if (ci->ci_flags & CPUF_AP) {
+		if ((error = i386_ipi_init(ci->ci_apicid)) != 0)
+			return (error);
+
+		delay(10000);
+
+		if (cpu_feature & CPUID_APIC) {
+			if ((error = i386_ipi(MP_TRAMPOLINE / PAGE_SIZE,
+			    ci->ci_apicid, LAPIC_DLMODE_STARTUP)) != 0)
+				return (error);
+			delay(200);
+
+			if ((error = i386_ipi(MP_TRAMPOLINE / PAGE_SIZE,
+			    ci->ci_apicid, LAPIC_DLMODE_STARTUP)) != 0)
+				return (error);
+			delay(200);
+		}
+	}
+#endif
+	return (0);
+}
+
+void
+mp_cpu_start_cleanup(struct cpu_info *ci)
+{
+	/*
+	 * Ensure the NVRAM reset byte contains something vaguely sane.
+	 */
+
+	outb(IO_RTC, NVRAM_RESET);
+	outb(IO_RTC+1, NVRAM_RESET_RST);
+}
+#endif
diff --git a/sys/arch/i386/i386/db_interface.c b/sys/arch/i386/i386/db_interface.c
index c9d55df065e..44e09f1c914 100644
--- a/sys/arch/i386/i386/db_interface.c
+++ b/sys/arch/i386/i386/db_interface.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_interface.c,v 1.12 2003/05/18 02:43:12 andreas Exp $	*/
+/*	$OpenBSD: db_interface.c,v 1.13 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: db_interface.c,v 1.22 1996/05/03 19:42:00 christos Exp $	*/
 
 /* 
@@ -53,11 +53,24 @@
 extern label_t	*db_recover;
 extern char *trap_type[];
 extern int trap_types;
+extern boolean_t db_cmd_loop_done;
+
+#ifdef MULTIPROCESSOR
+extern boolean_t	 db_switch_cpu;
+extern long		 db_switch_to_cpu;
+
+#endif
 
 int	db_active = 0;
 
 void kdbprinttrap(int, int);
 void db_sysregs_cmd(db_expr_t, int, db_expr_t, char *);
+#ifdef MULTIPROCESSOR
+void db_cpuinfo_cmd(db_expr_t, int, db_expr_t, char *);
+void db_startproc_cmd(db_expr_t, int, db_expr_t, char *);
+void db_stopproc_cmd(db_expr_t, int, db_expr_t, char *);
+void db_ddbproc_cmd(db_expr_t, int, db_expr_t, char *);
+#endif /* MULTIPROCESSOR */
 
 /*
  * Print trap reason.
@@ -109,7 +122,7 @@ kdb_trap(type, code, regs)
 		 * Kernel mode - esp and ss not saved
 		 */
 		ddb_regs.tf_esp = (int)&regs->tf_esp;	/* kernel stack pointer */
-		asm("movw %%ss,%w0" : "=r" (ddb_regs.tf_ss));
+		__asm__("movw %%ss,%w0" : "=r" (ddb_regs.tf_ss));
 	}
 
 	s = splhigh();
@@ -179,20 +192,141 @@ db_sysregs_cmd(addr, have_addr, count, modif)
 	db_printf("cr4:    0x%08x\n", cr);
 }
 
+#ifdef MULTIPROCESSOR
+void db_cpuinfo_cmd(addr, have_addr, count, modif)
+	db_expr_t addr;
+	int have_addr;
+	db_expr_t count;
+	char *modif;
+{
+	int i;
+
+	for (i = 0; i < I386_MAXPROCS; i++) {
+		if (cpu_info[i] != NULL) {
+			db_printf("%c%4d: ", (i == cpu_number()) ? '*' : ' ',
+			    i);
+			switch(cpu_info[i]->ci_ddb_paused) {
+			case CI_DDB_RUNNING:
+				db_printf("running\n");
+				break;
+			case CI_DDB_SHOULDSTOP:
+				db_printf("stopping\n");
+				break;
+			case CI_DDB_STOPPED:
+				db_printf("stopped\n");
+				break;
+			case CI_DDB_ENTERDDB:
+				db_printf("entering ddb\n");
+				break;
+			case CI_DDB_INDDB:
+				db_printf("ddb\n");
+				break;
+			default:
+				db_printf("? (%d)\n",
+				    cpu_info[i]->ci_ddb_paused);
+				break;
+			}
+		}
+	}
+}
+
+void db_startproc_cmd(addr, have_addr, count, modif)
+	db_expr_t addr;
+	int have_addr;
+	db_expr_t count;
+	char *modif;
+{
+	int i;
+
+	if (have_addr) {
+		if (addr >= 0 && addr < I386_MAXPROCS
+		    && cpu_info[addr] != NULL && addr != cpu_number())
+			db_startcpu(addr);
+		else
+			db_printf("Invalid cpu %d\n", (int)addr);
+	} else {
+		for (i = 0; i < I386_MAXPROCS; i++) {
+			if (cpu_info[i] != NULL && i != cpu_number()) {
+				db_startcpu(i);
+			}
+		}
+	}
+}
+
+void db_stopproc_cmd(addr, have_addr, count, modif)
+	db_expr_t addr;
+	int have_addr;
+	db_expr_t count;
+	char *modif;
+{
+	int i;
+
+	if (have_addr) {
+		if (addr >= 0 && addr < I386_MAXPROCS
+		    && cpu_info[addr] != NULL && addr != cpu_number())
+			db_stopcpu(addr);
+		else
+			db_printf("Invalid cpu %d\n", (int)addr);
+	} else {
+		for (i = 0; i < I386_MAXPROCS; i++) {
+			if (cpu_info[i] != NULL && i != cpu_number()) {
+				db_stopcpu(i);
+			}
+		}
+	}
+}
+
+void db_ddbproc_cmd(addr, have_addr, count, modif)
+	db_expr_t addr;
+	int have_addr;
+	db_expr_t count;
+	char *modif;
+{
+	if (have_addr) {
+		if (addr >= 0 && addr < I386_MAXPROCS
+		    && cpu_info[addr] != NULL && addr != cpu_number()) {
+			db_switch_to_cpu = addr;
+			db_switch_cpu = 1;
+			db_cmd_loop_done = 1;
+		} else {
+			db_printf("Invalid cpu %d\n", (int)addr);
+		}
+	} else {
+		db_printf("CPU not specified\n");
+	}
+}
+#endif /* MULTIPROCESSOR */
+
 struct db_command db_machine_command_table[] = {
 	{ "sysregs",	db_sysregs_cmd,		0,	0 },
+#ifdef MULTIPROCESSOR
+	{ "cpuinfo",	db_cpuinfo_cmd,		0,	0 },
+	{ "startcpu",	db_startproc_cmd,	0,	0 },
+	{ "stopcpu",	db_stopproc_cmd,	0,	0 },
+	{ "ddbcpu",	db_ddbproc_cmd,		0,	0 },
+#endif /* MULTIPROCESSOR */
 	{ (char *)0, }
 };
 
 void
 db_machine_init()
 {
+#ifdef MULTIPROCESSOR
+	int i;
+#endif /* MULTIPROCESSOR */
 
 	db_machine_commands_install(db_machine_command_table);
+#ifdef MULTIPROCESSOR
+	for (i = 0; i < I386_MAXPROCS; i++) {
+		if (cpu_info[i] != NULL)
+			cpu_info[i]->ci_ddb_paused = CI_DDB_RUNNING;
+	}
+	SIMPLE_LOCK_INIT(&ddb_mp_slock);
+#endif /* MULTIPROCESSOR */
 }
 
 void
 Debugger()
 {
-	asm("int $3");
+	__asm__("int $3");
 }
diff --git a/sys/arch/i386/i386/db_mp.c b/sys/arch/i386/i386/db_mp.c
new file mode 100644
index 00000000000..0a859f08360
--- /dev/null
+++ b/sys/arch/i386/i386/db_mp.c
@@ -0,0 +1,187 @@
+/*	$OpenBSD: db_mp.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+
+/*
+ * Copyright (c) 2003 Andreas Gunnarsson
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *    - Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    - Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/types.h>
+#include <sys/simplelock.h>
+
+#include <machine/db_machdep.h>
+
+#include <ddb/db_output.h>
+
+#define DDB_STATE_NOT_RUNNING	0
+#define DDB_STATE_RUNNING	1
+
+struct SIMPLELOCK ddb_mp_slock;
+
+volatile int ddb_state = DDB_STATE_NOT_RUNNING;	/* protected by ddb_mp_slock */
+volatile cpuid_t ddb_active_cpu;		/* protected by ddb_mp_slock */
+
+/*
+ * ddb_enter_ddb() is called when ddb is entered to stop the other
+ * CPUs. If another cpu is already in ddb we'll wait until it's finished.
+ */
+void
+db_enter_ddb()
+{
+	int s, i;
+
+	s = splhigh();
+	SIMPLE_LOCK(&ddb_mp_slock);
+
+	while (ddb_state == DDB_STATE_RUNNING
+	    && ddb_active_cpu != cpu_number()) {
+		db_printf("CPU %d waiting to enter ddb\n", cpu_number());
+		SIMPLE_UNLOCK(&ddb_mp_slock);
+		splx(s);
+
+		/* Busy wait without locking, we'll confirm with lock later */
+		while (ddb_state == DDB_STATE_RUNNING
+		   && ddb_active_cpu != cpu_number())
+			;	/* Do nothing */
+
+		s = splhigh();
+		SIMPLE_LOCK(&ddb_mp_slock);
+	}
+
+	ddb_state = DDB_STATE_RUNNING;
+	ddb_active_cpu = cpu_number();
+
+	for (i = 0; i < I386_MAXPROCS; i++) {
+		if (cpu_info[i] != NULL) {
+			if (i == cpu_number())
+				cpu_info[i]->ci_ddb_paused = CI_DDB_INDDB;
+			else if (cpu_info[i]->ci_ddb_paused
+			    != CI_DDB_STOPPED) {
+				cpu_info[i]->ci_ddb_paused = CI_DDB_SHOULDSTOP;
+				db_printf("Sending IPI to cpu %d\n", i);
+				i386_send_ipi(cpu_info[i], I386_IPI_DDB);
+			}
+		}
+	}
+	db_printf("CPU %d entering ddb\n", cpu_number());
+	SIMPLE_UNLOCK(&ddb_mp_slock);
+	splx(s);
+}
+
+void
+db_leave_ddb()
+{
+	int s, i;
+
+	s = splhigh();
+	SIMPLE_LOCK(&ddb_mp_slock);
+	db_printf("CPU %d leaving ddb\n", cpu_number());
+	for (i = 0; i < I386_MAXPROCS; i++) {
+		if (cpu_info[i] != NULL) {
+			cpu_info[i]->ci_ddb_paused = CI_DDB_RUNNING;
+		}
+	}
+	ddb_state = DDB_STATE_NOT_RUNNING;
+	SIMPLE_UNLOCK(&ddb_mp_slock);
+	splx(s);
+}
+
+void
+db_startcpu(int cpu)
+{
+	int s;
+
+	if (cpu != cpu_number() && cpu_info[cpu] != NULL) {
+		s = splhigh();
+		SIMPLE_LOCK(&ddb_mp_slock);
+		cpu_info[cpu]->ci_ddb_paused = CI_DDB_RUNNING;
+		SIMPLE_UNLOCK(&ddb_mp_slock);
+		splx(s);
+	}
+}
+
+void
+db_stopcpu(int cpu)
+{
+	int s;
+
+	if (cpu != cpu_number() && cpu_info[cpu] != NULL) {
+		s = splhigh();
+		SIMPLE_LOCK(&ddb_mp_slock);
+		cpu_info[cpu]->ci_ddb_paused = CI_DDB_SHOULDSTOP;
+		db_printf("Sending IPI to cpu %d\n", cpu);
+		SIMPLE_UNLOCK(&ddb_mp_slock);
+		splx(s);
+		i386_send_ipi(cpu_info[cpu], I386_IPI_DDB);
+	}
+}
+
+void
+db_movetocpu(int cpu)
+{
+	int s;
+
+	s = splhigh();
+	SIMPLE_LOCK(&ddb_mp_slock);
+	cpu_info[cpu]->ci_ddb_paused = CI_DDB_ENTERDDB;
+	db_printf("Sending IPI to cpu %d\n", cpu);
+	SIMPLE_UNLOCK(&ddb_mp_slock);
+	splx(s);
+	/* XXX If other CPU was running and IPI is lost, we lose. */
+	i386_send_ipi(cpu_info[cpu], I386_IPI_DDB);
+}
+
+void
+i386_ipi_db(struct cpu_info *ci)
+{
+	int s;
+
+	s = splhigh();
+	SIMPLE_LOCK(&ddb_mp_slock);
+	db_printf("CPU %d received ddb IPI\n", cpu_number());
+	while (ci->ci_ddb_paused == CI_DDB_SHOULDSTOP
+	    || ci->ci_ddb_paused == CI_DDB_STOPPED) {
+		if (ci->ci_ddb_paused == CI_DDB_SHOULDSTOP)
+			ci->ci_ddb_paused = CI_DDB_STOPPED;
+		SIMPLE_UNLOCK(&ddb_mp_slock);
+		while (ci->ci_ddb_paused == CI_DDB_STOPPED)
+			;	/* Do nothing */
+		SIMPLE_LOCK(&ddb_mp_slock);
+	}
+	if (ci->ci_ddb_paused == CI_DDB_ENTERDDB) {
+		ddb_state = DDB_STATE_RUNNING;
+		ddb_active_cpu = cpu_number();
+		ci->ci_ddb_paused = CI_DDB_INDDB;
+		db_printf("CPU %d grabbing ddb\n", cpu_number());
+		SIMPLE_UNLOCK(&ddb_mp_slock);
+		Debugger();
+		SIMPLE_LOCK(&ddb_mp_slock);
+		ci->ci_ddb_paused = CI_DDB_RUNNING;
+	}
+	db_printf("CPU %d leaving ddb IPI handler\n", cpu_number());
+	SIMPLE_UNLOCK(&ddb_mp_slock);
+	splx(s);
+}
diff --git a/sys/arch/i386/i386/gdt.c b/sys/arch/i386/i386/gdt.c
index 3bced92013f..5bb2775f9be 100644
--- a/sys/arch/i386/i386/gdt.c
+++ b/sys/arch/i386/i386/gdt.c
@@ -1,8 +1,8 @@
-/*	$OpenBSD: gdt.c,v 1.20 2003/11/08 05:38:33 nordin Exp $	*/
-/*	$NetBSD: gdt.c,v 1.8 1996/05/03 19:42:06 christos Exp $	*/
+/*	$OpenBSD: gdt.c,v 1.21 2004/06/13 21:49:15 niklas Exp $	*/
+/*	$NetBSD: gdt.c,v 1.28 2002/12/14 09:38:50 junyoung Exp $	*/
 
 /*-
- * Copyright (c) 1996 The NetBSD Foundation, Inc.
+ * Copyright (c) 1996, 1997 The NetBSD Foundation, Inc.
  * All rights reserved.
  *
  * This code is derived from software contributed to The NetBSD Foundation
@@ -27,8 +27,8 @@
  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
- * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE
- * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
@@ -37,27 +37,49 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+/*
+ * The GDT handling has two phases.  During the early lifetime of the
+ * kernel there is a static gdt which will be stored in bootstrap_gdt.
+ * Later, when the virtual memory is initialized, this will be
+ * replaced with a dynamically resizable GDT (although, we will only
+ * ever be growing it, there is almost no gain at all to compact it,
+ * and it has proven to be a complicated thing to do, considering
+ * parallel access, so it's just not worth the effort.
+ *
+ * The static GDT area will hold the initial requirement of NGDT descriptors.
+ * The dynamic GDT will have a statically sized virtual memory area of size
+ * GDTMAXPAGES, the physical area backing this will be allocated as needed
+ * starting with the size needed for holding a copy of the bootstrap gdt.
+ *
+ * Every CPU in a system has its own copy of the GDT.  The only real difference
+ * between the two are currently that there is a cpu-specific segment holding
+ * the struct cpu_info of the processor, for simplicity at getting cpu_info
+ * fields from assembly.  The boot processor will actually refer to the global
+ * copy of the GDT as pointed to by the gdt variable.
+ */
+
+#include <sys/cdefs.h>
+
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/proc.h>
+#include <sys/lock.h>
 #include <sys/user.h>
 
-#include <uvm/uvm_extern.h>
+#include <uvm/uvm.h>
 
 #include <machine/gdt.h>
 
-#define	MINGDTSIZ	512
-#define	MAXGDTSIZ	8192
+union descriptor bootstrap_gdt[NGDT];
+union descriptor *gdt = bootstrap_gdt;
 
-union descriptor *dynamic_gdt = gdt;
-int gdt_size = NGDT;		/* total number of GDT entries */
-int gdt_count = NGDT;		/* number of GDT entries in use */
-int gdt_next = NGDT;		/* next available slot for sweeping */
-int gdt_free = GNULL_SEL;	/* next free slot; terminated with GNULL_SEL */
+int gdt_size;		/* total number of GDT entries */
+int gdt_count;		/* number of GDT entries in use */
+int gdt_next;		/* next available slot for sweeping */
+int gdt_free;		/* next free slot; terminated with GNULL_SEL */
 
-int gdt_flags;
-#define	GDT_LOCKED	0x1
-#define	GDT_WANTED	0x2
+struct simplelock gdt_simplelock;
+struct lock gdt_lock_store;
 
 static __inline void gdt_lock(void);
 static __inline void gdt_unlock(void);
@@ -72,23 +94,31 @@ void gdt_put_slot(int);
 static __inline void
 gdt_lock()
 {
-
-	while ((gdt_flags & GDT_LOCKED) != 0) {
-		gdt_flags |= GDT_WANTED;
-		tsleep(&gdt_flags, PZERO, "gdtlck", 0);
-	}
-	gdt_flags |= GDT_LOCKED;
+	if (curproc != NULL)
+		lockmgr(&gdt_lock_store, LK_EXCLUSIVE, &gdt_simplelock,
+		    curproc);
 }
 
 static __inline void
 gdt_unlock()
 {
+	if (curproc != NULL)
+		lockmgr(&gdt_lock_store, LK_RELEASE, &gdt_simplelock, curproc);
+}
 
-	gdt_flags &= ~GDT_LOCKED;
-	if ((gdt_flags & GDT_WANTED) != 0) {
-		gdt_flags &= ~GDT_WANTED;
-		wakeup(&gdt_flags);
-	}
+/* XXX needs spinlocking if we ever mean to go finegrained. */
+void
+setgdt(int sel, void *base, size_t limit, int type, int dpl, int def32,
+    int gran)
+{
+	struct segment_descriptor *sd = &gdt[sel].sd;
+	CPU_INFO_ITERATOR cii;
+	struct cpu_info *ci;
+
+	setsegment(sd, base, limit, type, dpl, def32, gran);
+	for (CPU_INFO_FOREACH(cii, ci))
+		if (ci->ci_gdt != NULL && ci->ci_gdt != gdt)
+			ci->ci_gdt[sel].sd = *sd;
 }
 
 /*
@@ -98,18 +128,69 @@ void
 gdt_init()
 {
 	size_t max_len, min_len;
-	struct region_descriptor region;
+	struct vm_page *pg;
+	vaddr_t va;
+	struct cpu_info *ci = &cpu_info_primary;
+
+	simple_lock_init(&gdt_simplelock);
+	lockinit(&gdt_lock_store, PZERO, "gdtlck", 0, 0);
 
 	max_len = MAXGDTSIZ * sizeof(union descriptor);
 	min_len = MINGDTSIZ * sizeof(union descriptor);
+
 	gdt_size = MINGDTSIZ;
+	gdt_count = NGDT;
+	gdt_next = NGDT;
+	gdt_free = GNULL_SEL;
+
+	gdt = (union descriptor *)uvm_km_valloc(kernel_map, max_len);
+	for (va = (vaddr_t)gdt; va < (vaddr_t)gdt + min_len; va += PAGE_SIZE) {
+		pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_ZERO);
+		if (pg == NULL)
+			panic("gdt_init: no pages");
+		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg),
+		    VM_PROT_READ | VM_PROT_WRITE);
+	}
+	bcopy(bootstrap_gdt, gdt, NGDT * sizeof(union descriptor));
+	ci->ci_gdt = gdt;
+	setsegment(&ci->ci_gdt[GCPU_SEL].sd, ci, sizeof(struct cpu_info)-1,
+	    SDT_MEMRWA, SEL_KPL, 1, 1);
 
-	dynamic_gdt = (union descriptor *)uvm_km_valloc(kernel_map, max_len);
-	uvm_map_pageable(kernel_map, (vaddr_t)dynamic_gdt,
-	    (vaddr_t)dynamic_gdt + min_len, FALSE, FALSE);
-	bcopy(gdt, dynamic_gdt, NGDT * sizeof(union descriptor));
+	gdt_init_cpu(ci);
+}
 
-	setregion(&region, dynamic_gdt, max_len - 1);
+#ifdef MULTIPROCESSOR
+/*
+ * Allocate shadow GDT for a slave cpu.
+ */
+void
+gdt_alloc_cpu(struct cpu_info *ci)
+{
+	int max_len = MAXGDTSIZ * sizeof(union descriptor);
+	int min_len = MINGDTSIZ * sizeof(union descriptor);
+
+	ci->ci_gdt = (union descriptor *)uvm_km_valloc(kernel_map, max_len);
+	uvm_map_pageable(kernel_map, (vaddr_t)ci->ci_gdt,
+	    (vaddr_t)ci->ci_gdt + min_len, FALSE, FALSE);
+	bzero(ci->ci_gdt, min_len);
+	bcopy(gdt, ci->ci_gdt, gdt_count * sizeof(union descriptor));
+	setsegment(&ci->ci_gdt[GCPU_SEL].sd, ci, sizeof(struct cpu_info)-1,
+	    SDT_MEMRWA, SEL_KPL, 1, 1);
+}
+#endif	/* MULTIPROCESSOR */
+
+
+/*
+ * Load appropriate gdt descriptor; we better be running on *ci
+ * (for the most part, this is how a cpu knows who it is).
+ */
+void
+gdt_init_cpu(struct cpu_info *ci)
+{
+	struct region_descriptor region;
+
+	setregion(&region, ci->ci_gdt,
+	    MAXGDTSIZ * sizeof(union descriptor) - 1);
 	lgdt(&region);
 }
 
@@ -120,13 +201,29 @@ void
 gdt_grow()
 {
 	size_t old_len, new_len;
+	CPU_INFO_ITERATOR cii;
+	struct cpu_info *ci;
+	struct vm_page *pg;
+	vaddr_t va;
 
 	old_len = gdt_size * sizeof(union descriptor);
 	gdt_size <<= 1;
 	new_len = old_len << 1;
 
-	uvm_map_pageable(kernel_map, (vaddr_t)dynamic_gdt + old_len,
-	    (vaddr_t)dynamic_gdt + new_len, FALSE, FALSE);
+	for (CPU_INFO_FOREACH(cii, ci)) {
+		for (va = (vaddr_t)(ci->ci_gdt) + old_len;
+		     va < (vaddr_t)(ci->ci_gdt) + new_len;
+		     va += PAGE_SIZE) {
+			while (
+			    (pg =
+			    uvm_pagealloc(NULL, 0, NULL, UVM_PGA_ZERO)) ==
+			    NULL) {
+				uvm_wait("gdt_grow");
+			}
+			pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg),
+			    VM_PROT_READ | VM_PROT_WRITE);
+		}
+	}
 }
 
 /*
@@ -146,15 +243,13 @@ gdt_get_slot()
 
 	if (gdt_free != GNULL_SEL) {
 		slot = gdt_free;
-		gdt_free = dynamic_gdt[slot].gd.gd_selector;
+		gdt_free = gdt[slot].gd.gd_selector;
 	} else {
 		if (gdt_next != gdt_count)
 			panic("gdt_get_slot: gdt_next != gdt_count");
 		if (gdt_next >= gdt_size) {
 			if (gdt_size >= MAXGDTSIZ)
 				panic("gdt_get_slot: out of GDT descriptors");
-			if (dynamic_gdt == gdt)
-				panic("gdt_get_slot called before gdt_init");
 			gdt_grow();
 		}
 		slot = gdt_next++;
@@ -169,65 +264,56 @@ gdt_get_slot()
  * Deallocate a GDT slot, putting it on the free list.
  */
 void
-gdt_put_slot(slot)
-	int slot;
+gdt_put_slot(int slot)
 {
 
 	gdt_lock();
 	gdt_count--;
 
-	dynamic_gdt[slot].gd.gd_type = SDT_SYSNULL;
-	dynamic_gdt[slot].gd.gd_selector = gdt_free;
+	gdt[slot].gd.gd_type = SDT_SYSNULL;
+	gdt[slot].gd.gd_selector = gdt_free;
 	gdt_free = slot;
 
 	gdt_unlock();
 }
 
-void
-tss_alloc(pcb)
-	struct pcb *pcb;
+int
+tss_alloc(struct pcb *pcb)
 {
 	int slot;
 
 	slot = gdt_get_slot();
-	setsegment(&dynamic_gdt[slot].sd, &pcb->pcb_tss, sizeof(struct pcb) - 1,
+	setgdt(slot, &pcb->pcb_tss, sizeof(struct pcb) - 1,
 	    SDT_SYS386TSS, SEL_KPL, 0, 0);
-	pcb->pcb_tss_sel = GSEL(slot, SEL_KPL);
+	return GSEL(slot, SEL_KPL);
 }
 
 void
-tss_free(pcb)
-	struct pcb *pcb;
+tss_free(int sel)
 {
 
-	gdt_put_slot(IDXSEL(pcb->pcb_tss_sel));
+	gdt_put_slot(IDXSEL(sel));
 }
 
+/*
+ * Caller must have pmap locked for both of these functions.
+ */
 void
-ldt_alloc(pmap, ldt, len)
-	struct pmap *pmap;
-	union descriptor *ldt;
-	size_t len;
+ldt_alloc(struct pmap *pmap, union descriptor *ldt, size_t len)
 {
 	int slot;
 
 	slot = gdt_get_slot();
-	setsegment(&dynamic_gdt[slot].sd, ldt, len - 1, SDT_SYSLDT, SEL_KPL, 0,
-	    0);
-	simple_lock(&pmap->pm_lock);
+	setgdt(slot, ldt, len - 1, SDT_SYSLDT, SEL_KPL, 0, 0);
 	pmap->pm_ldt_sel = GSEL(slot, SEL_KPL);
-	simple_unlock(&pmap->pm_lock);
 }
 
 void
-ldt_free(pmap)
-	struct pmap *pmap;
+ldt_free(struct pmap *pmap)
 {
 	int slot;
 
-	simple_lock(&pmap->pm_lock);
 	slot = IDXSEL(pmap->pm_ldt_sel);
-	simple_unlock(&pmap->pm_lock);
 
 	gdt_put_slot(slot);
 }
diff --git a/sys/arch/i386/i386/genassym.cf b/sys/arch/i386/i386/genassym.cf
index e74f22cae5f..cc5149eb722 100644
--- a/sys/arch/i386/i386/genassym.cf
+++ b/sys/arch/i386/i386/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.13 2003/06/02 23:27:47 millert Exp $
+#	$OpenBSD: genassym.cf,v 1.14 2004/06/13 21:49:15 niklas Exp $
 #
 # Copyright (c) 1982, 1990 The Regents of the University of California.
 # All rights reserved.
@@ -57,12 +57,17 @@ if COMPAT_FREEBSD > 0
 include <machine/freebsd_machdep.h>
 endif
 
+if MULTIPROCESSOR
+include <machine/cpu.h>
+endif
+
 include "isa.h"
 if NISA > 0
 include <i386/isa/isa_machdep.h>
 endif
 
 export	SRUN
+export	SONPROC
 
 # values for page tables
 export	PDSLOT_KERN
@@ -84,6 +89,7 @@ member	p_stat
 member	p_wchan
 member	p_vmspace
 member	p_flag
+member	p_cpu
 
 export	P_SYSTEM
 
@@ -101,8 +107,8 @@ member	pcb_fs
 member	pcb_gs
 member	pcb_cr0
 member	pcb_ldt_sel
-member	pcb_tss_sel
 member	pcb_onfault
+member	pcb_fpcpu
 
 # frame definitions
 struct	trapframe
@@ -169,3 +175,26 @@ member	ih_arg
 member	ih_count
 member	ih_next
 endif
+
+define	P_MD_TSS_SEL		offsetof(struct proc, p_md.md_tss_sel)
+
+define	CPU_INFO_SELF		offsetof(struct cpu_info, ci_self)
+define	CPU_INFO_CURPROC	offsetof(struct cpu_info, ci_curproc)
+define	CPU_INFO_CURPCB		offsetof(struct cpu_info, ci_curpcb)
+define	CPU_INFO_NAME		offsetof(struct cpu_info, ci_dev.dv_xname)
+define	CPU_INFO_IDLE_PCB	offsetof(struct cpu_info, ci_idle_pcb)
+define	CPU_INFO_IDLE_TSS_SEL	offsetof(struct cpu_info, ci_idle_tss_sel)
+define	CPU_INFO_LEVEL		offsetof(struct cpu_info, ci_level)
+define	CPU_INFO_VENDOR		offsetof(struct cpu_info, ci_vendor[0])
+define	CPU_INFO_SIGNATURE	offsetof(struct cpu_info, ci_signature)
+define	CPU_INFO_RESCHED	offsetof(struct cpu_info, ci_want_resched)
+define	CPU_INFO_ASTPENDING	offsetof(struct cpu_info, ci_astpending)
+define	CPU_INFO_GDT		offsetof(struct cpu_info, ci_gdt)
+define	CPU_INFO_IPENDING	offsetof(struct cpu_info, ci_ipending)
+define	CPU_INFO_IMASK		offsetof(struct cpu_info, ci_imask)
+define	CPU_INFO_IUNMASK	offsetof(struct cpu_info, ci_iunmask)
+define	CPU_INFO_ILEVEL		offsetof(struct cpu_info, ci_ilevel)
+define	CPU_INFO_IDEPTH		offsetof(struct cpu_info, ci_idepth)
+define	CPU_INFO_ISOURCES	offsetof(struct cpu_info, ci_isources)
+
+define	SIZEOF_CPU_INFO		sizeof(struct cpu_info)
diff --git a/sys/arch/i386/i386/ioapic.c b/sys/arch/i386/i386/ioapic.c
new file mode 100644
index 00000000000..e5b329372f3
--- /dev/null
+++ b/sys/arch/i386/i386/ioapic.c
@@ -0,0 +1,736 @@
+/*	$OpenBSD: ioapic.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/* 	$NetBSD: ioapic.c,v 1.7 2003/07/14 22:32:40 lukem Exp $	*/
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+
+/*
+ * Copyright (c) 1999 Stefan Grefen
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed by the NetBSD
+ *      Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/device.h>
+#include <sys/malloc.h>
+
+#include <machine/bus.h>
+#include <machine/psl.h>
+
+#include <uvm/uvm_extern.h>
+
+#include <machine/i82093reg.h>
+#include <machine/i82093var.h>
+
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+
+#include <machine/pmap.h>
+
+#include <machine/mpbiosvar.h>
+
+#include "isa.h"
+
+/*
+ * XXX locking
+ */
+
+int     ioapic_match(struct device *, void *, void *);
+void    ioapic_attach(struct device *, struct device *, void *);
+
+/* XXX */
+extern int bus_mem_add_mapping(bus_addr_t, bus_size_t, int,
+    bus_space_handle_t *);
+
+void	apic_set_redir(struct ioapic_softc *, int);
+void	apic_vectorset(struct ioapic_softc *, int, int, int);
+
+int apic_verbose = 0;
+
+int ioapic_bsp_id = 0;
+int ioapic_cold = 1;
+
+struct ioapic_softc *ioapics;	 /* head of linked list */
+int nioapics = 0;	   	 /* number attached */
+
+/*
+ * Register read/write routines.
+ */
+static __inline u_int32_t
+ioapic_read(struct ioapic_softc *sc, int regid)
+{
+	u_int32_t val;
+
+	/*
+	 * XXX lock apic
+	 */
+	*(sc->sc_reg) = regid;
+	val = *sc->sc_data;
+
+	return (val);
+
+}
+
+static __inline void
+ioapic_write(struct ioapic_softc *sc, int regid, int val)
+{
+	/*
+	 * XXX lock apic
+	 */
+	*(sc->sc_reg) = regid;
+	*(sc->sc_data) = val;
+}
+
+struct ioapic_softc *
+ioapic_find(int apicid)
+{
+	struct ioapic_softc *sc;
+
+	if (apicid == MPS_ALL_APICS) {	/* XXX mpbios-specific */
+		/*
+		 * XXX kludge for all-ioapics interrupt support
+		 * on single ioapic systems
+		 */
+		if (nioapics <= 1)
+			return (ioapics);
+		panic("unsupported: all-ioapics interrupt with >1 ioapic");
+	}
+
+	for (sc = ioapics; sc != NULL; sc = sc->sc_next)
+		if (sc->sc_apicid == apicid)
+			return (sc);
+
+	return (NULL);
+}
+
+static __inline void
+ioapic_add(struct ioapic_softc *sc)
+{
+	sc->sc_next = ioapics;
+	ioapics = sc;
+	nioapics++;
+}
+
+void
+ioapic_print_redir(struct ioapic_softc *sc, char *why, int pin)
+{
+	u_int32_t redirlo = ioapic_read(sc, IOAPIC_REDLO(pin));
+	u_int32_t redirhi = ioapic_read(sc, IOAPIC_REDHI(pin));
+
+	apic_format_redir(sc->sc_dev.dv_xname, why, pin, redirhi, redirlo);
+}
+
+struct cfattach ioapic_ca = {
+	sizeof(struct ioapic_softc), ioapic_match, ioapic_attach
+};
+
+struct cfdriver ioapic_cd = {
+	NULL, "ioapic", DV_DULL /* XXX DV_CPU ? */
+};
+
+int
+ioapic_match(struct device *parent, void *matchv, void *aux)
+{
+        struct cfdata *match = (struct cfdata *)matchv;
+	struct apic_attach_args * aaa = (struct apic_attach_args *)aux;
+
+	if (strcmp(aaa->aaa_name, match->cf_driver->cd_name) == 0)
+		return (1);
+	return (0);
+}
+
+
+/*
+ * can't use bus_space_xxx as we don't have a bus handle ...
+ */
+void
+ioapic_attach(struct device *parent, struct device *self, void *aux)
+{
+	struct ioapic_softc *sc = (struct ioapic_softc *)self;
+	struct apic_attach_args  *aaa = (struct apic_attach_args *)aux;
+	int apic_id;
+	bus_space_handle_t bh;
+	u_int32_t ver_sz;
+	int i;
+
+	sc->sc_flags = aaa->flags;
+	sc->sc_apicid = aaa->apic_id;
+
+	printf(" apid %d", aaa->apic_id);
+
+	if (ioapic_find(aaa->apic_id) != NULL) {
+		printf(": duplicate apic id (ignored)\n");
+		return;
+	}
+
+	ioapic_add(sc);
+
+	printf(": pa 0x%lx", aaa->apic_address);
+
+	if (bus_mem_add_mapping(aaa->apic_address, PAGE_SIZE, 0, &bh) != 0) {
+		printf(", map failed\n");
+		return;
+	}
+	sc->sc_reg = (volatile u_int32_t *)(bh + IOAPIC_REG);
+	sc->sc_data = (volatile u_int32_t *)(bh + IOAPIC_DATA);
+
+	apic_id = (ioapic_read(sc,IOAPIC_ID) & IOAPIC_ID_MASK) >>
+	    IOAPIC_ID_SHIFT;
+	ver_sz = ioapic_read(sc, IOAPIC_VER);
+
+	sc->sc_apic_vers = (ver_sz & IOAPIC_VER_MASK) >> IOAPIC_VER_SHIFT;
+	sc->sc_apic_sz = (ver_sz & IOAPIC_MAX_MASK) >> IOAPIC_MAX_SHIFT;
+	sc->sc_apic_sz++;
+
+	if (mp_verbose) {
+		printf(", %s mode",
+		    aaa->flags & IOAPIC_PICMODE ? "PIC" : "virtual wire");
+	}
+
+	printf(", version %x, %d pins\n", sc->sc_apic_vers, sc->sc_apic_sz);
+
+	sc->sc_pins = malloc(sizeof(struct ioapic_pin) * sc->sc_apic_sz,
+	    M_DEVBUF, M_WAITOK);
+
+	for (i=0; i<sc->sc_apic_sz; i++) {
+		sc->sc_pins[i].ip_handler = NULL;
+		sc->sc_pins[i].ip_next = NULL;
+		sc->sc_pins[i].ip_map = NULL;
+		sc->sc_pins[i].ip_vector = 0;
+		sc->sc_pins[i].ip_type = 0;
+		sc->sc_pins[i].ip_minlevel = 0xff; /* XXX magic*/
+		sc->sc_pins[i].ip_maxlevel = 0;	/* XXX magic */
+	}
+
+	/*
+	 * In case the APIC is not initialized to the correct ID
+	 * do it now.
+	 * Maybe we should record the original ID for interrupt
+	 * mapping later ...
+	 */
+	if (apic_id != sc->sc_apicid) {
+		printf("%s: misconfigured as apic %d", sc->sc_dev.dv_xname,
+		    apic_id);
+
+		ioapic_write(sc, IOAPIC_ID,
+		    (ioapic_read(sc, IOAPIC_ID) & ~IOAPIC_ID_MASK)
+		    | (sc->sc_apicid << IOAPIC_ID_SHIFT));
+
+		apic_id = (ioapic_read(sc,IOAPIC_ID) & IOAPIC_ID_MASK) >>
+		    IOAPIC_ID_SHIFT;
+
+		if (apic_id != sc->sc_apicid)
+			printf(", can't remap to apid %d\n", sc->sc_apicid);
+		else
+			printf(", remapped to apic %d\n", sc->sc_apicid);
+	}
+#if 0
+	/* output of this was boring. */
+	if (mp_verbose)
+		for (i=0; i<sc->sc_apic_sz; i++)
+			ioapic_print_redir(sc, "boot", i);
+#endif
+}
+
+/*
+ * Interrupt mapping.
+ *
+ * Multiple handlers may exist for each pin, so there's an
+ * intrhand chain for each pin.
+ *
+ * Ideally, each pin maps to a single vector at the priority of the
+ * highest level interrupt for that pin.
+ *
+ * XXX in the event that there are more than 16 interrupt sources at a
+ * single level, some doubling-up may be needed.  This is not yet
+ * implemented.
+ *
+ * XXX we are wasting some space here because we only use a limited
+ * range of the vectors here.  (0x30..0xef)
+ */
+
+struct intrhand *apic_intrhand[256];
+int	apic_intrcount[256];
+
+
+/* XXX should check vs. softc max int number */
+#define	LEGAL_IRQ(x)	((x) >= 0 && (x) < APIC_ICU_LEN && (x) != 2)
+
+void
+apic_set_redir(struct ioapic_softc *sc, int pin)
+{
+	u_int32_t redlo;
+	u_int32_t redhi = 0;
+	int delmode;
+
+	struct ioapic_pin *pp;
+	struct mp_intr_map *map;
+
+	pp = &sc->sc_pins[pin];
+	map = pp->ip_map;
+	if (map == NULL) {
+		redlo = IOAPIC_REDLO_MASK;
+	} else {
+		redlo = map->redir;
+	}
+	delmode = (redlo & IOAPIC_REDLO_DEL_MASK) >> IOAPIC_REDLO_DEL_SHIFT;
+
+	/* XXX magic numbers */
+	if ((delmode != 0) && (delmode != 1))
+		;
+	else if (pp->ip_handler == NULL) {
+		redlo |= IOAPIC_REDLO_MASK;
+	} else {
+		redlo |= (pp->ip_vector & 0xff);
+		redlo |= (IOAPIC_REDLO_DEL_FIXED << IOAPIC_REDLO_DEL_SHIFT);
+		redlo &= ~IOAPIC_REDLO_DSTMOD;
+
+		/*
+		 * Destination: BSP CPU
+		 *
+		 * XXX will want to distribute interrupts across cpu's
+		 * eventually.  most likely, we'll want to vector each
+		 * interrupt to a specific CPU and load-balance across
+		 * cpu's.  but there's no point in doing that until after 
+		 * most interrupts run without the kernel lock.  
+		 */
+		redhi |= (ioapic_bsp_id << IOAPIC_REDHI_DEST_SHIFT);
+
+		/* XXX derive this bit from BIOS info */
+		if (pp->ip_type == IST_LEVEL)
+			redlo |= IOAPIC_REDLO_LEVEL;
+		else
+			redlo &= ~IOAPIC_REDLO_LEVEL;
+		if (map != NULL && ((map->flags & 3) == MPS_INTPO_DEF)) {
+			if (pp->ip_type == IST_LEVEL)
+				redlo |= IOAPIC_REDLO_ACTLO;
+			else
+				redlo &= ~IOAPIC_REDLO_ACTLO;
+		}
+	}
+	/* Do atomic write */
+	ioapic_write(sc, IOAPIC_REDLO(pin), IOAPIC_REDLO_MASK);
+	ioapic_write(sc, IOAPIC_REDHI(pin), redhi);
+	ioapic_write(sc, IOAPIC_REDLO(pin), redlo);
+	if (mp_verbose)
+		ioapic_print_redir(sc, "int", pin);
+}
+
+/*
+ * XXX To be really correct an NISA > 0 condition should check for these.
+ * However, the i386 port pretty much assumes isa is there anyway.
+ * For example, pci_intr_establish calls isa_intr_establish unconditionally.
+ */
+extern int fakeintr(void *); 	/* XXX headerify */
+extern char *isa_intr_typename(int); 	/* XXX headerify */
+
+/*
+ * apic_vectorset: allocate a vector for the given pin, based on
+ * the levels of the interrupts on that pin.
+ *
+ * XXX if the level of the pin changes while the pin is
+ * masked, need to do something special to prevent pending
+ * interrupts from being lost.
+ * (the answer may be to hang the interrupt chain off of both vectors
+ * until any interrupts from the old source have been handled.  the trouble
+ * is that we don't have a global view of what interrupts are pending.
+ *
+ * Deferring for now since MP systems are more likely servers rather
+ * than laptops or desktops, and thus will have relatively static
+ * interrupt configuration.
+ */
+
+void
+apic_vectorset(struct ioapic_softc *sc, int pin, int minlevel, int maxlevel)
+{
+	struct ioapic_pin *pp = &sc->sc_pins[pin];
+	int ovector = 0;
+	int nvector = 0;
+
+	ovector = pp->ip_vector;
+	
+	if (maxlevel == 0) {
+		/* no vector needed. */
+		pp->ip_minlevel = 0xff; /* XXX magic */
+		pp->ip_maxlevel = 0; /* XXX magic */
+		pp->ip_vector = 0;
+	} else if (maxlevel != pp->ip_maxlevel) {
+		if (minlevel != maxlevel) {
+			printf("%s: WARNING: sharing interrupt "
+			    "between different IPLs (currently broken)\n",
+			    sc->sc_dev.dv_xname);
+			printf("%s: pin %d, ipls %x..%x\n",
+			    sc->sc_dev.dv_xname,
+			    pin, minlevel, maxlevel);
+		}
+
+		/*
+		 * Allocate interrupt vector at the *lowest* priority level
+		 * of any of the handlers invoked by this pin.
+		 *
+		 * The interrupt handler will raise ipl higher than this
+		 * as appropriate.
+		 */
+		nvector = idt_vec_alloc(maxlevel, maxlevel+15);
+
+		if (nvector == 0) {
+			/*
+			 * XXX XXX we should be able to deal here..
+			 * need to double-up an existing vector
+			 * and install a slightly different handler.
+			 */
+			panic("%s: can't alloc vector for pin %d at level %x",
+			    sc->sc_dev.dv_xname, pin, maxlevel);
+		}
+		/* 
+		 * XXX want special handler for the maxlevel != minlevel
+		 * case here!
+		 */
+		idt_vec_set(nvector, apichandler[nvector & 0xf]);
+		pp->ip_vector = nvector;
+		pp->ip_minlevel = minlevel;
+		pp->ip_maxlevel = maxlevel;
+	}
+	apic_intrhand[pp->ip_vector] = pp->ip_handler;
+
+	if (ovector) {
+		/*
+		 * XXX should defer this until we're sure the old vector
+		 * doesn't have a pending interrupt on any processor.
+		 * do this by setting a counter equal to the number of CPU's,
+		 * and firing off a low-priority broadcast IPI to all cpu's.
+		 * each cpu then decrements the counter; when it
+		 * goes to zero, free the vector..
+		 * i.e., defer until all processors have run with a CPL
+		 * less than the level of the interrupt..
+		 *
+		 * this is only an issue for dynamic interrupt configuration
+		 * (e.g., cardbus or pcmcia).
+		 */
+		apic_intrhand[ovector] = NULL;
+		idt_vec_free(ovector);
+		printf("freed vector %x\n", ovector);
+	}
+
+	apic_set_redir(sc, pin);
+}
+
+/*
+ * Throw the switch and enable interrupts..
+ */
+
+void
+ioapic_enable(void)
+{
+	int p, maxlevel, minlevel;
+	struct ioapic_softc *sc;
+	struct intrhand *q;
+	extern void intr_calculatemasks(void); /* XXX */
+
+	intr_calculatemasks();	/* for softints, AST's */
+
+	ioapic_cold = 0;
+
+	if (ioapics == NULL)
+		return;
+
+#if 1 /* XXX Will probably get removed */
+	lapic_set_softvectors();
+	lapic_set_lvt();
+#endif
+
+	if (ioapics->sc_flags & IOAPIC_PICMODE) {
+		printf("%s: writing to IMCR to disable pics\n",
+		    ioapics->sc_dev.dv_xname);
+		outb(IMCR_ADDR, IMCR_REGISTER);
+		outb(IMCR_DATA, IMCR_APIC);
+	}
+
+#if 0 /* XXX Will be removed when we have intrsource. */
+	isa_nodefaultirq();
+#endif
+			
+	for (sc = ioapics; sc != NULL; sc = sc->sc_next) {
+		if (mp_verbose)
+			printf("%s: enabling\n", sc->sc_dev.dv_xname);
+
+		for (p=0; p<sc->sc_apic_sz; p++) {
+			maxlevel = 0;	 /* magic */
+			minlevel = 0xff; /* magic */
+				
+			for (q = sc->sc_pins[p].ip_handler; q != NULL;
+			     q = q->ih_next) {
+				if (q->ih_level > maxlevel)
+					maxlevel = q->ih_level;
+				if (q->ih_level < minlevel)
+					minlevel = q->ih_level;
+			}
+			apic_vectorset(sc, p, minlevel, maxlevel);
+		}
+	}
+}
+
+/*
+ * Interrupt handler management with the apic is radically different from the
+ * good old 8259.
+ *
+ * The APIC adds an additional level of indirection between interrupt
+ * signals and interrupt vectors in the IDT.
+ * It also encodes a priority into the high-order 4 bits of the IDT vector
+ * number.
+ *
+ *
+ * interrupt establishment:
+ *	-> locate interrupt pin.
+ *	-> locate or allocate vector for pin.
+ *	-> locate or allocate handler chain for vector.
+ *	-> chain interrupt into handler chain.
+ * 	#ifdef notyet
+ *	-> if level of handler chain increases, reallocate vector, move chain.
+ *	#endif
+ */
+
+void *
+apic_intr_establish(int irq, int type, int level, int (*ih_fun)(void *),
+    void *ih_arg, char *what)
+{
+	unsigned int ioapic = APIC_IRQ_APIC(irq);
+	unsigned int intr = APIC_IRQ_PIN(irq);
+	struct ioapic_softc *sc = ioapic_find(ioapic);
+	struct ioapic_pin *pin;
+	struct intrhand **p, *q, *ih;
+	static struct intrhand fakehand = {fakeintr};
+	extern int cold;
+	int minlevel, maxlevel;
+
+	if (sc == NULL)
+		panic("apic_intr_establish: unknown ioapic %d", ioapic);
+
+	if ((irq & APIC_INT_VIA_APIC) == 0)
+		panic("apic_intr_establish of non-apic interrupt 0x%x", irq);
+
+	pin = &sc->sc_pins[intr];
+	if (intr >= sc->sc_apic_sz || type == IST_NONE)
+		panic("apic_intr_establish: bogus intr or type");
+
+	/* no point in sleeping unless someone can free memory. */
+	ih = malloc(sizeof *ih, M_DEVBUF, cold ? M_NOWAIT : M_WAITOK);
+	if (ih == NULL)
+		panic("apic_intr_establish: can't malloc handler info");
+
+
+	switch (pin->ip_type) {
+	case IST_NONE:
+		pin->ip_type = type;
+		break;
+	case IST_EDGE:
+	case IST_LEVEL:
+		if (type == pin->ip_type)
+			break;
+	case IST_PULSE:
+		if (type != IST_NONE)
+			/* XXX should not panic here! */
+			panic("apic_intr_establish: "
+			      "intr %d can't share %s with %s",
+			      intr,
+			      isa_intr_typename(sc->sc_pins[intr].ip_type),
+			      isa_intr_typename(type));
+		break;
+	}
+
+	/*
+	 * Figure out where to put the handler.
+	 * This is O(N^2) to establish N interrupts, but we want to
+	 * preserve the order, and N is generally small.
+	 */
+	maxlevel = level;
+	minlevel = level;
+	for (p = &pin->ip_handler; (q = *p) != NULL; p = &q->ih_next) {
+		if (q->ih_level > maxlevel)
+			maxlevel = q->ih_level;
+		if (q->ih_level < minlevel)
+			minlevel = q->ih_level;
+	}
+
+	/*
+	 * Actually install a fake handler momentarily, since we might be doing
+	 * this with interrupts enabled and don't want the real routine called
+	 * until masking is set up.
+	 */
+	fakehand.ih_level = level;
+	*p = &fakehand;
+
+	/*
+	 * Fix up the vector for this pin.
+	 * (if cold, defer this until most interrupts have been established,
+	 * to avoid too much thrashing of the idt..)
+	 */
+
+	if (!ioapic_cold)
+		apic_vectorset(sc, intr, minlevel, maxlevel);
+
+#if 0
+	apic_calculatemasks();
+#endif
+
+	/*
+	 * Poke the real handler in now.
+	 */
+	ih->ih_fun = ih_fun;
+	ih->ih_arg = ih_arg;
+	ih->ih_count = 0;
+	ih->ih_next = NULL;
+	ih->ih_level = level;
+	ih->ih_irq = irq;
+	ih->ih_what = what;
+	*p = ih;
+
+	return (ih);
+}
+
+/*
+ * apic disestablish:
+ *	locate handler chain.
+ * 	dechain intrhand from handler chain
+ *	if chain empty {
+ *		reprogram apic for "safe" vector.
+ *		free vector (point at stray handler).
+ *	}
+ *	#ifdef notyet
+ *	else {
+ *		recompute level for current chain.
+ *		if changed, reallocate vector, move chain.
+ *	}
+ *	#endif
+ */
+
+void
+apic_intr_disestablish(void *arg)
+{
+	struct intrhand *ih = arg;
+	int irq = ih->ih_irq;
+	unsigned int ioapic = APIC_IRQ_APIC(irq);
+	unsigned int intr = APIC_IRQ_PIN(irq);
+	struct ioapic_softc *sc = ioapic_find(ioapic);
+	struct ioapic_pin *pin = &sc->sc_pins[intr];
+	struct intrhand **p, *q;
+	int minlevel, maxlevel;
+
+	if (sc == NULL)
+		panic("apic_intr_disestablish: unknown ioapic %d", ioapic);
+
+	if (intr >= sc->sc_apic_sz)
+		panic("apic_intr_disestablish: bogus irq");
+
+	/*
+	 * Remove the handler from the chain.
+	 * This is O(n^2), too.
+	 */
+	maxlevel = 0;
+	minlevel = 0xff;
+	for (p = &pin->ip_handler; (q = *p) != NULL && q != ih;
+	     p = &q->ih_next) {
+		if (q->ih_level > maxlevel)
+			maxlevel = q->ih_level;
+		if (q->ih_level < minlevel)
+			minlevel = q->ih_level;
+	}
+
+	if (q)
+		*p = q->ih_next;
+	else
+		panic("intr_disestablish: handler not registered");
+	for (; q != NULL; q = q->ih_next) {
+		if (q->ih_level > maxlevel)
+			maxlevel = q->ih_level;
+		if (q->ih_level < minlevel)
+			minlevel = q->ih_level;
+	}
+
+	if (!ioapic_cold)
+		apic_vectorset(sc, intr, minlevel, maxlevel);
+
+	free(ih, M_DEVBUF);
+}
+
+#ifdef DDB
+void ioapic_dump(void);
+
+void
+ioapic_dump(void)
+{
+	struct ioapic_softc *sc;
+	struct ioapic_pin *ip;
+	int p;
+
+	for (sc = ioapics; sc != NULL; sc = sc->sc_next) {
+		for (p = 0; p < sc->sc_apic_sz; p++) {
+			ip = &sc->sc_pins[p];
+			if (ip->ip_type != IST_NONE)
+				ioapic_print_redir(sc, "dump", p);
+		}
+	}
+}
+#endif
diff --git a/sys/arch/i386/i386/ipifuncs.c b/sys/arch/i386/i386/ipifuncs.c
new file mode 100644
index 00000000000..10256ca631a
--- /dev/null
+++ b/sys/arch/i386/i386/ipifuncs.c
@@ -0,0 +1,175 @@
+/*	$OpenBSD: ipifuncs.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/* $NetBSD: ipifuncs.c,v 1.1.2.3 2000/06/26 02:04:06 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+
+#include <sys/cdefs.h>			/* RCS ID & Copyright macro defns */
+
+/*
+ * Interprocessor interrupt handlers.
+ */
+
+#include "npx.h"
+
+#include <sys/param.h>
+#include <sys/device.h>
+#include <sys/systm.h>
+
+#include <machine/cpufunc.h>
+#include <machine/cpuvar.h>
+#include <machine/intr.h>
+#include <machine/atomic.h>
+#include <machine/i82093var.h>
+#include <machine/db_machdep.h>
+
+#include <uvm/uvm_extern.h>
+
+void i386_ipi_halt(struct cpu_info *);
+
+#if NNPX > 0
+void i386_ipi_synch_fpu(struct cpu_info *);
+void i386_ipi_flush_fpu(struct cpu_info *);
+#else
+#define i386_ipi_synch_fpu 0
+#define i386_ipi_flush_fpu 0
+#endif
+
+void (*ipifunc[I386_NIPI])(struct cpu_info *) = 
+{
+	i386_ipi_halt,
+#if 0 && (defined(I586_CPU) || defined(I686_CPU))
+	cc_microset,
+#else
+	0,
+#endif
+	i386_ipi_flush_fpu,
+	i386_ipi_synch_fpu,
+	pmap_do_tlb_shootdown,
+#if 0
+	i386_reload_mtrr,
+	gdt_reload_cpu,
+#else
+	0,
+	0,
+#endif
+	i386_ipi_db,
+};
+
+void
+i386_ipi_halt(struct cpu_info *ci)
+{
+	disable_intr();
+
+	printf("%s: shutting down\n", ci->ci_dev.dv_xname);
+	for(;;) {
+		asm volatile("hlt");
+	}
+}
+
+#if NNPX > 0
+void
+i386_ipi_flush_fpu(struct cpu_info *ci)
+{
+	npxsave_cpu(ci, 0);
+}
+
+void
+i386_ipi_synch_fpu(struct cpu_info *ci)
+{
+	npxsave_cpu(ci, 1);
+}
+#endif
+
+void
+i386_spurious(void)
+{
+	printf("spurious intr\n");
+}
+
+int
+i386_send_ipi(struct cpu_info *ci, int ipimask)
+{
+	int ret;
+
+	i386_atomic_setbits_l(&ci->ci_ipis, ipimask);
+
+	/* Don't send IPI to cpu which isn't (yet) running. */
+	if (!(ci->ci_flags & CPUF_RUNNING))
+		return ENOENT;
+
+	ret = i386_ipi(LAPIC_IPI_VECTOR, ci->ci_cpuid, LAPIC_DLMODE_FIXED);
+	if (ret != 0) {
+		printf("ipi of %x from %s to %s failed\n",
+		    ipimask, curcpu()->ci_dev.dv_xname, ci->ci_dev.dv_xname);
+	}
+
+	return ret;
+}
+
+void
+i386_self_ipi(int vector)
+{
+	i82489_writereg(LAPIC_ICRLO,
+	    vector | LAPIC_DLMODE_FIXED | LAPIC_LVL_ASSERT | LAPIC_DEST_SELF);
+}
+
+
+void
+i386_broadcast_ipi(int ipimask)
+{
+	panic("broadcast_ipi not implemented");
+}
+
+void
+i386_ipi_handler(void)
+{
+	struct cpu_info *ci = curcpu();
+	u_int32_t pending;
+	int bit;
+
+	pending = i386_atomic_testset_ul(&ci->ci_ipis, 0);
+
+	for (bit = 0; bit < I386_NIPI && pending; bit++) {
+		if (pending & (1<<bit)) {
+			pending &= ~(1<<bit);
+			(*ipifunc[bit])(ci);
+		}
+	}
+}
diff --git a/sys/arch/i386/i386/lapic.c b/sys/arch/i386/i386/lapic.c
new file mode 100644
index 00000000000..b4e4c36b5fa
--- /dev/null
+++ b/sys/arch/i386/i386/lapic.c
@@ -0,0 +1,474 @@
+/*	$OpenBSD: lapic.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/* $NetBSD: lapic.c,v 1.1.2.8 2000/02/23 06:10:50 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/param.h>
+#include <sys/proc.h>
+#include <sys/user.h>
+#include <sys/systm.h>
+#include <sys/device.h>
+
+#include <uvm/uvm_extern.h>
+
+#include <machine/cpu.h>
+#include <machine/cpufunc.h>
+#include <machine/cpuvar.h>
+#include <machine/pmap.h>
+#include <machine/vmparam.h>
+#include <machine/mpbiosvar.h>
+#include <machine/pcb.h>
+#include <machine/specialreg.h>
+#include <machine/segments.h>
+
+#include <machine/apicvar.h>
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+
+#include <i386/isa/timerreg.h>	/* XXX for TIMER_FREQ */
+
+void	lapic_delay(int);
+void	lapic_microtime(struct timeval *);
+static __inline u_int32_t lapic_gettick(void);
+void	lapic_clockintr(void *);
+void	lapic_initclocks(void);
+void 	lapic_map(paddr_t);
+
+void
+lapic_map(lapic_base)
+	paddr_t lapic_base;
+{
+	int s;
+	pt_entry_t *pte;
+	vaddr_t va = (vaddr_t)&local_apic;
+
+	disable_intr();
+	s = lapic_tpr;
+
+	/*
+	 * Map local apic.  If we have a local apic, it's safe to assume
+	 * we're on a 486 or better and can use invlpg and non-cacheable PTE's
+	 *
+	 * Whap the PTE "by hand" rather than calling pmap_kenter_pa because
+	 * the latter will attempt to invoke TLB shootdown code just as we
+	 * might have changed the value of cpu_number()..
+	 */
+
+	pte = kvtopte(va);
+	*pte = lapic_base | PG_RW | PG_V | PG_N;
+	invlpg(va);
+
+#ifdef MULTIPROCESSOR
+	cpu_init_first();	/* catch up to changed cpu_number() */
+#endif
+
+	lapic_tpr = s;
+	enable_intr();
+}
+
+/*
+ * enable local apic
+ */
+void
+lapic_enable()
+{
+	i82489_writereg(LAPIC_SVR, LAPIC_SVR_ENABLE | LAPIC_SPURIOUS_VECTOR);
+}
+
+extern struct mp_intr_map *lapic_ints[]; /* XXX header file? */
+
+void
+lapic_set_softvectors()
+{
+	idt_vec_set(LAPIC_SOFTCLOCK_VECTOR, Xintrsoftclock);
+	idt_vec_set(LAPIC_SOFTNET_VECTOR, Xintrsoftnet);
+	idt_vec_set(LAPIC_SOFTTTY_VECTOR, Xintrsofttty);
+}
+
+void
+lapic_set_lvt()
+{
+#ifdef MULTIPROCESSOR
+	struct cpu_info *ci = curcpu();
+
+	if (mp_verbose) {
+		apic_format_redir(ci->ci_dev.dv_xname, "prelint", 0, 0,
+		    i82489_readreg(LAPIC_LVINT0));
+		apic_format_redir(ci->ci_dev.dv_xname, "prelint", 1, 0,
+		    i82489_readreg(LAPIC_LVINT1));
+	}
+#endif
+	if (lapic_ints[0])
+		i82489_writereg(LAPIC_LVINT0, lapic_ints[0]->redir);
+	if (lapic_ints[1])
+		i82489_writereg(LAPIC_LVINT1, lapic_ints[1]->redir);
+
+#ifdef MULTIPROCESSOR
+	if (mp_verbose) {
+		apic_format_redir(ci->ci_dev.dv_xname, "timer", 0, 0,
+		    i82489_readreg(LAPIC_LVTT));
+		apic_format_redir(ci->ci_dev.dv_xname, "pcint", 0, 0,
+		    i82489_readreg(LAPIC_PCINT));
+		apic_format_redir(ci->ci_dev.dv_xname, "lint", 0, 0,
+		    i82489_readreg(LAPIC_LVINT0));
+		apic_format_redir(ci->ci_dev.dv_xname, "lint", 1, 0,
+		    i82489_readreg(LAPIC_LVINT1));
+		apic_format_redir(ci->ci_dev.dv_xname, "err", 0, 0,
+		    i82489_readreg(LAPIC_LVERR));
+	}
+#endif
+}
+
+/*
+ * Initialize fixed idt vectors for use by local apic.
+ */
+void
+lapic_boot_init(lapic_base)
+	paddr_t lapic_base;
+{
+	lapic_map(lapic_base);
+
+#ifdef MULTIPROCESSOR
+	idt_vec_set(LAPIC_IPI_VECTOR, Xintripi);
+#endif
+	idt_vec_set(LAPIC_SPURIOUS_VECTOR, Xintrspurious);
+	idt_vec_set(LAPIC_TIMER_VECTOR, Xintrltimer);
+}
+
+static __inline u_int32_t
+lapic_gettick()
+{
+	return (i82489_readreg(LAPIC_CCR_TIMER));
+}
+
+#include <sys/kernel.h>		/* for hz */
+
+int lapic_timer = 0;
+u_int32_t lapic_tval;
+
+/*
+ * this gets us up to a 4GHz busclock....
+ */
+u_int32_t lapic_per_second;
+u_int32_t lapic_frac_usec_per_cycle;
+u_int64_t lapic_frac_cycle_per_usec;
+u_int32_t lapic_delaytab[26];
+
+void
+lapic_clockintr(arg)
+	void *arg;
+{
+	struct clockframe *frame = arg;
+
+	hardclock(frame);
+}
+
+void
+lapic_initclocks()
+{
+	/*
+	 * Start local apic countdown timer running, in repeated mode.
+	 *
+	 * Mask the clock interrupt and set mode,
+	 * then set divisor,
+	 * then unmask and set the vector.
+	 */
+	i82489_writereg(LAPIC_LVTT, LAPIC_LVTT_TM|LAPIC_LVTT_M);
+	i82489_writereg(LAPIC_DCR_TIMER, LAPIC_DCRT_DIV1);
+	i82489_writereg(LAPIC_ICR_TIMER, lapic_tval);
+	i82489_writereg(LAPIC_LVTT, LAPIC_LVTT_TM|LAPIC_TIMER_VECTOR);
+}
+
+extern int gettick(void);	/* XXX put in header file */
+extern void (*initclock_func)(void); /* XXX put in header file */
+
+/*
+ * Calibrate the local apic count-down timer (which is running at
+ * bus-clock speed) vs. the i8254 counter/timer (which is running at
+ * a fixed rate).
+ *
+ * The Intel MP spec says: "An MP operating system may use the IRQ8
+ * real-time clock as a reference to determine the actual APIC timer clock
+ * speed."
+ *
+ * We're actually using the IRQ0 timer.  Hmm.
+ */
+void
+lapic_calibrate_timer(ci)
+	struct cpu_info *ci;
+{
+	unsigned int starttick, tick1, tick2, endtick;
+	unsigned int startapic, apic1, apic2, endapic;
+	u_int64_t dtick, dapic, tmp;
+	int i;
+	char tbuf[9];
+
+	if (mp_verbose)
+		printf("%s: calibrating local timer\n", ci->ci_dev.dv_xname);
+
+	/*
+	 * Configure timer to one-shot, interrupt masked,
+	 * large positive number.
+	 */
+	i82489_writereg(LAPIC_LVTT, LAPIC_LVTT_M);
+	i82489_writereg(LAPIC_DCR_TIMER, LAPIC_DCRT_DIV1);
+	i82489_writereg(LAPIC_ICR_TIMER, 0x80000000);
+
+	starttick = gettick();
+	startapic = lapic_gettick();
+
+	DELAY(2);		/* using "old" delay here.. */
+
+	for (i=0; i<hz; i++) {
+		do {
+			tick1 = gettick();
+			apic1 = lapic_gettick();
+		} while (tick1 < starttick);
+
+		do {
+			tick2 = gettick();
+			apic2 = lapic_gettick();
+		} while (tick2 > starttick);
+	}
+
+	endtick = gettick();
+	endapic = lapic_gettick();
+
+	dtick = hz * TIMER_DIV(hz) + (starttick-endtick);
+	dapic = startapic-endapic;
+
+	/*
+	 * there are TIMER_FREQ ticks per second.
+	 * in dtick ticks, there are dapic bus clocks.
+	 */
+	tmp = (TIMER_FREQ * dapic) / dtick;
+
+	lapic_per_second = tmp;
+
+#if 0
+	humanize_number(tbuf, sizeof(tbuf), tmp, "Hz", 1000);
+#else /* XXX: from NetBSD sources... sigh. */
+	{
+		/* prefixes are: (none), Kilo, Mega, Giga, Tera, Peta, Exa */
+		static const char prefixes[] = " KMGTPE";
+
+		int             i;
+		u_int64_t       max;
+		size_t          suffixlen;
+
+		if (tbuf == NULL)
+			goto out;
+		if (sizeof(tbuf) > 0)
+			tbuf[0] = '\0';
+		suffixlen = sizeof "Hz" - 1;
+		/* check if enough room for `x y' + suffix + `\0' */
+		if (sizeof(tbuf) < 4 + suffixlen)
+			goto out;
+
+		max = 1;
+		for (i = 0; i < sizeof(tbuf) - suffixlen - 3; i++)
+			max *= 10;
+		for (i = 0; tmp >= max && i < sizeof(prefixes); i++)
+			tmp /= 1000;
+
+		snprintf(tbuf, sizeof(tbuf), "%qu%s%c%s",
+		    (unsigned long long)tmp, i == 0 ? "" : " ", prefixes[i],
+		    "Hz");
+	out:
+		;
+	}
+#endif
+
+	printf("%s: apic clock running at %s\n", ci->ci_dev.dv_xname, tbuf);
+
+	if (lapic_per_second != 0) {
+		/*
+		 * reprogram the apic timer to run in periodic mode.
+		 * XXX need to program timer on other cpu's, too.
+		 */
+		lapic_tval = (lapic_per_second * 2) / hz;
+		lapic_tval = (lapic_tval / 2) + (lapic_tval & 0x1);
+
+		i82489_writereg(LAPIC_LVTT, LAPIC_LVTT_TM | LAPIC_LVTT_M |
+		    LAPIC_TIMER_VECTOR);
+		i82489_writereg(LAPIC_DCR_TIMER, LAPIC_DCRT_DIV1);
+		i82489_writereg(LAPIC_ICR_TIMER, lapic_tval);
+
+		/*
+		 * Compute fixed-point ratios between cycles and
+		 * microseconds to avoid having to do any division
+		 * in lapic_delay and lapic_microtime.
+		 */
+
+		tmp = (1000000 * (u_int64_t)1 << 32) / lapic_per_second;
+		lapic_frac_usec_per_cycle = tmp;
+
+		tmp = (lapic_per_second * (u_int64_t)1 << 32) / 1000000;
+
+		lapic_frac_cycle_per_usec = tmp;
+
+		/*
+		 * Compute delay in cycles for likely short delays in usec.
+		 */
+		for (i = 0; i < 26; i++)
+			lapic_delaytab[i] = (lapic_frac_cycle_per_usec * i) >>
+			    32;
+
+		/*
+		 * Now that the timer's calibrated, use the apic timer routines
+		 * for all our timing needs..
+		 */
+		delay_func = lapic_delay;
+		microtime_func = lapic_microtime;
+		initclock_func = lapic_initclocks;
+	}
+}
+
+/*
+ * delay for N usec.
+ */
+
+void lapic_delay(usec)
+	int usec;
+{
+	int32_t tick, otick;
+	int64_t deltat;		/* XXX may want to be 64bit */
+
+	otick = lapic_gettick();
+
+	if (usec <= 0)
+		return;
+	if (usec <= 25)
+		deltat = lapic_delaytab[usec];
+	else
+		deltat = (lapic_frac_cycle_per_usec * usec) >> 32;
+
+	while (deltat > 0) {
+		tick = lapic_gettick();
+		if (tick > otick)
+			deltat -= lapic_tval - (tick - otick);
+		else
+			deltat -= otick - tick;
+		otick = tick;
+	}
+}
+
+#define LAPIC_TICK_THRESH 200
+
+/*
+ * XXX need to make work correctly on other than cpu 0.
+ */
+
+void lapic_microtime(tv)
+	struct timeval *tv;
+{
+	struct timeval now;
+	u_int32_t tick;
+	u_int32_t usec;
+	u_int32_t tmp;
+
+	disable_intr();
+	tick = lapic_gettick();
+	now = time;
+	enable_intr();
+
+	tmp = lapic_tval - tick;
+	usec = ((u_int64_t)tmp * lapic_frac_usec_per_cycle) >> 32;
+
+	now.tv_usec += usec;
+	while (now.tv_usec >= 1000000) {
+		now.tv_sec += 1;
+		now.tv_usec -= 1000000;
+	}
+
+	*tv = now;
+}
+
+/*
+ * XXX the following belong mostly or partly elsewhere..
+ */
+
+int
+i386_ipi_init(target)
+	int target;
+{
+	unsigned j;
+
+	if ((target & LAPIC_DEST_MASK) == 0) {
+		i82489_writereg(LAPIC_ICRHI, target << LAPIC_ID_SHIFT);
+	}
+
+	i82489_writereg(LAPIC_ICRLO, (target & LAPIC_DEST_MASK) |
+	    LAPIC_DLMODE_INIT | LAPIC_LVL_ASSERT );
+
+	for (j = 100000; j > 0; j--)
+		if ((i82489_readreg(LAPIC_ICRLO) & LAPIC_DLSTAT_BUSY) == 0)
+			break;
+
+	delay(10000);
+
+	i82489_writereg(LAPIC_ICRLO, (target & LAPIC_DEST_MASK) |
+	     LAPIC_DLMODE_INIT | LAPIC_LVL_TRIG | LAPIC_LVL_DEASSERT);
+
+	for (j = 100000; j > 0; j--)
+		if ((i82489_readreg(LAPIC_ICRLO) & LAPIC_DLSTAT_BUSY) == 0)
+			break;
+
+	return (i82489_readreg(LAPIC_ICRLO) & LAPIC_DLSTAT_BUSY)?EBUSY:0;
+}
+
+int
+i386_ipi(vec,target,dl)
+	int vec,target,dl;
+{
+	unsigned j;
+
+	if ((target & LAPIC_DEST_MASK) == 0)
+		i82489_writereg(LAPIC_ICRHI, target << LAPIC_ID_SHIFT);
+
+	i82489_writereg(LAPIC_ICRLO,
+	    (target & LAPIC_DEST_MASK) | vec | dl | LAPIC_LVL_ASSERT);
+
+	for (j = 100000;
+	     j > 0 && (i82489_readreg(LAPIC_ICRLO) & LAPIC_DLSTAT_BUSY);
+	     j--)
+		;
+
+	return (i82489_readreg(LAPIC_ICRLO) & LAPIC_DLSTAT_BUSY) ? EBUSY : 0;
+}
diff --git a/sys/arch/i386/i386/lock_machdep.c b/sys/arch/i386/i386/lock_machdep.c
new file mode 100644
index 00000000000..0672d9eced4
--- /dev/null
+++ b/sys/arch/i386/i386/lock_machdep.c
@@ -0,0 +1,112 @@
+/*	$OpenBSD: lock_machdep.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/* $NetBSD: lock_machdep.c,v 1.1.2.3 2000/05/03 14:40:30 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 1998, 1999 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
+ * NASA Ames Research Center.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by the NetBSD
+ *	Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>			/* RCS ID & Copyright macro defns */
+
+/*
+ * Machine-dependent spin lock operations.
+ */
+
+#include <sys/param.h>
+#include <sys/lock.h>
+#include <sys/systm.h>
+
+#include <machine/atomic.h>
+#include <machine/lock.h>
+
+#include <ddb/db_output.h>
+
+#ifdef LOCKDEBUG
+
+void
+__cpu_simple_lock_init(lockp)
+	__cpu_simple_lock_t *lockp;
+{
+	*lockp = __SIMPLELOCK_UNLOCKED;
+}
+
+#if defined (DEBUG) && defined(DDB)
+int spin_limit = 10000000;
+#endif
+
+void
+__cpu_simple_lock(lockp)
+	__cpu_simple_lock_t *lockp;
+{
+#if defined (DEBUG) && defined(DDB)	
+	int spincount = 0;
+#endif
+	
+	while (i386_atomic_testset_i(lockp, __SIMPLELOCK_LOCKED)
+	    == __SIMPLELOCK_LOCKED) {
+#if defined(DEBUG) && defined(DDB)
+		spincount++;
+		if (spincount == spin_limit) {
+			extern int db_active;
+			db_printf("spundry\n");
+			if (db_active) {
+				db_printf("but already in debugger\n");
+			} else {
+				Debugger();
+			}
+		}
+#endif
+	}
+}
+
+int
+__cpu_simple_lock_try(lockp)
+	__cpu_simple_lock_t *lockp;
+{
+
+	if (i386_atomic_testset_i(lockp, __SIMPLELOCK_LOCKED)
+	    == __SIMPLELOCK_UNLOCKED)
+		return (1);
+	return (0);
+}
+
+void
+__cpu_simple_unlock(lockp)
+	__cpu_simple_lock_t *lockp;
+{
+	*lockp = __SIMPLELOCK_UNLOCKED;
+}
+
+#endif
diff --git a/sys/arch/i386/i386/locore.s b/sys/arch/i386/i386/locore.s
index 883e7f837c4..c738cee27c1 100644
--- a/sys/arch/i386/i386/locore.s
+++ b/sys/arch/i386/i386/locore.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.s,v 1.77 2004/02/01 19:05:23 deraadt Exp $	*/
+/*	$OpenBSD: locore.s,v 1.78 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: locore.s,v 1.145 1996/05/03 19:41:19 christos Exp $	*/
 
 /*-
@@ -39,6 +39,8 @@
 #include "npx.h"
 #include "assym.h"
 #include "apm.h"
+#include "lapic.h"
+#include "ioapic.h"
 #include "pctr.h"
 #include "ksyms.h"
 
@@ -63,6 +65,10 @@
 
 #include <dev/isa/isareg.h>
 
+#if NLAPIC > 0 
+#include <machine/i82489reg.h>
+#endif
+
 /*
  * override user-land alignment before including asm.h
  */
@@ -83,6 +89,57 @@
 #define	NOP	pushl %eax ; inb $0x84,%al ; inb $0x84,%al ; popl %eax
 #endif
 
+#define CPL _C_LABEL(lapic_tpr)
+
+#if defined(MULTIPROCESSOR)
+#include <machine/i82489reg.h>
+
+#define	GET_CPUINFO(reg)				\
+	movl	_C_LABEL(lapic_id),reg	;		\
+	shrl	$LAPIC_ID_SHIFT,reg	; 		\
+	movl	_C_LABEL(cpu_info)(,reg,4),reg
+#else
+#define	GET_CPUINFO(reg)				\
+	leal	_C_LABEL(cpu_info_primary),reg
+#endif
+
+#define	GET_CURPROC(reg, treg)				\
+	GET_CPUINFO(treg)			;	\
+	movl	CPU_INFO_CURPROC(treg),reg
+
+#define	PUSH_CURPROC(treg)				\
+	GET_CPUINFO(treg)			;	\
+	pushl	CPU_INFO_CURPROC(treg)
+       
+#define	CLEAR_CURPROC(treg)				\
+	GET_CPUINFO(treg)			;	\
+	movl	$0,CPU_INFO_CURPROC(treg)
+       
+#define	SET_CURPROC(proc,cpu)				\
+	GET_CPUINFO(cpu)			;	\
+	movl	proc,CPU_INFO_CURPROC(cpu)	;	\
+	movl	cpu,P_CPU(proc)
+       
+#define	GET_CURPCB(reg)					\
+	GET_CPUINFO(reg)			;	\
+	movl	CPU_INFO_CURPCB(reg),reg
+
+#define	SET_CURPCB(reg,treg)				\
+	GET_CPUINFO(treg)			;	\
+	movl	reg,CPU_INFO_CURPCB(treg)
+
+#define	CLEAR_RESCHED(treg)				\
+	GET_CPUINFO(treg)			;	\
+	xorl	%eax,%eax			;	\
+	movl	%eax,CPU_INFO_RESCHED(treg)
+
+#define	CHECK_ASTPENDING(treg)				\
+	GET_CPUINFO(treg)			;	\
+	cmpl	$0,CPU_INFO_ASTPENDING(treg)
+               
+#define	CLEAR_ASTPENDING(cireg)				\
+	movl	$0,CPU_INFO_ASTPENDING(cireg)
+
 /*
  * These are used on interrupt or trap entry or exit.
  */
@@ -143,11 +200,39 @@
 	.globl	_C_LABEL(cpu_feature), _C_LABEL(cpu_ecxfeature)
 	.globl	_C_LABEL(cpu_cache_eax), _C_LABEL(cpu_cache_ebx)
 	.globl	_C_LABEL(cpu_cache_ecx), _C_LABEL(cpu_cache_edx)
-	.globl	_C_LABEL(cold), _C_LABEL(esym)
+	.globl	_C_LABEL(cold), _C_LABEL(cnvmem), _C_LABEL(extmem)
+	.globl	_C_LABEL(esym)
 	.globl	_C_LABEL(boothowto), _C_LABEL(bootdev), _C_LABEL(atdevbase)
-	.globl	_C_LABEL(proc0paddr), _C_LABEL(curpcb), _C_LABEL(PTDpaddr)
-	.globl	_C_LABEL(dynamic_gdt)
+	.globl	_C_LABEL(proc0paddr), _C_LABEL(PTDpaddr)
+	.globl	_C_LABEL(gdt)
 	.globl	_C_LABEL(bootapiver), _C_LABEL(bootargc), _C_LABEL(bootargv)
+#ifndef MULTIPROCESSOR
+	.globl	_C_LABEL(curpcb)
+#endif
+	.globl	_C_LABEL(lapic_tpr)
+
+#if NLAPIC > 0
+#ifdef __ELF__
+	.align NBPG
+#else
+	.align 12
+#endif
+	.globl _C_LABEL(local_apic), _C_LABEL(lapic_id)
+_C_LABEL(local_apic):
+	.space	LAPIC_ID
+_C_LABEL(lapic_id):
+	.long	0x00000000
+	.space	LAPIC_TPRI-(LAPIC_ID+4)
+_C_LABEL(lapic_tpr):
+	.space	LAPIC_PPRI-LAPIC_TPRI
+_C_LABEL(lapic_ppr):
+	.space	LAPIC_ISR-LAPIC_PPRI
+_C_LABEL(lapic_isr):
+	.space	NBPG-LAPIC_ISR
+#else
+_C_LABEL(lapic_tpr):
+	.long	0
+#endif
 
 _C_LABEL(cpu):		.long	0	# are we 386, 386sx, 486, 586 or 686
 _C_LABEL(cpu_id):	.long	0	# saved from 'cpuid' instruction
@@ -162,6 +247,8 @@ _C_LABEL(cpu_vendor): .space 16	# vendor string returned by 'cpuid' instruction
 _C_LABEL(cpu_brandstr):	.space 48 # brand string returned by 'cpuid'
 _C_LABEL(cold):		.long	1	# cold till we are not
 _C_LABEL(esym):		.long	0	# ptr to end of syms
+_C_LABEL(cnvmem):	.long	0	# conventional memory size
+_C_LABEL(extmem):	.long	0	# extended memory size
 _C_LABEL(boothowto):	.long	0	# boot flags
 _C_LABEL(atdevbase):	.long	0	# location of start of iomem in virtual
 _C_LABEL(bootapiver):	.long	0	# /boot API version
@@ -625,6 +712,10 @@ begin:
 	call	_C_LABEL(main)
 
 NENTRY(proc_trampoline)
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(proc_trampoline_mp)
+#endif
+	movl	$IPL_NONE,CPL
 	pushl	%ebx
 	call	*%esi
 	addl	$4,%esp
@@ -759,20 +850,32 @@ ENTRY(fillw)
 	popl	%edi
 	ret
 
+
+/* Frame pointer reserve on stack. */
+#ifdef DDB
+#define FPADD 4
+#else
+#define FPADD 0
+#endif
+
 /*
  * kcopy(caddr_t from, caddr_t to, size_t len);
  * Copy len bytes, abort on fault.
  */
 ENTRY(kcopy)
+#ifdef DDB
+	pushl	%ebp
+	movl	%esp,%ebp
+#endif
 	pushl	%esi
 	pushl	%edi
-	movl	_C_LABEL(curpcb),%eax	# load curpcb into eax and set on-fault
+	GET_CURPCB(%eax)		# load curpcb into eax and set on-fault
 	pushl	PCB_ONFAULT(%eax)
 	movl	$_C_LABEL(copy_fault), PCB_ONFAULT(%eax)
 
-	movl	16(%esp),%esi
-	movl	20(%esp),%edi
-	movl	24(%esp),%ecx
+	movl	16+FPADD(%esp),%esi
+	movl	20+FPADD(%esp),%edi
+	movl	24+FPADD(%esp),%ecx
 	movl	%edi,%eax
 	subl	%esi,%eax
 	cmpl	%ecx,%eax		# overlapping?
@@ -781,16 +884,19 @@ ENTRY(kcopy)
 	shrl	$2,%ecx			# copy by 32-bit words
 	rep
 	movsl
-	movl	24(%esp),%ecx
+	movl	24+FPADD(%esp),%ecx
 	andl	$3,%ecx			# any bytes left?
 	rep
 	movsb
 
-	movl	_C_LABEL(curpcb),%edx
+	GET_CURPCB(%edx)		# XXX save curpcb?
 	popl	PCB_ONFAULT(%edx)
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
+#ifdef DDB
+	leave
+#endif
 	ret
 
 	ALIGN_TEXT
@@ -802,7 +908,7 @@ ENTRY(kcopy)
 	decl	%esi
 	rep
 	movsb
-	movl	24(%esp),%ecx		# copy remainder by 32-bit words
+	movl	24+FPADD(%esp),%ecx	# copy remainder by 32-bit words
 	shrl	$2,%ecx
 	subl	$3,%esi
 	subl	$3,%edi
@@ -810,11 +916,14 @@ ENTRY(kcopy)
 	movsl
 	cld
 
-	movl	_C_LABEL(curpcb),%edx
+	GET_CURPCB(%edx)
 	popl	PCB_ONFAULT(%edx)
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
+#ifdef DDB
+	leave
+#endif
 	ret
 	
 /*
@@ -885,13 +994,17 @@ ENTRY(memcpy)
  * Copy len bytes into the user's address space.
  */
 ENTRY(copyout)
+#ifdef DDB
+	pushl	%ebp
+	movl	%esp,%ebp
+#endif
 	pushl	%esi
 	pushl	%edi
 	pushl	$0	
 	
-	movl	16(%esp),%esi
-	movl	20(%esp),%edi
-	movl	24(%esp),%eax
+	movl	16+FPADD(%esp),%esi
+	movl	20+FPADD(%esp),%edi
+	movl	24+FPADD(%esp),%eax
 
 	/*
 	 * We check that the end of the destination buffer is not past the end
@@ -930,7 +1043,7 @@ ENTRY(copyout)
 	/* Compute PTE offset for start address. */
 	shrl	$PGSHIFT,%edi
 
-	movl	_C_LABEL(curpcb), %edx
+	GET_CURPCB(%edx)
 	movl	$2f, PCB_ONFAULT(%edx)
 
 1:	/* Check PTE for each page. */
@@ -941,8 +1054,8 @@ ENTRY(copyout)
 	decl	%ecx
 	jns	1b
 
-	movl	20(%esp),%edi
-	movl	24(%esp),%eax
+	movl	20+FPADD(%esp),%edi
+	movl	24+FPADD(%esp),%eax
 	jmp	3f
 
 2:	/* Simulate a trap. */
@@ -958,7 +1071,7 @@ ENTRY(copyout)
 	jmp	_C_LABEL(copy_fault)
 #endif /* I386_CPU */
 
-3:	movl	_C_LABEL(curpcb),%edx
+3:	GET_CURPCB(%edx)
 	movl	$_C_LABEL(copy_fault),PCB_ONFAULT(%edx)
 
 	/* bcopy(%esi, %edi, %eax); */
@@ -976,6 +1089,9 @@ ENTRY(copyout)
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
+#ifdef DDB
+	leave
+#endif
 	ret
 
 /*
@@ -983,15 +1099,19 @@ ENTRY(copyout)
  * Copy len bytes from the user's address space.
  */
 ENTRY(copyin)
+#ifdef DDB
+	pushl	%ebp
+	movl	%esp,%ebp
+#endif
 	pushl	%esi
 	pushl	%edi
-	movl	_C_LABEL(curpcb),%eax
+	GET_CURPCB(%eax)
 	pushl	$0
 	movl	$_C_LABEL(copy_fault),PCB_ONFAULT(%eax)
 	
-	movl	16(%esp),%esi
-	movl	20(%esp),%edi
-	movl	24(%esp),%eax
+	movl	16+FPADD(%esp),%esi
+	movl	20+FPADD(%esp),%edi
+	movl	24+FPADD(%esp),%eax
 
 	/*
 	 * We check that the end of the destination buffer is not past the end
@@ -1015,19 +1135,25 @@ ENTRY(copyin)
 	rep
 	movsb
 
-	movl	_C_LABEL(curpcb),%edx
+	GET_CURPCB(%edx)
 	popl	PCB_ONFAULT(%edx)
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
+#ifdef DDB
+	leave
+#endif
 	ret
 
 ENTRY(copy_fault)
-	movl	_C_LABEL(curpcb),%edx
+	GET_CURPCB(%edx)
 	popl	PCB_ONFAULT(%edx)
 	popl	%edi
 	popl	%esi
 	movl	$EFAULT,%eax
+#ifdef DDB
+	leave
+#endif
 	ret
 
 /*
@@ -1038,12 +1164,16 @@ ENTRY(copy_fault)
  * return 0 or EFAULT.
  */
 ENTRY(copyoutstr)
+#ifdef DDB
+	pushl	%ebp
+	movl	%esp,%ebp
+#endif
 	pushl	%esi
 	pushl	%edi
 
-	movl	12(%esp),%esi		# esi = from
-	movl	16(%esp),%edi		# edi = to
-	movl	20(%esp),%edx		# edx = maxlen
+	movl	12+FPADD(%esp),%esi		# esi = from
+	movl	16+FPADD(%esp),%edi		# edi = to
+	movl	20+FPADD(%esp),%edx		# edx = maxlen
 
 #if defined(I386_CPU)
 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
@@ -1057,8 +1187,8 @@ ENTRY(copyoutstr)
 	movl	$NBPG,%ecx
 	subl	%eax,%ecx		# ecx = NBPG - (src % NBPG)
 
-	movl	_C_LABEL(curpcb), %eax
-	movl	$6f, PCB_ONFAULT(%eax)
+	GET_CURPCB(%ecx)
+	movl	$6f, PCB_ONFAULT(%ecx)
 
 1:	/*
 	 * Once per page, check that we are still within the bounds of user
@@ -1112,7 +1242,7 @@ ENTRY(copyoutstr)
 #endif /* I386_CPU */
 
 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
-5:	movl	_C_LABEL(curpcb), %eax
+5:	GET_CURPCB(%eax)
 	movl	$_C_LABEL(copystr_fault),PCB_ONFAULT(%eax)
 	/*
 	 * Get min(%edx, VM_MAXUSER_ADDRESS-%edi).
@@ -1122,7 +1252,7 @@ ENTRY(copyoutstr)
 	cmpl	%edx,%eax
 	jae	1f
 	movl	%eax,%edx
-	movl	%eax,20(%esp)
+	movl	%eax,20+FPADD(%esp)
 
 1:	incl	%edx
 	cld
@@ -1154,14 +1284,18 @@ ENTRY(copyoutstr)
  * return 0 or EFAULT.
  */
 ENTRY(copyinstr)
+#ifdef DDB
+	pushl	%ebp
+	movl	%esp,%ebp
+#endif
 	pushl	%esi
 	pushl	%edi
-	movl	_C_LABEL(curpcb),%ecx
+	GET_CURPCB(%ecx)
 	movl	$_C_LABEL(copystr_fault),PCB_ONFAULT(%ecx)
 
-	movl	12(%esp),%esi		# %esi = from
-	movl	16(%esp),%edi		# %edi = to
-	movl	20(%esp),%edx		# %edx = maxlen
+	movl	12+FPADD(%esp),%esi		# %esi = from
+	movl	16+FPADD(%esp),%edi		# %edi = to
+	movl	20+FPADD(%esp),%edx		# %edx = maxlen
 
 	/*
 	 * Get min(%edx, VM_MAXUSER_ADDRESS-%esi).
@@ -1171,7 +1305,7 @@ ENTRY(copyinstr)
 	cmpl	%edx,%eax
 	jae	1f
 	movl	%eax,%edx
-	movl	%eax,20(%esp)
+	movl	%eax,20+FPADD(%esp)
 
 1:	incl	%edx
 	cld
@@ -1199,17 +1333,20 @@ ENTRY(copystr_fault)
 
 copystr_return:
 	/* Set *lencopied and return %eax. */
-	movl	_C_LABEL(curpcb),%ecx
+	GET_CURPCB(%ecx)
 	movl	$0,PCB_ONFAULT(%ecx)
-	movl	20(%esp),%ecx
+	movl	20+FPADD(%esp),%ecx
 	subl	%edx,%ecx
-	movl	24(%esp),%edx
+	movl	24+FPADD(%esp),%edx
 	testl	%edx,%edx
 	jz	8f
 	movl	%ecx,(%edx)
 
 8:	popl	%edi
 	popl	%esi
+#ifdef DDB
+	leave
+#endif
 	ret
 
 /*
@@ -1219,12 +1356,16 @@ copystr_return:
  * string is too long, return ENAMETOOLONG; else return 0.
  */
 ENTRY(copystr)
+#ifdef DDB
+	pushl	%ebp
+	movl	%esp,%ebp
+#endif
 	pushl	%esi
 	pushl	%edi
 
-	movl	12(%esp),%esi		# esi = from
-	movl	16(%esp),%edi		# edi = to
-	movl	20(%esp),%edx		# edx = maxlen
+	movl	12+FPADD(%esp),%esi		# esi = from
+	movl	16+FPADD(%esp),%edi		# edi = to
+	movl	20+FPADD(%esp),%edx		# edx = maxlen
 	incl	%edx
 	cld
 
@@ -1244,15 +1385,18 @@ ENTRY(copystr)
 	movl	$ENAMETOOLONG,%eax
 
 6:	/* Set *lencopied and return %eax. */
-	movl	20(%esp),%ecx
+	movl	20+FPADD(%esp),%ecx
 	subl	%edx,%ecx
-	movl	24(%esp),%edx
+	movl	24+FPADD(%esp),%edx
 	testl	%edx,%edx
 	jz	7f
 	movl	%ecx,(%edx)
 
 7:	popl	%edi
 	popl	%esi
+#ifdef DDB
+	leave
+#endif
 	ret
 
 /*****************************************************************************/
@@ -1277,6 +1421,8 @@ NENTRY(lgdt)
 	movw	%ax,%ds
 	movw	%ax,%es
 	movw	%ax,%ss
+	movl	$GSEL(GCPU_SEL, SEL_KPL),%eax
+	movw	%ax,%fs
 	/* Reload code selector by doing intersegment return. */
 	popl	%eax
 	pushl	$GSEL(GCODE_SEL, SEL_KPL)
@@ -1389,11 +1535,85 @@ NENTRY(remrunqueue)
  * something to come ready.
  */
 ENTRY(idle)
+	/* Skip context saving if we have none. */
+	testl %esi,%esi
+	jz	1f
+
+	/*
+	 * idling:	save old context.
+	 *
+	 * Registers:
+	 *   %eax, %ebx, %ecx - scratch
+	 *   %esi - old proc, then old pcb
+	 *   %edi - idle pcb
+	 *   %edx - idle TSS selector
+	 */
+
+	pushl	%esi
+	call	_C_LABEL(pmap_deactivate)	# pmap_deactivate(oldproc)
+	addl	$4,%esp
+
+	movl	P_ADDR(%esi),%esi
+
+	/* Save stack pointers. */
+	movl	%esp,PCB_ESP(%esi)
+	movl	%ebp,PCB_EBP(%esi)
+
+	/* Find idle PCB for this CPU */
+#ifndef MULTIPROCESSOR
+	movl	$_C_LABEL(proc0),%ebx
+	movl	P_ADDR(%ebx),%edi
+	movl	P_MD_TSS_SEL(%ebx),%edx
+#else
+	GET_CPUINFO(%ebx)
+	movl	CPU_INFO_IDLE_PCB(%ebx),%edi
+	movl	CPU_INFO_IDLE_TSS_SEL(%ebx),%edx
+#endif
+
+	/* Restore the idle context (avoid interrupts) */
 	cli
-	movl	_C_LABEL(whichqs),%ecx
-	testl	%ecx,%ecx
-	jnz	sw1
+
+	/* Restore stack pointers. */
+	movl	PCB_ESP(%edi),%esp
+	movl	PCB_EBP(%edi),%ebp
+
+
+	/* Switch address space. */
+	movl	PCB_CR3(%edi),%ecx
+	movl	%ecx,%cr3
+
+	/* Switch TSS. Reset "task busy" flag before loading. */
+#ifdef MULTIPROCESSOR
+	movl	CPU_INFO_GDT(%ebx),%eax
+#else
+	movl	_C_LABEL(gdt),%eax
+#endif
+	andl	$~0x0200,4-SEL_KPL(%eax,%edx,1)
+	ltr	%dx
+
+	/* We're always in the kernel, so we don't need the LDT. */
+
+	/* Restore cr0 (including FPU state). */
+	movl	PCB_CR0(%edi),%ecx
+	movl	%ecx,%cr0
+
+	/* Record new pcb. */
+	SET_CURPCB(%edi,%ecx)
+
+	xorl	%esi,%esi
 	sti
+
+1:
+#if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)	
+	call	_C_LABEL(sched_unlock_idle)
+#endif
+
+	movl	$IPL_NONE,CPL		# spl0()
+	call	_C_LABEL(Xspllower)	# process pending interrupts
+
+ENTRY(idle_loop)
+	cmpl	$0,_C_LABEL(whichqs)
+	jnz	_C_LABEL(idle_exit)
 #if NAPM > 0
 	call	_C_LABEL(apm_cpu_idle)
 	cmpl	$0,_C_LABEL(apm_dobusy)
@@ -1407,8 +1627,24 @@ ENTRY(idle)
 #else
 	hlt
 #endif
-	jmp	_C_LABEL(idle)
+	jmp	_C_LABEL(idle_loop)
 
+ENTRY(idle_exit)
+	movl	$IPL_HIGH,CPL		# splhigh
+#if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)	
+	call	_C_LABEL(sched_lock_idle)
+#endif
+#if 0
+	GET_CPUINFO(%ebx)
+	leal	CPU_INFO_NAME(%ebx),%ebx
+	pushl	%ebx
+	pushl	$1f
+	call	_C_LABEL(printf)
+	addl	$8,%esp
+#endif
+	jmp	switch_search
+1:	.asciz	"%s: unidle\n"
+		
 #ifdef DIAGNOSTIC
 NENTRY(switch_error)
 	pushl	$1f
@@ -1427,9 +1663,9 @@ ENTRY(cpu_switch)
 	pushl	%ebx
 	pushl	%esi
 	pushl	%edi
-	pushl	_C_LABEL(cpl)
+	pushl	CPL
 
-	movl	_C_LABEL(curproc),%esi
+	GET_CURPROC(%esi,%ecx)
 
 	/*
 	 * Clear curproc so that we don't accumulate system time while idle.
@@ -1438,10 +1674,7 @@ ENTRY(cpu_switch)
 	 * below and changes the priority.  (See corresponding comment in
 	 * userret()).
 	 */
-	movl	$0,_C_LABEL(curproc)
-
-	movl	$IPL_NONE,_C_LABEL(cpl)	# spl0()
-	call	_C_LABEL(Xspllower)	# process pending interrupts
+	CLEAR_CURPROC(%ecx)
 
 switch_search:
 	/*
@@ -1457,14 +1690,10 @@ switch_search:
 	 */
 
 	/* Wait for new process. */
-	cli				# splhigh doesn't do a cli
 	movl	_C_LABEL(whichqs),%ecx
-
-sw1:	bsfl	%ecx,%ebx		# find a full q
+	bsfl	%ecx,%ebx		# find a full q
 	jz	_C_LABEL(idle)		# if none, idle
-
 	leal	_C_LABEL(qs)(,%ebx,8),%eax	# select q
-
 	movl	P_FORW(%eax),%edi	# unlink from front of process q
 #ifdef	DIAGNOSTIC
 	cmpl	%edi,%eax		# linked to self (i.e. nothing queued)?
@@ -1481,8 +1710,7 @@ sw1:	bsfl	%ecx,%ebx		# find a full q
 	movl	%ecx,_C_LABEL(whichqs)	# update q status
 
 3:	/* We just did it. */
-	xorl	%eax,%eax
-	movl	%eax,_C_LABEL(want_resched)
+	CLEAR_RESCHED(%ecx)
 
 #ifdef	DIAGNOSTIC
 	cmpl	%eax,P_WCHAN(%edi)	# Waiting for something?
@@ -1495,10 +1723,8 @@ sw1:	bsfl	%ecx,%ebx		# find a full q
 	movl	%eax,P_BACK(%edi)
 
 	/* Record new process. */
-	movl	%edi,_C_LABEL(curproc)
-
-	/* It's okay to take interrupts here. */
-	sti
+	movb	$SONPROC,P_STAT(%edi)	# p->p_stat = SONPROC
+	SET_CURPROC(%edi,%ecx)
 
 	/* Skip context switch if same process. */
 	cmpl	%edi,%esi
@@ -1517,6 +1743,10 @@ sw1:	bsfl	%ecx,%ebx		# find a full q
 	 *   %edi - new process
 	 */
 
+	pushl	%esi
+	call	_C_LABEL(pmap_deactivate)
+	addl	$4,%esp
+
 	movl	P_ADDR(%esi),%esi
 
 	/* Save segment registers. */
@@ -1557,13 +1787,24 @@ switch_exited:
 	jnz	switch_restored
 #endif
 
+	/*
+	 * Activate the address space.  We're curproc, so %cr3 will
+	 * be reloaded, but we're not yet curpcb, so the LDT won't
+	 * be reloaded, although the PCB copy of the selector will
+	 * be refreshed from the pmap.
+	 */
+	pushl	%edi
+	call	_C_LABEL(pmap_activate)
+	addl	$4,%esp
+	
 	/* Load TSS info. */
-	movl	_C_LABEL(dynamic_gdt),%eax
-	movl	PCB_TSS_SEL(%esi),%edx
-
-	/* Switch address space. */
-	movl	PCB_CR3(%esi),%ecx
-	movl	%ecx,%cr3
+#ifdef MULTIPROCESSOR
+	GET_CPUINFO(%ebx)
+	movl	CPU_INFO_GDT(%ebx),%eax
+#else
+	movl	_C_LABEL(gdt),%eax
+#endif
+	movl	P_MD_TSS_SEL(%edi),%edx
 
 	/* Switch TSS. */
 	andl	$~0x0200,4-SEL_KPL(%eax,%edx,1)
@@ -1590,51 +1831,87 @@ switch_exited:
 switch_restored:
 	/* Restore cr0 (including FPU state). */
 	movl	PCB_CR0(%esi),%ecx
+#ifdef MULTIPROCESSOR
+	/* 
+	 * If our floating point registers are on a different CPU,
+	 * clear CR0_TS so we'll trap rather than reuse bogus state.
+	 */
+	GET_CPUINFO(%ebx)
+	cmpl	PCB_FPCPU(%esi),%ebx
+	jz	1f
+	orl	$CR0_TS,%ecx
+1:	
+#endif	
 	movl	%ecx,%cr0
 
 	/* Record new pcb. */
-	movl	%esi,_C_LABEL(curpcb)
+	SET_CURPCB(%esi, %ecx)
 
 	/* Interrupts are okay again. */
 	sti
 
 switch_return:
+#if 0
+	pushl	%edi
+	GET_CPUINFO(%ebx)
+	leal	CPU_INFO_NAME(%ebx),%ebx
+	pushl	%ebx
+	pushl	$1f
+	call	_C_LABEL(printf)
+	addl	$0xc,%esp
+#endif
+#if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)     
+	call    _C_LABEL(sched_unlock_idle)
+#endif
 	/*
 	 * Restore old cpl from stack.  Note that this is always an increase,
 	 * due to the spl0() on entry.
 	 */
-	popl	_C_LABEL(cpl)
+	popl	CPL
 
 	movl	%edi,%eax		# return (p);
 	popl	%edi
 	popl	%esi
 	popl	%ebx
 	ret
-
+1:	.asciz	"%s: scheduled %x\n"
 /*
  * switch_exit(struct proc *p);
- * Switch to proc0's saved context and deallocate the address space and kernel
- * stack for p.  Then jump into cpu_switch(), as if we were in proc0 all along.
+ * Switch to the appropriate idle context (proc0's if uniprocessor; the cpu's if
+ * multiprocessor) and deallocate the address space and kernel stack for p.
+ * Then jump into cpu_switch(), as if we were in the idle proc all along.
  */
+#ifndef MULTIPROCESSOR
 	.globl	_C_LABEL(proc0)
+#endif
 ENTRY(switch_exit)
 	movl	4(%esp),%edi		# old process
+#ifndef MULTIPROCESSOR
 	movl	$_C_LABEL(proc0),%ebx
+	movl	P_ADDR(%ebx),%esi
+	movl	P_MD_TSS_SEL(%ebx),%edx
+#else
+	GET_CPUINFO(%ebx)
+	movl	CPU_INFO_IDLE_PCB(%ebx),%esi
+	movl	CPU_INFO_IDLE_TSS_SEL(%ebx),%edx
+#endif
 
 	/* In case we fault... */
-	movl	$0,_C_LABEL(curproc)
+	CLEAR_CURPROC(%ecx)
 
-	/* Restore proc0's context. */
+	/* Restore the idle context. */
 	cli
-	movl	P_ADDR(%ebx),%esi
 
 	/* Restore stack pointers. */
 	movl	PCB_ESP(%esi),%esp
 	movl	PCB_EBP(%esi),%ebp
 
 	/* Load TSS info. */
-	movl	_C_LABEL(dynamic_gdt),%eax
-	movl	PCB_TSS_SEL(%esi),%edx
+#ifdef MULTIPROCESSOR
+	movl	CPU_INFO_GDT(%ebx),%eax
+#else
+	movl	_C_LABEL(gdt),%eax
+#endif
 
 	/* Switch address space. */
 	movl	PCB_CR3(%esi),%ecx
@@ -1656,7 +1933,7 @@ ENTRY(switch_exit)
 	movl	%ecx,%cr0
 
 	/* Record new pcb. */
-	movl	%esi,_C_LABEL(curpcb)
+	SET_CURPCB(%esi, %ecx)
 
 	/* Interrupts are okay again. */
 	sti
@@ -1669,8 +1946,8 @@ ENTRY(switch_exit)
 	addl	$4,%esp
 
 	/* Jump into cpu_switch() with the right state. */
-	movl	%ebx,%esi
-	movl	$0,_C_LABEL(curproc)
+	xorl	%esi,%esi
+	CLEAR_CURPROC(%ecx)
 	jmp	switch_search
 
 /*
@@ -1746,7 +2023,12 @@ IDTVEC(dna)
 	pushl	$0			# dummy error code
 	pushl	$T_DNA
 	INTRENTRY
-	pushl	_C_LABEL(curproc)
+#ifdef MULTIPROCESSOR
+	GET_CPUINFO(%eax)
+	pushl	%eax
+#else
+	pushl	$_C_LABEL(cpu_info_primary)
+#endif
 	call	*_C_LABEL(npxdna_func)
 	addl	$4,%esp
 	testl	%eax,%eax
@@ -1784,6 +2066,16 @@ IDTVEC(page)
 	TRAP(T_PAGEFLT)
 IDTVEC(rsvd)
 	ZTRAP(T_RESERVED)
+IDTVEC(intrspurious)
+	/*
+	 * The Pentium Pro local APIC may erroneously call this vector for a
+	 * default IR7.  Just ignore it.
+	 *
+	 * (The local APIC does this when CPL is raised while it's on the 
+	 * way to delivering an interrupt.. presumably enough has been set 
+	 * up that it's inconvenient to abort delivery completely..)
+	 */
+	iret
 IDTVEC(fpu)
 #if NNPX > 0
 	/*
@@ -1794,7 +2086,7 @@ IDTVEC(fpu)
 	pushl	$0			# dummy error code
 	pushl	$T_ASTFLT
 	INTRENTRY
-	pushl	_C_LABEL(cpl)		# if_ppl in intrframe
+	pushl	CPL			# if_ppl in intrframe
 	pushl	%esp			# push address of intrframe
 	incl	_C_LABEL(uvmexp)+V_TRAP
 	call	_C_LABEL(npxintr)
@@ -1826,12 +2118,12 @@ NENTRY(alltraps)
 	INTRENTRY
 calltrap:
 #ifdef DIAGNOSTIC
-	movl	_C_LABEL(cpl),%ebx
+	movl	CPL,%ebx
 #endif /* DIAGNOSTIC */
 	call	_C_LABEL(trap)
 2:	/* Check for ASTs on exit to user mode. */
 	cli
-	cmpb	$0,_C_LABEL(astpending)
+	CHECK_ASTPENDING(%ecx)
 	je	1f
 	testb	$SEL_RPL,TF_CS(%esp)
 #ifdef VM86
@@ -1839,15 +2131,15 @@ calltrap:
 	testl	$PSL_VM,TF_EFLAGS(%esp)
 #endif
 	jz	1f
-5:	movb	$0,_C_LABEL(astpending)
+5:	CLEAR_ASTPENDING(%ecx)
 	sti
 	movl	$T_ASTFLT,TF_TRAPNO(%esp)
 	call	_C_LABEL(trap)
 	jmp	2b
 #ifndef DIAGNOSTIC
 1:	INTRFASTEXIT
-#else /* DIAGNOSTIC */
-1:	cmpl	_C_LABEL(cpl),%ebx
+#else 
+1:	cmpl	CPL,%ebx
 	jne	3f
 	INTRFASTEXIT
 3:	sti
@@ -1857,7 +2149,7 @@ calltrap:
 #if defined(DDB) && 0
 	int	$3
 #endif /* DDB */
-	movl	%ebx,_C_LABEL(cpl)
+	movl	%ebx,CPL
 	jmp	2b
 4:	.asciz	"WARNING: SPL NOT LOWERED ON TRAP EXIT\n"
 #endif /* DIAGNOSTIC */
@@ -1888,17 +2180,17 @@ syscall1:
 	call	_C_LABEL(syscall)
 2:	/* Check for ASTs on exit to user mode. */
 	cli
-	cmpb	$0,_C_LABEL(astpending)
+	CHECK_ASTPENDING(%ecx)
 	je	1f
 	/* Always returning to user mode here. */
-	movb	$0,_C_LABEL(astpending)
+	CLEAR_ASTPENDING(%ecx)
 	sti
 	/* Pushed T_ASTFLT into tf_trapno on entry. */
 	call	_C_LABEL(trap)
 	jmp	2b
 1:	INTRFASTEXIT
 
-#include <i386/isa/vector.s>
+#include <i386/i386/vector.s>
 #include <i386/isa/icu.s>
 
 /*
@@ -2039,3 +2331,7 @@ ENTRY(i686_pagezero)
 	popl	%edi
 	ret
 #endif
+
+#if NLAPIC > 0 
+#include <i386/i386/apicvec.s>
+#endif
diff --git a/sys/arch/i386/i386/machdep.c b/sys/arch/i386/i386/machdep.c
index 07ac282f32d..2bd03934ac6 100644
--- a/sys/arch/i386/i386/machdep.c
+++ b/sys/arch/i386/i386/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.296 2004/06/09 16:01:48 tedu Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.297 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: machdep.c,v 1.214 1996/11/10 03:16:17 thorpej Exp $	*/
 
 /*-
@@ -115,6 +115,7 @@
 
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
+#include <machine/cpuvar.h>
 #include <machine/gdt.h>
 #include <machine/pio.h>
 #include <machine/bus.h>
@@ -248,6 +249,9 @@ int p4_model;
 int p3_step;
 int setperf_prio = 0;		/* for concurrent handlers */
 
+void (*delay_func)(int) = i8254_delay;
+void (*microtime_func)(struct timeval *) = i8254_microtime;
+void (*initclock_func)(void) = i8254_initclocks;
 void (*update_cpuspeed)(void) = NULL;
 
 /*
@@ -272,7 +276,7 @@ caddr_t	allocsys(caddr_t);
 void	setup_buffers(vaddr_t *);
 void	dumpsys(void);
 int	cpu_dump(void);
-void	identifycpu(void);
+void	old_identifycpu(void);
 void	init386(paddr_t);
 void	consinit(void);
 void	(*cpuresetfn)(void);
@@ -397,7 +401,8 @@ cpu_startup()
 	printf("%s", version);
 	startrtclock();
 
-	identifycpu();
+	/* XXX Merge with identifycpu */
+	old_identifycpu();
 	printf("real mem  = %u (%uK)\n", ctob(physmem), ctob(physmem)/1024);
 
 	/*
@@ -458,10 +463,11 @@ cpu_startup()
 void
 i386_proc0_tss_ldt_init()
 {
-	struct pcb *pcb;
 	int x;
+	struct pcb *pcb;
 
 	curpcb = pcb = &proc0.p_addr->u_pcb;
+
 	pcb->pcb_tss.tss_ioopt =
 	    ((caddr_t)pcb->pcb_iomap - (caddr_t)&pcb->pcb_tss) << 16;
 	for (x = 0; x < sizeof(pcb->pcb_iomap) / 4; x++)
@@ -472,14 +478,33 @@ i386_proc0_tss_ldt_init()
 	pcb->pcb_cr0 = rcr0();
 	pcb->pcb_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
 	pcb->pcb_tss.tss_esp0 = (int)proc0.p_addr + USPACE - 16;
-	tss_alloc(pcb);
+	proc0.p_md.md_regs = (struct trapframe *)pcb->pcb_tss.tss_esp0 - 1;
+	proc0.p_md.md_tss_sel = tss_alloc(pcb);
 
-	ltr(pcb->pcb_tss_sel);
+	ltr(proc0.p_md.md_tss_sel);
 	lldt(pcb->pcb_ldt_sel);
-
-	proc0.p_md.md_regs = (struct trapframe *)pcb->pcb_tss.tss_esp0 - 1;
 }
 
+#ifdef MULTIPROCESSOR
+void
+i386_init_pcb_tss_ldt(struct cpu_info *ci)
+{
+	int x;
+	struct pcb *pcb = ci->ci_idle_pcb;
+
+	pcb->pcb_tss.tss_ioopt =
+	    ((caddr_t)pcb->pcb_iomap - (caddr_t)&pcb->pcb_tss) << 16;
+	for (x = 0; x < sizeof(pcb->pcb_iomap) / 4; x++)
+		pcb->pcb_iomap[x] = 0xffffffff;
+	pcb->pcb_iomap_pad = 0xff;
+
+	pcb->pcb_ldt_sel = pmap_kernel()->pm_ldt_sel = GSEL(GLDT_SEL, SEL_KPL);
+	pcb->pcb_cr0 = rcr0();
+	ci->ci_idle_tss_sel = tss_alloc(pcb);
+}  
+#endif	/* MULTIPROCESSOR */
+
+
 /*
  * Allocate space for system data structures.  We are given
  * a starting virtual address and we return a final virtual
@@ -1102,13 +1127,12 @@ winchip_cpu_setup(cpu_device, model, step)
 {
 #if defined(I586_CPU)
 
-	switch (model) {
+  	switch ((curcpu()->ci_signature >> 4) & 15) { /* model */
 	case 4: /* WinChip C6 */
-		cpu_feature &= ~CPUID_TSC;
+		curcpu()->ci_feature_flags &= ~CPUID_TSC;
 		/* Disable RDTSC instruction from user-level. */
 		lcr4(rcr4() | CR4_TSD);
-
-		printf("%s: TSC disabled\n", cpu_device);
+		printf("%s: TSC disabled\n", curcpu()->ci_dev.dv_xname);
 		break;
 	}
 #endif
@@ -1212,7 +1236,7 @@ cyrix6x86_cpu_setup(cpu_device, model, step)
 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
 	extern int clock_broken_latch;
 
-	switch (model) {
+  	switch ((curcpu()->ci_signature >> 4) & 15) { /* model */
 	case -1: /* M1 w/o cpuid */
 	case 2:	/* M1 */
 		/* set up various cyrix registers */
@@ -1228,11 +1252,13 @@ cyrix6x86_cpu_setup(cpu_device, model, step)
 		/* disable access to ccr4/ccr5 */
 		cyrix_write_reg(0xC3, cyrix_read_reg(0xC3) & ~0x10);
 
-		printf("%s: xchg bug workaround performed\n", cpu_device);
+		printf("%s: xchg bug workaround performed\n", 
+		       curcpu()->ci_dev.dv_xname);
 		break;	/* fallthrough? */
-	case 4:
+	case 4:	/* GXm */
+		/* Unset the TSC bit until calibrate_delay() gets fixed. */
 		clock_broken_latch = 1;
-		cpu_feature &= ~CPUID_TSC;
+		curcpu()->ci_feature_flags &= ~CPUID_TSC;
 		printf("%s: TSC disabled\n", cpu_device);
 		break;
 	}
@@ -1278,8 +1304,11 @@ intel586_cpu_setup(cpu_device, model, step)
 	int model, step;
 {
 #if defined(I586_CPU)
-	fix_f00f();
-	printf("%s: F00F bug workaround installed\n", cpu_device);
+	if (!cpu_f00f_bug) {
+		fix_f00f();
+		printf("%s: F00F bug workaround installed\n", 
+		    curcpu()->ci_dev.dv_xname);
+	}
 #endif
 }
 
@@ -1365,6 +1394,9 @@ intel686_common_cpu_setup(const char *cpu_device, int model, int step)
 void
 intel686_cpu_setup(const char *cpu_device, int model, int step)
 {
+        struct cpu_info *ci = curcpu();
+  	/* XXX SMP int model = (ci->ci_signature >> 4) & 15; */
+	/* XXX SMP int step = ci->ci_signature & 15; */
 	u_quad_t msr119;
 
 	intel686_common_cpu_setup(cpu_device, model, step);
@@ -1374,19 +1406,20 @@ intel686_cpu_setup(const char *cpu_device, int model, int step)
 	 * From Intel Application Note #485.
 	 */
 	if ((model == 1) && (step < 3))
-		cpu_feature &= ~CPUID_SEP;
+		ci->ci_feature_flags &= ~CPUID_SEP;
 
 	/*
 	 * Disable the Pentium3 serial number.
 	 */
-	if ((model == 7) && (cpu_feature & CPUID_SER)) {
+	if ((model == 7) && (ci->ci_feature_flags & CPUID_SER)) {
 		msr119 = rdmsr(MSR_BBL_CR_CTL);
 		msr119 |= 0x0000000000200000LL;
 		wrmsr(MSR_BBL_CR_CTL, msr119);
 
-		printf("%s: disabling processor serial number\n", cpu_device);
-		cpu_feature &= ~CPUID_SER;
-		cpuid_level = 2;
+		printf("%s: disabling processor serial number\n", 
+		       ci->ci_dev.dv_xname);
+		ci->ci_feature_flags &= ~CPUID_SER;
+		ci->ci_level = 2;
 	}
 
 #if !defined(SMALL_KERNEL) && defined(I686_CPU)
@@ -1474,6 +1507,249 @@ cyrix3_cpu_name(model, step)
 	return name;
 }
 
+/* XXXSMP: must be shared with UP */
+#ifdef MULTIPROCESSOR
+/*
+ * Print identification for the given CPU.
+ * XXX XXX
+ * This is not as clean as one might like, because it references
+ *
+ * the "cpuid_level" and "cpu_vendor" globals.
+ * cpuid_level isn't so bad, since both CPU's will hopefully
+ * be of the same level.
+ *
+ * The Intel multiprocessor spec doesn't give us the cpu_vendor
+ * information; however, the chance of multi-vendor SMP actually
+ * ever *working* is sufficiently low that it's probably safe to assume
+ * all processors are of the same vendor.
+ */
+ 
+void
+identifycpu(struct cpu_info *ci)
+{
+	extern char cpu_vendor[];
+#ifdef CPUDEBUG
+	extern int cpu_cache_eax, cpu_cache_ebx, cpu_cache_ecx, cpu_cache_edx;
+#else
+	extern int cpu_cache_edx;
+#endif
+	const char *name, *modifier, *vendorname, *token;
+	int class = CPUCLASS_386, vendor, i, max;
+	int family, model, step, modif, cachesize;
+	const struct cpu_cpuid_nameclass *cpup = NULL;
+
+	char *cpu_device = ci->ci_dev.dv_xname;
+	/* XXX SMP XXX void (*cpu_setup)(const char *, int, int); */
+
+	if (cpuid_level == -1) {
+#ifdef DIAGNOSTIC
+		if (cpu < 0 || cpu >=
+		    (sizeof i386_nocpuid_cpus/sizeof(struct cpu_nocpuid_nameclass)))
+			panic("unknown cpu type %d", cpu);
+#endif
+		name = i386_nocpuid_cpus[cpu].cpu_name;
+		vendor = i386_nocpuid_cpus[cpu].cpu_vendor;
+		vendorname = i386_nocpuid_cpus[cpu].cpu_vendorname;
+		model = -1;
+		step = -1;
+		class = i386_nocpuid_cpus[cpu].cpu_class;
+		ci->cpu_setup = i386_nocpuid_cpus[cpu].cpu_setup;
+		modifier = "";
+		token = "";
+	} else {
+		max = sizeof (i386_cpuid_cpus) / sizeof (i386_cpuid_cpus[0]);
+		modif = (ci->ci_signature >> 12) & 3;
+		family = (ci->ci_signature >> 8) & 15;
+		if (family < CPU_MINFAMILY)
+			panic("identifycpu: strange family value");
+		model = (ci->ci_signature >> 4) & 15;
+		step = ci->ci_signature & 15;
+#ifdef CPUDEBUG
+		printf("%s: family %x model %x step %x\n", cpu_device, family,
+			model, step);
+		printf("%s: cpuid level %d cache eax %x ebx %x ecx %x edx %x\n",
+			cpu_device, cpuid_level, cpu_cache_eax, cpu_cache_ebx,
+			cpu_cache_ecx, cpu_cache_edx);
+#endif
+
+		for (i = 0; i < max; i++) {
+			if (!strncmp(cpu_vendor,
+			    i386_cpuid_cpus[i].cpu_id, 12)) {
+				cpup = &i386_cpuid_cpus[i];
+				break;
+			}
+		}
+
+		if (cpup == NULL) {
+			vendor = CPUVENDOR_UNKNOWN;
+			if (cpu_vendor[0] != '\0')
+				vendorname = &cpu_vendor[0];
+			else
+				vendorname = "Unknown";
+			if (family > CPU_MAXFAMILY)
+				family = CPU_MAXFAMILY;
+			class = family - 3;
+			modifier = "";
+			name = "";
+			token = "";
+			ci->cpu_setup = NULL;
+		} else {
+			token = cpup->cpu_id;
+			vendor = cpup->cpu_vendor;
+			vendorname = cpup->cpu_vendorname;
+			modifier = modifiers[modif];
+			if (family > CPU_MAXFAMILY) {
+				family = CPU_MAXFAMILY;
+				model = CPU_DEFMODEL;
+			} else if (model > CPU_MAXMODEL)
+				model = CPU_DEFMODEL;
+			i = family - CPU_MINFAMILY;
+
+			/* Special hack for the PentiumII/III series. */
+			if (vendor == CPUVENDOR_INTEL && family == 6 &&
+			    (model == 5 || model == 7)) {
+				name = intel686_cpu_name(model);
+			} else
+				name = cpup->cpu_family[i].cpu_models[model];
+			if (name == NULL)
+				name = cpup->cpu_family[i].cpu_models[CPU_DEFMODEL];
+			class = cpup->cpu_family[i].cpu_class;
+			ci->cpu_setup = cpup->cpu_family[i].cpu_setup;
+		}
+	}
+
+	/* Find the amount of on-chip L2 cache.  Add support for AMD K6-3...*/
+	cachesize = -1;
+	if (vendor == CPUVENDOR_INTEL && cpuid_level >= 2 && family < 0xf) {
+		int intel_cachetable[] = { 0, 128, 256, 512, 1024, 2048 };
+		if ((cpu_cache_edx & 0xFF) >= 0x40 &&
+		    (cpu_cache_edx & 0xFF) <= 0x45)
+			cachesize = intel_cachetable[(cpu_cache_edx & 0xFF) - 0x40];
+	}
+
+	if ((ci->ci_flags & CPUF_BSP) == 0) {
+		if (cachesize > -1) {
+			snprintf(cpu_model, sizeof(cpu_model),
+				"%s %s%s (%s%s%s%s-class, %dKB L2 cache)",
+				vendorname, modifier, name,
+				((*token) ? "\"" : ""), ((*token) ? token : ""),
+				((*token) ? "\" " : ""), classnames[class], cachesize);
+		} else {
+			snprintf(cpu_model, sizeof(cpu_model),
+				"%s %s%s (%s%s%s%s-class)",
+				vendorname, modifier, name,
+				((*token) ? "\"" : ""), ((*token) ? token : ""),
+				((*token) ? "\" " : ""), classnames[class]);
+		}
+
+		printf("%s: %s", cpu_device, cpu_model);
+	}
+
+#if defined(I586_CPU) || defined(I686_CPU)
+	if (ci->ci_feature_flags && (ci->ci_feature_flags & CPUID_TSC)) {
+		/* Has TSC */
+		calibrate_cyclecounter();
+		if (pentium_mhz > 994) {
+			int ghz, fr;
+
+			ghz = (pentium_mhz + 9) / 1000;
+			fr = ((pentium_mhz + 9) / 10 ) % 100;
+			if ((ci->ci_flags & CPUF_BSP) == 0) {
+				if (fr)
+					printf(" %d.%02d GHz", ghz, fr);
+				else
+					printf(" %d GHz", ghz);
+			}
+		} else {
+			if ((ci->ci_flags & CPUF_BSP) == 0) {
+				printf(" %d MHz", pentium_mhz);
+			}
+		}
+	}
+#endif
+	if ((ci->ci_flags & CPUF_BSP) == 0) {
+		printf("\n");
+
+		if (ci->ci_feature_flags) {
+			int numbits = 0;
+	
+			printf("%s: ", cpu_device);
+			max = sizeof(i386_cpuid_features)
+				/ sizeof(i386_cpuid_features[0]);
+			for (i = 0; i < max; i++) {
+				if (ci->ci_feature_flags & 
+				    i386_cpuid_features[i].feature_bit) {
+					printf("%s%s", (numbits == 0 ? "" : ","),
+					    i386_cpuid_features[i].feature_name);
+					numbits++;
+				}
+			}
+			printf("\n");
+		}
+	}
+
+	cpu_class = class;
+	ci->cpu_class = class;
+
+	/*
+	 * Now that we have told the user what they have,
+	 * let them know if that machine type isn't configured.
+	 */
+	switch (cpu_class) {
+#if !defined(I386_CPU) && !defined(I486_CPU) && !defined(I586_CPU) && !defined(I686_CPU)
+#error No CPU classes configured.
+#endif
+#ifndef I686_CPU
+	case CPUCLASS_686:
+		printf("NOTICE: this kernel does not support Pentium Pro CPU class\n");
+#ifdef I586_CPU
+		printf("NOTICE: lowering CPU class to i586\n");
+		cpu_class = CPUCLASS_586;
+		break;
+#endif
+#endif
+#ifndef I586_CPU
+	case CPUCLASS_586:
+		printf("NOTICE: this kernel does not support Pentium CPU class\n");
+#ifdef I486_CPU
+		printf("NOTICE: lowering CPU class to i486\n");
+		cpu_class = CPUCLASS_486;
+		break;
+#endif
+#endif
+#ifndef I486_CPU
+	case CPUCLASS_486:
+		printf("NOTICE: this kernel does not support i486 CPU class\n");
+#ifdef I386_CPU
+		printf("NOTICE: lowering CPU class to i386\n");
+		cpu_class = CPUCLASS_386;
+		break;
+#endif
+#endif
+#ifndef I386_CPU
+	case CPUCLASS_386:
+		printf("NOTICE: this kernel does not support i386 CPU class\n");
+		panic("no appropriate CPU class available");
+#endif
+	default:
+		break;
+	}
+
+	if (cpu == CPU_486DLC) {
+#ifndef CYRIX_CACHE_WORKS
+		printf("WARNING: CYRIX 486DLC CACHE UNCHANGED.\n");
+#else
+#ifndef CYRIX_CACHE_REALLY_WORKS
+		printf("WARNING: CYRIX 486DLC CACHE ENABLED IN HOLD-FLUSH MODE.\n");
+#else
+		printf("WARNING: CYRIX 486DLC CACHE ENABLED.\n");
+#endif
+#endif
+	}
+
+}
+#endif	/* MULTIPROCESSOR */
+
 char *
 tm86_cpu_name(model)
 	int model;
@@ -1495,7 +1771,7 @@ tm86_cpu_name(model)
 }
 
 void
-identifycpu()
+old_identifycpu()
 {
 	extern char cpu_vendor[];
 	extern char cpu_brandstr[];
@@ -2402,8 +2678,8 @@ setregs(p, pack, stack, retval)
 
 #if NNPX > 0
 	/* If we were using the FPU, forget about it. */
-	if (npxproc == p)
-		npxdrop();
+	if (pcb->pcb_fpcpu != NULL)
+		npxsave_proc(p, 0);
 #endif
 
 #ifdef USER_LDT
@@ -2437,7 +2713,6 @@ setregs(p, pack, stack, retval)
  * Initialize segments and descriptor tables
  */
 
-union descriptor gdt[NGDT];
 union descriptor ldt[NLDT];
 struct gate_descriptor idt_region[NIDT];
 struct gate_descriptor *idt = idt_region;
@@ -2462,6 +2737,20 @@ setgate(gd, func, args, type, dpl, seg)
 }
 
 void
+unsetgate(gd)
+	struct gate_descriptor *gd;
+{
+	gd->gd_p = 0;
+	gd->gd_hioffset = 0;
+	gd->gd_looffset = 0;
+	gd->gd_selector = 0;
+	gd->gd_xx = 0;
+	gd->gd_stkcpy = 0;
+	gd->gd_type = 0;
+	gd->gd_dpl = 0;
+}
+
+void
 setregion(rd, base, limit)
 	struct region_descriptor *rd;
 	void *base;
@@ -2521,8 +2810,8 @@ fix_f00f(void)
 	idt = p;
 
 	/* Fix up paging redirect */
-	setgate(&idt[ 14], &IDTVEC(f00f_redirect), 0, SDT_SYS386TGT,
-		SEL_KPL, GCODE_SEL);
+	setgate(&idt[ 14], &IDTVEC(f00f_redirect), 0, SDT_SYS386TGT, SEL_KPL,
+	    GCODE_SEL);
 
 	/* Map first page RO */
 	pte = PTE_BASE + i386_btop(va);
@@ -2537,6 +2826,16 @@ fix_f00f(void)
 }
 #endif
 
+#ifdef MULTIPROCESSOR
+void
+cpu_init_idt()
+{
+	struct region_descriptor region;
+	setregion(&region, idt, NIDT * sizeof(idt[0]) - 1);
+	lidt(&region);
+}
+#endif	/* MULTIPROCESSOR */
+
 void
 init386(paddr_t first_avail)
 {
@@ -2565,18 +2864,20 @@ init386(paddr_t first_avail)
 	    (caddr_t)iomem_ex_storage, sizeof(iomem_ex_storage),
 	    EX_NOCOALESCE|EX_NOWAIT);
 
-	/* make gdt gates and memory segments */
+	/* make bootstrap gdt gates and memory segments */
 	setsegment(&gdt[GCODE_SEL].sd, 0, 0xfffff, SDT_MEMERA, SEL_KPL, 1, 1);
 	setsegment(&gdt[GICODE_SEL].sd, 0, 0xfffff, SDT_MEMERA, SEL_KPL, 1, 1);
 	setsegment(&gdt[GDATA_SEL].sd, 0, 0xfffff, SDT_MEMRWA, SEL_KPL, 1, 1);
-	setsegment(&gdt[GLDT_SEL].sd, ldt, sizeof(ldt) - 1, SDT_SYSLDT, SEL_KPL,
-	    0, 0);
+	setsegment(&gdt[GLDT_SEL].sd, ldt, sizeof(ldt) - 1, SDT_SYSLDT,
+	    SEL_KPL, 0, 0);
 	setsegment(&gdt[GUCODE1_SEL].sd, 0, i386_btop(VM_MAXUSER_ADDRESS) - 1,
 	    SDT_MEMERA, SEL_UPL, 1, 1);
 	setsegment(&gdt[GUCODE_SEL].sd, 0, i386_btop(I386_MAX_EXE_ADDR) - 1,
 	    SDT_MEMERA, SEL_UPL, 1, 1);
 	setsegment(&gdt[GUDATA_SEL].sd, 0, i386_btop(VM_MAXUSER_ADDRESS) - 1,
 	    SDT_MEMRWA, SEL_UPL, 1, 1);
+	setsegment(&gdt[GCPU_SEL].sd, &cpu_info_primary,
+	    sizeof(struct cpu_info)-1, SDT_MEMRWA, SEL_KPL, 1, 1);
 
 	/* make ldt gates and memory segments */
 	setgate(&ldt[LSYS5CALLS_SEL].gd, &IDTVEC(osyscall), 1, SDT_SYS386CGT,
@@ -2606,11 +2907,13 @@ init386(paddr_t first_avail)
 	setgate(&idt[ 16], &IDTVEC(fpu),     0, SDT_SYS386TGT, SEL_KPL, GCODE_SEL);
 	setgate(&idt[ 17], &IDTVEC(align),   0, SDT_SYS386TGT, SEL_KPL, GCODE_SEL);
 	setgate(&idt[ 18], &IDTVEC(rsvd),    0, SDT_SYS386TGT, SEL_KPL, GCODE_SEL);
-	for (i = 19; i < NIDT; i++)
+	for (i = 19; i < NRSVIDT; i++)
 		setgate(&idt[i], &IDTVEC(rsvd), 0, SDT_SYS386TGT, SEL_KPL, GCODE_SEL);
+	for (i = NRSVIDT; i < NIDT; i++)
+		unsetgate(&idt[i]);
 	setgate(&idt[128], &IDTVEC(syscall), 0, SDT_SYS386TGT, SEL_UPL, GCODE_SEL);
 
-	setregion(&region, gdt, sizeof(gdt) - 1);
+	setregion(&region, gdt, NGDT * sizeof(union descriptor) - 1);
 	lgdt(&region);
 	setregion(&region, idt, sizeof(idt_region) - 1);
 	lidt(&region);
@@ -2661,6 +2964,14 @@ init386(paddr_t first_avail)
 		panic("no BIOS memory map supplied");
 #endif
 
+#if defined(MULTIPROCESSOR)
+	/* install the page after boot args as PT page for first 4M */
+	pmap_enter(pmap_kernel(), (u_long)vtopte(0),
+	   i386_round_page(bootargv + bootargc), VM_PROT_READ|VM_PROT_WRITE,
+	   VM_PROT_READ|VM_PROT_WRITE|PMAP_WIRED);
+	memset(vtopte(0), 0, NBPG);  /* make sure it is clean before using */
+#endif
+
 	/*
 	 * account all the memory passed in the map from /boot
 	 * calculate avail_end and count the physmem.
@@ -2677,8 +2988,8 @@ init386(paddr_t first_avail)
 			a = i386_round_page(im->addr);
 			e = i386_trunc_page(im->addr + im->size);
 			/* skip first four pages */
-			if (a < 4 * NBPG)
-				a = 4 * NBPG;
+			if (a < 5 * NBPG)
+				a = 5 * NBPG;
 #ifdef DEBUG
 			printf(" %u-%u", a, e);
 #endif
@@ -2937,6 +3248,48 @@ cpu_reset()
 	for (;;);
 }
 
+void
+cpu_initclocks()
+{
+	(*initclock_func)();
+}
+
+void
+need_resched(struct cpu_info *ci)
+{
+	ci->ci_want_resched = 1;
+	ci->ci_astpending = 1;
+}
+
+#ifdef MULTIPROCESSOR
+/* Allocate an IDT vector slot within the given range.
+ * XXX needs locking to avoid MP allocation races.
+ */
+
+int
+idt_vec_alloc(int low, int high)
+{
+	int vec;
+
+	for (vec = low; vec <= high; vec++)
+		if (idt[vec].gd_p == 0)
+			return (vec);
+	return (0);
+}
+
+void
+idt_vec_set(int vec, void (*function)(void))
+{
+	setgate(&idt[vec], function, 0, SDT_SYS386IGT, SEL_KPL, GCODE_SEL);
+}
+
+void
+idt_vec_free(int vec)
+{
+	unsetgate(&idt[vec]);
+}
+#endif	/* MULTIPROCESSOR */
+
 /*  
  * machine dependent system variables.
  */ 
@@ -2998,7 +3351,7 @@ cpu_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
 	case CPU_CPUID:
 		return (sysctl_rdint(oldp, oldlenp, newp, cpu_id));
 	case CPU_CPUFEATURE:
-		return (sysctl_rdint(oldp, oldlenp, newp, cpu_feature));
+		return (sysctl_rdint(oldp, oldlenp, newp, curcpu()->ci_feature_flags));
 #if NAPM > 0
 	case CPU_APMWARN:
 		return (sysctl_int(oldp, oldlenp, newp, newlen, &cpu_apmwarn));
@@ -3204,6 +3557,9 @@ bus_mem_add_mapping(bpa, size, cacheable, bshp)
 	vaddr_t va;
 	pt_entry_t *pte;
 	bus_size_t map_size;
+#ifdef MULTIPROCESSOR
+	u_int32_t cpumask = 0;
+#endif
 
 	pa = i386_trunc_page(bpa);
 	endpa = i386_round_page(bpa + size);
@@ -3236,9 +3592,17 @@ bus_mem_add_mapping(bpa, size, cacheable, bshp)
 				*pte &= ~PG_N;
 			else
 				*pte |= PG_N;
+#ifdef MULTIPROCESSOR
+			pmap_tlb_shootdown(pmap_kernel(), va, *pte,
+			    &cpumask);
+#else
 			pmap_update_pg(va);
+#endif
 		}
 	}
+#ifdef MULTIPROCESSOR
+	pmap_tlb_shootnow(cpumask);
+#endif
 	pmap_update(pmap_kernel());
 
 	return 0;
@@ -3945,9 +4309,52 @@ _bus_dmamem_alloc_range(t, size, alignment, boundary, segs, nsegs, rsegs,
 void
 splassert_check(int wantipl, const char *func)
 {
-	if (cpl < wantipl) {
-		splassert_fail(wantipl, cpl, func);
-	}
+	if (lapic_tpr < wantipl)
+		splassert_fail(wantipl, lapic_tpr, func);
+}
+#endif
+
+#ifdef MULTIPROCESSOR
+void
+i386_intlock(struct intrframe iframe)
+{
+	if (iframe.if_ppl < IPL_SCHED)
+#ifdef notdef
+		spinlockmgr(&kernel_lock, LK_EXCLUSIVE|LK_CANRECURSE, 0);
+#else
+		__mp_lock(&kernel_lock);
+#endif
+}
+
+void
+i386_intunlock(struct intrframe iframe)
+{
+	if (iframe.if_ppl < IPL_SCHED)
+#ifdef notdef
+		spinlockmgr(&kernel_lock, LK_RELEASE, 0);
+#else
+		__mp_unlock(&kernel_lock);
+#endif
+}
+
+void
+i386_softintlock(void)
+{
+#ifdef notdef
+	spinlockmgr(&kernel_lock, LK_EXCLUSIVE|LK_CANRECURSE, 0);
+#else
+	__mp_lock(&kernel_lock);
+#endif
+}
+
+void
+i386_softintunlock(void)
+{
+#ifdef notdef
+	spinlockmgr(&kernel_lock, LK_RELEASE, 0);
+#else
+	__mp_unlock(&kernel_lock);
+#endif
 }
 #endif
 
@@ -3957,11 +4364,15 @@ splassert_check(int wantipl, const char *func)
  * We hand-code this to ensure that it's atomic.
  */
 void
-softintr(mask)
-	int mask;
+softintr(sir, vec)
+	int sir;
+	int vec;
 {
-	__asm __volatile("orl %1, %0" : "=m"(ipending) : "ir" (mask));
-
+	__asm __volatile("orl %1, %0" : "=m" (ipending) : "ir" (sir));
+#ifdef MULTIPROCESSOR
+	i82489_writereg(LAPIC_ICRLO,
+	    vec | LAPIC_DLMODE_FIXED | LAPIC_LVL_ASSERT | LAPIC_DEST_SELF);
+#endif
 }
 
 /*
@@ -3971,10 +4382,10 @@ int
 splraise(ncpl)
 	int ncpl;
 {
-	int ocpl = cpl;
+	int ocpl = lapic_tpr;
 
 	if (ncpl > ocpl)
-		cpl = ncpl;
+		lapic_tpr = ncpl;
 	return (ocpl);
 }
 
@@ -3986,7 +4397,7 @@ void
 splx(ncpl)
 	int ncpl;
 {
-	cpl = ncpl;
+	lapic_tpr = ncpl;
 	if (ipending & IUNMASK(ncpl))
 		Xspllower();
 }
@@ -3999,8 +4410,9 @@ int
 spllower(ncpl)
 	int ncpl;
 {
-	int ocpl = cpl;
+	int ocpl = lapic_tpr;
 
 	splx(ncpl);
 	return (ocpl);
 }
+
diff --git a/sys/arch/i386/i386/mainbus.c b/sys/arch/i386/i386/mainbus.c
index 47577e0a0b9..c242f14e5da 100644
--- a/sys/arch/i386/i386/mainbus.c
+++ b/sys/arch/i386/i386/mainbus.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mainbus.c,v 1.15 2002/03/14 01:26:32 millert Exp $	*/
+/*	$OpenBSD: mainbus.c,v 1.16 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: mainbus.c,v 1.21 1997/06/06 23:14:20 thorpej Exp $	*/
 
 /*
@@ -49,11 +49,22 @@
 #include "isa.h"
 #include "apm.h"
 #include "bios.h"
+#include "mpbios.h"
+
+#include <machine/cpuvar.h>
+#include <machine/i82093var.h>
+#include <machine/mpbiosvar.h>
 
 #if NBIOS > 0
 #include <machine/biosvar.h>
 #endif
 
+#if 0
+#ifdef SMP /* XXX MULTIPROCESSOR */
+#include <machine/mp.h>
+#endif
+#endif
+
 int	mainbus_match(struct device *, void *, void *);
 void	mainbus_attach(struct device *, struct device *, void *);
 
@@ -75,6 +86,8 @@ union mainbus_attach_args {
 #if NBIOS > 0
 	struct bios_attach_args mba_bios;
 #endif
+	struct cpu_attach_args mba_caa;
+	struct apic_attach_args	aaa_caa;
 };
 
 /*
@@ -104,6 +117,7 @@ mainbus_attach(parent, self, aux)
 	void *aux;
 {
 	union mainbus_attach_args mba;
+	extern int cpu_id, cpu_feature;
 
 	printf("\n");
 
@@ -116,6 +130,40 @@ mainbus_attach(parent, self, aux)
 	}
 #endif
 
+#if NMPBIOS > 0
+	if (mpbios_probe(self))
+		mpbios_scan(self);
+	else
+#endif
+	{
+		struct cpu_attach_args caa;
+		
+		memset(&caa, 0, sizeof(caa));
+		caa.caa_name = "cpu";
+		caa.cpu_number = 0;
+		caa.cpu_role = CPU_ROLE_SP;
+		caa.cpu_func = 0;
+		caa.cpu_signature = cpu_id;
+		caa.feature_flags = cpu_feature;
+
+		config_found(self, &caa, mainbus_print);
+	}
+
+#if 0
+#ifdef SMP
+	if (bios_smpinfo != NULL) {
+		struct mp_float *mp = bios_smpinfo;
+
+		printf("%s: MP 1.%d configuration %d\n", self->dv_xname,
+		    mp->revision, mp->feature1);
+	}
+#ifdef CPU_DEBUG
+	else
+		printf ("%s: No MP configuration found.", self->dv_xname);
+#endif
+#endif
+#endif
+
 	/*
 	 * XXX Note also that the presence of a PCI bus should
 	 * XXX _always_ be checked, and if present the bus should be
diff --git a/sys/arch/i386/i386/microtime.s b/sys/arch/i386/i386/microtime.s
index 7523e48b73a..ac1b1742cb4 100644
--- a/sys/arch/i386/i386/microtime.s
+++ b/sys/arch/i386/i386/microtime.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: microtime.s,v 1.19 2003/06/04 16:36:14 deraadt Exp $	*/
+/*	$OpenBSD: microtime.s,v 1.20 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: microtime.s,v 1.16 1995/04/17 12:06:47 cgd Exp $	*/
 
 /*-
@@ -42,7 +42,7 @@
  * overridden (i.e. it is 100Hz).
  */
 #ifndef HZ
-ENTRY(microtime)
+ENTRY(i8254_microtime)
 
 #if defined(I586_CPU) || defined(I686_CPU)
 	movl	_C_LABEL(pentium_mhz), %ecx
diff --git a/sys/arch/i386/i386/mpbios.c b/sys/arch/i386/i386/mpbios.c
new file mode 100644
index 00000000000..689b9ace3f5
--- /dev/null
+++ b/sys/arch/i386/i386/mpbios.c
@@ -0,0 +1,1105 @@
+/*	$OpenBSD: mpbios.c,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+/*	$NetBSD: mpbios.c,v 1.2 2002/10/01 12:56:57 fvdl Exp $	*/
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Copyright (c) 1999 Stefan Grefen
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed by the NetBSD
+ *      Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Derived from FreeBSD's mp_machdep.c
+ */
+/*
+ * Copyright (c) 1996, by Steve Passe
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. The name of the developer may NOT be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+/*
+ * The Intel MP-stuff is just one way of x86 SMP systems
+ * so only Intel MP specific stuff is here.
+ */
+
+#include "mpbios.h"
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/device.h>
+#include <sys/malloc.h>
+
+#include <uvm/uvm_extern.h>
+
+#include <machine/specialreg.h>
+#include <machine/cputypes.h>
+#include <machine/cpuvar.h>
+#include <machine/bus.h>
+#include <machine/mpbiosreg.h>
+#include <machine/mpbiosvar.h>
+
+#include <machine/i82093reg.h>
+#include <machine/i82093var.h>
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+#include <dev/isa/isareg.h>
+
+#include <dev/eisa/eisavar.h>	/* for ELCR* def'ns */
+
+
+static struct mpbios_ioapic default_ioapic = {
+    2, 0, 1, IOAPICENTRY_FLAG_EN, (caddr_t)IOAPIC_BASE_DEFAULT
+};
+
+/* descriptions of MP basetable entries */
+struct mpbios_baseentry {
+	u_int8_t  	type;
+	u_int8_t  	length;
+	u_int16_t	count;
+	const char    	*name;
+};
+
+static const char *loc_where[] = {
+	"extended bios data area",
+	"last page of base memory",
+	"bios"
+};
+
+struct mp_map
+{
+	vaddr_t 	baseva;
+	int	 	vsize;
+	paddr_t 	pa;
+	paddr_t 	pg;
+	int		psize;
+};
+
+int	mp_print(void *, const char *);
+int	mp_match(struct device *, void *, void *);
+int	mpbios_cpu_start(struct cpu_info *);
+const void *mpbios_search(struct device *, paddr_t, int,
+    struct mp_map *);
+static __inline int mpbios_cksum(const void *, int);
+
+void	mp_cfg_special_intr(const struct mpbios_int *, u_int32_t *);
+void	mp_cfg_pci_intr(const struct mpbios_int *, u_int32_t *);
+void	mp_cfg_eisa_intr(const struct mpbios_int *, u_int32_t *);
+void	mp_cfg_isa_intr(const struct mpbios_int *, u_int32_t *);
+void	mp_print_special_intr (int);
+void	mp_print_pci_intr (int);
+void	mp_print_eisa_intr (int);
+void	mp_print_isa_intr (int);
+
+void	mpbios_cpu(const u_int8_t *, struct device *);
+void	mpbios_bus(const u_int8_t *, struct device *);
+void	mpbios_ioapic(const u_int8_t *, struct device *);
+void	mpbios_int(const u_int8_t *, int, struct mp_intr_map *);
+
+const void *mpbios_map(paddr_t, int, struct mp_map *);
+static __inline void mpbios_unmap(struct mp_map *);
+
+/*
+ * globals to help us bounce our way through parsing the config table.
+ */
+
+static struct mp_map mp_cfg_table_map;
+static struct mp_map mp_fp_map;
+const struct mpbios_cth	*mp_cth;
+const struct mpbios_fps	*mp_fps;
+
+#ifdef MPVERBOSE
+int mp_verbose = 1;
+#else
+int mp_verbose = 0;
+#endif
+
+int
+mp_print(aux, pnp)
+	void *aux;
+	const char *pnp;
+{
+	struct cpu_attach_args * caa = (struct cpu_attach_args *) aux;
+	if (pnp)
+		printf("%s at %s:", caa->caa_name, pnp);
+	return (UNCONF);
+}
+
+int
+mp_match(parent, cfv, aux)
+	struct device *parent;
+	void *cfv;
+	void *aux;
+{
+        struct cfdata *cf = (struct cfdata *)cfv;
+	struct cpu_attach_args * caa = (struct cpu_attach_args *) aux;
+	if (strcmp(caa->caa_name, cf->cf_driver->cd_name))
+		return 0;
+
+	return ((*cf->cf_attach->ca_match)(parent, cf, aux));
+}
+
+/*
+ * Map a chunk of memory read-only and return an appropriately
+ * const'ed pointer.
+ */
+const void *
+mpbios_map(pa, len, handle)
+	paddr_t pa;
+	int len;
+	struct mp_map *handle;
+{
+	paddr_t pgpa = i386_trunc_page(pa);
+	paddr_t endpa = i386_round_page(pa + len);
+	vaddr_t va = uvm_km_valloc(kernel_map, endpa - pgpa);
+	vaddr_t retva = va + (pa & PGOFSET);
+
+	handle->pa = pa;
+	handle->pg = pgpa;
+	handle->psize = len;
+	handle->baseva = va;
+	handle->vsize = endpa-pgpa;
+
+	do {
+#if 1
+		pmap_kenter_pa(va, pgpa, VM_PROT_READ);
+#else
+		pmap_enter(pmap_kernel(), va, pgpa, VM_PROT_READ, TRUE,
+		    VM_PROT_READ);
+#endif
+		va += NBPG;
+		pgpa += NBPG;
+	} while (pgpa < endpa);
+
+	return ((const void *)retva);
+}
+
+static __inline void
+mpbios_unmap(handle)
+	struct mp_map *handle;
+{
+#if 1
+  	pmap_kremove(handle->baseva, handle->vsize);
+#else
+  	pmap_extract(pmap_kernel(), handle->baseva, NULL);
+#endif
+	uvm_km_free(kernel_map, handle->baseva, handle->vsize);
+}
+
+/*
+ * Look for an Intel MP spec table, indicating SMP capable hardware.
+ */
+int
+mpbios_probe(self)
+	struct device *self;
+{
+	paddr_t  	ebda, memtop;
+
+	paddr_t		cthpa;
+	int		cthlen;
+	const u_int8_t	*mpbios_page;
+	int 		scan_loc;
+
+	struct		mp_map t;
+
+	/* see if EBDA exists */
+
+	mpbios_page = mpbios_map(0, NBPG, &t);
+
+	/* XXX Ugly magic constants below. */
+	ebda = *(const u_int16_t *)(&mpbios_page[0x40e]);
+	ebda <<= 4;
+
+	memtop = *(const u_int16_t *)(&mpbios_page[0x413]);
+	memtop <<= 10;
+
+	mpbios_page = NULL;
+	mpbios_unmap(&t);
+
+	scan_loc = 0;
+
+	if (ebda && ebda < IOM_BEGIN ) {
+		mp_fps = mpbios_search(self, ebda, 1024, &mp_fp_map);
+		if (mp_fps != NULL)
+			goto found;
+	}
+
+	scan_loc = 1;
+
+	if (memtop && memtop <= IOM_BEGIN ) {
+		mp_fps = mpbios_search(self, memtop - 1024, 1024, &mp_fp_map);
+		if (mp_fps != NULL)
+			goto found;
+	}
+
+	scan_loc = 2;
+
+	mp_fps = mpbios_search(self, BIOS_BASE, BIOS_COUNT, &mp_fp_map);
+	if (mp_fps != NULL)
+		goto found;
+
+	/* nothing found */
+	return (0);
+
+ found:
+	if (mp_verbose)
+		printf("%s: MP floating pointer found in %s at 0x%lx\n",
+		    self->dv_xname, loc_where[scan_loc], mp_fp_map.pa);
+
+	if (mp_fps->pap == 0) {
+		if (mp_fps->mpfb1 == 0) {
+			printf("%s: MP fps invalid: "
+			    "no default config and no configuration table\n",
+			    self->dv_xname);
+
+			goto err;
+		}
+		printf("%s: MP default configuration %d\n",
+		    self->dv_xname, mp_fps->mpfb1);
+		return (10);
+	}
+
+	cthpa = mp_fps->pap;
+
+	mp_cth = mpbios_map(cthpa, sizeof (*mp_cth), &mp_cfg_table_map);
+	cthlen = mp_cth->base_len;
+	mpbios_unmap(&mp_cfg_table_map);
+
+	mp_cth = mpbios_map(cthpa, cthlen, &mp_cfg_table_map);
+
+	if (mp_verbose)
+		printf("%s: MP config table at 0x%lx, %d bytes long\n",
+		    self->dv_xname, cthpa, cthlen);
+
+	if (mp_cth->signature != MP_CT_SIG) {
+		printf("%s: MP signature mismatch (%x vs %x)\n",
+		    self->dv_xname,
+		    MP_CT_SIG, mp_cth->signature);
+		goto err;
+	}
+
+	if (mpbios_cksum(mp_cth, cthlen)) {
+		printf ("%s: MP Configuration Table checksum mismatch\n",
+		    self->dv_xname);
+		goto err;
+	}
+	return (10);
+
+ err:
+	if (mp_fps) {
+		mp_fps = NULL;
+		mpbios_unmap(&mp_fp_map);
+	}
+	if (mp_cth) {
+		mp_cth = NULL;
+		mpbios_unmap(&mp_cfg_table_map);
+	}
+	return (0);
+}
+
+
+/*
+ * Simple byte checksum used on config tables.
+ */
+
+static __inline int
+mpbios_cksum (start, len)
+	const void *start;
+	int len;
+{
+	unsigned char res=0;
+	const char *p = start;
+	const char *end = p + len;
+
+	while (p < end)
+		res += *p++;
+
+	return res;
+}
+
+
+/*
+ * Look for the MP floating pointer signature in the given physical
+ * address range.
+ *
+ * We map the memory, scan through it, and unmap it.
+ * If we find it, remap the floating pointer structure and return it.
+ */
+
+const void *
+mpbios_search (self, start, count, map)
+	struct device *self;
+	paddr_t start;
+	int count;
+	struct mp_map *map;
+{
+	struct mp_map t;
+
+	int i, len;
+	const struct mpbios_fps *m;
+	int end = count - sizeof(*m);
+	const u_int8_t *base = mpbios_map(start, count, &t);
+
+	if (mp_verbose)
+		printf("%s: scanning 0x%lx to 0x%lx for MP signature\n",
+		    self->dv_xname, start, start + count - sizeof(*m));
+
+	for (i = 0; i <= end; i += 4) {
+		m = (struct mpbios_fps *)&base[i];
+
+		if ((m->signature == MP_FP_SIG) &&
+		    ((len = m->length << 4) != 0) &&
+		    mpbios_cksum(m, (m->length << 4)) == 0) {
+			mpbios_unmap(&t);
+
+			return (mpbios_map(start + i, len, map));
+		}
+	}
+	mpbios_unmap(&t);
+
+	return (0);
+}
+
+/*
+ * MP configuration table parsing.
+ */
+
+static struct mpbios_baseentry mp_conf[] =
+{
+	{0, 20, 0, "cpu"},
+	{1, 8, 0, "bus"},
+	{2, 8, 0, "ioapic"},
+	{3, 8, 0, "ioint"},
+	{4, 8, 0, "lint"},
+};
+
+struct mp_bus *mp_busses;
+int mp_nbus;
+struct mp_intr_map *mp_intrs;
+
+struct mp_intr_map *lapic_ints[2]; /* XXX */
+int mp_isa_bus = -1;		/* XXX */
+int mp_eisa_bus = -1;		/* XXX */
+
+static struct mp_bus extint_bus = {
+	"ExtINT",
+	-1,
+	mp_print_special_intr,
+	mp_cfg_special_intr,
+	0
+};
+static struct mp_bus smi_bus = {
+	"SMI",
+	-1,
+	mp_print_special_intr,
+	mp_cfg_special_intr,
+	0
+};
+static struct mp_bus nmi_bus = {
+	"NMI",
+	-1,
+	mp_print_special_intr,
+	mp_cfg_special_intr,
+	0
+};
+
+
+/*
+ * 1st pass on BIOS's Intel MP specification table.
+ *
+ * initializes:
+ *	mp_ncpus = 1
+ *
+ * determines:
+ *	cpu_apic_address (common to all CPUs)
+ *	ioapic_address[N]
+ *	mp_naps
+ *	mp_nbusses
+ *	mp_napics
+ *	nintrs
+ */
+void
+mpbios_scan(self)
+	struct device *self;
+{
+	const u_int8_t 	*position, *end;
+	int		count;
+	int		type;
+	int		intr_cnt, cur_intr;
+	paddr_t		lapic_base;
+
+	printf("%s: Intel MP Specification ", self->dv_xname);
+
+	switch (mp_fps->spec_rev) {
+	case 1:
+		printf("(Version 1.1)");
+		break;
+	case 4:
+		printf("(Version 1.4)");
+		break;
+	default:
+		printf("(unrecognized rev %d)", mp_fps->spec_rev);
+	}
+
+	/*
+	 * looks like we've got a MP system.  start setting up
+	 * infrastructure..
+	 * XXX is this the right place??
+	 */
+
+	lapic_base = LAPIC_BASE;
+	if (mp_cth != NULL)
+		lapic_base = (paddr_t)mp_cth->apic_address;
+
+	lapic_boot_init(lapic_base);
+
+	/* check for use of 'default' configuration */
+	if (mp_fps->mpfb1 != 0) {
+		struct mpbios_proc pe;
+
+		printf("\n%s: MP default configuration %d\n",
+		    self->dv_xname, mp_fps->mpfb1);
+
+		/* use default addresses */
+		pe.apic_id = cpu_number();
+		pe.cpu_flags = PROCENTRY_FLAG_EN|PROCENTRY_FLAG_BP;
+		pe.cpu_signature = cpu_info_primary.ci_signature;
+		pe.feature_flags = cpu_info_primary.ci_feature_flags;
+
+		mpbios_cpu((u_int8_t *)&pe, self);
+
+		pe.apic_id = 1 - cpu_number();
+		pe.cpu_flags = PROCENTRY_FLAG_EN;
+
+		mpbios_cpu((u_int8_t *)&pe, self);
+
+		mpbios_ioapic((u_int8_t *)&default_ioapic, self);
+
+		/* XXX */
+		printf("%s: WARNING: interrupts not configured\n",
+		    self->dv_xname);
+		panic("lazy bum");
+		return;
+	} else {
+		/*
+		 * should not happen; mp_probe returns 0 in this case,
+		 * but..
+		 */
+		if (mp_cth == NULL)
+			panic("mpbios_scan: no config (can't happen?)");
+
+		printf(" (%8.8s %12.12s)\n",
+		    mp_cth->oem_id, mp_cth->product_id);
+
+		/*
+		 * Walk the table once, counting items
+		 */
+		position = (const u_int8_t *)(mp_cth);
+		end = position + mp_cth->base_len;
+		position += sizeof(*mp_cth);
+
+		count = mp_cth->entry_count;
+		intr_cnt = 0;
+
+		while ((count--) && (position < end)) {
+			type = *position;
+			if (type >= MPS_MCT_NTYPES) {
+				printf("%s: unknown entry type %x"
+				    " in MP config table\n",
+				    self->dv_xname, type);
+				break;
+			}
+			mp_conf[type].count++;
+			if (type == MPS_MCT_BUS) {
+				const struct mpbios_bus *bp =
+				    (const struct mpbios_bus *)position;
+				if (bp->bus_id >= mp_nbus)
+					mp_nbus = bp->bus_id + 1;
+			}
+			/*
+			 * Count actual interrupt instances.
+			 * dst_apic_id of MPS_ALL_APICS means "wired to all
+			 * apics of this type".
+			 */
+			if ((type == MPS_MCT_IOINT) ||
+			    (type == MPS_MCT_LINT)) {
+				const struct mpbios_int *ie =
+				    (const struct mpbios_int *)position;
+				if (ie->dst_apic_id != MPS_ALL_APICS)
+					intr_cnt++;
+				else if (type == MPS_MCT_IOINT)
+					intr_cnt +=
+					    mp_conf[MPS_MCT_IOAPIC].count;
+				else
+					intr_cnt += mp_conf[MPS_MCT_CPU].count;
+			}
+			position += mp_conf[type].length;
+		}
+
+		mp_busses = malloc(sizeof(struct mp_bus) * mp_nbus,
+		    M_DEVBUF, M_NOWAIT);
+		memset(mp_busses, 0, sizeof(struct mp_bus) * mp_nbus);
+		mp_intrs = malloc(sizeof(struct mp_intr_map) * intr_cnt,
+		    M_DEVBUF, M_NOWAIT);
+
+		/* re-walk the table, recording info of interest */
+		position = (const u_int8_t *)mp_cth + sizeof(*mp_cth);
+		count = mp_cth->entry_count;
+		cur_intr = 0;
+
+		while ((count--) && (position < end)) {
+			switch (type = *(u_char *)position) {
+			case MPS_MCT_CPU:
+				mpbios_cpu(position, self);
+				break;
+			case MPS_MCT_BUS:
+				mpbios_bus(position, self);
+				break;
+			case MPS_MCT_IOAPIC:
+				mpbios_ioapic(position, self);
+				break;
+			case MPS_MCT_IOINT:
+			case MPS_MCT_LINT:
+				mpbios_int(position, type,
+				    &mp_intrs[cur_intr]);
+				cur_intr++;
+				break;
+			default:
+				printf("%s: unknown entry type %x "
+				    "in MP config table\n",
+				    self->dv_xname, type);
+				/* NOTREACHED */
+				return;
+			}
+
+			(u_char*)position += mp_conf[type].length;
+		}
+		if (mp_verbose && mp_cth->ext_len)
+			printf("%s: MP WARNING: %d "
+			    "bytes of extended entries not examined\n",
+			    self->dv_xname, mp_cth->ext_len);
+	}
+	/* Clean up. */
+	mp_fps = NULL;
+	mpbios_unmap(&mp_fp_map);
+	if (mp_cth != NULL) {
+		mp_cth = NULL;
+		mpbios_unmap(&mp_cfg_table_map);
+	}
+}
+
+void
+mpbios_cpu(ent, self)
+	const u_int8_t *ent;
+	struct device *self;
+{
+	const struct mpbios_proc *entry = (const struct mpbios_proc *)ent;
+	struct cpu_attach_args caa;
+
+	/* XXX move this into the CPU attachment goo. */
+	/* check for usability */
+	if (!(entry->cpu_flags & PROCENTRY_FLAG_EN))
+		return;
+
+	/* check for BSP flag */
+	if (entry->cpu_flags & PROCENTRY_FLAG_BP)
+		caa.cpu_role = CPU_ROLE_BP;
+	else
+		caa.cpu_role = CPU_ROLE_AP;
+
+	caa.caa_name = "cpu";
+	caa.cpu_number = entry->apic_id;
+	caa.cpu_func = &mp_cpu_funcs;
+#if 1 /* XXX Will be removed when the real stuff is probed */
+	caa.cpu_signature = entry->cpu_signature;
+
+	/*
+	 * XXX this is truncated to just contain the low-order 16 bits
+	 * of the flags on at least some MP bioses
+	 */
+	caa.feature_flags = entry->feature_flags;
+#endif
+
+	config_found_sm(self, &caa, mp_print, mp_match);
+}
+
+/*
+ * The following functions conspire to compute base ioapic redirection
+ * table entry for a given interrupt line.
+ *
+ * Fill in: trigger mode, polarity, and possibly delivery mode.
+ */
+void
+mp_cfg_special_intr(entry, redir)
+	const struct mpbios_int *entry;
+	u_int32_t *redir;
+{
+
+	/*
+	 * All of these require edge triggered, zero vector,
+	 * appropriate delivery mode.
+	 * see page 13 of the 82093AA datasheet.
+	 */
+	*redir &= ~IOAPIC_REDLO_DEL_MASK;
+	*redir &= ~IOAPIC_REDLO_VECTOR_MASK;
+	*redir &= ~IOAPIC_REDLO_LEVEL;
+
+	switch (entry->int_type) {
+	case MPS_INTTYPE_NMI:
+		*redir |= (IOAPIC_REDLO_DEL_NMI<<IOAPIC_REDLO_DEL_SHIFT);
+		break;
+
+	case MPS_INTTYPE_SMI:
+		*redir |= (IOAPIC_REDLO_DEL_SMI<<IOAPIC_REDLO_DEL_SHIFT);
+		break;
+	case MPS_INTTYPE_ExtINT:
+		/*
+		 * We are using the ioapic in "native" mode.
+		 * This indicates where the 8259 is wired to the ioapic
+		 * and/or local apic..
+		 */
+		*redir |= (IOAPIC_REDLO_DEL_EXTINT<<IOAPIC_REDLO_DEL_SHIFT);
+		*redir |= (IOAPIC_REDLO_MASK);
+		break;
+	default:
+		panic("unknown MPS interrupt type %d", entry->int_type);
+	}
+}
+
+/* XXX too much duplicated code here. */
+
+void
+mp_cfg_pci_intr(entry, redir)
+	const struct mpbios_int *entry;
+	u_int32_t *redir;
+{
+	int mpspo = entry->int_flags & 0x03; /* XXX magic */
+	int mpstrig = (entry->int_flags >> 2) & 0x03; /* XXX magic */
+
+	*redir &= ~IOAPIC_REDLO_DEL_MASK;
+	switch (mpspo) {
+	case MPS_INTPO_ACTHI:
+		*redir &= ~IOAPIC_REDLO_ACTLO;
+		break;
+	case MPS_INTPO_DEF:
+	case MPS_INTPO_ACTLO:
+		*redir |= IOAPIC_REDLO_ACTLO;
+		break;
+	default:
+		panic("unknown MPS interrupt polarity %d", mpspo);
+	}
+
+	if (entry->int_type != MPS_INTTYPE_INT) {
+		mp_cfg_special_intr(entry, redir);
+		return;
+	}
+	*redir |= (IOAPIC_REDLO_DEL_LOPRI<<IOAPIC_REDLO_DEL_SHIFT);
+
+	switch (mpstrig) {
+	case MPS_INTTR_DEF:
+	case MPS_INTTR_LEVEL:
+		*redir |= IOAPIC_REDLO_LEVEL;
+		break;
+	case MPS_INTTR_EDGE:
+		*redir &= ~IOAPIC_REDLO_LEVEL;
+		break;
+	default:
+		panic("unknown MPS interrupt trigger %d", mpstrig);
+	}
+}
+
+void
+mp_cfg_eisa_intr (entry, redir)
+	const struct mpbios_int *entry;
+	u_int32_t *redir;
+{
+	int mpspo = entry->int_flags & 0x03; /* XXX magic */
+	int mpstrig = (entry->int_flags >> 2) & 0x03; /* XXX magic */
+
+	*redir &= ~IOAPIC_REDLO_DEL_MASK;
+	switch (mpspo) {
+	case MPS_INTPO_DEF:
+	case MPS_INTPO_ACTHI:
+		*redir &= ~IOAPIC_REDLO_ACTLO;
+		break;
+	case MPS_INTPO_ACTLO:
+		*redir |= IOAPIC_REDLO_ACTLO;
+		break;
+	default:
+		panic("unknown MPS interrupt polarity %d", mpspo);
+	}
+
+	if (entry->int_type != MPS_INTTYPE_INT) {
+		mp_cfg_special_intr(entry, redir);
+		return;
+	}
+	*redir |= (IOAPIC_REDLO_DEL_LOPRI<<IOAPIC_REDLO_DEL_SHIFT);
+
+	switch (mpstrig) {
+	case MPS_INTTR_LEVEL:
+		*redir |= IOAPIC_REDLO_LEVEL;
+		break;
+	case MPS_INTTR_EDGE:
+		*redir &= ~IOAPIC_REDLO_LEVEL;
+		break;
+	case MPS_INTTR_DEF:
+		/*
+		 * Set "default" setting based on ELCR value snagged
+		 * earlier.
+		 */
+		if (mp_busses[entry->src_bus_id].mb_data &
+		    (1<<entry->src_bus_irq)) {
+			*redir |= IOAPIC_REDLO_LEVEL;
+		} else {
+			*redir &= ~IOAPIC_REDLO_LEVEL;
+		}
+		break;
+	default:
+		panic("unknown MPS interrupt trigger %d", mpstrig);
+	}
+}
+
+
+void
+mp_cfg_isa_intr(entry, redir)
+	const struct mpbios_int *entry;
+	u_int32_t *redir;
+{
+	int mpspo = entry->int_flags & 0x03; /* XXX magic */
+	int mpstrig = (entry->int_flags >> 2) & 0x03; /* XXX magic */
+
+	*redir &= ~IOAPIC_REDLO_DEL_MASK;
+	switch (mpspo) {
+	case MPS_INTPO_DEF:
+	case MPS_INTPO_ACTHI:
+		*redir &= ~IOAPIC_REDLO_ACTLO;
+		break;
+	case MPS_INTPO_ACTLO:
+		*redir |= IOAPIC_REDLO_ACTLO;
+		break;
+	default:
+		panic("unknown MPS interrupt polarity %d", mpspo);
+	}
+
+	if (entry->int_type != MPS_INTTYPE_INT) {
+		mp_cfg_special_intr(entry, redir);
+		return;
+	}
+	*redir |= (IOAPIC_REDLO_DEL_LOPRI << IOAPIC_REDLO_DEL_SHIFT);
+
+	switch (mpstrig) {
+	case MPS_INTTR_LEVEL:
+		*redir |= IOAPIC_REDLO_LEVEL;
+		break;
+	case MPS_INTTR_DEF:
+	case MPS_INTTR_EDGE:
+		*redir &= ~IOAPIC_REDLO_LEVEL;
+		break;
+	default:
+		panic("unknown MPS interrupt trigger %d", mpstrig);
+	}
+}
+
+
+void
+mp_print_special_intr(intr)
+	int intr;
+{
+}
+
+void
+mp_print_pci_intr(intr)
+	int intr;
+{
+	printf(" device %d INT_%c", (intr >> 2) & 0x1f, 'A' + (intr & 0x3));
+}
+
+void
+mp_print_isa_intr(intr)
+	int intr;
+{
+	printf(" irq %d", intr);
+}
+
+void
+mp_print_eisa_intr(intr)
+	int intr;
+{
+	printf(" EISA irq %d", intr);
+}
+
+
+
+#define TAB_UNIT	4
+#define TAB_ROUND(a)	_TAB_ROUND(a, TAB_UNIT)
+
+#define _TAB_ROUND(a,u)	(((a) + (u - 1)) & ~(u - 1))
+#define EXTEND_TAB(a,u)	(!(_TAB_ROUND(a, u) == _TAB_ROUND((a + 1), u)))
+
+void
+mpbios_bus(ent, self)
+	const u_int8_t *ent;
+	struct device *self;
+{
+	const struct mpbios_bus *entry = (const struct mpbios_bus *)ent;
+	int bus_id = entry->bus_id;
+
+	printf("%s: bus %d is type %6.6s\n", self->dv_xname,
+	    bus_id, entry->bus_type);
+
+#ifdef DIAGNOSTIC
+	/*
+	 * This "should not happen" unless the table changes out
+	 * from underneath us
+	 */
+	if (bus_id >= mp_nbus) {
+		panic("%s: bus number %d out of range?? (type %6.6s)\n",
+		    self->dv_xname, bus_id, entry->bus_type);
+	}
+#endif
+
+	mp_busses[bus_id].mb_intrs = NULL;
+
+	if (memcmp(entry->bus_type, "PCI   ", 6) == 0) {
+		mp_busses[bus_id].mb_name = "pci";
+		mp_busses[bus_id].mb_idx = bus_id;
+		mp_busses[bus_id].mb_intr_print = mp_print_pci_intr;
+		mp_busses[bus_id].mb_intr_cfg = mp_cfg_pci_intr;
+	} else if (memcmp(entry->bus_type, "EISA  ", 6) == 0) {
+		mp_busses[bus_id].mb_name = "eisa";
+		mp_busses[bus_id].mb_idx = bus_id;
+		mp_busses[bus_id].mb_intr_print = mp_print_eisa_intr;
+		mp_busses[bus_id].mb_intr_cfg = mp_cfg_eisa_intr;
+
+		mp_busses[bus_id].mb_data = inb(ELCR0) | (inb(ELCR1) << 8);
+
+		if (mp_eisa_bus != -1)
+			printf("%s: multiple eisa busses?\n",
+			    self->dv_xname);
+		else
+			mp_eisa_bus = bus_id;
+	} else if (memcmp(entry->bus_type, "ISA   ", 6) == 0) {
+		mp_busses[bus_id].mb_name = "isa";
+		mp_busses[bus_id].mb_idx = 0; /* XXX */
+		mp_busses[bus_id].mb_intr_print = mp_print_isa_intr;
+		mp_busses[bus_id].mb_intr_cfg = mp_cfg_isa_intr;
+		if (mp_isa_bus != -1)
+			printf("%s: multiple isa busses?\n",
+			    self->dv_xname);
+		else
+			mp_isa_bus = bus_id;
+	} else {
+		printf("%s: unsupported bus type %6.6s\n", self->dv_xname,
+		    entry->bus_type);
+	}
+}
+
+
+void
+mpbios_ioapic(ent, self)
+	const u_int8_t *ent;
+	struct device *self;
+{
+	const struct mpbios_ioapic *entry = (const struct mpbios_ioapic *)ent;
+	struct apic_attach_args aaa;
+
+	/* XXX let flags checking happen in ioapic driver.. */
+	if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN))
+		return;
+
+	aaa.aaa_name = "ioapic";
+	aaa.apic_id = entry->apic_id;
+	aaa.apic_version = entry->apic_version;
+	aaa.apic_address = (paddr_t)entry->apic_address;
+	aaa.flags =  (mp_fps->mpfb2 & 0x80) ? IOAPIC_PICMODE : IOAPIC_VWIRE;
+
+	config_found_sm(self, &aaa, mp_print, mp_match);
+}
+
+static const char inttype_fmt[] = "\177\020"
+		"f\0\2type\0" "=\1NMI\0" "=\2SMI\0" "=\3ExtINT\0";
+
+static const char flagtype_fmt[] = "\177\020"
+		"f\0\2pol\0" "=\1Act Hi\0" "=\3Act Lo\0"
+		"f\2\2trig\0" "=\1Edge\0" "=\3Level\0";
+
+void
+mpbios_int(ent, enttype, mpi)
+	const u_int8_t *ent;
+	int enttype;
+	struct mp_intr_map *mpi;
+{
+	const struct mpbios_int *entry = (const struct mpbios_int *)ent;
+	struct ioapic_softc *sc = NULL;
+
+	struct mp_intr_map *altmpi;
+	struct mp_bus *mpb;
+
+	u_int32_t id = entry->dst_apic_id;
+	u_int32_t pin = entry->dst_apic_int;
+	u_int32_t bus = entry->src_bus_id;
+	u_int32_t dev = entry->src_bus_irq;
+	u_int32_t type = entry->int_type;
+	u_int32_t flags = entry->int_flags;
+
+	switch (type) {
+	case MPS_INTTYPE_INT:
+		mpb = &(mp_busses[bus]);
+		break;
+	case MPS_INTTYPE_ExtINT:
+		mpb = &extint_bus;
+		break;
+	case MPS_INTTYPE_SMI:
+		mpb = &smi_bus;
+		break;
+	case MPS_INTTYPE_NMI:
+		mpb = &nmi_bus;
+		break;
+	}
+	mpi->next = mpb->mb_intrs;
+	mpb->mb_intrs = mpi;
+	mpi->bus = mpb;
+	mpi->bus_pin = dev;
+
+	mpi->ioapic_ih = APIC_INT_VIA_APIC |
+	    ((id << APIC_INT_APIC_SHIFT) | ((pin << APIC_INT_PIN_SHIFT)));
+
+	mpi->type = type;
+	mpi->flags = flags;
+	mpi->redir = 0;
+	if (mpb->mb_intr_cfg == NULL) {
+		printf("mpbios: can't find bus %d for apic %d pin %d\n",
+		    bus, id, pin);
+		return;
+	}
+
+	(*mpb->mb_intr_cfg)(entry, &mpi->redir);
+
+	if (enttype == MPS_MCT_IOINT) {
+		sc = ioapic_find(id);
+		if (sc == NULL) {
+			printf("mpbios: can't find ioapic %d\n", id);
+			return;
+		}
+
+		mpi->ioapic = sc;
+		mpi->ioapic_pin = pin;
+
+		altmpi = sc->sc_pins[pin].ip_map;
+
+		if (altmpi != NULL) {
+			if ((altmpi->type != type) ||
+			    (altmpi->flags != flags)) {
+				printf(
+				    "%s: conflicting map entries for pin %d\n",
+				    sc->sc_dev.dv_xname, pin);
+			}
+		} else {
+			sc->sc_pins[pin].ip_map = mpi;
+		}
+	} else {
+		if (id != MPS_ALL_APICS)
+			panic("can't deal with not-all-lapics interrupt yet!");
+		if (pin >= 2)
+			printf("pin %d of local apic doesn't exist!\n", pin);
+		else {
+			mpi->ioapic = NULL;
+			mpi->ioapic_pin = pin;
+			lapic_ints[pin] = mpi;
+		}
+	}
+	if (mp_verbose) {
+		printf("%s: int%d attached to %s",
+		    sc ? sc->sc_dev.dv_xname : "local apic", pin,
+		    mpb->mb_name);
+		if (mpb->mb_idx != -1)
+			printf("%d", mpb->mb_idx);
+
+		if (mpb != NULL)
+
+		(*(mpb->mb_intr_print))(dev);
+
+		printf(" (type 0x%x flags 0x%x)\n", type, flags);
+	}
+}
diff --git a/sys/arch/i386/i386/mptramp.s b/sys/arch/i386/i386/mptramp.s
new file mode 100644
index 00000000000..d35b6560c89
--- /dev/null
+++ b/sys/arch/i386/i386/mptramp.s
@@ -0,0 +1,265 @@
+/*	$OpenBSD: mptramp.s,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+	
+/*
+ * Copyright (c) 1999 Stefan Grefen
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed by the NetBSD 
+ *      Foundation, Inc. and its contributors.  
+ * 4. Neither the name of The NetBSD Foundation nor the names of its 
+ *    contributors may be used to endorse or promote products derived  
+ *    from this software without specific prior written permission.   
+ *
+ * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * MP startup ...
+ * the stuff from cpu_spinup_trampoline to mp_startup 
+ * is copied into the first 640 KB
+ *
+ * We startup the processors now when the kthreads become ready.
+ * The steps are:
+ *        1)   Get the processors running kernel-code from a special
+ *                  page-table and stack page, do chip identification.
+ *        2)   halt the processors waiting for them to be enabled
+ *              by a idle-thread 
+ */
+	
+#include "assym.h"
+#include <machine/param.h>
+#include <machine/asm.h>
+#include <machine/specialreg.h>
+#include <machine/segments.h>
+#include <machine/gdt.h>
+#include <machine/mpbiosvar.h>
+#include <machine/i82489reg.h>
+
+#define GDTE(a,b)               .byte   0xff,0xff,0x0,0x0,0x0,a,b,0x0
+#define _RELOC(x)       ((x) - KERNBASE)
+#define RELOC(x)        _RELOC(_C_LABEL(x))
+
+#define _TRMP_LABEL(a)  a = . - _C_LABEL(cpu_spinup_trampoline) + MP_TRAMPOLINE
+
+/*
+ * Debug code to stop aux. processors in various stages based on the
+ * value in cpu_trace.
+ *
+ * %edi points at cpu_trace;  cpu_trace[0] is the "hold point";
+ * cpu_trace[1] is the point which the cpu has reached.
+ * cpu_trace[2] is the last value stored by HALTT.
+ */
+	
+	
+#ifdef MPDEBUG
+#define HALT(x)	1: movl (%edi),%ebx;cmpl $ x,%ebx ; jle 1b ; movl $x,4(%edi)
+#define HALTT(x,y)	movl y,8(%edi); HALT(x)
+#else
+#define HALT(x)	/**/
+#define HALTT(x,y) /**/
+#endif
+
+        .globl  _C_LABEL(cpu),_C_LABEL(cpu_id),_C_LABEL(cpu_vendor)
+	.globl  _C_LABEL(cpuid_level),_C_LABEL(cpu_feature)
+
+	.global _C_LABEL(cpu_spinup_trampoline)
+	.global _C_LABEL(cpu_spinup_trampoline_end)
+	.global _C_LABEL(cpu_hatch)
+	.global _C_LABEL(mp_pdirpa)
+	.global _C_LABEL(gdt), _C_LABEL(local_apic)
+
+	.text
+	.align 4,0x0
+	.code16
+_C_LABEL(cpu_spinup_trampoline):
+	cli
+	xorw    %ax, %ax
+	movw    %ax, %ds
+	movw    %ax, %es
+	movw    %ax, %ss
+	data32 addr32 lgdt    (gdt_desc)      # load flat descriptor table
+	movl    %cr0, %eax      # get cr0
+	orl     $0x1, %eax      # enable protected mode
+	movl    %eax, %cr0      # doit
+	ljmp    $0x8, $mp_startup
+
+_TRMP_LABEL(mp_startup)
+	.code32
+
+	movl    $0x10, %eax     # data segment
+	movw    %ax, %ds
+	movw    %ax, %ss
+	movw    %ax, %es
+	movw    %ax, %fs
+	movw    %ax, %gs
+	movl    $(MP_TRAMPOLINE+NBPG-16),%esp	# bootstrap stack end,
+						# with scratch space..
+
+#ifdef MPDEBUG
+	leal    RELOC(cpu_trace),%edi       
+#endif
+
+	HALT(0x1)
+	/* First, reset the PSL. */
+	pushl	$PSL_MBO
+	popfl
+	
+	movl	RELOC(mp_pdirpa),%ecx
+	HALTT(0x5,%ecx)
+	
+	/* Load base of page directory and enable mapping. */
+	movl	%ecx,%cr3		# load ptd addr into mmu
+        movl    %cr0,%eax               # get control word
+                                        # enable paging & NPX emulation
+        orl     $(CR0_PE|CR0_PG|CR0_NE|CR0_TS|CR0_EM|CR0_MP|CR0_WP),%eax
+        movl    %eax,%cr0               # and let's page NOW!
+
+#ifdef MPDEBUG
+	leal    _C_LABEL(cpu_trace),%edi
+#endif
+	HALT(0x6)
+	
+# ok, we're now running with paging enabled and sharing page tables with cpu0.
+# figure out which processor we really are, what stack we should be on, etc.
+
+	movl	_C_LABEL(local_apic)+LAPIC_ID,%ecx
+	shrl	$LAPIC_ID_SHIFT,%ecx
+	leal	0(,%ecx,4),%ecx
+	movl	_C_LABEL(cpu_info)(%ecx),%ecx
+	
+	HALTT(0x7, %ecx)
+
+# %ecx points at our cpu_info structure..
+
+	movw    $((MAXGDTSIZ*8) - 1), 6(%esp)	# prepare segment descriptor
+	movl    CPU_INFO_GDT(%ecx), %eax	# for real gdt
+	movl    %eax, 8(%esp)
+	HALTT(0x8, %eax)	
+	lgdt	6(%esp)
+	HALT(0x9)	
+	jmp	1f
+	nop
+1:	
+	HALT(0xa)
+	movl    $GSEL(GDATA_SEL, SEL_KPL),%eax 	#switch to new segment
+	HALTT(0x10, %eax)
+	movw    %ax,%ds
+	HALT(0x11)
+	movw    %ax,%es
+	HALT(0x12)
+	movw    %ax,%ss
+	HALT(0x13)
+	pushl   $GSEL(GCODE_SEL, SEL_KPL)
+	pushl	$mp_cont
+	HALT(0x14)
+	lret
+	.align 4,0x0
+_TRMP_LABEL(gdt_table)   
+	.word   0x0,0x0,0x0,0x0  # null GDTE
+	 GDTE(0x9f,0xcf)         # Kernel text
+	 GDTE(0x93,0xcf)         # Kernel data
+_TRMP_LABEL(gdt_desc)	
+	.word   0x17             # limit 3 entries
+	.long   gdt_table              # where is is gdt
+
+_C_LABEL(cpu_spinup_trampoline_end):	#end of code copied to MP_TRAMPOLINE
+mp_cont:
+
+	movl	CPU_INFO_IDLE_PCB(%ecx),%esi
+	
+# %esi now points at our PCB.
+	
+	HALTT(0x19, %esi)
+
+	movl	PCB_ESP(%esi),%esp
+	movl	PCB_EBP(%esi),%ebp
+	
+	HALT(0x20)	
+	/* Switch address space. */
+	movl	PCB_CR3(%esi),%eax
+	HALTT(0x22, %eax)		
+	movl	%eax,%cr3
+	HALT(0x25)
+	/* Load segment registers. */
+	movl	$GSEL(GCPU_SEL, SEL_KPL),%eax
+	HALTT(0x26,%eax)	
+	movl	%eax,%fs		
+	xorl	%eax,%eax
+	HALTT(0x27,%eax)		
+	movl	%eax,%gs
+	movl    PCB_CR0(%esi),%eax
+	HALTT(0x28,%eax)		
+	movl    %eax,%cr0
+	HALTT(0x30,%ecx)	
+	pushl	%ecx
+	call	_C_LABEL(cpu_hatch)
+	HALT(0x33)	
+	xorl	%esi,%esi
+	jmp	_C_LABEL(idle_loop)
+	
+	.data
+_C_LABEL(mp_pdirpa):
+	.long	0
+#ifdef MPDEBUG
+	.global _C_LABEL(cpu_trace)
+_C_LABEL(cpu_trace):
+	.long  0x40
+	.long  0xff
+	.long  0xff		
+#endif
diff --git a/sys/arch/i386/i386/pctr.c b/sys/arch/i386/i386/pctr.c
index 9ed49ebe0bd..c1c63ceb2f5 100644
--- a/sys/arch/i386/i386/pctr.c
+++ b/sys/arch/i386/i386/pctr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pctr.c,v 1.18 2003/09/29 17:33:01 mickey Exp $	*/
+/*	$OpenBSD: pctr.c,v 1.19 2004/06/13 21:49:15 niklas Exp $	*/
 
 /*
  * Pentium performance counter driver for OpenBSD.
@@ -9,6 +9,7 @@
  * OpenBSD project by leaving this copyright notice intact.
  */
 
+#include <sys/param.h>
 #include <sys/types.h>
 #include <sys/errno.h>
 #include <sys/fcntl.h>
diff --git a/sys/arch/i386/i386/pmap.c b/sys/arch/i386/i386/pmap.c
index 8dc8115009e..dd34d9b7625 100644
--- a/sys/arch/i386/i386/pmap.c
+++ b/sys/arch/i386/i386/pmap.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.c,v 1.75 2004/02/01 12:26:45 grange Exp $	*/
+/*	$OpenBSD: pmap.c,v 1.76 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: pmap.c,v 1.91 2000/06/02 17:46:37 thorpej Exp $	*/
 
 /*
@@ -70,6 +70,7 @@
 
 #include <uvm/uvm.h>
 
+#include <machine/atomic.h>
 #include <machine/cpu.h>
 #include <machine/specialreg.h>
 #include <machine/gdt.h>
@@ -128,8 +129,6 @@
  *  - pv_page/pv_page_info: pv_entry's are allocated out of pv_page's.
  *      if we run out of pv_entry's we allocate a new pv_page and free
  *      its pv_entrys.
- * - pmap_remove_record: a list of virtual addresses whose mappings
- *	have been changed.   used for TLB flushing.
  */
 
 /*
@@ -225,35 +224,18 @@
  *   this lock protects the list of active pmaps (headed by "pmaps").
  *   we lock it when adding or removing pmaps from this list.
  *
- * - pmap_copy_page_lock
- *   locks the tmp kernel PTE mappings we used to copy data
- *
- * - pmap_zero_page_lock
- *   locks the tmp kernel PTE mapping we use to zero a page
- *
- * - pmap_tmpptp_lock
- *   locks the tmp kernel PTE mapping we use to look at a PTP
- *   in another process
- *
- * XXX: would be nice to have per-CPU VAs for the above 4
  */
 
 /*
  * locking data structures
  */
 
-#ifdef __OpenBSD__
-/* XXX */
-#define spinlockinit(lock, name, flags) /* nada */
-#define spinlockmgr(lock, flags, slock) /* nada */
-#endif
-
-struct lock pmap_main_lock;
 struct simplelock pvalloc_lock;
 struct simplelock pmaps_lock;
-struct simplelock pmap_copy_page_lock;
-struct simplelock pmap_zero_page_lock;
-struct simplelock pmap_tmpptp_lock;
+
+#if defined(MULTIPROCESSOR) && 0
+
+struct lock pmap_main_lock;
 
 #define PMAP_MAP_TO_HEAD_LOCK() \
      spinlockmgr(&pmap_main_lock, LK_SHARED, (void *) 0)
@@ -265,6 +247,59 @@ struct simplelock pmap_tmpptp_lock;
 #define PMAP_HEAD_TO_MAP_UNLOCK() \
      spinlockmgr(&pmap_main_lock, LK_RELEASE, (void *) 0)
 
+#else
+
+#define PMAP_MAP_TO_HEAD_LOCK()		/* null */
+#define PMAP_MAP_TO_HEAD_UNLOCK()	/* null */
+
+#define PMAP_HEAD_TO_MAP_LOCK()		/* null */
+#define PMAP_HEAD_TO_MAP_UNLOCK()	/* null */
+
+#endif
+
+/*
+ * TLB Shootdown:
+ *
+ * When a mapping is changed in a pmap, the TLB entry corresponding to
+ * the virtual address must be invalidated on all processors.  In order
+ * to accomplish this on systems with multiple processors, messages are
+ * sent from the processor which performs the mapping change to all
+ * processors on which the pmap is active.  For other processors, the
+ * ASN generation numbers for that processor is invalidated, so that
+ * the next time the pmap is activated on that processor, a new ASN
+ * will be allocated (which implicitly invalidates all TLB entries).
+ *
+ * Shootdown job queue entries are allocated using a simple special-
+ * purpose allocator for speed.
+ */
+struct pmap_tlb_shootdown_job {
+	TAILQ_ENTRY(pmap_tlb_shootdown_job) pj_list;
+	vaddr_t pj_va;			/* virtual address */
+	pmap_t pj_pmap;			/* the pmap which maps the address */
+	pt_entry_t pj_pte;		/* the PTE bits */
+	struct pmap_tlb_shootdown_job *pj_nextfree;
+};
+
+struct pmap_tlb_shootdown_q {
+	TAILQ_HEAD(, pmap_tlb_shootdown_job) pq_head;
+	int pq_pte;			/* aggregate PTE bits */
+	int pq_count;			/* number of pending requests */
+	struct SIMPLELOCK pq_slock;	/* spin lock on queue */
+	int pq_flushg;		/* pending flush global */
+	int pq_flushu;		/* pending flush user */
+} pmap_tlb_shootdown_q[I386_MAXPROCS];
+
+#define	PMAP_TLB_MAXJOBS	16
+
+void	pmap_tlb_shootdown_q_drain(struct pmap_tlb_shootdown_q *);
+struct pmap_tlb_shootdown_job *pmap_tlb_shootdown_job_get(
+    struct pmap_tlb_shootdown_q *);
+void	pmap_tlb_shootdown_job_put(struct pmap_tlb_shootdown_q *,
+    struct pmap_tlb_shootdown_job *);
+
+struct SIMPLELOCK pmap_tlb_shootdown_job_lock;
+struct pmap_tlb_shootdown_job *pj_page, *pj_free;
+
 /*
  * global data structures
  */
@@ -348,6 +383,20 @@ static struct pmap *pmaps_hand = NULL;	/* used by pmap_steal_ptp */
 struct pool pmap_pmap_pool;
 
 /*
+ * MULTIPROCESSOR: special VA's/ PTE's are actually allocated inside a
+ * I386_MAXPROCS*NPTECL array of PTE's, to avoid cache line thrashing
+ * due to false sharing.
+ */
+
+#ifdef MULTIPROCESSOR
+#define PTESLEW(pte, id) ((pte)+(id)*NPTECL)
+#define VASLEW(va,id) ((va)+(id)*NPTECL*NBPG)
+#else
+#define PTESLEW(pte, id) (pte)
+#define VASLEW(va,id) (va)
+#endif
+
+/*
  * special VAs and the PTEs that map them
  */
 
@@ -373,41 +422,42 @@ extern vaddr_t pentium_idt_vaddr;
  * local prototypes
  */
 
-static struct pv_entry	*pmap_add_pvpage(struct pv_page *, boolean_t);
-static struct vm_page	*pmap_alloc_ptp(struct pmap *, int, boolean_t);
-static struct pv_entry	*pmap_alloc_pv(struct pmap *, int); /* see codes below */
+struct pv_entry	*pmap_add_pvpage(struct pv_page *, boolean_t);
+struct vm_page	*pmap_alloc_ptp(struct pmap *, int, boolean_t);
+struct pv_entry	*pmap_alloc_pv(struct pmap *, int); /* see codes below */
 #define ALLOCPV_NEED	0	/* need PV now */
 #define ALLOCPV_TRY	1	/* just try to allocate, don't steal */
 #define ALLOCPV_NONEED	2	/* don't need PV, just growing cache */
-static struct pv_entry	*pmap_alloc_pvpage(struct pmap *, int);
-static void		 pmap_enter_pv(struct pv_head *,
+struct pv_entry	*pmap_alloc_pvpage(struct pmap *, int);
+void		 pmap_enter_pv(struct pv_head *,
 					    struct pv_entry *, struct pmap *,
 					    vaddr_t, struct vm_page *);
-static void		 pmap_free_pv(struct pmap *, struct pv_entry *);
-static void		 pmap_free_pvs(struct pmap *, struct pv_entry *);
-static void		 pmap_free_pv_doit(struct pv_entry *);
-static void		 pmap_free_pvpage(void);
-static struct vm_page	*pmap_get_ptp(struct pmap *, int, boolean_t);
-static boolean_t	 pmap_is_curpmap(struct pmap *);
-static pt_entry_t	*pmap_map_ptes(struct pmap *);
-static struct pv_entry	*pmap_remove_pv(struct pv_head *, struct pmap *,
-					     vaddr_t);
-static boolean_t	 pmap_remove_pte(struct pmap *, struct vm_page *,
-					      pt_entry_t *, vaddr_t);
-static void		 pmap_remove_ptes(struct pmap *,
-					       struct pmap_remove_record *,
-					       struct vm_page *, vaddr_t,
-					       vaddr_t, vaddr_t);
-static struct vm_page	*pmap_steal_ptp(struct uvm_object *,
+void		 pmap_free_pv(struct pmap *, struct pv_entry *);
+void		 pmap_free_pvs(struct pmap *, struct pv_entry *);
+void		 pmap_free_pv_doit(struct pv_entry *);
+void		 pmap_free_pvpage(void);
+struct vm_page	*pmap_get_ptp(struct pmap *, int, boolean_t);
+boolean_t	 pmap_is_curpmap(struct pmap *);
+boolean_t	 pmap_is_active(struct pmap *, int);
+pt_entry_t	*pmap_map_ptes(struct pmap *);
+struct pv_entry	*pmap_remove_pv(struct pv_head *, struct pmap *,
 					     vaddr_t);
-static vaddr_t		 pmap_tmpmap_pa(paddr_t);
-static pt_entry_t	*pmap_tmpmap_pvepte(struct pv_entry *);
-static void		 pmap_tmpunmap_pa(void);
-static void		 pmap_tmpunmap_pvepte(struct pv_entry *);
-static boolean_t	 pmap_try_steal_pv(struct pv_head *,
+boolean_t	 pmap_remove_pte(struct pmap *, struct vm_page *, pt_entry_t *,
+    vaddr_t, int32_t *);
+void		 pmap_remove_ptes(struct pmap *, struct vm_page *, vaddr_t,
+    vaddr_t, vaddr_t, int32_t *);
+struct vm_page	*pmap_steal_ptp(struct uvm_object *, vaddr_t);
+vaddr_t		 pmap_tmpmap_pa(paddr_t);
+pt_entry_t	*pmap_tmpmap_pvepte(struct pv_entry *);
+void		 pmap_tmpunmap_pa(void);
+void		 pmap_tmpunmap_pvepte(struct pv_entry *);
+void		 pmap_apte_flush(struct pmap *);
+boolean_t	 pmap_try_steal_pv(struct pv_head *,
 						struct pv_entry *,
 						struct pv_entry *);
-static void		pmap_unmap_ptes(struct pmap *);
+void		pmap_unmap_ptes(struct pmap *);
+void		pmap_exec_account(struct pmap *, vaddr_t, pt_entry_t,
+		    pt_entry_t);
 
 void			pmap_pinit(pmap_t);
 void			pmap_release(pmap_t);
@@ -423,7 +473,7 @@ void			pmap_zero_phys(paddr_t);
  *		of course the kernel is always loaded
  */
 
-__inline static boolean_t
+boolean_t
 pmap_is_curpmap(pmap)
 	struct pmap *pmap;
 {
@@ -432,50 +482,72 @@ pmap_is_curpmap(pmap)
 }
 
 /*
+ * pmap_is_active: is this pmap loaded into the specified processor's %cr3?
+ */
+
+boolean_t
+pmap_is_active(pmap, cpu_id)
+	struct pmap *pmap;
+	int cpu_id;
+{
+
+	return (pmap == pmap_kernel() ||
+	    (pmap->pm_cpus & (1U << cpu_id)) != 0);
+}
+
+/*
  * pmap_tmpmap_pa: map a page in for tmp usage
- *
- * => returns with pmap_tmpptp_lock held
  */
 
-__inline static vaddr_t
+vaddr_t
 pmap_tmpmap_pa(pa)
 	paddr_t pa;
 {
-	simple_lock(&pmap_tmpptp_lock);
+#ifdef MULTIPROCESSOR
+	int id = cpu_number();
+#endif
+	pt_entry_t *ptpte = PTESLEW(ptp_pte, id);
+	caddr_t ptpva = VASLEW(ptpp, id);
 #if defined(DIAGNOSTIC)
-	if (*ptp_pte)
+	if (*ptpte)
 		panic("pmap_tmpmap_pa: ptp_pte in use?");
 #endif
-	*ptp_pte = PG_V | PG_RW | pa;		/* always a new mapping */
-	return((vaddr_t)ptpp);
+	*ptpte = PG_V | PG_RW | pa;		/* always a new mapping */
+	return((vaddr_t)ptpva);
 }
 
 /*
  * pmap_tmpunmap_pa: unmap a tmp use page (undoes pmap_tmpmap_pa)
- *
- * => we release pmap_tmpptp_lock
  */
 
-__inline static void
+void
 pmap_tmpunmap_pa()
 {
+#ifdef MULTIPROCESSOR
+	int id = cpu_number();
+#endif
+	pt_entry_t *ptpte = PTESLEW(ptp_pte, id);
+	caddr_t ptpva = VASLEW(ptpp, id);
 #if defined(DIAGNOSTIC)
-	if (!pmap_valid_entry(*ptp_pte))
+	if (!pmap_valid_entry(*ptpte))
 		panic("pmap_tmpunmap_pa: our pte invalid?");
 #endif
-	*ptp_pte = 0;		/* zap! */
-	pmap_update_pg((vaddr_t)ptpp);
-	simple_unlock(&pmap_tmpptp_lock);
+	*ptpte = 0;		/* zap! */
+	pmap_update_pg((vaddr_t)ptpva);
+#ifdef MULTIPROCESSOR
+	/*
+	 * No need for tlb shootdown here, since ptp_pte is per-CPU.
+	 */
+#endif
 }
 
 /*
  * pmap_tmpmap_pvepte: get a quick mapping of a PTE for a pv_entry
  *
  * => do NOT use this on kernel mappings [why?  because pv_ptp may be NULL]
- * => we may grab pmap_tmpptp_lock and return with it held
  */
 
-__inline static pt_entry_t *
+pt_entry_t *
 pmap_tmpmap_pvepte(pve)
 	struct pv_entry *pve;
 {
@@ -494,11 +566,9 @@ pmap_tmpmap_pvepte(pve)
 
 /*
  * pmap_tmpunmap_pvepte: release a mapping obtained with pmap_tmpmap_pvepte
- *
- * => we will release pmap_tmpptp_lock if we hold it
  */
 
-__inline static void
+void
 pmap_tmpunmap_pvepte(pve)
 	struct pv_entry *pve;
 {
@@ -509,6 +579,41 @@ pmap_tmpunmap_pvepte(pve)
 	pmap_tmpunmap_pa();
 }
 
+void
+pmap_apte_flush(struct pmap *pmap)
+{
+#if defined(MULTIPROCESSOR)
+	struct pmap_tlb_shootdown_q *pq;
+	struct cpu_info *ci, *self = curcpu();
+	CPU_INFO_ITERATOR cii;
+	int s;
+#endif
+
+	tlbflush();		/* flush TLB on current processor */
+#if defined(MULTIPROCESSOR)
+	/*
+	 * Flush the APTE mapping from all other CPUs that
+	 * are using the pmap we are using (who's APTE space
+	 * is the one we've just modified).
+	 *
+	 * XXXthorpej -- find a way to defer the IPI.
+	 */
+	for (CPU_INFO_FOREACH(cii, ci)) {
+		if (ci == self)
+			continue;
+		if (pmap_is_active(pmap, ci->ci_cpuid)) {
+			pq = &pmap_tlb_shootdown_q[ci->ci_cpuid];
+			s = splipi();
+			SIMPLE_LOCK(&pq->pq_slock);
+			pq->pq_flushu++;
+			SIMPLE_UNLOCK(&pq->pq_slock);
+			splx(s);
+			i386_send_ipi(ci, I386_IPI_TLB);
+		}
+	}
+#endif
+}
+
 /*
  * pmap_map_ptes: map a pmap's PTEs into KVM and lock them in
  *
@@ -516,7 +621,7 @@ pmap_tmpunmap_pvepte(pve)
  * => must be undone with pmap_unmap_ptes before returning
  */
 
-__inline static pt_entry_t *
+pt_entry_t *
 pmap_map_ptes(pmap)
 	struct pmap *pmap;
 {
@@ -547,7 +652,7 @@ pmap_map_ptes(pmap)
 	if (!pmap_valid_entry(opde) || (opde & PG_FRAME) != pmap->pm_pdirpa) {
 		*APDP_PDE = (pd_entry_t) (pmap->pm_pdirpa | PG_RW | PG_V);
 		if (pmap_valid_entry(opde))
-			tlbflush();
+			pmap_apte_flush(pmap);
 	}
 	return(APTE_BASE);
 }
@@ -556,7 +661,7 @@ pmap_map_ptes(pmap)
  * pmap_unmap_ptes: unlock the PTE mapping of "pmap"
  */
 
-__inline static void
+void
 pmap_unmap_ptes(pmap)
 	struct pmap *pmap;
 {
@@ -566,12 +671,16 @@ pmap_unmap_ptes(pmap)
 	if (pmap_is_curpmap(pmap)) {
 		simple_unlock(&pmap->pm_obj.vmobjlock);
 	} else {
+#if defined(MULTIPROCESSOR)
+		*APDP_PDE = 0;
+		pmap_apte_flush(curpcb->pcb_pmap);
+#endif
 		simple_unlock(&pmap->pm_obj.vmobjlock);
 		simple_unlock(&curpcb->pcb_pmap->pm_obj.vmobjlock);
 	}
 }
 
-__inline static void
+void
 pmap_exec_account(struct pmap *pm, vaddr_t va,
     pt_entry_t opte, pt_entry_t npte)
 {
@@ -579,9 +688,18 @@ pmap_exec_account(struct pmap *pm, vaddr_t va,
 	    pm != vm_map_pmap(&curproc->p_vmspace->vm_map))
 		return;
 
-	if ((opte ^ npte) & PG_X)
+	if ((opte ^ npte) & PG_X) {
+#ifdef MULTIPROCESSOR
+		int32_t cpumask = 0;
+
+		pmap_tlb_shootdown(pm, va, opte, &cpumask);
+		pmap_tlb_shootnow(cpumask);
+#else
+		/* Don't bother deferring in the single CPU case. */
 		pmap_update_pg(va);
-		
+#endif
+	}
+			
 	/*
 	 * Executability was removed on the last executable change.
 	 * Reset the code segment to something conservative and
@@ -656,14 +774,22 @@ pmap_kenter_pa(va, pa, prot)
 	paddr_t pa;
 	vm_prot_t prot;
 {
-	pt_entry_t *pte, opte;
+	pt_entry_t *pte, opte, npte;
 
 	pte = vtopte(va);
-	opte = *pte;
-	*pte = pa | ((prot & VM_PROT_WRITE)? PG_RW : PG_RO) |
-		PG_V | pmap_pg_g;	/* zap! */
-	if (pmap_valid_entry(opte))
+	npte = pa | ((prot & VM_PROT_WRITE)? PG_RW : PG_RO) | PG_V | pmap_pg_g;
+	opte = i386_atomic_testset_ul(pte, npte); /* zap! */
+	if (pmap_valid_entry(opte)) {
+#ifdef MULTIPROCESSOR
+		int32_t cpumask = 0;
+
+		pmap_tlb_shootdown(pmap_kernel(), va, opte, &cpumask);
+		pmap_tlb_shootnow(cpumask);
+#else
+		/* Don't bother deferring in the single CPU case. */
 		pmap_update_pg(va);
+#endif
+	}
 }
 
 /*
@@ -673,8 +799,6 @@ pmap_kenter_pa(va, pa, prot)
  * => caller must dispose of any vm_page mapped in the va range
  * => note: not an inline function
  * => we assume the va is page aligned and the len is a multiple of PAGE_SIZE
- * => we assume kernel only unmaps valid addresses and thus don't bother
- *    checking the valid bit before doing TLB flushing
  */
 
 void
@@ -682,25 +806,31 @@ pmap_kremove(va, len)
 	vaddr_t va;
 	vsize_t len;
 {
-	pt_entry_t *pte;
+	pt_entry_t *pte, opte;
+#ifdef MULTIPROCESSOR
+	int32_t cpumask = 0;
+#endif
 
 	len >>= PAGE_SHIFT;
-	for ( /* null */ ; len ; len--, va += NBPG) {
-		pte = vtopte(va);
+	for ( /* null */ ; len ; len--, va += PAGE_SIZE) {
+		if (va < VM_MIN_KERNEL_ADDRESS)
+			pte = vtopte(va);
+		else
+			pte = kvtopte(va);
+		opte = i386_atomic_testset_ul(pte, 0); /* zap! */
 #ifdef DIAGNOSTIC
-		if (*pte & PG_PVLIST)
-			panic("pmap_kremove: PG_PVLIST mapping for 0x%lx",
-			      va);
-#endif
-		*pte = 0;		/* zap! */
-#if defined(I386_CPU)
-		if (cpu_class != CPUCLASS_386)
+		if (opte & PG_PVLIST)
+			panic("pmap_kremove: PG_PVLIST mapping for 0x%lx", va);
 #endif
+		if ((opte & (PG_V | PG_U)) == (PG_V | PG_U))
+#ifdef MULTIPROCESSOR
+			pmap_tlb_shootdown(pmap_kernel(), va, opte, &cpumask);
+#else
 			pmap_update_pg(va);
+#endif
 	}
-#if defined(I386_CPU)
-	if (cpu_class == CPUCLASS_386)
-		tlbflush();
+#ifdef MULTIPROCESSOR
+	pmap_tlb_shootnow(cpumask);
 #endif
 }
 
@@ -729,6 +859,7 @@ pmap_bootstrap(kva_start)
 	struct pmap *kpm;
 	vaddr_t kva;
 	pt_entry_t *pte;
+	int i;
 
 	/*
 	 * set the page size (default value is 4K which is ok)
@@ -831,6 +962,26 @@ pmap_bootstrap(kva_start)
 
 	pte = PTE_BASE + i386_btop(virtual_avail);
 
+#ifdef MULTIPROCESSOR
+	/*
+	 * Waste some VA space to avoid false sharing of cache lines
+	 * for page table pages: Give each possible CPU a cache line
+	 * of PTE's (8) to play with, though we only need 4.  We could
+	 * recycle some of this waste by putting the idle stacks here
+	 * as well; we could waste less space if we knew the largest
+	 * CPU ID beforehand.
+	 */
+	csrcp = (caddr_t) virtual_avail;  csrc_pte = pte;
+
+	cdstp = (caddr_t) virtual_avail+PAGE_SIZE;  cdst_pte = pte+1;
+
+	zerop = (caddr_t) virtual_avail+PAGE_SIZE*2;  zero_pte = pte+2;
+
+	ptpp = (caddr_t) virtual_avail+PAGE_SIZE*3;  ptp_pte = pte+3;
+
+	virtual_avail += PAGE_SIZE * I386_MAXPROCS * NPTECL;
+	pte += I386_MAXPROCS * NPTECL;
+#else
 	csrcp = (caddr_t) virtual_avail;  csrc_pte = pte;	/* allocate */
 	virtual_avail += PAGE_SIZE; pte++;			/* advance */
 
@@ -842,10 +993,11 @@ pmap_bootstrap(kva_start)
 
 	ptpp = (caddr_t) virtual_avail;  ptp_pte = pte;
 	virtual_avail += PAGE_SIZE; pte++;
+#endif
 
 	/* XXX: vmmap used by mem.c... should be uvm_map_reserve */
 	vmmap = (char *)virtual_avail;			/* don't need pte */
-	virtual_avail += PAGE_SIZE; pte++;
+	virtual_avail += PAGE_SIZE;
 
 #ifdef __NetBSD
 	msgbuf_vaddr = virtual_avail;			/* don't need pte */
@@ -883,12 +1035,11 @@ pmap_bootstrap(kva_start)
 	 * init the static-global locks and global lists.
 	 */
 
+#if defined(MULTIPROCESSOR) && 0
 	spinlockinit(&pmap_main_lock, "pmaplk", 0);
+#endif
 	simple_lock_init(&pvalloc_lock);
 	simple_lock_init(&pmaps_lock);
-	simple_lock_init(&pmap_copy_page_lock);
-	simple_lock_init(&pmap_zero_page_lock);
-	simple_lock_init(&pmap_tmpptp_lock);
 	LIST_INIT(&pmaps);
 	TAILQ_INIT(&pv_freepages);
 	TAILQ_INIT(&pv_unusedpgs);
@@ -900,6 +1051,17 @@ pmap_bootstrap(kva_start)
 	pool_init(&pmap_pmap_pool, sizeof(struct pmap), 0, 0, 0, "pmappl",
 	    &pool_allocator_nointr);
 
+	/*
+	 * Initialize the TLB shootdown queues.
+	 */
+
+	SIMPLE_LOCK_INIT(&pmap_tlb_shootdown_job_lock);
+
+	for (i = 0; i < I386_MAXPROCS; i++) {
+		TAILQ_INIT(&pmap_tlb_shootdown_q[i].pq_head);
+		SIMPLE_LOCK_INIT(&pmap_tlb_shootdown_q[i].pq_slock);
+	}
+
 #ifdef __NetBSD__
 	/*
 	 * we must call uvm_page_physload() after we are done playing with
@@ -1009,6 +1171,15 @@ pmap_init()
 	pv_nfpvents = 0;
 	(void) pmap_add_pvpage(pv_initpage, FALSE);
 
+	pj_page = (void *)uvm_km_alloc(kernel_map, PAGE_SIZE);
+	if (pj_page == NULL)
+		panic("pmap_init: pj_page");
+
+	for (i = 0; i < PAGE_SIZE / sizeof *pj_page - 1; i++)
+		pj_page[i].pj_nextfree = &pj_page[i + 1];
+	pj_page[i].pj_nextfree = NULL;
+	pj_free = &pj_page[0];
+
 	/*
 	 * done: pmap module is up (and ready for business)
 	 */
@@ -1043,7 +1214,7 @@ pmap_init()
  * "try" is for optional functions like pmap_copy().
  */
 
-__inline static struct pv_entry *
+struct pv_entry *
 pmap_alloc_pv(pmap, mode)
 	struct pmap *pmap;
 	int mode;
@@ -1100,7 +1271,7 @@ pmap_alloc_pv(pmap, mode)
  * => we assume that the caller holds pvalloc_lock
  */
 
-static struct pv_entry *
+struct pv_entry *
 pmap_alloc_pvpage(pmap, mode)
 	struct pmap *pmap;
 	int mode;
@@ -1233,12 +1404,15 @@ steal_one:
  * => return true if we did it!
  */
 
-static boolean_t
+boolean_t
 pmap_try_steal_pv(pvh, cpv, prevpv)
 	struct pv_head *pvh;
 	struct pv_entry *cpv, *prevpv;
 {
-	pt_entry_t *ptep;	/* pointer to a PTE */
+	pt_entry_t *ptep, opte;
+#ifdef MULTIPROCESSOR
+	int32_t cpumask = 0;
+#endif
 
 	/*
 	 * we never steal kernel mappings or mappings from pmaps we can't lock
@@ -1257,9 +1431,15 @@ pmap_try_steal_pv(pvh, cpv, prevpv)
 	if (*ptep & PG_W) {
 		ptep = NULL;	/* wired page, avoid stealing this one */
 	} else {
-		*ptep = 0;		/* zap! */
+		opte = i386_atomic_testset_ul(ptep, 0); /* zap! */
+#ifdef MULTIPROCESSOR
+		pmap_tlb_shootdown(cpv->pv_pmap, cpv->pv_va, opte, &cpumask);
+		pmap_tlb_shootnow(cpumask);
+#else
+		/* Don't bother deferring in the single CPU case. */
 		if (pmap_is_curpmap(cpv->pv_pmap))
 			pmap_update_pg(cpv->pv_va);
+#endif
 		pmap_tmpunmap_pvepte(cpv);
 	}
 	if (ptep == NULL) {
@@ -1296,7 +1476,7 @@ pmap_try_steal_pv(pvh, cpv, prevpv)
  * => if need_entry is true, we allocate and return one pv_entry
  */
 
-static struct pv_entry *
+struct pv_entry *
 pmap_add_pvpage(pvp, need_entry)
 	struct pv_page *pvp;
 	boolean_t need_entry;
@@ -1329,7 +1509,7 @@ pmap_add_pvpage(pvp, need_entry)
  * => we must be holding pvalloc_lock
  */
 
-__inline static void
+void
 pmap_free_pv_doit(pv)
 	struct pv_entry *pv;
 {
@@ -1364,7 +1544,7 @@ pmap_free_pv_doit(pv)
  * => we gain the pvalloc_lock
  */
 
-__inline static void
+void
 pmap_free_pv(pmap, pv)
 	struct pmap *pmap;
 	struct pv_entry *pv;
@@ -1389,7 +1569,7 @@ pmap_free_pv(pmap, pv)
  * => we gain the pvalloc_lock
  */
 
-__inline static void
+void
 pmap_free_pvs(pmap, pvs)
 	struct pmap *pmap;
 	struct pv_entry *pvs;
@@ -1426,7 +1606,7 @@ pmap_free_pvs(pmap, pvs)
  *	holding kmem_object's lock.
  */
 
-static void
+void
 pmap_free_pvpage()
 {
 	int s;
@@ -1488,7 +1668,7 @@ pmap_free_pvpage()
  * => caller should adjust ptp's wire_count before calling
  */
 
-__inline static void
+void
 pmap_enter_pv(pvh, pve, pmap, va, ptp)
 	struct pv_head *pvh;
 	struct pv_entry *pve;	/* preallocated pve for us to use */
@@ -1515,7 +1695,7 @@ pmap_enter_pv(pvh, pve, pmap, va, ptp)
  * => we return the removed pve
  */
 
-__inline static struct pv_entry *
+struct pv_entry *
 pmap_remove_pv(pvh, pmap, va)
 	struct pv_head *pvh;
 	struct pmap *pmap;
@@ -1554,7 +1734,7 @@ pmap_remove_pv(pvh, pmap, va)
  * 	from another pmap (e.g. during optional functions like pmap_copy)
  */
 
-__inline static struct vm_page *
+struct vm_page *
 pmap_alloc_ptp(pmap, pde_index, just_try)
 	struct pmap *pmap;
 	int pde_index;
@@ -1596,7 +1776,7 @@ pmap_alloc_ptp(pmap, pde_index, just_try)
  *	enough free pages around for us to allocate off the free page list]
  */
 
-static struct vm_page *
+struct vm_page *
 pmap_steal_ptp(obj, offset)
 	struct uvm_object *obj;
 	vaddr_t offset;
@@ -1607,6 +1787,7 @@ pmap_steal_ptp(obj, offset)
 	pt_entry_t *ptes;
 	int idx, lcv;
 	boolean_t caller_locked, we_locked;
+	int32_t cpumask = 0;
 
 	simple_lock(&pmaps_lock);
 	if (pmaps_hand == NULL)
@@ -1644,10 +1825,9 @@ pmap_steal_ptp(obj, offset)
 					    (PG_V|PG_W))
 						break;
 				if (lcv == PTES_PER_PTP)
-					pmap_remove_ptes(pmaps_hand, NULL, ptp,
-							 (vaddr_t)ptes,
-							 ptp_i2v(idx),
-							 ptp_i2v(idx+1));
+					pmap_remove_ptes(pmaps_hand, ptp,
+					    (vaddr_t)ptes, ptp_i2v(idx),
+					    ptp_i2v(idx+1), &cpumask);
 				pmap_tmpunmap_pa();
 
 				if (lcv != PTES_PER_PTP)
@@ -1660,14 +1840,17 @@ pmap_steal_ptp(obj, offset)
 
 				pmaps_hand->pm_pdir[idx] = 0;	/* zap! */
 				pmaps_hand->pm_stats.resident_count--;
+#ifdef MULTIPROCESSOR
+				pmap_apte_flush(pmaps_hand);
+#else
 				if (pmap_is_curpmap(pmaps_hand))
-					tlbflush();
+					pmap_apte_flush(pmaps_hand);
 				else if (pmap_valid_entry(*APDP_PDE) &&
-					 (*APDP_PDE & PG_FRAME) ==
-					 pmaps_hand->pm_pdirpa) {
+				    (*APDP_PDE & PG_FRAME) ==
+				    pmaps_hand->pm_pdirpa)
 					pmap_update_pg(((vaddr_t)APTE_BASE) +
 						       ptp->offset);
-				}
+#endif
 
 				/* put it in our pmap! */
 				uvm_pagerealloc(ptp, obj, offset);
@@ -1687,6 +1870,7 @@ pmap_steal_ptp(obj, offset)
 	} while (ptp == NULL && pmaps_hand != firstpmap);
 
 	simple_unlock(&pmaps_lock);
+	pmap_tlb_shootnow(cpumask);
 	return(ptp);
 }
 
@@ -1697,7 +1881,7 @@ pmap_steal_ptp(obj, offset)
  * => pmap should be locked
  */
 
-static struct vm_page *
+struct vm_page *
 pmap_get_ptp(pmap, pde_index, just_try)
 	struct pmap *pmap;
 	int pde_index;
@@ -1873,7 +2057,10 @@ pmap_release(pmap)
 		uvm_pagefree(pg);
 	}
 
-	/* XXX: need to flush it out of other processor's APTE space? */
+	/*
+	 * MULTIPROCESSOR -- no need to flush out of other processors'
+	 * APTE space because we do that in pmap_unmap_ptes().
+	 */
 	uvm_km_free(kernel_map, (vaddr_t)pmap->pm_pdir, NBPG);
 
 #ifdef USER_LDT
@@ -1993,22 +2180,31 @@ pmap_activate(p)
 	pcb->pcb_pmap = pmap;
 	pcb->pcb_ldt_sel = pmap->pm_ldt_sel;
 	pcb->pcb_cr3 = pmap->pm_pdirpa;
-	if (p == curproc)
+	if (p == curproc) {
 		lcr3(pcb->pcb_cr3);
-	if (pcb == curpcb)
 		lldt(pcb->pcb_ldt_sel);
+
+		/*
+		 * mark the pmap in use by this processor.
+		 */
+		i386_atomic_setbits_l(&pmap->pm_cpus, (1U << cpu_number()));
+	}
 }
 
 /*
  * pmap_deactivate: deactivate a process' pmap
- *
- * => XXX: what should this do, if anything?
  */
 
 void
 pmap_deactivate(p)
 	struct proc *p;
 {
+	struct pmap *pmap = p->p_vmspace->vm_map.pmap;
+
+	/*
+	 * mark the pmap no longer in use by this processor.
+	 */
+	i386_atomic_clearbits_l(&pmap->pm_cpus, (1U << cpu_number()));
 }
 
 /*
@@ -2065,19 +2261,7 @@ void (*pagezero)(void *, size_t) = bzero;
 void
 pmap_zero_page(struct vm_page *pg)
 {
-	paddr_t pa = VM_PAGE_TO_PHYS(pg);
-
-	simple_lock(&pmap_zero_page_lock);
-#ifdef DIAGNOSTIC
-	if (*zero_pte)
-		panic("pmap_zero_page: lock botch");
-#endif
-
-	*zero_pte = (pa & PG_FRAME) | PG_V | PG_RW;	/* map in */
-	pagezero(zerop, PAGE_SIZE);				/* zero */
-	*zero_pte = 0;				/* zap! */
-	pmap_update_pg((vaddr_t)zerop);		/* flush TLB */
-	simple_unlock(&pmap_zero_page_lock);
+	pmap_zero_phys(VM_PAGE_TO_PHYS(pg));
 }
 
 /*
@@ -2087,17 +2271,21 @@ pmap_zero_page(struct vm_page *pg)
 void
 pmap_zero_phys(paddr_t pa)
 {
-	simple_lock(&pmap_zero_page_lock);
+#ifdef MULTIPROCESSOR
+	int id = cpu_number();
+#endif
+	pt_entry_t *zpte = PTESLEW(zero_pte, id);
+	caddr_t zerova = VASLEW(zerop, id);
+
 #ifdef DIAGNOSTIC
-	if (*zero_pte)
-		panic("pmap_zero_page: lock botch");
+	if (*zpte)
+		panic("pmap_zero_phys: lock botch");
 #endif
 
-	*zero_pte = (pa & PG_FRAME) | PG_V | PG_RW;	/* map in */
-	pagezero(zerop, PAGE_SIZE);				/* zero */
-	*zero_pte = 0;				/* zap! */
-	pmap_update_pg((vaddr_t)zerop);		/* flush TLB */
-	simple_unlock(&pmap_zero_page_lock);
+	*zpte = (pa & PG_FRAME) | PG_V | PG_RW;	/* map in */
+	pmap_update_pg((vaddr_t)zerova);	/* flush TLB */
+	pagezero(zerova, PAGE_SIZE);		/* zero */
+	*zpte = 0;				/* zap! */
 }
 
 /*
@@ -2108,18 +2296,22 @@ boolean_t
 pmap_zero_page_uncached(pa)
 	paddr_t pa;
 {
-	simple_lock(&pmap_zero_page_lock);
+#ifdef MULTIPROCESSOR
+	int id = cpu_number();
+#endif
+	pt_entry_t *zpte = PTESLEW(zero_pte, id);
+	caddr_t zerova = VASLEW(zerop, id);
+
 #ifdef DIAGNOSTIC
-	if (*zero_pte)
+	if (*zpte)
 		panic("pmap_zero_page_uncached: lock botch");
 #endif
 
-	*zero_pte = (pa & PG_FRAME) | PG_V | PG_RW |	/* map in */
+	*zpte = (pa & PG_FRAME) | PG_V | PG_RW |	/* map in */
 	    ((cpu_class != CPUCLASS_386) ? PG_N : 0);
-	pagezero(zerop, PAGE_SIZE);				/* zero */
-	*zero_pte = 0;					/* zap! */
-	pmap_update_pg((vaddr_t)zerop);			/* flush TLB */
-	simple_unlock(&pmap_zero_page_lock);
+	pmap_update_pg((vaddr_t)zerova);		/* flush TLB */
+	pagezero(zerova, PAGE_SIZE);				/* zero */
+	*zpte = 0;					/* zap! */
 
 	return (TRUE);
 }
@@ -2133,19 +2325,28 @@ pmap_copy_page(struct vm_page *srcpg, struct vm_page *dstpg)
 {
 	paddr_t srcpa = VM_PAGE_TO_PHYS(srcpg);
 	paddr_t dstpa = VM_PAGE_TO_PHYS(dstpg);
+#ifdef MULTIPROCESSOR
+	int id = cpu_number();
+#endif
+	pt_entry_t *spte = PTESLEW(csrc_pte,id);
+	pt_entry_t *dpte = PTESLEW(cdst_pte,id);
+	caddr_t csrcva = VASLEW(csrcp, id);
+	caddr_t cdstva = VASLEW(cdstp, id);
 
-	simple_lock(&pmap_copy_page_lock);
 #ifdef DIAGNOSTIC
-	if (*csrc_pte || *cdst_pte)
+	if (*spte || *dpte)
 		panic("pmap_copy_page: lock botch");
 #endif
 
-	*csrc_pte = (srcpa & PG_FRAME) | PG_V | PG_RW;
-	*cdst_pte = (dstpa & PG_FRAME) | PG_V | PG_RW;
-	bcopy(csrcp, cdstp, PAGE_SIZE);
-	*csrc_pte = *cdst_pte = 0;			/* zap! */
-	pmap_update_2pg((vaddr_t)csrcp, (vaddr_t)cdstp);
-	simple_unlock(&pmap_copy_page_lock);
+	*spte = (srcpa & PG_FRAME) | PG_V | PG_RW;
+	*dpte = (dstpa & PG_FRAME) | PG_V | PG_RW;
+	pmap_update_2pg((vaddr_t)csrcva, (vaddr_t)cdstva);
+	bcopy(csrcva, cdstva, PAGE_SIZE);
+	*spte = *dpte = 0;			/* zap! */
+	pmap_update_2pg((vaddr_t)csrcva, (vaddr_t)cdstva);
+#ifdef MULTIPROCESSOR
+	/* Using per-cpu VA; no shootdown required here. */
+#endif
 }
 
 /*
@@ -2163,13 +2364,13 @@ pmap_copy_page(struct vm_page *srcpg, struct vm_page *dstpg)
  * => PTP should be null if pmap == pmap_kernel()
  */
 
-static void
-pmap_remove_ptes(pmap, pmap_rr, ptp, ptpva, startva, endva)
+void
+pmap_remove_ptes(pmap, ptp, ptpva, startva, endva, cpumaskp)
 	struct pmap *pmap;
-	struct pmap_remove_record *pmap_rr;
 	struct vm_page *ptp;
 	vaddr_t ptpva;
 	vaddr_t startva, endva;
+	int32_t *cpumaskp;
 {
 	struct pv_entry *pv_tofree = NULL;	/* list of pv_entrys to free */
 	struct pv_entry *pve;
@@ -2191,29 +2392,23 @@ pmap_remove_ptes(pmap, pmap_rr, ptp, ptpva, startva, endva)
 		if (!pmap_valid_entry(*pte))
 			continue;			/* VA not mapped */
 
-		opte = *pte;		/* save the old PTE */
-		*pte = 0;			/* zap! */
+		/* atomically save the old PTE and zap! it */
+		opte = i386_atomic_testset_ul(pte, 0);
+
 		if (opte & PG_W)
 			pmap->pm_stats.wired_count--;
 		pmap->pm_stats.resident_count--;
 
-		if (pmap_rr) {		/* worried about tlb flushing? */
-			if (opte & PG_G) {
-				/* PG_G requires this */
-				pmap_update_pg(startva);
-			} else {
-				if (pmap_rr->prr_npages < PMAP_RR_MAX) {
-					pmap_rr->prr_vas[pmap_rr->prr_npages++]
-						= startva;
-				} else {
-					if (pmap_rr->prr_npages == PMAP_RR_MAX)
-						/* signal an overflow */
-						pmap_rr->prr_npages++;
-				}
-			}
-		}
-		if (ptp)
+		if (opte & PG_U)
+			pmap_tlb_shootdown(pmap, startva, opte, cpumaskp);
+
+		if (ptp) {
 			ptp->wire_count--;		/* dropping a PTE */
+			/* Make sure that the PDE is flushed */
+			if ((ptp->wire_count <= 1) && !(opte & PG_U))
+				pmap_tlb_shootdown(pmap, startva, opte,
+				    cpumaskp);
+		}
 
 		/*
 		 * if we are not on a pv_head list we are done.
@@ -2266,12 +2461,13 @@ pmap_remove_ptes(pmap, pmap_rr, ptp, ptpva, startva, endva)
  * => returns true if we removed a mapping
  */
 
-static boolean_t
-pmap_remove_pte(pmap, ptp, pte, va)
+boolean_t
+pmap_remove_pte(pmap, ptp, pte, va, cpumaskp)
 	struct pmap *pmap;
 	struct vm_page *ptp;
 	pt_entry_t *pte;
 	vaddr_t va;
+	int32_t *cpumaskp;
 {
 	pt_entry_t opte;
 	int bank, off;
@@ -2289,11 +2485,16 @@ pmap_remove_pte(pmap, ptp, pte, va)
 		pmap->pm_stats.wired_count--;
 	pmap->pm_stats.resident_count--;
 
-	if (ptp)
+	if (opte & PG_U)
+		pmap_tlb_shootdown(pmap, va, opte, cpumaskp);
+
+	if (ptp) {
 		ptp->wire_count--;		/* dropping a PTE */
+		/* Make sure that the PDE is flushed */
+		if ((ptp->wire_count <= 1) && !(opte & PG_U))
+			pmap_tlb_shootdown(pmap, va, opte, cpumaskp);
 
-	if (pmap_is_curpmap(pmap))
-		pmap_update_pg(va);		/* flush TLB */
+	}
 
 	/*
 	 * if we are not on a pv_head list we are done.
@@ -2338,18 +2539,18 @@ pmap_remove(pmap, sva, eva)
 	struct pmap *pmap;
 	vaddr_t sva, eva;
 {
-	pt_entry_t *ptes;
+	pt_entry_t *ptes, opte;
 	boolean_t result;
 	paddr_t ptppa;
 	vaddr_t blkendva;
 	struct vm_page *ptp;
-	struct pmap_remove_record pmap_rr, *prr;
+	int32_t cpumask = 0;
 
 	/*
 	 * we lock in the pmap => pv_head direction
 	 */
 
-	PMAP_MAP_TO_HEAD_LOCK();
+ 	PMAP_MAP_TO_HEAD_LOCK();
 	ptes = pmap_map_ptes(pmap);	/* locks pmap */
 
 	/*
@@ -2385,7 +2586,7 @@ pmap_remove(pmap, sva, eva)
 
 			/* do it! */
 			result = pmap_remove_pte(pmap, ptp,
-						 &ptes[i386_btop(sva)], sva);
+			    &ptes[i386_btop(sva)], sva, &cpumask);
 
 			/*
 			 * if mapping removed and the PTP is no longer
@@ -2393,15 +2594,30 @@ pmap_remove(pmap, sva, eva)
 			 */
 
 			if (result && ptp && ptp->wire_count <= 1) {
-				pmap->pm_pdir[pdei(sva)] = 0;	/* zap! */
-#if defined(I386_CPU)
-				/* already dumped whole TLB on i386 */
-				if (cpu_class != CPUCLASS_386)
+				/* zap! */
+				opte = i386_atomic_testset_ul(
+				    &pmap->pm_pdir[pdei(sva)], 0);
+#ifdef MULTIPROCESSOR
+				/*
+				 * XXXthorpej Redundant shootdown can happen
+				 * here if we're using APTE space.
+				 */
+#endif
+				pmap_tlb_shootdown(curpcb->pcb_pmap,
+				    ((vaddr_t)ptes) + ptp->offset, opte,
+				    &cpumask);
+#ifdef MULTIPROCESSOR
+				/*
+				 * Always shoot down the pmap's self-mapping
+				 * of the PTP.
+				 * XXXthorpej Redundant shootdown can happen
+				 * here if pmap == curpcb->pcb_pmap (not APTE
+				 * space).
+				 */
+				pmap_tlb_shootdown(pmap,
+				    ((vaddr_t)PTE_BASE) + ptp->offset, opte,
+				    &cpumask);
 #endif
-				{
-					pmap_update_pg(((vaddr_t) ptes) +
-						       ptp->offset);
-				}
 				pmap->pm_stats.resident_count--;
 				if (pmap->pm_ptphint == ptp)
 					pmap->pm_ptphint =
@@ -2410,26 +2626,12 @@ pmap_remove(pmap, sva, eva)
 				uvm_pagefree(ptp);
 			}
 		}
-
+		pmap_tlb_shootnow(cpumask);
 		pmap_unmap_ptes(pmap);		/* unlock pmap */
 		PMAP_MAP_TO_HEAD_UNLOCK();
 		return;
 	}
 
-	/*
-	 * removing a range of pages: we unmap in PTP sized blocks (4MB)
-	 *
-	 * if we are the currently loaded pmap, we use prr to keep track
-	 * of the VAs we unload so that we can flush them out of the tlb.
-	 */
-
-	if (pmap_is_curpmap(pmap)) {
-		prr = &pmap_rr;
-		prr->prr_npages = 0;
-	} else {
-		prr = NULL;
-	}
-
 	for (/* null */ ; sva < eva ; sva = blkendva) {
 
 		/* determine range of block */
@@ -2479,17 +2681,31 @@ pmap_remove(pmap, sva, eva)
 #endif
 			}
 		}
-		pmap_remove_ptes(pmap, prr, ptp,
-				 (vaddr_t)&ptes[i386_btop(sva)], sva, blkendva);
+		pmap_remove_ptes(pmap, ptp, (vaddr_t)&ptes[i386_btop(sva)],
+		    sva, blkendva, &cpumask);
 
 		/* if PTP is no longer being used, free it! */
 		if (ptp && ptp->wire_count <= 1) {
-			pmap->pm_pdir[pdei(sva)] = 0;	/* zap! */
-			pmap_update_pg( ((vaddr_t) ptes) + ptp->offset);
-#if defined(I386_CPU)
-			/* cancel possible pending pmap update on i386 */
-			if (cpu_class == CPUCLASS_386 && prr)
-				prr->prr_npages = 0;
+			/* zap! */
+			opte = i386_atomic_testset_ul(
+			    &pmap->pm_pdir[pdei(sva)], 0);
+#if defined(MULTIPROCESSOR)
+			/*
+			 * XXXthorpej Redundant shootdown can happen here
+			 * if we're using APTE space.
+			 */
+#endif
+			pmap_tlb_shootdown(curpcb->pcb_pmap,
+			    ((vaddr_t)ptes) + ptp->offset, opte, &cpumask);
+#if defined(MULTIPROCESSOR)
+			/*
+			 * Always shoot down the pmap's self-mapping
+			 * of the PTP.
+			 * XXXthorpej Redundant shootdown can happen here
+			 * if pmap == curpcb->pcb_pmap (not APTE space).
+			 */
+			pmap_tlb_shootdown(pmap,
+			    ((vaddr_t)PTE_BASE) + ptp->offset, opte, &cpumask);
 #endif
 			pmap->pm_stats.resident_count--;
 			if (pmap->pm_ptphint == ptp)	/* update hint? */
@@ -2500,27 +2716,7 @@ pmap_remove(pmap, sva, eva)
 		}
 	}
 
-	/*
-	 * if we kept a removal record and removed some pages update the TLB
-	 */
-
-	if (prr && prr->prr_npages) {
-#if defined(I386_CPU)
-		if (cpu_class == CPUCLASS_386) {
-			tlbflush();
-		} else
-#endif
-		{ /* not I386 */
-			if (prr->prr_npages > PMAP_RR_MAX) {
-				tlbflush();
-			} else {
-				while (prr->prr_npages) {
-					pmap_update_pg(
-					    prr->prr_vas[--prr->prr_npages]);
-				}
-			}
-		} /* not I386 */
-	}
+	pmap_tlb_shootnow(cpumask);
 	pmap_unmap_ptes(pmap);
 	PMAP_MAP_TO_HEAD_UNLOCK();
 }
@@ -2540,9 +2736,7 @@ pmap_page_remove(pg)
 	struct pv_head *pvh;
 	struct pv_entry *pve;
 	pt_entry_t *ptes, opte;
-#if defined(I386_CPU)
-	boolean_t needs_update = FALSE;
-#endif
+	int32_t cpumask = 0;
 
 	/* XXX: vm_page should either contain pv_head or have a pointer to it */
 	bank = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), &off);
@@ -2590,14 +2784,10 @@ pmap_page_remove(pg)
 			pve->pv_pmap->pm_stats.wired_count--;
 		pve->pv_pmap->pm_stats.resident_count--;
 
-		if (pmap_is_curpmap(pve->pv_pmap)) {
-#if defined(I386_CPU)
-			if (cpu_class == CPUCLASS_386)
-				needs_update = TRUE;
-			else
-#endif
-				pmap_update_pg(pve->pv_va);
-		}
+		/* Shootdown only if referenced */
+		if (opte & PG_U)
+			pmap_tlb_shootdown(pve->pv_pmap, pve->pv_va, opte,
+			    &cpumask);
 
 		/* sync R/M bits */
 		vm_physmem[bank].pmseg.attrs[off] |= (opte & (PG_U|PG_M));
@@ -2606,12 +2796,29 @@ pmap_page_remove(pg)
 		if (pve->pv_ptp) {
 			pve->pv_ptp->wire_count--;
 			if (pve->pv_ptp->wire_count <= 1) {
+				/*
+				 * Do we have to shootdown the page just to
+				 * get the pte out of the TLB ?
+				 */
+				if(!(opte & PG_U))
+					pmap_tlb_shootdown(pve->pv_pmap,
+					    pve->pv_va, opte, &cpumask);
+
 				/* zap! */
-				pve->pv_pmap->pm_pdir[pdei(pve->pv_va)] = 0;
-				pmap_update_pg(((vaddr_t)ptes) +
-					       pve->pv_ptp->offset);
-#if defined(I386_CPU)
-				needs_update = FALSE;
+				opte = i386_atomic_testset_ul(
+				    &pve->pv_pmap->pm_pdir[pdei(pve->pv_va)],
+				    0);
+				pmap_tlb_shootdown(curpcb->pcb_pmap,
+				    ((vaddr_t)ptes) + pve->pv_ptp->offset,
+				    opte, &cpumask);
+#if defined(MULTIPROCESSOR)
+				/*
+				 * Always shoot down the other pmap's
+				 * self-mapping of the PTP.
+				 */
+				pmap_tlb_shootdown(pve->pv_pmap,
+				    ((vaddr_t)PTE_BASE) + pve->pv_ptp->offset,
+				    opte, &cpumask);
 #endif
 				pve->pv_pmap->pm_stats.resident_count--;
 				/* update hint? */
@@ -2628,10 +2835,7 @@ pmap_page_remove(pg)
 	pvh->pvh_list = NULL;
 	simple_unlock(&pvh->pvh_lock);
 	PMAP_HEAD_TO_MAP_UNLOCK();
-#if defined(I386_CPU)
-	if (needs_update)
-		tlbflush();
-#endif
+	pmap_tlb_shootnow(cpumask);
 }
 
 /*
@@ -2719,11 +2923,9 @@ pmap_change_attrs(pg, setbits, clearbits)
 	int bank, off;
 	struct pv_head *pvh;
 	struct pv_entry *pve;
-	pt_entry_t *ptes, npte;
+	pt_entry_t *ptes, npte, opte;
 	char *myattrs;
-#if defined(I386_CPU)
-	boolean_t needs_update = FALSE;
-#endif
+	int32_t cpumask = 0;
 
 	/* XXX: vm_page should either contain pv_head or have a pointer to it */
 	bank = vm_physseg_find(atop(VM_PAGE_TO_PHYS(pg)), &off);
@@ -2753,27 +2955,18 @@ pmap_change_attrs(pg, setbits, clearbits)
 		result |= (npte & clearbits);
 		npte = (npte | setbits) & ~clearbits;
 		if (ptes[i386_btop(pve->pv_va)] != npte) {
-			ptes[i386_btop(pve->pv_va)] = npte;	/* zap! */
-
-			if (pmap_is_curpmap(pve->pv_pmap)) {
-#if defined(I386_CPU)
-				if (cpu_class == CPUCLASS_386)
-					needs_update = TRUE;
-				else
-#endif
-					pmap_update_pg(pve->pv_va);
-			}
+			opte = i386_atomic_testset_ul(
+			    &ptes[i386_btop(pve->pv_va)], npte);
+			pmap_tlb_shootdown(pve->pv_pmap,
+			    i386_btop(pve->pv_va), opte, &cpumask);
 		}
 		pmap_unmap_ptes(pve->pv_pmap);		/* unlocks pmap */
 	}
 
 	simple_unlock(&pvh->pvh_lock);
 	PMAP_HEAD_TO_MAP_UNLOCK();
+	pmap_tlb_shootnow(cpumask);
 
-#if defined(I386_CPU)
-	if (needs_update)
-		tlbflush();
-#endif
 	return(result != 0);
 }
 
@@ -2809,20 +3002,12 @@ pmap_write_protect(pmap, sva, eva, prot)
 	vm_prot_t prot;
 {
 	pt_entry_t *ptes, *spte, *epte, npte;
-	struct pmap_remove_record pmap_rr, *prr;
-	vaddr_t blockend, va;
+	vaddr_t blockend;
 	u_int32_t md_prot;
+	int32_t cpumask = 0;
 
 	ptes = pmap_map_ptes(pmap);		/* locks pmap */
 
-	/* need to worry about TLB? [TLB stores protection bits] */
-	if (pmap_is_curpmap(pmap)) {
-		prr = &pmap_rr;
-		prr->prr_npages = 0;
-	} else {
-		prr = NULL;
-	}
-
 	/* should be ok, but just in case ... */
 	sva &= PG_FRAME;
 	eva &= PG_FRAME;
@@ -2869,53 +3054,13 @@ pmap_write_protect(pmap, sva, eva, prot)
 
 			if (npte != *spte) {
 				pmap_exec_account(pmap, sva, *spte, npte);
-
-				*spte = npte;		/* zap! */
-
-				if (prr) {    /* worried about tlb flushing? */
-					va = i386_ptob(spte - ptes);
-					if (npte & PG_G) {
-						/* PG_G requires this */
-						pmap_update_pg(va);
-					} else {
-						if (prr->prr_npages <
-						    PMAP_RR_MAX) {
-							prr->prr_vas[
-							    prr->prr_npages++] =
-								va;
-						} else {
-						    if (prr->prr_npages ==
-							PMAP_RR_MAX)
-							/* signal an overflow */
-							    prr->prr_npages++;
-						}
-					}
-				}	/* if (prr) */
-			}	/* npte != *spte */
-		}	/* for loop */
-	}
-
-	/*
-	 * if we kept a removal record and removed some pages update the TLB
-	 */
-
-	if (prr && prr->prr_npages) {
-#if defined(I386_CPU)
-		if (cpu_class == CPUCLASS_386) {
-			tlbflush();
-		} else
-#endif
-		{ /* not I386 */
-			if (prr->prr_npages > PMAP_RR_MAX) {
-				tlbflush();
-			} else {
-				while (prr->prr_npages) {
-					pmap_update_pg(prr->prr_vas[
-						       --prr->prr_npages]);
-				}
+				i386_atomic_testset_ul(spte, npte); /* zap! */
+				pmap_tlb_shootdown(pmap, sva, *spte, &cpumask);
 			}
-		} /* not I386 */
+		}
 	}
+
+	pmap_tlb_shootnow(cpumask);
 	pmap_unmap_ptes(pmap);		/* unlocks pmap */
 }
 
@@ -3179,8 +3324,18 @@ enter_now:
 
 	ptes[i386_btop(va)] = npte;		/* zap! */
 
-	if ((opte & ~(PG_M|PG_U)) != npte && pmap_is_curpmap(pmap))
-		pmap_update_pg(va);
+	if ((opte & ~(PG_M|PG_U)) != npte) {
+#ifdef MULTIPROCESSOR
+		int32_t cpumask = 0;
+
+		pmap_tlb_shootdown(pmap, va, opte, &cpumask);
+		pmap_tlb_shootnow(cpumask);
+#else
+		/* Don't bother deferring in the single CPU case. */
+		if (pmap_is_curpmap(pmap))
+			pmap_update_pg(va);
+#endif
+	}
 
 	error = 0;
 
@@ -3330,3 +3485,295 @@ pmap_dump(pmap, sva, eva)
 	PMAP_MAP_TO_HEAD_UNLOCK();
 }
 #endif
+
+
+/******************** TLB shootdown code ********************/
+
+void
+pmap_tlb_shootnow(int32_t cpumask)
+{
+#ifdef MULTIPROCESSOR
+	struct cpu_info *ci, *self;
+	CPU_INFO_ITERATOR cii;
+	int s;
+#ifdef DIAGNOSTIC
+	int count = 0;
+#endif
+#endif
+
+	if (cpumask == 0)
+		return;
+
+#ifdef MULTIPROCESSOR
+	self = curcpu();
+	s = splipi();
+	self->ci_tlb_ipi_mask = cpumask;
+#endif
+
+	pmap_do_tlb_shootdown(0);	/* do *our* work. */
+
+#ifdef MULTIPROCESSOR
+	splx(s);
+
+	/*
+	 * Send the TLB IPI to other CPUs pending shootdowns.
+	 */
+	for (CPU_INFO_FOREACH(cii, ci)) {
+		if (ci == self)
+			continue;
+		if (cpumask & (1U << ci->ci_cpuid))
+			if (i386_send_ipi(ci, I386_IPI_TLB) != 0)
+				i386_atomic_clearbits_l(&self->ci_tlb_ipi_mask,
+				    (1U << ci->ci_cpuid));
+	}
+
+	while (self->ci_tlb_ipi_mask != 0)
+#ifdef DIAGNOSTIC
+		if (count++ > 100000000)
+			panic("TLB IPI rendezvous failed (mask %x)",
+			    self->ci_tlb_ipi_mask);
+#else
+		/* XXX insert pause instruction */
+		;
+#endif
+#endif
+}
+
+/*
+ * pmap_tlb_shootdown:
+ *
+ *	Cause the TLB entry for pmap/va to be shot down.
+ */
+void
+pmap_tlb_shootdown(pmap, va, pte, cpumaskp)
+	pmap_t pmap;
+	vaddr_t va;
+	pt_entry_t pte;
+	int32_t *cpumaskp;
+{
+	struct cpu_info *ci, *self;
+	struct pmap_tlb_shootdown_q *pq;
+	struct pmap_tlb_shootdown_job *pj;
+	CPU_INFO_ITERATOR cii;
+	int s;
+
+	if (pmap_initialized == FALSE) {
+		pmap_update_pg(va);
+		return;
+	}
+
+	self = curcpu();
+
+	s = splipi();
+#if 0
+	printf("dshootdown %lx\n", va);
+#endif
+
+	for (CPU_INFO_FOREACH(cii, ci)) {
+		/* Note: we queue shootdown events for ourselves here! */
+		if (pmap_is_active(pmap, ci->ci_cpuid) == 0)
+			continue;
+		if (ci != self && !(ci->ci_flags & CPUF_RUNNING))
+			continue;
+		pq = &pmap_tlb_shootdown_q[ci->ci_cpuid];
+		SIMPLE_LOCK(&pq->pq_slock);
+
+		/*
+		 * If there's a global flush already queued, or a
+		 * non-global flush, and this pte doesn't have the G
+		 * bit set, don't bother.
+		 */
+		if (pq->pq_flushg > 0 ||
+		    (pq->pq_flushu > 0 && (pte & pmap_pg_g) == 0)) {
+			SIMPLE_UNLOCK(&pq->pq_slock);
+			continue;
+		}
+
+#ifdef I386_CPU
+		/*
+		 * i386 CPUs can't invalidate a single VA, only
+		 * flush the entire TLB, so don't bother allocating
+		 * jobs for them -- just queue a `flushu'.
+		 *
+		 * XXX note that this can be executed for non-i386
+		 * when called early (before identifycpu() has set
+		 * cpu_class)
+		 */
+		if (cpu_class == CPUCLASS_386) {
+			pq->pq_flushu++;
+			*cpumaskp |= 1U << ci->ci_cpuid;
+			SIMPLE_UNLOCK(&pq->pq_slock);
+			continue;
+		}
+#endif
+
+		pj = pmap_tlb_shootdown_job_get(pq);
+		pq->pq_pte |= pte;
+		if (pj == NULL) {
+			/*
+			 * Couldn't allocate a job entry.
+			 * Kill it now for this cpu, unless the failure
+			 * was due to too many pending flushes; otherwise,
+			 * tell other cpus to kill everything..
+			 */
+			if (ci == self && pq->pq_count < PMAP_TLB_MAXJOBS) {
+				pmap_update_pg(va);
+				SIMPLE_UNLOCK(&pq->pq_slock);
+				continue;
+			} else {
+				if (pq->pq_pte & pmap_pg_g)
+					pq->pq_flushg++;
+				else
+					pq->pq_flushu++;
+				/*
+				 * Since we've nailed the whole thing,
+				 * drain the job entries pending for that
+				 * processor.
+				 */
+				pmap_tlb_shootdown_q_drain(pq);
+				*cpumaskp |= 1U << ci->ci_cpuid;
+			}
+		} else {
+			pj->pj_pmap = pmap;
+			pj->pj_va = va;
+			pj->pj_pte = pte;
+			TAILQ_INSERT_TAIL(&pq->pq_head, pj, pj_list);
+			*cpumaskp |= 1U << ci->ci_cpuid;
+		}
+		SIMPLE_UNLOCK(&pq->pq_slock);
+	}
+	splx(s);
+}
+
+/*
+ * pmap_do_tlb_shootdown:
+ *
+ *	Process pending TLB shootdown operations for this processor.
+ */
+void
+pmap_do_tlb_shootdown(struct cpu_info *self)
+{
+	u_long cpu_id = cpu_number();
+	struct pmap_tlb_shootdown_q *pq = &pmap_tlb_shootdown_q[cpu_id];
+	struct pmap_tlb_shootdown_job *pj;
+	int s;
+#ifdef MULTIPROCESSOR
+	struct cpu_info *ci;
+	CPU_INFO_ITERATOR cii;
+#endif
+
+	s = splipi();
+
+	SIMPLE_LOCK(&pq->pq_slock);
+
+	if (pq->pq_flushg) {
+		tlbflushg();
+		pq->pq_flushg = 0;
+		pq->pq_flushu = 0;
+		pmap_tlb_shootdown_q_drain(pq);
+	} else {
+		/*
+		 * TLB flushes for PTEs with PG_G set may be in the queue
+		 * after a flushu, they need to be dealt with.
+		 */
+		if (pq->pq_flushu) {
+			tlbflush();
+		}
+		while ((pj = TAILQ_FIRST(&pq->pq_head)) != NULL) {
+			TAILQ_REMOVE(&pq->pq_head, pj, pj_list);
+
+			if ((!pq->pq_flushu && pmap_is_curpmap(pj->pj_pmap)) ||
+			    (pj->pj_pte & pmap_pg_g))
+				pmap_update_pg(pj->pj_va);
+
+			pmap_tlb_shootdown_job_put(pq, pj);
+		}
+
+		pq->pq_flushu = pq->pq_pte = 0;
+	}
+
+#ifdef MULTIPROCESSOR
+	for (CPU_INFO_FOREACH(cii, ci))
+		i386_atomic_clearbits_l(&ci->ci_tlb_ipi_mask,
+		    (1U << cpu_id));
+#endif
+	SIMPLE_UNLOCK(&pq->pq_slock);
+
+	splx(s);
+}
+
+/*
+ * pmap_tlb_shootdown_q_drain:
+ *
+ *	Drain a processor's TLB shootdown queue.  We do not perform
+ *	the shootdown operations.  This is merely a convenience
+ *	function.
+ *
+ *	Note: We expect the queue to be locked.
+ */
+void
+pmap_tlb_shootdown_q_drain(pq)
+	struct pmap_tlb_shootdown_q *pq;
+{
+	struct pmap_tlb_shootdown_job *pj;
+
+	while ((pj = TAILQ_FIRST(&pq->pq_head)) != NULL) {
+		TAILQ_REMOVE(&pq->pq_head, pj, pj_list);
+		pmap_tlb_shootdown_job_put(pq, pj);
+	}
+	pq->pq_pte = 0;
+}
+
+/*
+ * pmap_tlb_shootdown_job_get:
+ *
+ *	Get a TLB shootdown job queue entry.  This places a limit on
+ *	the number of outstanding jobs a processor may have.
+ *
+ *	Note: We expect the queue to be locked.
+ */
+struct pmap_tlb_shootdown_job *
+pmap_tlb_shootdown_job_get(pq)
+	struct pmap_tlb_shootdown_q *pq;
+{
+	struct pmap_tlb_shootdown_job *pj;
+
+	if (pq->pq_count >= PMAP_TLB_MAXJOBS)
+		return (NULL);
+
+	SIMPLE_LOCK(&pmap_tlb_shootdown_job_lock);
+	if (pj_free == NULL) {
+		SIMPLE_UNLOCK(&pmap_tlb_shootdown_job_lock);
+		return NULL;
+	}
+	pj = pj_free;
+	pj_free = pj_free->pj_nextfree;
+	SIMPLE_UNLOCK(&pmap_tlb_shootdown_job_lock);
+
+	pq->pq_count++;
+	return (pj);
+}
+
+/*
+ * pmap_tlb_shootdown_job_put:
+ *
+ *	Put a TLB shootdown job queue entry onto the free list.
+ *
+ *	Note: We expect the queue to be locked.
+ */
+void
+pmap_tlb_shootdown_job_put(pq, pj)
+	struct pmap_tlb_shootdown_q *pq;
+	struct pmap_tlb_shootdown_job *pj;
+{
+#ifdef DIAGNOSTIC
+	if (pq->pq_count == 0)
+		panic("pmap_tlb_shootdown_job_put: queue length inconsistency");
+#endif
+	SIMPLE_LOCK(&pmap_tlb_shootdown_job_lock);
+	pj->pj_nextfree = pj_free;
+	pj_free = pj;
+	SIMPLE_UNLOCK(&pmap_tlb_shootdown_job_lock);
+
+	pq->pq_count--;
+}
diff --git a/sys/arch/i386/i386/process_machdep.c b/sys/arch/i386/i386/process_machdep.c
index 1667b8e223d..fbf8c250220 100644
--- a/sys/arch/i386/i386/process_machdep.c
+++ b/sys/arch/i386/i386/process_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: process_machdep.c,v 1.15 2004/02/05 01:06:33 deraadt Exp $	*/
+/*	$OpenBSD: process_machdep.c,v 1.16 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: process_machdep.c,v 1.22 1996/05/03 19:42:25 christos Exp $	*/
 
 /*
@@ -211,8 +211,7 @@ process_read_fpregs(p, regs)
 		union savefpu *frame = process_fpframe(p);
 
 #if NNPX > 0
-		if (npxproc == p)
-			npxsave();
+		npxsave_proc(p, 1);
 #endif
 
 		if (i386_use_fxsave) {
@@ -249,21 +248,22 @@ process_write_regs(p, regs)
 	} else
 #endif
 	{
+#if 0
 		extern int gdt_size;
-		extern union descriptor *dynamic_gdt;
 
 #define	verr_ldt(slot)	(slot < pcb->pcb_ldt_len && \
 			 (pcb->pcb_ldt[slot].sd.sd_type & SDT_MEMRO) != 0 && \
 			 pcb->pcb_ldt[slot].sd.sd_dpl == SEL_UPL && \
 			 pcb->pcb_ldt[slot].sd.sd_p == 1)
 #define	verr_gdt(slot)	(slot < gdt_size && \
-			 (dynamic_gdt[slot].sd.sd_type & SDT_MEMRO) != 0 && \
-			 dynamic_gdt[slot].sd.sd_dpl == SEL_UPL && \
-			 dynamic_gdt[slot].sd.sd_p == 1)
+			 (gdt[slot].sd.sd_type & SDT_MEMRO) != 0 && \
+			 gdt[slot].sd.sd_dpl == SEL_UPL && \
+			 gdt[slot].sd.sd_p == 1)
 #define	verr(sel)	(ISLDT(sel) ? verr_ldt(IDXSEL(sel)) : \
 				      verr_gdt(IDXSEL(sel)))
 #define	valid_sel(sel)	(ISPL(sel) == SEL_UPL && verr(sel))
 #define	null_sel(sel)	(!ISLDT(sel) && IDXSEL(sel) == 0)
+#endif
 
 		/*
 		 * Check for security violations.
@@ -272,11 +272,14 @@ process_write_regs(p, regs)
 		    !USERMODE(regs->r_cs, regs->r_eflags))
 			return (EINVAL);
 
+		/* XXX Is this safe to remove. */
+#if 0
 		if ((regs->r_gs != pcb->pcb_gs && \
 		     !valid_sel(regs->r_gs) && !null_sel(regs->r_gs)) ||
 		    (regs->r_fs != pcb->pcb_fs && \
 		     !valid_sel(regs->r_fs) && !null_sel(regs->r_fs)))
 			return (EINVAL);
+#endif
 
 		pcb->pcb_gs = regs->r_gs & 0xffff;
 		pcb->pcb_fs = regs->r_fs & 0xffff;
@@ -308,8 +311,7 @@ process_write_fpregs(p, regs)
 
 	if (p->p_md.md_flags & MDP_USEDFPU) {
 #if NNPX > 0
-		if (npxproc == p)
-			npxdrop();
+		npxsave_proc(p, 0);
 #endif
 	} else
 		p->p_md.md_flags |= MDP_USEDFPU;
diff --git a/sys/arch/i386/i386/trap.c b/sys/arch/i386/i386/trap.c
index 5c25f2ba8da..223c6bf3544 100644
--- a/sys/arch/i386/i386/trap.c
+++ b/sys/arch/i386/i386/trap.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: trap.c,v 1.62 2004/04/15 00:22:42 tedu Exp $	*/
+/*	$OpenBSD: trap.c,v 1.63 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: trap.c,v 1.95 1996/05/05 06:50:02 mycroft Exp $	*/
 
 /*-
@@ -132,7 +132,7 @@ userret(p, pc, oticks)
 		addupc_task(p, pc, (int)(p->p_sticks - oticks) * psratio);
 	}                   
 
-	curpriority = p->p_priority;
+	p->p_cpu->ci_schedstate.spc_curpriority = p->p_priority;
 }
 
 char	*trap_type[] = {
@@ -186,6 +186,7 @@ trap(frame)
 	vm_prot_t vftype, ftype;
 	union sigval sv;
 	caddr_t onfault;
+	uint32_t cr2;
 
 	uvmexp.traps++;
 
@@ -200,7 +201,7 @@ trap(frame)
 	if (trapdebug) {
 		printf("trap %d code %x eip %x cs %x eflags %x cr2 %x cpl %x\n",
 		    frame.tf_trapno, frame.tf_err, frame.tf_eip, frame.tf_cs,
-		    frame.tf_eflags, rcr2(), cpl);
+		    frame.tf_eflags, rcr2(), lapic_tpr);
 		printf("curproc %p\n", curproc);
 	}
 #endif
@@ -261,7 +262,7 @@ trap(frame)
 			printf("unknown trap %d", frame.tf_trapno);
 		printf(" in %s mode\n", (type & T_USER) ? "user" : "supervisor");
 		printf("trap type %d code %x eip %x cs %x eflags %x cr2 %x cpl %x\n",
-		    type, frame.tf_err, frame.tf_eip, frame.tf_cs, frame.tf_eflags, rcr2(), cpl);
+		    type, frame.tf_err, frame.tf_eip, frame.tf_cs, frame.tf_eflags, rcr2(), lapic_tpr);
 
 		panic("trap type %d, code=%x, pc=%x",
 		    type, frame.tf_err, frame.tf_eip);
@@ -335,35 +336,47 @@ trap(frame)
 
 	case T_TSSFLT|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGBUS, vftype, BUS_OBJERR, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_SEGNPFLT|T_USER:
 	case T_STKFLT|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGSEGV, vftype, SEGV_MAPERR, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_ALIGNFLT|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGBUS, vftype, BUS_ADRALN, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_PRIVINFLT|T_USER:	/* privileged instruction fault */
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGILL, type &~ T_USER, ILL_PRVOPC, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_FPOPFLT|T_USER:		/* coprocessor operand fault */
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGILL, type &~ T_USER, ILL_COPROC, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_ASTFLT|T_USER:		/* Allow process switch */
 		uvmexp.softs++;
 		if (p->p_flag & P_OWEUPC) {
 			p->p_flag &= ~P_OWEUPC;
+			KERNEL_PROC_LOCK(p);
 			ADDUPROF(p);
+			KERNEL_PROC_UNLOCK(p);
 		}
 		goto out;
 
@@ -376,55 +389,84 @@ trap(frame)
 			return;
 		}
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, rv, type &~ T_USER, FPE_FLTINV, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 #else
 		printf("pid %d killed due to lack of floating point\n",
 		    p->p_pid);
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGKILL, type &~ T_USER, FPE_FLTINV, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 #endif
 	}
 
 	case T_BOUND|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGFPE, type &~ T_USER, FPE_FLTSUB, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 	case T_OFLOW|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGFPE, type &~ T_USER, FPE_INTOVF, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 	case T_DIVIDE|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGFPE, type &~ T_USER, FPE_INTDIV, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_ARITHTRAP|T_USER:
 		sv.sival_int = frame.tf_eip;
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGFPE, frame.tf_err, FPE_INTOVF, sv);
+		KERNEL_PROC_UNLOCK(p);
 		goto out;
 
 	case T_PAGEFLT:			/* allow page faults in kernel mode */
 		if (p == 0 || p->p_addr == 0)
 			goto we_re_toast;
+#ifdef LOCKDEBUG
+		/* If we page-fault while in scheduler, we're doomed. */
+#ifdef notyet
+		if (simple_lock_held(&sched_lock))
+#else
+		if (__mp_lock_held(&sched_lock))
+#endif
+			goto we_re_toast;
+#endif
+
 		pcb = &p->p_addr->u_pcb;
 #if 0
 		/* XXX - check only applies to 386's and 486's with WP off */
 		if (frame.tf_err & PGEX_P)
 			goto we_re_toast;
 #endif
-		/* FALLTHROUGH */
+		cr2 = rcr2();
+		KERNEL_LOCK(LK_CANRECURSE|LK_EXCLUSIVE);
+		goto faultcommon;
+
 	case T_PAGEFLT|T_USER: {	/* page fault */
 		vaddr_t va, fa;
-		struct vmspace *vm = p->p_vmspace;
+		struct vmspace *vm;
 		struct vm_map *map;
 		int rv;
 		unsigned nss;
 
+		cr2 = rcr2();
+		KERNEL_PROC_LOCK(p);
+	faultcommon:
+		vm = p->p_vmspace;
 		if (vm == NULL)
 			goto we_re_toast;
-		fa = (vaddr_t)rcr2();
+		fa = (vaddr_t)cr2;
 		va = trunc_page(fa);
 		/*
 		 * It is only a kernel address space fault iff:
@@ -472,20 +514,26 @@ trap(frame)
 		if (rv == 0) {
 			if (nss > vm->vm_ssize)
 				vm->vm_ssize = nss;
-			if (type == T_PAGEFLT)
+			if (type == T_PAGEFLT) {
+				KERNEL_UNLOCK();
 				return;
+			}
+			KERNEL_PROC_UNLOCK(p);
 			goto out;
 		}
 
 		if (type == T_PAGEFLT) {
-			if (pcb->pcb_onfault != 0)
+			if (pcb->pcb_onfault != 0) {
+				KERNEL_UNLOCK();
 				goto copyfault;
+			}
 			printf("uvm_fault(%p, 0x%lx, 0, %d) -> %x\n",
 			    map, va, ftype, rv);
 			goto we_re_toast;
 		}
 		sv.sival_int = fa;
 		trapsignal(p, SIGSEGV, vftype, SEGV_MAPERR, sv);
+		KERNEL_PROC_UNLOCK(p);
 		break;
 	}
 
@@ -500,14 +548,18 @@ trap(frame)
 
 	case T_BPTFLT|T_USER:		/* bpt instruction fault */
 		sv.sival_int = rcr2();
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGTRAP, type &~ T_USER, TRAP_BRKPT, sv);
+		KERNEL_PROC_UNLOCK(p);
 		break;
 	case T_TRCTRAP|T_USER:		/* trace trap */
 #if defined(GPL_MATH_EMULATE)
 	trace:
 #endif
 		sv.sival_int = rcr2();
+		KERNEL_PROC_LOCK(p);
 		trapsignal(p, SIGTRAP, type &~ T_USER, TRAP_TRACE, sv);
+		KERNEL_PROC_UNLOCK(p);
 		break;
 
 #if	NISA > 0
@@ -591,7 +643,7 @@ syscall(frame)
 	register_t code, args[8], rval[2];
 	u_quad_t sticks;
 #ifdef DIAGNOSTIC
-	int ocpl = cpl;
+	int ocpl = lapic_tpr;
 #endif
 
 	uvmexp.syscalls++;
@@ -716,12 +768,14 @@ syscall(frame)
 		goto bad;
 	rval[0] = 0;
 	rval[1] = frame.tf_edx;
+	KERNEL_PROC_LOCK(p);
 #if NSYSTRACE > 0
 	if (ISSET(p->p_flag, P_SYSTRACE))
 		orig_error = error = systrace_redirect(code, p, args, rval);
 	else
 #endif
 		orig_error = error = (*callp->sy_call)(p, args, rval);
+	KERNEL_PROC_UNLOCK(p);
 	switch (error) {
 	case 0:
 		/*
@@ -758,15 +812,18 @@ syscall(frame)
 #endif
 	userret(p, frame.tf_eip, sticks);
 #ifdef KTRACE
-	if (KTRPOINT(p, KTR_SYSRET))
+	if (KTRPOINT(p, KTR_SYSRET)) {
+		KERNEL_PROC_LOCK(p);
 		ktrsysret(p, code, orig_error, rval[0]);
+		KERNEL_PROC_UNLOCK(p);
+	}
 #endif
 #ifdef DIAGNOSTIC
-	if (cpl != ocpl) {
+	if (lapic_tpr != ocpl) {
 		printf("WARNING: SPL (0x%x) NOT LOWERED ON "
 		    "syscall(0x%x, 0x%x, 0x%x, 0x%x...) EXIT, PID %d\n",
-		    cpl, code, args[0], args[1], args[2], p->p_pid);
-		cpl = ocpl;
+		    lapic_tpr, code, args[0], args[1], args[2], p->p_pid);
+		lapic_tpr = ocpl;
 	}
 #endif
 }
@@ -781,9 +838,15 @@ child_return(arg)
 	tf->tf_eax = 0;
 	tf->tf_eflags &= ~PSL_C;
 
+#ifdef notyet
+	KERNEL_PROC_UNLOCK(p);
+#endif
+
 	userret(p, tf->tf_eip, 0);
 #ifdef KTRACE
 	if (KTRPOINT(p, KTR_SYSRET))
+		KERNEL_PROC_LOCK(p);
 		ktrsysret(p, SYS_fork, 0, 0);
+		KERNEL_PROC_UNLOCK(p);
 #endif
 }
diff --git a/sys/arch/i386/isa/vector.s b/sys/arch/i386/i386/vector.s
index 93a7bab51da..b7ddd3f0021 100644
--- a/sys/arch/i386/isa/vector.s
+++ b/sys/arch/i386/i386/vector.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vector.s,v 1.16 2003/04/17 03:42:14 drahn Exp $	*/
+/*	$OpenBSD: vector.s,v 1.2 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: vector.s,v 1.32 1996/01/07 21:29:47 mycroft Exp $	*/
 
 /*
@@ -30,85 +30,12 @@
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <i386/isa/icu.h>
+#include <machine/i8259.h>
 #include <dev/isa/isareg.h>
 
-#define ICU_HARDWARE_MASK
-
 #define MY_COUNT _C_LABEL(uvmexp)
 
 /*
- * These macros are fairly self explanatory.  If ICU_SPECIAL_MASK_MODE is
- * defined, we try to take advantage of the ICU's `special mask mode' by only
- * EOIing the interrupts on return.  This avoids the requirement of masking and
- * unmasking.  We can't do this without special mask mode, because the ICU
- * would also hold interrupts that it thinks are of lower priority.
- *
- * Many machines do not support special mask mode, so by default we don't try
- * to use it.
- */
-
-#define	IRQ_BIT(irq_num)	(1 << ((irq_num) % 8))
-#define	IRQ_BYTE(irq_num)	((irq_num) / 8)
-
-#ifdef ICU_SPECIAL_MASK_MODE
-
-#define	ACK1(irq_num)
-#define	ACK2(irq_num) \
-	movb	$(0x60|IRQ_SLAVE),%al	/* specific EOI for IRQ2 */	;\
-	outb	%al,$IO_ICU1
-#define	MASK(irq_num, icu)
-#define	UNMASK(irq_num, icu) \
-	movb	$(0x60|(irq_num%8)),%al	/* specific EOI */		;\
-	outb	%al,$icu
-
-#else /* ICU_SPECIAL_MASK_MODE */
-
-#ifndef	AUTO_EOI_1
-#define	ACK1(irq_num) \
-	movb	$(0x60|(irq_num%8)),%al	/* specific EOI */		;\
-	outb	%al,$IO_ICU1
-#else
-#define	ACK1(irq_num)
-#endif
-
-#ifndef AUTO_EOI_2
-#define	ACK2(irq_num) \
-	movb	$(0x60|(irq_num%8)),%al	/* specific EOI */		;\
-	outb	%al,$IO_ICU2		/* do the second ICU first */	;\
-	movb	$(0x60|IRQ_SLAVE),%al	/* specific EOI for IRQ2 */	;\
-	outb	%al,$IO_ICU1
-#else
-#define	ACK2(irq_num)
-#endif
-
-#ifdef ICU_HARDWARE_MASK
-
-#define	MASK(irq_num, icu) \
-	movb	_C_LABEL(imen) + IRQ_BYTE(irq_num),%al				;\
-	orb	$IRQ_BIT(irq_num),%al					;\
-	movb	%al,_C_LABEL(imen) + IRQ_BYTE(irq_num)				;\
-	FASTER_NOP							;\
-	outb	%al,$(icu+1)
-#define	UNMASK(irq_num, icu) \
-	cli								;\
-	movb	_C_LABEL(imen) + IRQ_BYTE(irq_num),%al				;\
-	andb	$~IRQ_BIT(irq_num),%al					;\
-	movb	%al,_C_LABEL(imen) + IRQ_BYTE(irq_num)				;\
-	FASTER_NOP							;\
-	outb	%al,$(icu+1)						;\
-	sti
-
-#else /* ICU_HARDWARE_MASK */
-
-#define	MASK(irq_num, icu)
-#define	UNMASK(irq_num, icu)
-
-#endif /* ICU_HARDWARE_MASK */
-
-#endif /* ICU_SPECIAL_MASK_MODE */
-
-/*
  * Macros for interrupt entry, call to handler, and exit.
  *
  * XXX
@@ -132,6 +59,16 @@
 
 	.globl	_C_LABEL(isa_strayintr)
 
+#ifdef MULTIPROCESSOR
+#define LOCK_KERNEL	call _C_LABEL(i386_intlock)
+#define UNLOCK_KERNEL	call _C_LABEL(i386_intunlock)
+#else
+#define LOCK_KERNEL
+#define UNLOCK_KERNEL
+#endif
+
+#define voidop(num)
+
 /*
  * Normal vectors.
  *
@@ -148,34 +85,35 @@
  *
  * On exit, we jump to Xdoreti(), to process soft interrupts and ASTs.
  */
-#define	INTR(irq_num, icu, ack) \
-IDTVEC(recurse/**/irq_num)						;\
+#define	INTRSTUB(name, num, early_ack, late_ack, mask, unmask, level_mask) \
+IDTVEC(recurse_/**/name/**/num)						;\
 	pushfl								;\
 	pushl	%cs							;\
 	pushl	%esi							;\
 	cli								;\
-_C_LABEL(Xintr)/**/irq_num/**/:						;\
+_C_LABEL(Xintr_/**/name/**/num):					;\
 	pushl	$0			/* dummy error code */		;\
 	pushl	$T_ASTFLT		/* trap # for doing ASTs */	;\
 	INTRENTRY							;\
 	MAKE_FRAME							;\
-	MASK(irq_num, icu)		/* mask it in hardware */	;\
-	ack(irq_num)			/* and allow other intrs */	;\
+	mask(num)			/* mask it in hardware */	;\
+	early_ack(num)			/* and allow other intrs */	;\
 	incl	MY_COUNT+V_INTR		/* statistical info */		;\
-	movl	_C_LABEL(iminlevel) + (irq_num) * 4, %eax		;\
-	movzbl	_C_LABEL(cpl),%ebx					;\
+	movl	_C_LABEL(iminlevel) + (num) * 4, %eax			;\
+	movl	CPL,%ebx						;\
 	cmpl	%eax,%ebx						;\
-	jae	_C_LABEL(Xhold/**/irq_num)/* currently masked; hold it */;\
-_C_LABEL(Xresume)/**/irq_num/**/:					;\
-	movzbl	_C_LABEL(cpl),%eax	/* cpl to restore on exit */	;\
+	jae	_C_LABEL(Xhold_/**/name/**/num)/* currently masked; hold it */;\
+Xresume_/**/name/**/num/**/:						;\
+	movl	CPL,%eax		/* cpl to restore on exit */	;\
 	pushl	%eax							;\
-	movl	_C_LABEL(imaxlevel) + (irq_num) * 4,%eax		;\
-	movl	%eax,_C_LABEL(cpl)	/* block enough for this irq */	;\
+	movl	_C_LABEL(imaxlevel) + (num) * 4,%eax			;\
+	movl	%eax,CPL		/* block enough for this irq */	;\
 	sti				/* safe to take intrs now */	;\
-	movl	_C_LABEL(intrhand) + (irq_num) * 4,%ebx	/* head of chain */	;\
+	movl	_C_LABEL(intrhand) + (num) * 4,%ebx	/* head of chain */ ;\
 	testl	%ebx,%ebx						;\
-	jz	_C_LABEL(Xstray)/**/irq_num	/* no handlears; we're stray */	;\
+	jz	_C_LABEL(Xstray_/**/name/**/num)	/* no handlears; we're stray */	;\
 	STRAY_INITIALIZE		/* nobody claimed it yet */	;\
+	LOCK_KERNEL							;\
 7:	movl	IH_ARG(%ebx),%eax	/* get handler arg */		;\
 	testl	%eax,%eax						;\
 	jnz	4f							;\
@@ -190,16 +128,18 @@ _C_LABEL(Xresume)/**/irq_num/**/:					;\
 5:	movl	IH_NEXT(%ebx),%ebx	/* next handler in chain */	;\
 	testl	%ebx,%ebx						;\
 	jnz	7b							;\
+	UNLOCK_KERNEL							;\
 	STRAY_TEST			/* see if it's a stray */	;\
-6:	UNMASK(irq_num, icu)		/* unmask it in hardware */	;\
+6:	unmask(num)			/* unmask it in hardware */	;\
+	late_ack(num)							;\
 	jmp	_C_LABEL(Xdoreti)	/* lower spl and do ASTs */	;\
-IDTVEC(stray/**/irq_num)						;\
-	pushl	$irq_num						;\
+IDTVEC(stray_/**/name/**/num)						;\
+	pushl	$num							;\
 	call	_C_LABEL(isa_strayintr)					;\
 	addl	$4,%esp							;\
 	jmp	6b							;\
-IDTVEC(hold/**/irq_num)							;\
-	orb	$IRQ_BIT(irq_num),_C_LABEL(ipending) + IRQ_BYTE(irq_num)	;\
+IDTVEC(hold_/**/name/**/num)						;\
+	orb	$IRQ_BIT(num),_C_LABEL(ipending) + IRQ_BYTE(num)	;\
 	INTRFASTEXIT
 
 #if defined(DEBUG) && defined(notdef)
@@ -209,7 +149,7 @@ IDTVEC(hold/**/irq_num)							;\
 	orl	%eax,%esi
 #define	STRAY_TEST \
 	testl	%esi,%esi						;\
-	jz	_C_LABEL(Xstray)/**/irq_num
+	jz	_C_LABEL(Xstray_/**/name/**/num)
 #else /* !DEBUG */
 #define	STRAY_INITIALIZE
 #define	STRAY_INTEGRATE
@@ -223,66 +163,92 @@ IDTVEC(hold/**/irq_num)							;\
 #define	MAKE_FRAME
 #endif /* DDB */
 
-INTR(0, IO_ICU1, ACK1)
-INTR(1, IO_ICU1, ACK1)
-INTR(2, IO_ICU1, ACK1)
-INTR(3, IO_ICU1, ACK1)
-INTR(4, IO_ICU1, ACK1)
-INTR(5, IO_ICU1, ACK1)
-INTR(6, IO_ICU1, ACK1)
-INTR(7, IO_ICU1, ACK1)
-INTR(8, IO_ICU2, ACK2)
-INTR(9, IO_ICU2, ACK2)
-INTR(10, IO_ICU2, ACK2)
-INTR(11, IO_ICU2, ACK2)
-INTR(12, IO_ICU2, ACK2)
-INTR(13, IO_ICU2, ACK2)
-INTR(14, IO_ICU2, ACK2)
-INTR(15, IO_ICU2, ACK2)
+#define ICUADDR IO_ICU1
+
+INTRSTUB(legacy,0, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,1, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,2, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,3, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,4, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,5, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,6, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,7, i8259_asm_ack1, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+
+#undef ICUADDR
+#define ICUADDR IO_ICU2
+
+INTRSTUB(legacy,8, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,9, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,10, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,11, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,12, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,13, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,14, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
+INTRSTUB(legacy,15, i8259_asm_ack2, voidop, i8259_asm_mask, i8259_asm_unmask,
+    voidop)
 
 /*
  * These tables are used by the ISA configuration code.
  */
 /* interrupt service routine entry points */
 IDTVEC(intr)
-	.long   _C_LABEL(Xintr0), _C_LABEL(Xintr1), _C_LABEL(Xintr2)
-	.long	_C_LABEL(Xintr3), _C_LABEL(Xintr4), _C_LABEL(Xintr5)
-	.long	_C_LABEL(Xintr6), _C_LABEL(Xintr7), _C_LABEL(Xintr8)
-	.long	_C_LABEL(Xintr9), _C_LABEL(Xintr10), _C_LABEL(Xintr11)
-	.long	_C_LABEL(Xintr12), _C_LABEL(Xintr13)
-	.long	_C_LABEL(Xintr14), _C_LABEL(Xintr15)
+	.long   _C_LABEL(Xintr_legacy0), _C_LABEL(Xintr_legacy1)
+	.long	_C_LABEL(Xintr_legacy2), _C_LABEL(Xintr_legacy3)
+	.long	_C_LABEL(Xintr_legacy4), _C_LABEL(Xintr_legacy5)
+	.long	_C_LABEL(Xintr_legacy6), _C_LABEL(Xintr_legacy7)
+	.long	_C_LABEL(Xintr_legacy8), _C_LABEL(Xintr_legacy9)
+	.long	_C_LABEL(Xintr_legacy10), _C_LABEL(Xintr_legacy11)
+	.long	_C_LABEL(Xintr_legacy12), _C_LABEL(Xintr_legacy13)
+	.long	_C_LABEL(Xintr_legacy14), _C_LABEL(Xintr_legacy15)
 
 /*
  * These tables are used by Xdoreti() and Xspllower().
  */
 /* resume points for suspended interrupts */
 IDTVEC(resume)
-	.long	_C_LABEL(Xresume0), _C_LABEL(Xresume1)
-	.long	_C_LABEL(Xresume2), _C_LABEL(Xresume3)
-	.long	_C_LABEL(Xresume4), _C_LABEL(Xresume5)
-	.long	_C_LABEL(Xresume6), _C_LABEL(Xresume7)
-	.long	_C_LABEL(Xresume8), _C_LABEL(Xresume9)
-	.long	_C_LABEL(Xresume10), _C_LABEL(Xresume11)
-	.long	_C_LABEL(Xresume12), _C_LABEL(Xresume13)
-	.long	_C_LABEL(Xresume14), _C_LABEL(Xresume15)
+	.long	_C_LABEL(Xresume_legacy0), _C_LABEL(Xresume_legacy1)
+	.long	_C_LABEL(Xresume_legacy2), _C_LABEL(Xresume_legacy3)
+	.long	_C_LABEL(Xresume_legacy4), _C_LABEL(Xresume_legacy5)
+	.long	_C_LABEL(Xresume_legacy6), _C_LABEL(Xresume_legacy7)
+	.long	_C_LABEL(Xresume_legacy8), _C_LABEL(Xresume_legacy9)
+	.long	_C_LABEL(Xresume_legacy10), _C_LABEL(Xresume_legacy11)
+	.long	_C_LABEL(Xresume_legacy12), _C_LABEL(Xresume_legacy13)
+	.long	_C_LABEL(Xresume_legacy14), _C_LABEL(Xresume_legacy15)
 	/* for soft interrupts */
-	.long	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+	.long	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
 	.long	_C_LABEL(Xsofttty), _C_LABEL(Xsoftnet)
 	.long	_C_LABEL(Xsoftclock)
+	.long	0, 0
 /* fake interrupts to resume from splx() */
 IDTVEC(recurse)
-	.long	_C_LABEL(Xrecurse0), _C_LABEL(Xrecurse1)
-	.long	_C_LABEL(Xrecurse2), _C_LABEL(Xrecurse3)
-	.long	_C_LABEL(Xrecurse4), _C_LABEL(Xrecurse5)
-	.long	_C_LABEL(Xrecurse6), _C_LABEL(Xrecurse7)
-	.long	_C_LABEL(Xrecurse8), _C_LABEL(Xrecurse9)
-	.long	_C_LABEL(Xrecurse10), _C_LABEL(Xrecurse11)
-	.long	_C_LABEL(Xrecurse12), _C_LABEL(Xrecurse13)
-	.long	_C_LABEL(Xrecurse14), _C_LABEL(Xrecurse15)
+	.long	_C_LABEL(Xrecurse_legacy0), _C_LABEL(Xrecurse_legacy1)
+	.long	_C_LABEL(Xrecurse_legacy2), _C_LABEL(Xrecurse_legacy3)
+	.long	_C_LABEL(Xrecurse_legacy4), _C_LABEL(Xrecurse_legacy5)
+	.long	_C_LABEL(Xrecurse_legacy6), _C_LABEL(Xrecurse_legacy7)
+	.long	_C_LABEL(Xrecurse_legacy8), _C_LABEL(Xrecurse_legacy9)
+	.long	_C_LABEL(Xrecurse_legacy10), _C_LABEL(Xrecurse_legacy11)
+	.long	_C_LABEL(Xrecurse_legacy12), _C_LABEL(Xrecurse_legacy13)
+	.long	_C_LABEL(Xrecurse_legacy14), _C_LABEL(Xrecurse_legacy15)
 	/* for soft interrupts */
-	.long	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+	.long	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
 	.long	_C_LABEL(Xsofttty), _C_LABEL(Xsoftnet)
 	.long	_C_LABEL(Xsoftclock)
+	.long	0, 0
 
 /* Some bogus data, to keep vmstat happy, for now. */
 	.globl	_C_LABEL(intrnames), _C_LABEL(eintrnames)
diff --git a/sys/arch/i386/i386/vm_machdep.c b/sys/arch/i386/i386/vm_machdep.c
index 20cda037a0b..3f5eeb74d62 100644
--- a/sys/arch/i386/i386/vm_machdep.c
+++ b/sys/arch/i386/i386/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.39 2003/06/02 23:27:47 millert Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.40 2004/06/13 21:49:15 niklas Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.61 1996/05/03 19:42:35 christos Exp $	*/
 
 /*-
@@ -62,9 +62,6 @@
 #include <machine/specialreg.h>
 
 #include "npx.h"
-#if NNPX > 0
-extern struct proc *npxproc;
-#endif
 
 /*
  * Finish a fork operation, with process p2 nearly set up.
@@ -88,37 +85,36 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	struct switchframe *sf;
 
 #if NNPX > 0
-	/*
-	 * If npxproc != p1, then the npx h/w state is irrelevant and the
-	 * state had better already be in the pcb.  This is true for forks
-	 * but not for dumps.
-	 *
-	 * If npxproc == p1, then we have to save the npx h/w state to
-	 * p1's pcb so that we can copy it.
-	 */
-	if (npxproc == p1)
-		npxsave();
+	npxsave_proc(p1, 1);
 #endif
 
 	p2->p_md.md_flags = p1->p_md.md_flags;
 
-	/* Sync curpcb (which is presumably p1's PCB) and copy it to p2. */
-	savectx(curpcb);
+	/* Copy pcb from proc p1 to p2. */
+	if (p1 == curproc) {
+		/* Sync the PCB before we copy it. */
+		savectx(curpcb);
+	}
+#ifdef DIAGNOSTIC
+	else if (p1 != &proc0)
+		panic("cpu_fork: curproc");
+#endif
 	*pcb = p1->p_addr->u_pcb;
+
 	/*
 	 * Preset these so that gdt_compact() doesn't get confused if called
 	 * during the allocations below.
+	 *
+	 * Note: pcb_ldt_sel is handled in the pmap_activate() call when
+	 * we run the new process.
 	 */
-	pcb->pcb_tss_sel = GSEL(GNULL_SEL, SEL_KPL);
-	/*
-	 * Activate the addres space.  Note this will refresh pcb_ldt_sel.
-	 */
-	pmap_activate(p2);
+	p2->p_md.md_tss_sel = GSEL(GNULL_SEL, SEL_KPL);
 
 	/* Fix up the TSS. */
 	pcb->pcb_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
 	pcb->pcb_tss.tss_esp0 = (int)p2->p_addr + USPACE - 16;
-	tss_alloc(pcb);
+
+	p2->p_md.md_tss_sel = tss_alloc(pcb);
 
 	/*
 	 * Copy the trapframe, and arrange for the child to return directly
@@ -150,8 +146,7 @@ cpu_swapout(p)
 	/*
 	 * Make sure we save the FP state before the user area vanishes.
 	 */
-	if (npxproc == p)
-		npxsave();
+	npxsave_proc(p, 1);
 #endif
 }
 
@@ -169,8 +164,8 @@ cpu_exit(p)
 {
 #if NNPX > 0
 	/* If we were using the FPU, forget about it. */
-	if (npxproc == p)
-		npxproc = 0;
+	if (p->p_addr->u_pcb.pcb_fpcpu != NULL)
+		npxsave_proc(p, 0);
 #endif
 
 	uvmexp.swtch++;
@@ -181,10 +176,7 @@ void
 cpu_wait(p)
 	struct proc *p;
 {
-	struct pcb *pcb;
-
-	pcb = &p->p_addr->u_pcb;
-	tss_free(pcb);
+	tss_free(p->p_md.md_tss_sel);
 }
 
 /*
@@ -251,6 +243,9 @@ pagemove(from, to, size)
 {
 	pt_entry_t *fpte, *tpte;
 	pt_entry_t ofpte, otpte;
+#ifdef MULTIPROCESSOR
+	u_int32_t cpumask = 0;
+#endif
 
 #ifdef DIAGNOSTIC
 	if ((size & PAGE_MASK) != 0)
@@ -263,24 +258,38 @@ pagemove(from, to, size)
 		otpte = *tpte;
 		*tpte++ = *fpte;
 		*fpte++ = 0;
-#if defined(I386_CPU)
+#if defined(I386_CPU) && !defined(MULTIPROCESSOR)
 		if (cpu_class != CPUCLASS_386)
 #endif
 		{
 			if (otpte & PG_V)
-				pmap_update_pg((vaddr_t) to);
+#ifdef MULTIPROCESSOR
+				pmap_tlb_shootdown(pmap_kernel(), (vaddr_t)to,
+				    otpte, &cpumask);
+#else
+				pmap_update_pg((vaddr_t)to);
+#endif
 			if (ofpte & PG_V)
-				pmap_update_pg((vaddr_t) from);
+#ifdef MULTIPROCESSOR
+				pmap_tlb_shootdown(pmap_kernel(),
+				    (vaddr_t)from, ofpte, &cpumask);
+#else
+				pmap_update_pg((vaddr_t)from);
+#endif
 		}
 
 		from += PAGE_SIZE;
 		to += PAGE_SIZE;
 		size -= PAGE_SIZE;
 	}
+#ifdef MULTIPROCESSOR
+	pmap_tlb_shootnow(cpumask);
+#else
 #if defined(I386_CPU)
 	if (cpu_class == CPUCLASS_386)
 		tlbflush();		
 #endif
+#endif
 }
 
 /*
diff --git a/sys/arch/i386/include/apicvar.h b/sys/arch/i386/include/apicvar.h
new file mode 100644
index 00000000000..aac452fca40
--- /dev/null
+++ b/sys/arch/i386/include/apicvar.h
@@ -0,0 +1,57 @@
+/*	$OpenBSD: apicvar.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: apicvar.h,v 1.1.2.3 2000/02/27 20:25:00 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _I386_APICVAR_H_
+#define _I386_APICVAR_H_
+
+struct apic_attach_args {
+	const char *aaa_name;
+	int apic_id;
+	int apic_version;
+	int flags;
+#define IOAPIC_PICMODE		0x01
+#define IOAPIC_VWIRE		0x02
+	paddr_t apic_address;
+};
+
+void apic_format_redir(char *, char *, int, u_int32_t, u_int32_t);
+
+#endif /* !_I386_APICVAR_H_ */
diff --git a/sys/arch/i386/include/asm.h b/sys/arch/i386/include/asm.h
index 2126271a519..f2da5755a33 100644
--- a/sys/arch/i386/include/asm.h
+++ b/sys/arch/i386/include/asm.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: asm.h,v 1.7 2003/06/02 23:27:47 millert Exp $	*/
+/*	$OpenBSD: asm.h,v 1.8 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: asm.h,v 1.7 1994/10/27 04:15:56 cgd Exp $	*/
 
 /*-
@@ -61,6 +61,16 @@
 #define _C_LABEL(name)	name
 #define	_ASM_LABEL(x)	x
 
+#define CVAROFF(x, y)	_C_LABEL(x) + y
+
+#ifdef __STDC__
+# define __CONCAT(x,y)	x ## y
+# define __STRING(x)	#x
+#else
+# define __CONCAT(x,y)	x/**/y
+# define __STRING(x)	"x"
+#endif
+
 /*
  * WEAK ALIAS: create a weak alias
  */
diff --git a/sys/arch/i386/include/atomic.h b/sys/arch/i386/include/atomic.h
new file mode 100644
index 00000000000..e3be6b68b1b
--- /dev/null
+++ b/sys/arch/i386/include/atomic.h
@@ -0,0 +1,72 @@
+/*	$OpenBSD: atomic.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: atomic.h,v 1.1.2.2 2000/02/21 18:54:07 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _ATOMIC_H_
+#define _ATOMIC_H_
+
+#ifndef _LOCORE
+
+static __inline u_int32_t
+i386_atomic_testset_ul (volatile u_int32_t *ptr, unsigned long val) {
+    __asm__ volatile ("xchgl %0,(%2)" :"=r" (val):"0" (val),"r" (ptr));
+    return val;
+}
+
+static __inline int
+i386_atomic_testset_i (volatile int *ptr, unsigned long val) {
+    __asm__ volatile ("xchgl %0,(%2)" :"=r" (val):"0" (val),"r" (ptr));
+    return val;
+}
+
+static __inline void 
+i386_atomic_setbits_l (volatile u_int32_t *ptr, unsigned long bits) {
+    __asm __volatile("lock ; orl %1,%0" :  "=m" (*ptr) : "ir" (bits));
+}
+
+static __inline void 
+i386_atomic_clearbits_l (volatile u_int32_t *ptr, unsigned long bits) {
+    bits = ~bits;
+    __asm __volatile("lock ; and %1,%0" :  "=m" (*ptr) : "ir" (bits));
+}
+
+#endif
+#endif
+
diff --git a/sys/arch/i386/include/biosvar.h b/sys/arch/i386/include/biosvar.h
index 4fc00866e87..7e4d98933b4 100644
--- a/sys/arch/i386/include/biosvar.h
+++ b/sys/arch/i386/include/biosvar.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: biosvar.h,v 1.42 2003/11/27 00:31:55 espie Exp $	*/
+/*	$OpenBSD: biosvar.h,v 1.43 2004/06/13 21:49:16 niklas Exp $	*/
 
 /*
  * Copyright (c) 1997-1999 Michael Shalayeff
@@ -192,6 +192,8 @@ typedef struct _bios_consdev {
 	int	conspeed;
 } bios_consdev_t;
 
+#define BOOTARG_SMPINFO 6		/* struct mp_float[] */
+
 #if defined(_KERNEL) || defined (_STANDALONE)
 
 #ifdef _LOCORE
@@ -244,6 +246,8 @@ int  bios32_service(u_int32_t, bios32_entry_t, bios32_entry_info_t);
 
 extern u_int bootapiver;
 extern bios_memmap_t *bios_memmap;
+extern void *bios_smpinfo;
+extern bios_pciinfo_t *bios_pciinfo;
 
 #endif /* _KERNEL */
 #endif /* _LOCORE */
diff --git a/sys/arch/i386/include/cpu.h b/sys/arch/i386/include/cpu.h
index cc690de9219..9d66a9c496c 100644
--- a/sys/arch/i386/include/cpu.h
+++ b/sys/arch/i386/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.60 2004/06/06 17:34:37 grange Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.61 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.35 1996/05/05 19:29:26 christos Exp $	*/
 
 /*-
@@ -41,9 +41,19 @@
 /*
  * Definitions unique to i386 cpu support.
  */
-#include <machine/psl.h>
 #include <machine/frame.h>
+#include <machine/psl.h>
 #include <machine/segments.h>
+#include <machine/intrdefs.h>
+
+#ifdef MULTIPROCESSOR
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+
+/* XXX for now... */
+#define NLAPIC 1
+
+#endif
 
 /*
  * definitions of cpu-dependent requirements
@@ -60,16 +70,156 @@
  */
 #define clockframe intrframe
 
-#define	CLKF_USERMODE(frame)	USERMODE((frame)->if_cs, (frame)->if_eflags)
-#define	CLKF_PC(frame)		((frame)->if_eip)
-#define	CLKF_INTR(frame)	(IDXSEL((frame)->if_cs) == GICODE_SEL)
+#include <sys/device.h>
+#include <sys/lock.h>                  /* will also get LOCKDEBUG */
+#include <sys/proc.h>
+
+struct intrsource;
+
+/* XXX stuff to move to cpuvar.h later */
+struct cpu_info {
+	struct device ci_dev;		/* our device */
+	struct cpu_info *ci_self;	/* pointer to this structure */
+	struct schedstate_percpu ci_schedstate; /* scheduler state */
+	struct cpu_info *ci_next;	/* next cpu */
+	
+	/* 
+	 * Public members. 
+	 */
+	struct proc *ci_curproc; 	/* current owner of the processor */
+	struct simplelock ci_slock;	/* lock on this data structure */
+	cpuid_t ci_cpuid; 		/* our CPU ID */
+	u_int ci_apicid;		/* our APIC ID */
+#if defined(DIAGNOSTIC) || defined(LOCKDEBUG)
+	u_long ci_spin_locks;		/* # of spin locks held */
+	u_long ci_simple_locks;		/* # of simple locks held */
+#endif
+
+	/*
+	 * Private members.
+	 */
+	struct proc *ci_fpcurproc;	/* current owner of the FPU */
+	int ci_fpsaving;		/* save in progress */
+
+	volatile u_int32_t ci_tlb_ipi_mask;
+
+	struct pcb *ci_curpcb;		/* VA of current HW PCB */
+	struct pcb *ci_idle_pcb;	/* VA of current PCB */
+	int ci_idle_tss_sel;		/* TSS selector of idle PCB */
+
+	struct intrsource *ci_isources[MAX_INTR_SOURCES];
+	u_int32_t	ci_ipending;
+	int		ci_ilevel;
+	int		ci_idepth;
+	u_int32_t	ci_imask[NIPL];
+	u_int32_t	ci_iunmask[NIPL];
+
+	paddr_t ci_idle_pcb_paddr;	/* PA of idle PCB */
+	u_long ci_flags;		/* flags; see below */
+	u_int32_t ci_ipis; 		/* interprocessor interrupts pending */
+	int sc_apic_version;  		/* local APIC version */
+	
+	u_int32_t	ci_level;
+	u_int32_t	ci_vendor[4];
+	u_int32_t	ci_signature;		/* X86 cpuid type */
+	u_int32_t	ci_feature_flags;	/* X86 CPUID feature bits */
+	u_int32_t	cpu_class;		/* CPU class */
+
+	struct cpu_functions *ci_func;	/* start/stop functions */
+	void (*cpu_setup)(const char *, int, int);	/* proc-dependant init */
+
+	int		ci_want_resched;
+	int		ci_astpending;
+
+	union descriptor *ci_gdt;
+
+	volatile int ci_ddb_paused;	/* paused due to other proc in ddb */
+#define CI_DDB_RUNNING		0
+#define CI_DDB_SHOULDSTOP	1
+#define CI_DDB_STOPPED		2
+#define CI_DDB_ENTERDDB		3
+#define CI_DDB_INDDB		4
+};
+
+/*
+ * Processor flag notes: The "primary" CPU has certain MI-defined
+ * roles (mostly relating to hardclock handling); we distinguish
+ * betwen the processor which booted us, and the processor currently
+ * holding the "primary" role just to give us the flexibility later to
+ * change primaries should we be sufficiently twisted.  
+ */
+
+#define	CPUF_BSP	0x0001		/* CPU is the original BSP */
+#define	CPUF_AP		0x0002		/* CPU is an AP */
+#define	CPUF_SP		0x0004		/* CPU is only processor */
+#define	CPUF_PRIMARY	0x0008		/* CPU is active primary processor */
+#define	CPUF_APIC_CD	0x0010		/* CPU has apic configured */
+
+#define	CPUF_PRESENT	0x1000		/* CPU is present */
+#define	CPUF_RUNNING	0x2000		/* CPU is running */
 
 /*
- * Preempt the current process if in interrupt from user mode,
+ * We statically allocate the CPU info for the primary CPU (or,
+ * the only CPU on uniprocessors), and the primary CPU is the
+ * first CPU on the CPU info list.
+ */
+extern struct cpu_info cpu_info_primary;
+extern struct cpu_info *cpu_info_list;
+
+#define	CPU_INFO_ITERATOR		int
+#define	CPU_INFO_FOREACH(cii, ci)	cii = 0, ci = cpu_info_list; \
+					ci != NULL; ci = ci->ci_next
+
+#ifdef MULTIPROCESSOR
+
+#define I386_MAXPROCS		32	/* because we use a bitmask */
+
+#define CPU_STARTUP(_ci)	((_ci)->ci_func->start(_ci))
+#define CPU_STOP(_ci)		((_ci)->ci_func->stop(_ci))
+#define CPU_START_CLEANUP(_ci)	((_ci)->ci_func->cleanup(_ci))
+
+#define cpu_number()		(i82489_readreg(LAPIC_ID)>>LAPIC_ID_SHIFT)
+#define	curcpu()		(cpu_info[cpu_number()])
+
+#define CPU_IS_PRIMARY(ci)	((ci)->ci_flags & CPUF_PRIMARY)
+
+extern struct cpu_info	*cpu_info[I386_MAXPROCS];
+extern u_long		 cpus_running;
+
+extern void cpu_boot_secondary_processors(void);
+extern void cpu_init_idle_pcbs(void);
+
+#else /* MULTIPROCESSOR */
+
+#define I386_MAXPROCS		1
+
+#define cpu_number()		0
+#define	curcpu()		(&cpu_info_primary)
+
+#define CPU_IS_PRIMARY(ci)	1
+
+/*
+ * definitions of cpu-dependent requirements
+ * referenced in generic code
+ */
+#define	cpu_swapin(p)			/* nothing */
+
+#endif
+
+#define curpcb			curcpu()->ci_curpcb
+
+#define want_resched (curcpu()->ci_want_resched)
+#define astpending (curcpu()->ci_astpending)
+
+/*
+ * Preemt the current process if in interrupt from user monre,
  * or after the current trap/syscall if in system mode.
  */
-int	want_resched;		/* resched() was called */
-#define	need_resched()		(want_resched = 1, setsoftast())
+extern void need_resched(struct cpu_info *);
+
+#define	CLKF_USERMODE(frame)	USERMODE((frame)->if_cs, (frame)->if_eflags)
+#define	CLKF_PC(frame)		((frame)->if_eip)
+#define	CLKF_INTR(frame)	(IDXSEL((frame)->if_cs) == GICODE_SEL)
 
 /*
  * Give a profiling tick to the current process when the user profiling
@@ -87,8 +237,13 @@ int	want_resched;		/* resched() was called */
 /*
  * We need a machine-independent name for this.
  */
-#define	DELAY(x)		delay(x)
-void	delay(int);
+extern void (*delay_func)(int);
+struct timeval;
+extern void (*microtime_func)(struct timeval *);
+
+#define	DELAY(x)		(*delay_func)(x)
+#define delay(x)		(*delay_func)(x)
+#define microtime(tv)		(*microtime_func)(tv)
 
 #if defined(I586_CPU) || defined(I686_CPU)
 /*
@@ -173,6 +328,7 @@ extern void (*update_cpuspeed)(void);
 void	dumpconf(void);
 void	cpu_reset(void);
 void	i386_proc0_tss_ldt_init(void);
+void	i386_init_pcb_tss_ldt(struct cpu_info *);
 void	cpuid(u_int32_t, u_int32_t *);
 
 /* locore.s */
@@ -189,6 +345,9 @@ void	proc_trampoline(void);
 void	initrtclock(void);
 void	startrtclock(void);
 void	rtcdrain(void *);
+void	i8254_delay(int);
+void	i8254_microtime(struct timeval *);
+void	i8254_initclocks(void);
 
 /* est.c */
 #if !defined(SMALL_KERNEL) && defined(I686_CPU)
@@ -213,8 +372,9 @@ int	k6_powernow_setperf(int);
 
 
 /* npx.c */
-void	npxdrop(void);
-void	npxsave(void);
+void	npxdrop(struct proc *);
+void	npxsave_proc(struct proc *, int);
+void	npxsave_cpu(struct cpu_info *, int);
 
 #if defined(GPL_MATH_EMULATE)
 /* math_emulate.c */
@@ -230,6 +390,7 @@ int	i386_set_ldt(struct proc *, void *, register_t *);
 
 /* isa_machdep.c */
 void	isa_defaultirq(void);
+void	isa_nodefaultirq(void);
 int	isa_nmi(void);
 
 /* pmap.c */
@@ -291,4 +452,12 @@ void	setconf(void);
 	{ "xcrypt", CTLTYPE_INT }, \
 }
 
+/*
+ * This needs to be included late since it relies on definitions higher
+ * up in this file.
+ */
+#if defined(MULTIPROCESSOR) && defined(_KERNEL)
+#include <sys/mplock.h>
+#endif
+
 #endif /* !_I386_CPU_H_ */
diff --git a/sys/arch/i386/include/cpufunc.h b/sys/arch/i386/include/cpufunc.h
index d90a120429b..832e4271142 100644
--- a/sys/arch/i386/include/cpufunc.h
+++ b/sys/arch/i386/include/cpufunc.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpufunc.h,v 1.11 2003/10/28 13:22:44 avsm Exp $	*/
+/*	$OpenBSD: cpufunc.h,v 1.12 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: cpufunc.h,v 1.8 1994/10/27 04:15:59 cgd Exp $	*/
 
 /*
@@ -43,6 +43,8 @@
 #include <sys/cdefs.h>
 #include <sys/types.h>
 
+#include <machine/specialreg.h>
+
 static __inline void invlpg(u_int);
 static __inline void lidt(void *);
 static __inline void lldt(u_short);
@@ -55,6 +57,7 @@ static __inline u_int rcr3(void);
 static __inline void lcr4(u_int);
 static __inline u_int rcr4(void);
 static __inline void tlbflush(void);
+static __inline void tlbflushg(void);
 static __inline void disable_intr(void);
 static __inline void enable_intr(void);
 static __inline u_int read_eflags(void);
@@ -146,6 +149,39 @@ tlbflush(void)
 	__asm __volatile("movl %0,%%cr3" : : "r" (val));
 }
 
+static __inline void
+tlbflushg(void)
+{
+	/*
+	 * Big hammer: flush all TLB entries, including ones from PTE's
+	 * with the G bit set.  This should only be necessary if TLB
+	 * shootdown falls far behind.
+	 *
+	 * Intel Architecture Software Developer's Manual, Volume 3,
+	 *	System Programming, section 9.10, "Invalidating the
+	 * Translation Lookaside Buffers (TLBS)":
+	 * "The following operations invalidate all TLB entries, irrespective
+	 * of the setting of the G flag:
+	 * ...
+	 * "(P6 family processors only): Writing to control register CR4 to
+	 * modify the PSE, PGE, or PAE flag."
+	 *
+	 * (the alternatives not quoted above are not an option here.)
+	 *
+	 * If PGE is not in use, we reload CR3 for the benefit of
+	 * pre-P6-family processors.
+	 */
+
+#if defined(I686_CPU)
+	if (cpu_feature & CPUID_PGE) {
+		u_int cr4 = rcr4();
+		lcr4(cr4 & ~CR4_PGE);
+		lcr4(cr4);
+	} else
+#endif
+		tlbflush();
+}
+
 #ifdef notyet
 void	setidt(int idx, /*XXX*/caddr_t func, int typ, int dpl);
 #endif
diff --git a/sys/arch/i386/include/cpuvar.h b/sys/arch/i386/include/cpuvar.h
new file mode 100644
index 00000000000..5bf9f3342ae
--- /dev/null
+++ b/sys/arch/i386/include/cpuvar.h
@@ -0,0 +1,108 @@
+/*	$OpenBSD: cpuvar.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: cpuvar.h,v 1.1.2.3 2000/02/21 18:54:07 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Copyright (c) 1999 Stefan Grefen
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed by the NetBSD
+ *      Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.  
+ *
+ * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+struct cpu_functions {
+    int (*start)(struct cpu_info *);
+    int (*stop)(struct cpu_info *);
+    void (*cleanup)(struct cpu_info *);
+};
+
+extern struct cpu_functions mp_cpu_funcs;
+
+#define CPU_ROLE_SP	0
+#define CPU_ROLE_BP	1
+#define CPU_ROLE_AP	2
+
+struct cpu_attach_args {
+	const char *caa_name;
+	int cpu_number;
+	int cpu_role;
+	struct cpu_functions *cpu_func;
+	int cpu_signature;
+	int feature_flags;
+};
+
+#define MP_PICMODE	0x00000001      /* System booted in picmode */
+
+#ifdef _KERNEL
+
+int i386_ipi(int,int,int);
+void i386_self_ipi(int);
+int i386_ipi_init(int);
+
+void identifycpu(struct cpu_info *);
+void cpu_init(struct cpu_info *);
+void cpu_init_first(void);
+
+#endif
+
diff --git a/sys/arch/i386/include/db_machdep.h b/sys/arch/i386/include/db_machdep.h
index 69fb1d2cb30..c1f5b9b6dfe 100644
--- a/sys/arch/i386/include/db_machdep.h
+++ b/sys/arch/i386/include/db_machdep.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_machdep.h,v 1.9 2003/05/18 02:43:13 andreas Exp $	*/
+/*	$OpenBSD: db_machdep.h,v 1.10 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: db_machdep.h,v 1.9 1996/05/03 19:23:59 christos Exp $	*/
 
 /* 
@@ -120,5 +120,13 @@ void		db_task_name(/* task_t */);
 int kdb_trap(int, int, db_regs_t *);
 
 void db_machine_init(void);
+void db_enter_ddb(void);
+void db_leave_ddb(void);
+void db_startcpu(int cpu);
+void db_stopcpu(int cpu);
+void db_movetocpu(int cpu);
+void i386_ipi_db(struct cpu_info *);
+
+extern struct SIMPLELOCK ddb_mp_slock;
 
 #endif	/* _I386_DB_MACHDEP_H_ */
diff --git a/sys/arch/i386/include/gdt.h b/sys/arch/i386/include/gdt.h
index 1568b573a86..ba741e2c637 100644
--- a/sys/arch/i386/include/gdt.h
+++ b/sys/arch/i386/include/gdt.h
@@ -1,8 +1,8 @@
-/*	$OpenBSD: gdt.h,v 1.9 2002/03/14 01:26:33 millert Exp $	*/
-/*	$NetBSD: gdt.h,v 1.3 1996/02/27 22:32:11 jtc Exp $	*/
+/*	$OpenBSD: gdt.h,v 1.10 2004/06/13 21:49:16 niklas Exp $	*/
+/*	$NetBSD: gdt.h,v 1.7.10.6 2002/08/19 01:22:36 sommerfeld Exp $	*/
 
 /*-
- * Copyright (c) 1996 The NetBSD Foundation, Inc.
+ * Copyright (c) 1996, 1997 The NetBSD Foundation, Inc.
  * All rights reserved.
  *
  * This code is derived from software contributed to The NetBSD Foundation
@@ -37,10 +37,35 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifdef _KERNEL
+#ifndef _LOCORE
+
+struct cpu_info;
+struct pcb;
+struct pmap;
+union descriptor;
+
+void gdt_alloc_cpu(struct cpu_info *);
+int gdt_get_slot(void);
 void gdt_init(void);
-void tss_alloc(struct pcb *);
-void tss_free(struct pcb *);
+void gdt_init_cpu(struct cpu_info *);
+void gdt_reload_cpu(/* XXX struct cpu_info * */ void);
 void ldt_alloc(struct pmap *, union descriptor *, size_t);
 void ldt_free(struct pmap *);
+int tss_alloc(struct pcb *);
+void tss_free(int);
+void setgdt(int, void *, size_t, int, int, int, int);
 #endif
+
+/*
+ * The initial GDT size (as a descriptor count), and the maximum
+ * GDT size possible.
+ *
+ * These are actually not arbitrary.  To start with, they have to be
+ * multiples of 512 and at least 512, in order to work with the
+ * allocation strategy set forth by gdt_init and gdt_grow.  Then, the
+ * max cannot exceed 65536 since the selector field of a descriptor is
+ * just 16 bits, and used as free list link.
+ */
+
+#define MINGDTSIZ 512
+#define MAXGDTSIZ 8192
diff --git a/sys/arch/i386/include/i82093reg.h b/sys/arch/i386/include/i82093reg.h
new file mode 100644
index 00000000000..07ec03d8991
--- /dev/null
+++ b/sys/arch/i386/include/i82093reg.h
@@ -0,0 +1,124 @@
+/*	$OpenBSD: i82093reg.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: i82093reg.h,v 1.1.2.2 2000/02/21 18:54:07 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Typically, the first apic lives here.
+ */
+#define IOAPIC_BASE_DEFAULT	0xfec00000
+
+/*
+ * Memory-space registers.
+ */
+
+/*
+ * The externally visible registers are all 32 bits wide;
+ * store the register number of interest in IOAPIC_REG, and store/fetch
+ * the real value in IOAPIC_DATA.
+ */
+   
+
+
+#define	IOAPIC_REG		0x0000
+#define IOAPIC_DATA		0x0010
+
+/*
+ * Internal I/O APIC registers.
+ */
+
+#define IOAPIC_ID		0x00
+
+#define 	IOAPIC_ID_SHIFT		24
+#define		IOAPIC_ID_MASK		0x0f000000
+
+/* Version, and maximum interrupt pin number. */
+  
+#define IOAPIC_VER		0x01
+
+#define		IOAPIC_VER_SHIFT		0
+#define		IOAPIC_VER_MASK			0x000000ff
+
+#define		IOAPIC_MAX_SHIFT	       	16
+#define		IOAPIC_MAX_MASK	       	0x00ff0000
+
+/*
+ * Arbitration ID.  Same format as IOAPIC_ID register.
+ */
+#define IOAPIC_ARB		0x02
+
+/*
+ * Redirection table registers.
+ */
+
+#define IOAPIC_REDHI(pin)	(0x11 + ((pin)<<1))
+#define IOAPIC_REDLO(pin)	(0x10 + ((pin)<<1))
+
+#define IOAPIC_REDHI_DEST_SHIFT		24	   /* destination. */
+#define IOAPIC_REDHI_DEST_MASK		0xff000000
+
+#define IOAPIC_REDLO_MASK		0x00010000 /* 0=enabled; 1=masked */
+
+#define IOAPIC_REDLO_LEVEL		0x00008000 /* 0=edge, 1=level */
+#define IOAPIC_REDLO_RIRR		0x00004000 /* remote IRR; read only */
+#define IOAPIC_REDLO_ACTLO		0x00002000 /* 0=act. hi; 1=act. lo */
+#define IOAPIC_REDLO_DELSTS		0x00001000 /* 0=idle; 1=send pending */
+#define IOAPIC_REDLO_DSTMOD		0x00000800 /* 0=physical; 1=logical */
+
+#define IOAPIC_REDLO_DEL_MASK		0x00000700 /* del. mode mask */
+#define IOAPIC_REDLO_DEL_SHIFT		8
+
+#define IOAPIC_REDLO_DEL_FIXED		0
+#define IOAPIC_REDLO_DEL_LOPRI		1
+#define IOAPIC_REDLO_DEL_SMI		2
+#define IOAPIC_REDLO_DEL_NMI		4
+#define IOAPIC_REDLO_DEL_INIT		5
+#define IOAPIC_REDLO_DEL_EXTINT		7
+
+#define IOAPIC_REDLO_VECTOR_MASK	0x000000ff /* delivery vector */
+
+#define IMCR_ADDR		0x22
+#define IMCR_DATA		0x23
+
+#define IMCR_REGISTER		0x70
+#define		IMCR_PIC	0x00
+#define 	IMCR_APIC	0x01
+
+#define ioapic_asm_ack(num) \
+    movl	$0,_C_LABEL(local_apic) + LAPIC_EOI
diff --git a/sys/arch/i386/include/i82093var.h b/sys/arch/i386/include/i82093var.h
new file mode 100644
index 00000000000..c3d7ce586f3
--- /dev/null
+++ b/sys/arch/i386/include/i82093var.h
@@ -0,0 +1,103 @@
+/*	$OpenBSD: i82093var.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: i82093var.h,v 1.1 2003/02/26 21:26:10 fvdl Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _I386_I82093VAR_H_
+#define _I386_I82093VAR_H_
+
+#include <machine/apicvar.h>
+
+struct ioapic_pin 
+{
+	struct intrhand		*ip_handler; 
+	struct ioapic_pin	*ip_next; /* next pin on this vector */
+	struct mp_intr_map 	*ip_map;
+	int			ip_vector; /* IDT vector */
+	int			ip_type;
+	int			ip_minlevel;
+	int			ip_maxlevel;
+};
+
+struct ioapic_softc {
+	struct 			device sc_dev;	/* generic device glue */
+	struct ioapic_softc	*sc_next;
+	int			sc_apicid;
+	int			sc_apic_vers;
+	int			sc_apic_sz;	/* apic size*/
+	int			sc_flags;
+	paddr_t			sc_pa;		/* PA of ioapic */
+	volatile u_int32_t	*sc_reg;	/* KVA of ioapic addr */
+	volatile u_int32_t	*sc_data;	/* KVA of ioapic data */
+	struct ioapic_pin	*sc_pins;	/* sc_apic_sz entries */
+};      
+
+/*
+ * MP: intr_handle_t is bitfielded.
+ * ih&0xff -> line number.
+ * ih&0x10000000 -> if 0, old-style isa irq; if 1, routed via ioapic.
+ * (ih&0xff0000)>>16 -> ioapic id.
+ * (ih&0x00ff00)>>8 -> ioapic line.
+ */
+
+#define APIC_INT_VIA_APIC	0x10000000
+#define APIC_INT_APIC_MASK	0x00ff0000
+#define APIC_INT_APIC_SHIFT	16
+#define APIC_INT_PIN_MASK	0x0000ff00
+#define APIC_INT_PIN_SHIFT	8
+
+#define APIC_IRQ_APIC(x) ((x & APIC_INT_APIC_MASK) >> APIC_INT_APIC_SHIFT)
+#define APIC_IRQ_PIN(x) ((x & APIC_INT_PIN_MASK) >> APIC_INT_PIN_SHIFT)
+
+void   *apic_intr_establish(int, int, int, int (*)(void *), void *, char *); 
+void	apic_intr_disestablish(void *);
+
+void	ioapic_print_redir(struct ioapic_softc *, char *, int);
+void	ioapic_format_redir(char *, char *, int, u_int32_t, u_int32_t);
+struct ioapic_softc *ioapic_find(int);
+struct ioapic_softc *ioapic_find_bybase(int);
+
+void	ioapic_enable(void);
+void	lapic_vectorset(void); /* XXX */
+
+extern int ioapic_bsp_id;
+extern int nioapics;
+extern struct ioapic_softc *ioapics;
+
+#endif /* !_I386_I82093VAR_H_ */
diff --git a/sys/arch/i386/include/i82489reg.h b/sys/arch/i386/include/i82489reg.h
new file mode 100644
index 00000000000..9b850519370
--- /dev/null
+++ b/sys/arch/i386/include/i82489reg.h
@@ -0,0 +1,150 @@
+/*	$OpenBSD: i82489reg.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/*	$NetBSD: i82489reg.h,v 1.1.2.1 2000/02/20 16:30:27 sommerfeld Exp $	*/
+
+/*-
+ * Copyright (c) 1998 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Frank van der Linden.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by the NetBSD
+ *	Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+
+/*
+ * Registers and constants for the 82489DX and Pentium (and up) integrated
+ * "local" APIC.
+ */
+
+#define	LAPIC_ID		0x020		/* ID. RW */
+#	define LAPIC_ID_MASK		0x0f000000
+#	define LAPIC_ID_SHIFT		24
+
+#define LAPIC_VERS		0x030		/* Version. R */
+#	define LAPIC_VERSION_MASK	0x000000ff
+#	define LAPIC_VERSION_LVT_MASK	0x00ff0000
+#	define LAPIC_VERSION_LVT_SHIFT	16
+
+#define LAPIC_TPRI		0x080		/* Task Prio. RW */
+#	define LAPIC_TPRI_MASK		0x000000ff
+#	define LAPIC_TPRI_INT_MASK	0x000000f0
+#	define LAPIC_TPRI_SUB_MASK	0x0000000f
+
+#define LAPIC_APRI		0x090		/* Arbitration prio R */
+#	define LAPIC_APRI_MASK		0x000000ff
+
+#define LAPIC_PPRI		0x0a0		/* Processor prio. R */
+#define LAPIC_EOI		0x0b0		/* End Int. W */
+#define LAPIC_RRR		0x0c0		/* Remote read R */
+#define LAPIC_LDR		0x0d0		/* Logical dest. RW */
+#define LAPIC_DFR		0x0e0		/* Dest. format RW */
+
+#define LAPIC_SVR		0x0f0		/* Spurious intvec RW */
+#	define LAPIC_SVR_VECTOR_MASK	0x000000ff
+#	define LAPIC_SVR_VEC_FIX	0x0000000f
+#	define LAPIC_SVR_VEC_PROG	0x000000f0
+#	define LAPIC_SVR_ENABLE		0x00000100
+#	define LAPIC_SVR_SWEN		0x00000100
+#	define LAPIC_SVR_FOCUS		0x00000200
+#	define LAPIC_SVR_FDIS		0x00000200
+
+#define LAPIC_ISR	0x100			/* Int. status. R */
+#define LAPIC_TMR	0x180
+#define LAPIC_IRR	0x200
+#define LAPIC_ESR	0x280			/* Err status. R */
+
+#define LAPIC_ICRLO	0x300			/* Int. cmd. RW */
+#	define LAPIC_DLMODE_MASK	0x00000700
+#	define LAPIC_DLMODE_FIXED	0x00000000
+#	define LAPIC_DLMODE_LOW		0x00000100
+#	define LAPIC_DLMODE_SMI		0x00000200
+#	define LAPIC_DLMODE_RR		0x00000300
+#	define LAPIC_DLMODE_NMI		0x00000400
+#	define LAPIC_DLMODE_INIT	0x00000500
+#	define LAPIC_DLMODE_STARTUP	0x00000600
+
+#	define LAPIC_DSTMODE_LOG	0x00000800
+
+#	define LAPIC_DLSTAT_BUSY	0x00001000
+
+#	define LAPIC_LVL_ASSERT		0x00004000
+#	define LAPIC_LVL_DEASSERT	0x00000000
+
+#	define LAPIC_LVL_TRIG		0x00008000
+
+#	define LAPIC_RRSTAT_MASK	0x00030000
+#	define LAPIC_RRSTAT_INPROG	0x00010000
+#	define LAPIC_RRSTAT_VALID	0x00020000
+
+#	define LAPIC_DEST_MASK		0x000c0000
+#	define LAPIC_DEST_SELF		0x00040000
+#	define LAPIC_DEST_ALLINCL	0x00080000
+#	define LAPIC_DEST_ALLEXCL	0x000c0000
+
+#	define LAPIC_RESV2_MASK		0xfff00000
+
+
+#define LAPIC_ICRHI	0x310			/* Int. cmd. RW */
+#	define LAPIC_ID_MASK		0x0f000000
+#	define LAPIC_ID_SHIFT		24
+
+#define LAPIC_LVTT	0x320			/* Loc.vec.(timer) RW */
+#	define LAPIC_LVTT_VEC_MASK	0x000000ff
+#	define LAPIC_LVTT_DS		0x00001000
+#	define LAPIC_LVTT_M		0x00010000
+#	define LAPIC_LVTT_TM		0x00020000
+
+#define LAPIC_PCINT	0x340
+#define LAPIC_LVINT0	0x350			/* Loc.vec (LINT0) RW */
+#	define LAPIC_LVT_PERIODIC	0x00020000
+#	define LAPIC_LVT_MASKED		0x00010000
+#	define LAPIC_LVT_LEVTRIG	0x00008000
+#	define LAPIC_LVT_REMOTE_IRR	0x00004000
+#	define LAPIC_INP_POL		0x00002000
+#	define LAPIC_PEND_SEND		0x00001000
+
+#define LAPIC_LVINT1	0x360			/* Loc.vec (LINT1) RW */
+#define LAPIC_LVERR	0x370			/* Loc.vec (ERROR) RW */
+#define LAPIC_ICR_TIMER	0x380			/* Initial count RW */
+#define LAPIC_CCR_TIMER	0x390			/* Current count RO */
+
+#define LAPIC_DCR_TIMER	0x3e0			/* Divisor config register */
+#	define LAPIC_DCRT_DIV1		0x0b
+#	define LAPIC_DCRT_DIV2		0x00
+#	define LAPIC_DCRT_DIV4		0x01
+#	define LAPIC_DCRT_DIV8		0x02
+#	define LAPIC_DCRT_DIV16		0x03
+#	define LAPIC_DCRT_DIV32		0x08
+#	define LAPIC_DCRT_DIV64		0x09
+#	define LAPIC_DCRT_DIV128	0x0a
+
+#define LAPIC_BASE		0xfee00000
+
+#define LAPIC_IRQ_MASK(i)	(1 << ((i) + 1))
diff --git a/sys/arch/i386/include/i82489var.h b/sys/arch/i386/include/i82489var.h
new file mode 100644
index 00000000000..3c4632872ce
--- /dev/null
+++ b/sys/arch/i386/include/i82489var.h
@@ -0,0 +1,113 @@
+/*	$OpenBSD: i82489var.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/*	$NetBSD: i82489var.h,v 1.1.2.2 2000/02/21 18:46:14 sommerfeld Exp $	*/
+
+/*-
+ * Copyright (c) 1998 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Frank van der Linden.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by the NetBSD
+ *	Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _I386_I82489VAR_H_
+#define _I386_I82489VAR_H_
+
+static __inline__ u_int32_t i82489_readreg(int);
+static __inline__ void i82489_writereg(int, u_int32_t);
+
+#ifdef _KERNEL
+extern volatile u_int32_t local_apic[];
+#endif
+
+static __inline__ u_int32_t
+i82489_readreg(reg)
+	int reg;
+{
+	return *((volatile u_int32_t *)(((volatile u_int8_t *)local_apic)
+	    + reg));
+}
+
+static __inline__ void
+i82489_writereg(reg, val)
+	int reg;
+	u_int32_t val;
+{
+	*((volatile u_int32_t *)(((volatile u_int8_t *)local_apic) + reg)) =
+	    val;
+}
+
+/*
+ * "spurious interrupt vector"; vector used by interrupt which was
+ * aborted because the CPU masked it after it happened but before it
+ * was delivered.. "Oh, sorry, i caught you at a bad time".
+ * Low-order 4 bits must be all ones.
+ */
+extern void i386_spurious(void);
+extern void Xintrspurious(void);
+#define LAPIC_SPURIOUS_VECTOR		0xef
+
+/*
+ * Vector used for inter-processor interrupts.
+ */
+extern void Xintripi(void);
+#define LAPIC_IPI_VECTOR		IPL_IPI
+
+/*
+ * Vector used for local apic timer interrupts.
+ */
+
+extern void Xintrltimer(void);
+#define LAPIC_TIMER_VECTOR		IPL_CLOCK
+
+/*
+ * Vectors to be used for self-soft-interrupts.
+ */
+
+#define LAPIC_SOFTCLOCK_VECTOR		IPL_SOFTCLOCK
+#define LAPIC_SOFTNET_VECTOR		IPL_SOFTNET
+#define LAPIC_SOFTTTY_VECTOR		IPL_SOFTTTY
+
+extern void Xintrsoftclock(void);
+extern void Xintrsoftnet(void);
+extern void Xintrsofttty(void);
+
+extern void (*apichandler[])(void);
+
+struct cpu_info;
+
+extern void lapic_boot_init(paddr_t);
+extern void lapic_initclocks(void);
+extern void lapic_set_lvt(void);
+extern void lapic_set_softvectors(void);
+extern void lapic_enable(void);
+extern void lapic_calibrate_timer(struct cpu_info *);
+
+#endif
diff --git a/sys/arch/i386/include/i8259.h b/sys/arch/i386/include/i8259.h
new file mode 100644
index 00000000000..f1c4462e4d0
--- /dev/null
+++ b/sys/arch/i386/include/i8259.h
@@ -0,0 +1,154 @@
+/*	$OpenBSD: i8259.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/*	$NetBSD: i8259.h,v 1.3 2003/05/04 22:01:56 fvdl Exp $	*/
+
+/*-
+ * Copyright (c) 1990 The Regents of the University of California.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to Berkeley by
+ * William Jolitz.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *	@(#)icu.h	5.6 (Berkeley) 5/9/91
+ */
+
+#ifndef	_I386_I8259_H_
+#define	_I386_I8259_H_
+
+#include <dev/isa/isareg.h>
+
+#ifndef	_LOCORE
+
+/*
+ * Interrupt "level" mechanism variables, masks, and macros
+ */
+extern	unsigned imen;		/* interrupt mask enable */
+extern unsigned i8259_setmask(unsigned);
+
+#define SET_ICUS()	(outb(IO_ICU1 + 1, imen), outb(IO_ICU2 + 1, imen >> 8))
+
+extern void i8259_default_setup(void);
+extern void i8259_reinit(void);
+
+#endif /* !_LOCORE */
+
+/*
+ * Interrupt enable bits -- in order of priority
+ */
+#define	IRQ_SLAVE	2
+
+/*
+ * Interrupt Control offset into Interrupt descriptor table (IDT)
+ */
+#define	ICU_OFFSET	32		/* 0-31 are processor exceptions */
+#define	ICU_LEN		16		/* 32-47 are ISA interrupts */
+
+
+#define ICU_HARDWARE_MASK
+
+/*
+ * These macros are fairly self explanatory.  If ICU_SPECIAL_MASK_MODE is
+ * defined, we try to take advantage of the ICU's `special mask mode' by only
+ * EOIing the interrupts on return.  This avoids the requirement of masking and
+ * unmasking.  We can't do this without special mask mode, because the ICU
+ * would also hold interrupts that it thinks are of lower priority.
+ *
+ * Many machines do not support special mask mode, so by default we don't try
+ * to use it.
+ */
+
+#define	IRQ_BIT(num)	(1 << ((num) % 8))
+#define	IRQ_BYTE(num)	((num) >> 3)
+
+#define i8259_late_ack(num)
+
+#ifdef ICU_SPECIAL_MASK_MODE
+
+#define	i8259_asm_ack1(num)
+#define	i8259_asm_ack2(num) \
+	movb	$(0x60|IRQ_SLAVE),%al	/* specific EOI for IRQ2 */	;\
+	outb	%al,$IO_ICU1
+#define	i8259_asm_mask(num)
+#define	i8259_asm_unmask(num) \
+	movb	$(0x60|(num%8)),%al	/* specific EOI */		;\
+	outb	%al,$ICUADDR
+
+#else /* ICU_SPECIAL_MASK_MODE */
+
+#ifndef	AUTO_EOI_1
+#define	i8259_asm_ack1(num) \
+	movb	$(0x60|(num%8)),%al	/* specific EOI */		;\
+	outb	%al,$IO_ICU1
+#else
+#define	i8259_asm_ack1(num)
+#endif
+
+#ifndef AUTO_EOI_2
+#define	i8259_asm_ack2(num) \
+	movb	$(0x60|(num%8)),%al	/* specific EOI */		;\
+	outb	%al,$IO_ICU2		/* do the second ICU first */	;\
+	movb	$(0x60|IRQ_SLAVE),%al	/* specific EOI for IRQ2 */	;\
+	outb	%al,$IO_ICU1
+#else
+#define	i8259_asm_ack2(num)
+#endif
+
+#ifndef DUMMY_NOPS
+#define PIC_MASKDELAY
+#endif
+
+#ifdef PIC_MASKDELAY
+#define MASKDELAY	pushl %eax ; inb $0x84,%al ; popl %eax
+#else
+#define MASKDELAY
+#endif
+
+#ifdef ICU_HARDWARE_MASK
+
+#define	i8259_asm_mask(num) \
+	movb	CVAROFF(imen, IRQ_BYTE(num)),%al			;\
+	orb	$IRQ_BIT(num),%al					;\
+	movb	%al,CVAROFF(imen, IRQ_BYTE(num))			;\
+	MASKDELAY							;\
+	outb	%al,$(ICUADDR+1)
+#define	i8259_asm_unmask(num) \
+	cli								;\
+	movb	CVAROFF(imen, IRQ_BYTE(num)),%al			;\
+	andb	$~IRQ_BIT(num),%al					;\
+	movb	%al,CVAROFF(imen, IRQ_BYTE(num))			;\
+	MASKDELAY							;\
+	outb	%al,$(ICUADDR+1)					;\
+	sti
+
+#else /* ICU_HARDWARE_MASK */
+
+#define	i8259_asm_mask(num)
+#define	i8259_asm_unmask(num)
+
+#endif /* ICU_HARDWARE_MASK */
+#endif /* ICU_SPECIAL_MASK_MODE */
+
+#endif /* !_X86_I8259_H_ */
diff --git a/sys/arch/i386/include/intr.h b/sys/arch/i386/include/intr.h
index 00ba2cea665..9d7a7816cf0 100644
--- a/sys/arch/i386/include/intr.h
+++ b/sys/arch/i386/include/intr.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: intr.h,v 1.20 2004/05/23 00:06:01 tedu Exp $	*/
+/*	$OpenBSD: intr.h,v 1.21 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: intr.h,v 1.5 1996/05/13 06:11:28 mycroft Exp $	*/
 
 /*
@@ -33,85 +33,31 @@
 #ifndef _I386_INTR_H_
 #define _I386_INTR_H_
 
-/*
- * Intel APICs (advanced programmable interrupt controllers) have
- * bytesized priority registers where the upper nibble is the actual
- * interrupt priority level (a.k.a. IPL).  Interrupt vectors are
- * closely tied to these levels as interrupts whose vectors' upper
- * nibble is lower than or equal to the current level are blocked.
- * Not all 256 possible vectors are available for interrupts in
- * APIC systems, only
- *
- * For systems where instead the older ICU (interrupt controlling
- * unit, a.k.a. PIC or 82C59) is used, the IPL is not directly useful,
- * since the interrupt blocking is handled via interrupt masks instead
- * of levels.  However the IPL is easily used as an offset into arrays
- * of masks.
- */
-#define IPLSHIFT 4	/* The upper nibble of vectors is the IPL.	*/
-#define NIPL 16		/* Four bits of information gives as much.	*/
-#define IPL(level) ((level) >> IPLSHIFT)	/* Extract the IPL.	*/
-/* XXX Maybe this IDTVECOFF definition should be elsewhere? */
-#define IDTVECOFF 0x20	/* The lower 32 IDT vectors are reserved.	*/
+#include <machine/intrdefs.h>
 
-/*
- * This macro is only defined for 0 <= x < 14, i.e. there are fourteen
- * distinct priority levels available for interrupts.
- */
-#define MAKEIPL(priority) (IDTVECOFF + ((priority) << IPLSHIFT))
+#ifndef _LOCORE
 
-/*
- * Interrupt priority levels.
- * XXX We are somewhat sloppy about what we mean by IPLs, sometimes
- * XXX we refer to the eight-bit value suitable for storing into APICs'
- * XXX priority registers, other times about the four-bit entity found
- * XXX in the former values' upper nibble, which can be used as offsets
- * XXX in various arrays of our implementation.  We are hoping that
- * XXX the context will provide enough information to not make this
- * XXX sloppy naming a real problem.
- */
-#define	IPL_NONE	0		/* nothing */
-#define	IPL_SOFTCLOCK	MAKEIPL(0)	/* timeouts */
-#define	IPL_SOFTNET	MAKEIPL(1)	/* protocol stacks */
-#define	IPL_BIO		MAKEIPL(2)	/* block I/O */
-#define	IPL_NET		MAKEIPL(3)	/* network */
-#define	IPL_SOFTTTY	MAKEIPL(4)	/* delayed terminal handling */
-#define	IPL_TTY		MAKEIPL(5)	/* terminal */
-#define	IPL_VM		MAKEIPL(6)	/* memory allocation */
-#define IPL_IMP		IPL_VM		/* XXX - should not be here. */
-#define	IPL_AUDIO	MAKEIPL(7)	/* audio */
-#define	IPL_CLOCK	MAKEIPL(8)	/* clock */
-#define	IPL_STATCLOCK	MAKEIPL(9)	/* statclock */
-#define	IPL_HIGH	MAKEIPL(9)	/* everything */
-
-/* Interrupt sharing types. */
-#define	IST_NONE	0	/* none */
-#define	IST_PULSE	1	/* pulsed */
-#define	IST_EDGE	2	/* edge-triggered */
-#define	IST_LEVEL	3	/* level-triggered */
-
-/* Soft interrupt masks. */
-#define	SIR_CLOCK	31
-#define	SIR_NET		30
-#define	SIR_TTY		29
+#ifdef MULTIPROCESSOR
+#include <machine/i82489reg.h>
+#include <machine/i82489var.h>
+#include <machine/cpu.h>
+#endif
 
-#ifndef _LOCORE
+extern volatile u_int32_t lapic_tpr;	/* Current interrupt priority level. */
 
-volatile int cpl;	/* Current interrupt priority level.		*/
-volatile int ipending;	/* Interrupts pending.				*/
-volatile int astpending;/* Asynchronous software traps (softints) pending. */
-int imask[NIPL];	/* Bitmasks telling what interrupts are blocked. */
-int iunmask[NIPL];	/* Bitmasks telling what interrupts are accepted. */
+extern volatile u_int32_t ipending;	/* Interrupts pending. */
+extern int imask[];	/* Bitmasks telling what interrupts are blocked. */
+extern int iunmask[];	/* Bitmasks telling what interrupts are accepted. */
 
 #define IMASK(level) imask[IPL(level)]
 #define IUNMASK(level) iunmask[IPL(level)]
 
 extern void Xspllower(void);
 
-int splraise(int);
-int spllower(int);
-void splx(int);
-void softintr(int);
+extern int splraise(int);
+extern int spllower(int);
+extern void splx(int);
+extern void softintr(int, int);
 
 /* SPL asserts */
 #ifdef DIAGNOSTIC
@@ -140,6 +86,7 @@ void splassert_check(int, const char *);
 #define	splaudio()	splraise(IPL_AUDIO)
 #define	splclock()	splraise(IPL_CLOCK)
 #define	splstatclock()	splhigh()
+#define splipi()	splraise(IPL_IPI)
 
 /*
  * Software interrupt masks
@@ -158,12 +105,40 @@ void splassert_check(int, const char *);
 #define	splvm()		splraise(IPL_VM)
 #define splimp()	splvm()
 #define	splhigh()	splraise(IPL_HIGH)
+#define	splsched()	splraise(IPL_SCHED)
+#define spllock() 	splhigh()
 #define	spl0()		spllower(IPL_NONE)
 
 #define	setsoftast()	(astpending = 1)
-#define	setsoftclock()	softintr(1 << SIR_CLOCK)
-#define	setsoftnet()	softintr(1 << SIR_NET)
-#define	setsofttty()	softintr(1 << SIR_TTY)
+#define	setsoftclock()	softintr(1 << SIR_CLOCK, IPL_SOFTCLOCK)
+#define	setsoftnet()	softintr(1 << SIR_NET, IPL_SOFTNET)
+#define	setsofttty()	softintr(1 << SIR_TTY, IPL_SOFTTTY)
+
+#define I386_IPI_HALT		0x00000001
+#define I386_IPI_MICROSET	0x00000002
+#define I386_IPI_FLUSH_FPU	0x00000004
+#define I386_IPI_SYNCH_FPU	0x00000008
+#define I386_IPI_TLB		0x00000010
+#define I386_IPI_MTRR		0x00000020
+#define I386_IPI_GDT		0x00000040
+#define I386_IPI_DDB		0x00000080	/* synchronize while in ddb */
+
+#define I386_NIPI	8
+
+struct cpu_info;
+
+#ifdef MULTIPROCESSOR
+int i386_send_ipi(struct cpu_info *, int);
+void i386_broadcast_ipi(int);
+void i386_multicast_ipi(int, int);
+void i386_ipi_handler(void);
+void i386_intlock(struct intrframe);
+void i386_intunlock(struct intrframe);
+void i386_softintlock(void);
+void i386_softintunlock(void);
+
+extern void (*ipifunc[I386_NIPI])(struct cpu_info *);
+#endif
 
 #endif /* !_LOCORE */
 
diff --git a/sys/arch/i386/include/intrdefs.h b/sys/arch/i386/include/intrdefs.h
new file mode 100644
index 00000000000..c2c998b76b3
--- /dev/null
+++ b/sys/arch/i386/include/intrdefs.h
@@ -0,0 +1,127 @@
+/*	$OpenBSD: intrdefs.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/*	$NetBSD: intrdefs.h,v 1.2 2003/05/04 22:01:56 fvdl Exp $	*/
+
+#ifndef _i386_INTRDEFS_H
+#define _i386_INTRDEFS_H
+
+/*
+ * Intel APICs (advanced programmable interrupt controllers) have
+ * bytesized priority registers where the upper nibble is the actual
+ * interrupt priority level (a.k.a. IPL).  Interrupt vectors are
+ * closely tied to these levels as interrupts whose vectors' upper
+ * nibble is lower than or equal to the current level are blocked.
+ * Not all 256 possible vectors are available for interrupts in
+ * APIC systems, only
+ *
+ * For systems where instead the older ICU (interrupt controlling
+ * unit, a.k.a. PIC or 82C59) is used, the IPL is not directly useful,
+ * since the interrupt blocking is handled via interrupt masks instead
+ * of levels.  However the IPL is easily used as an offset into arrays
+ * of masks.
+ */
+#define IPLSHIFT 4	/* The upper nibble of vectors is the IPL.	*/
+#define NIPL 16		/* Four bits of information gives as much.	*/
+#define IPL(level) ((level) >> IPLSHIFT)	/* Extract the IPL.	*/
+/* XXX Maybe this IDTVECOFF definition should be elsewhere? */
+#define IDTVECOFF 0x20	/* The lower 32 IDT vectors are reserved.	*/
+
+/*
+ * This macro is only defined for 0 <= x < 14, i.e. there are fourteen
+ * distinct priority levels available for interrupts.
+ */
+#define MAKEIPL(priority) (IDTVECOFF + ((priority) << IPLSHIFT))
+
+/*
+ * Interrupt priority levels.
+ *
+ * XXX We are somewhat sloppy about what we mean by IPLs, sometimes
+ * XXX we refer to the eight-bit value suitable for storing into APICs'
+ * XXX priority registers, other times about the four-bit entity found
+ * XXX in the former values' upper nibble, which can be used as offsets
+ * XXX in various arrays of our implementation.  We are hoping that
+ * XXX the context will provide enough information to not make this
+ * XXX sloppy naming a real problem.
+ * 
+ * There are tty, network and disk drivers that use free() at interrupt
+ * time, so imp > (tty | net | bio).
+ *
+ * Since run queues may be manipulated by both the statclock and tty,
+ * network, and disk drivers, clock > imp.
+ *
+ * IPL_HIGH must block everything that can manipulate a run queue.
+ *
+ * XXX Ultimately we may need serial drivers to run at the absolute highest
+ * XXX priority to avoid overruns, then we must make serial > high.
+ *
+ * The level numbers are picked to fit into APIC vector priorities.
+ */
+#define	IPL_NONE	0		/* nothing */
+#define	IPL_SOFTCLOCK	MAKEIPL(0)	/* timeouts */
+#define	IPL_SOFTNET	MAKEIPL(1)	/* protocol stacks */
+#define	IPL_BIO		MAKEIPL(2)	/* block I/O */
+#define	IPL_NET		MAKEIPL(3)	/* network */
+#define	IPL_SOFTTTY	MAKEIPL(4)	/* delayed terminal handling */
+#define	IPL_TTY		MAKEIPL(5)	/* terminal */
+#define	IPL_VM		MAKEIPL(6)	/* memory allocation */
+#define IPL_IMP		IPL_VM		/* XXX - should not be here. */
+#define	IPL_AUDIO	MAKEIPL(7)	/* audio */
+#define	IPL_CLOCK	MAKEIPL(8)	/* clock */
+#define	IPL_SCHED	IPL_CLOCK
+#define	IPL_STATCLOCK	MAKEIPL(9)	/* statclock */
+#define	IPL_HIGH	MAKEIPL(9)	/* everything */
+#define	IPL_IPI		MAKEIPL(10)	/* interprocessor interrupt */
+
+/* Interrupt sharing types. */
+#define	IST_NONE	0	/* none */
+#define	IST_PULSE	1	/* pulsed */
+#define	IST_EDGE	2	/* edge-triggered */
+#define	IST_LEVEL	3	/* level-triggered */
+
+/*
+ * Local APIC masks. Must not conflict with SIR_* below, and must
+ * be >= NUM_LEGACY_IRQs. Note that LIR_IPI must be first.
+ */
+#define LIR_IPI		31
+#define LIR_TIMER	30
+
+/* Soft interrupt masks. */
+#define	SIR_CLOCK	29
+#define	SIR_NET		28
+#define	SIR_TTY		27
+
+
+/*
+ * Maximum # of interrupt sources per CPU. 32 to fit in one word.
+ * ioapics can theoretically produce more, but it's not likely to
+ * happen. For multiple ioapics, things can be routed to different
+ * CPUs.
+ */
+#define MAX_INTR_SOURCES	32
+#define NUM_LEGACY_IRQS		16
+
+/*
+ * Low and high boundaries between which interrupt gates will
+ * be allocated in the IDT.
+ */
+#define IDT_INTR_LOW	(0x20 + NUM_LEGACY_IRQS)
+#define IDT_INTR_HIGH	0xef
+
+#define I386_IPI_HALT		0x00000001
+#define I386_IPI_MICROSET	0x00000002
+#define I386_IPI_FLUSH_FPU	0x00000004
+#define I386_IPI_SYNCH_FPU	0x00000008
+#define I386_IPI_TLB		0x00000010
+#define I386_IPI_MTRR		0x00000020
+#define I386_IPI_GDT		0x00000040
+#define I386_IPI_DDB		0x00000080	/* synchronize while in ddb */
+
+#define I386_NIPI	8
+
+#define I386_IPI_NAMES { "halt IPI", "timeset IPI", "FPU flush IPI", \
+			 "FPU synch IPI", "TLB shootdown IPI", \
+			 "MTRR update IPI", "GDT update IPI", \
+			 "DDB IPI" }
+
+#define IREENT_MAGIC	0x18041969
+
+#endif /* _I386_INTRDEFS_H */
diff --git a/sys/arch/i386/include/lock.h b/sys/arch/i386/include/lock.h
new file mode 100644
index 00000000000..a647999a532
--- /dev/null
+++ b/sys/arch/i386/include/lock.h
@@ -0,0 +1,119 @@
+/*	$OpenBSD: lock.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/*	$NetBSD: lock.h,v 1.1.2.2 2000/05/03 14:40:55 sommerfeld Exp $	*/
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Jason R. Thorpe.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by the NetBSD
+ *	Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * Machine-dependent spin lock operations.
+ */
+
+#ifndef _I386_LOCK_H_
+#define	_I386_LOCK_H_
+
+typedef	__volatile int		__cpu_simple_lock_t;
+
+#define	__SIMPLELOCK_LOCKED	1
+#define	__SIMPLELOCK_UNLOCKED	0
+
+/*
+ * compiler barrier: prevent reordering of instructions.
+ * XXX something similar will move to <sys/cdefs.h>
+ * or thereabouts.
+ * This prevents the compiler from reordering code around
+ * this "instruction", acting as a sequence point for code generation.
+ */
+
+#define __lockbarrier() __asm __volatile("": : :"memory")
+
+#ifdef LOCKDEBUG
+
+extern void __cpu_simple_lock_init(__cpu_simple_lock_t *);
+extern void __cpu_simple_lock(__cpu_simple_lock_t *);
+extern int __cpu_simple_lock_try(__cpu_simple_lock_t *);
+extern void __cpu_simple_unlock(__cpu_simple_lock_t *);
+
+#else
+
+#include <machine/atomic.h>
+
+static __inline void __cpu_simple_lock_init(__cpu_simple_lock_t *)
+	__attribute__((__unused__));
+static __inline void __cpu_simple_lock(__cpu_simple_lock_t *)
+	__attribute__((__unused__));
+static __inline int __cpu_simple_lock_try(__cpu_simple_lock_t *)
+	__attribute__((__unused__));
+static __inline void __cpu_simple_unlock(__cpu_simple_lock_t *)
+	__attribute__((__unused__));
+
+static __inline void
+__cpu_simple_lock_init(__cpu_simple_lock_t *lockp)
+{
+	*lockp = __SIMPLELOCK_UNLOCKED;
+	__lockbarrier();
+}
+
+static __inline void
+__cpu_simple_lock(__cpu_simple_lock_t *lockp)
+{
+	while (i386_atomic_testset_i(lockp, __SIMPLELOCK_LOCKED)
+	    == __SIMPLELOCK_LOCKED) {
+		continue;	/* spin */
+	}
+	__lockbarrier();
+}
+
+static __inline int
+__cpu_simple_lock_try(__cpu_simple_lock_t *lockp)
+{
+	int r = (i386_atomic_testset_i(lockp, __SIMPLELOCK_LOCKED)
+	    == __SIMPLELOCK_UNLOCKED);
+
+	__lockbarrier();
+
+	return (r);
+}
+
+static __inline void
+__cpu_simple_unlock(__cpu_simple_lock_t *lockp)
+{
+	__lockbarrier();
+	*lockp = __SIMPLELOCK_UNLOCKED;
+}
+
+#endif /* !LOCKDEBUG */
+
+#endif /* _I386_LOCK_H_ */
diff --git a/sys/arch/i386/include/mp.h b/sys/arch/i386/include/mp.h
new file mode 100644
index 00000000000..d12e8a54aca
--- /dev/null
+++ b/sys/arch/i386/include/mp.h
@@ -0,0 +1,222 @@
+/*	$OpenBSD: mp.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+
+/*-
+ * Copyright (c) 1996 SigmaSoft, Th. Lockert <tholo@sigmasoft.com>
+ * Copyright (c) 2000 Niklas Hallqvist.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by SigmaSoft, Th.  Lockert.
+ * 4. The name of the author may not be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES,
+ * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL
+ * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
+ * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _MACHINE_MP_H
+#define _MACHINE_MP_H
+
+/*
+ * Configuration structures as defined in the Intel MP specification,
+ * version 1.4
+ */
+
+/*
+ * MP Floating Pointer structure; must be located on 16-byte boundary
+ */
+struct mp_float {
+	u_int8_t	signature[4];
+#define	MPF_SIGNATURE	"_MP_"
+	u_int32_t	pointer;
+	u_int8_t	length;
+	u_int8_t	revision;
+	u_int8_t	checksum;
+	u_int8_t	feature1;
+#define	MP_CONF_EXTENDED	0
+#define	MP_CONF_2_ISA		1
+#define	MP_CONF_2_EISA_NO_8	2
+#define	MP_CONF_2_EISA		3
+#define	MP_CONF_2_MCA		4
+#define	MP_CONF_2_ISA_PCI	5
+#define	MP_CONF_2_EISA_PCI	6
+#define	MP_CONF_2_MCA_PCI	7
+	u_int8_t	feature2;
+#define	MP_IMCR		0x80
+	u_int8_t	feature3;
+	u_int8_t	feature4;
+	u_int8_t	feature5;
+};
+
+/*
+ * MP configuration table header
+ */
+struct mp_conf {
+	u_int8_t	signature[4];
+#define	MPC_SIGNATURE	"PCMP"
+	u_int16_t	length;
+	u_int8_t	revision;
+	u_int8_t	checksum;
+	u_int8_t	oem[8];
+	u_int8_t	product[12];
+	u_int32_t	oem_pointer;
+	u_int16_t	oem_length;
+	u_int16_t	entry_count;
+	u_int32_t	local_apic;
+	u_int16_t	ext_length;
+	u_int8_t	et_checksum;
+	u_int8_t	reserved;
+};
+
+/*
+ * Processor entry
+ */
+struct mp_proc {
+	u_int8_t	type;
+#define	MP_PROCESSOR	0
+	u_int8_t	local_apic;
+	u_int8_t	apic_version;
+	u_int8_t	flags;
+#define	MP_ENABLE	0x01
+#define	MP_BOOTCPU	0x02
+	u_int32_t	cpu_signature;
+#define	MP_STEPPING	0x0000000F
+#define	MP_MODEL	0x000000F0
+#define	MP_FAMILY	0x00000F00
+	u_int32_t	feature_flags;
+#define	MP_FP		0x00000001
+#define	MP_MCE		0x00000080
+#define	MP_CX8		0x00000100
+#define	MP_APIC		0x00000200
+	u_int32_t	reserved1;
+	u_int32_t	reserved2;
+};
+
+/*
+ * Bus entry
+ */
+struct mp_bus {
+	u_int8_t	type;
+#define	MP_BUS		1
+	u_int8_t	bus_id;
+	u_int8_t	bustype[6]	__attribute((packed));
+#define	MP_BUS_CBUS	"CBUS  "
+#define	MP_BUS_CBUSII	"CBUSII"
+#define	MP_BUS_EISA	"EISA  "
+#define	MP_BUS_FUTURE	"FUTURE"
+#define	MP_BUS_INTERN	"INTERN"
+#define	MP_BUS_ISA	"ISA   "
+#define	MP_BUS_MBI	"MBI   "
+#define	MP_BUS_MBII	"MBII  "
+#define	MP_BUS_MCA	"MCA   "
+#define	MP_BUS_MPI	"MPI   "
+#define	MP_BUS_MPSA	"MPSA  "
+#define	MP_BUS_NUBUS	"NUBUS "
+#define	MP_BUS_PCI	"PCI   "
+#define	MP_BUS_PCCARD	"PCMCIA"
+#define	MP_BUS_TC	"TC    "
+#define	MP_BUS_VLB	"VL    "
+#define	MP_BUS_VME	"VME   "
+#define	MP_BUS_XPRESS	"XPRESS"
+};
+
+/*
+ * I/O APIC entry
+ */
+struct mp_apic {
+	u_int8_t	type;
+#define	MP_IOAPIC	2
+	u_int8_t	apic_id;
+	u_int8_t	apic_version;
+	u_int8_t	apic_flags;
+#define	MP_APIC_ENABLE	0x80
+	u_int32_t	apic_address;
+};
+
+/*
+ * I/O Interrupt Assignment entry
+ * Local Interrupt Assignment entry
+ */
+struct mp_irq {
+	u_int8_t	type;
+#define	MP_INTSRC	3
+#define	MP_LOCINTSRC	4
+	u_int8_t	irqtype;
+#define	MP_INT_NORMAL	0
+#define	MP_INT_NMI	1
+#define	MP_INT_SMI	2
+#define	MP_INT_EXT	3
+	u_int16_t	irqflags;
+	u_int8_t	bus_id;
+	u_int8_t	source_irq;
+	u_int8_t	destination_apic;
+#define	MP_ALL_APIC	0xFF
+	u_int8_t	apic_intr;
+};
+
+/*
+ * System Address Space Mapping entry
+ */
+struct mp_map {
+	u_int8_t	type;
+#define	MP_SYSMAP	128
+	u_int8_t	length;
+	u_int8_t	bus;
+	u_int8_t	address_type;
+#define	MP_ADDR_IO	0
+#define	MP_ADDR_MEM	1
+#define	MP_ADDR_PRE	2
+	u_int64_t	address_base;
+	u_int64_t	address_length;
+};
+
+/*
+ * Bus Hierarchy Descriptor entry
+ */
+struct mp_bushier {
+	u_int8_t	type;
+#define	MP_BUSHIER	129
+	u_int8_t	length;
+	u_int8_t	bus_id;
+	u_int8_t	bus_info;
+#define	MP_BUS_SUB	0x01
+	u_int8_t	parent;
+	u_int8_t	reserved1;
+	u_int16_t	reserved2;
+};
+
+/*
+ * Compatibility Bus Address Space Modifier entry
+ */
+struct mp_buscompat {
+	u_int8_t	type;
+#define	MP_BUSCOMPAT	130
+	u_int8_t	length;
+	u_int8_t	bus_id;
+	u_int8_t	modifier;
+#define	MP_COMPAT_SUB	0x01
+	u_int32_t	range;
+};
+
+#ifdef _KERNEL
+extern int napics;
+#endif /* _KERNEL */
+
+#endif /* _MACHINE_MP_H */
diff --git a/sys/arch/i386/include/mpbiosreg.h b/sys/arch/i386/include/mpbiosreg.h
new file mode 100644
index 00000000000..d42aed04cdc
--- /dev/null
+++ b/sys/arch/i386/include/mpbiosreg.h
@@ -0,0 +1,155 @@
+/*	$OpenBSD: mpbiosreg.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: mpbiosreg.h,v 1.1.2.3 2000/02/29 13:17:51 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _I386_MPBIOSREG_H_
+#define _I386_MPBIOSREG_H_
+
+#define BIOS_BASE		(0xf0000)
+#define BIOS_SIZE		(0x10000)
+#define BIOS_COUNT		(BIOS_SIZE)
+
+/*
+ * Multiprocessor config table entry types.
+ */
+
+#define MPS_MCT_CPU	0
+#define MPS_MCT_BUS	1
+#define MPS_MCT_IOAPIC	2
+#define MPS_MCT_IOINT	3
+#define MPS_MCT_LINT	4
+
+#define MPS_MCT_NTYPES	5
+
+/*
+ * Interrupt typess
+ */
+
+#define MPS_INTTYPE_INT		0
+#define MPS_INTTYPE_NMI		1
+#define MPS_INTTYPE_SMI		2
+#define MPS_INTTYPE_ExtINT	3
+
+#define MPS_INTPO_DEF		0
+#define MPS_INTPO_ACTHI		1
+#define MPS_INTPO_ACTLO		3
+
+#define MPS_INTTR_DEF		0
+#define MPS_INTTR_EDGE		1
+#define MPS_INTTR_LEVEL		3
+
+
+/* MP Floating Pointer Structure */
+struct mpbios_fps {
+	u_int32_t	signature;
+/* string defined by the Intel MP Spec as identifying the MP table */
+#define MP_FP_SIG		0x5f504d5f	/* _MP_ */
+	
+	u_int32_t 	pap;
+	u_int8_t  	length;
+	u_int8_t  	spec_rev;
+	u_int8_t  	checksum;
+	u_int8_t  	mpfb1;	/* system configuration */
+	u_int8_t  	mpfb2;	/* flags */
+#define MPFPS_FLAG_IMCR		0x80	/* IMCR present */
+	u_int8_t  	mpfb3;	/* unused */
+	u_int8_t  	mpfb4;	/* unused */
+	u_int8_t  	mpfb5;	/* unused */
+};
+
+/* MP Configuration Table Header */
+struct mpbios_cth {
+	u_int32_t	signature;
+#define MP_CT_SIG		0x504d4350 	/* PCMP */
+	
+	u_int16_t 	base_len;
+	u_int8_t  	spec_rev;
+	u_int8_t  	checksum;
+	u_int8_t  	oem_id[8];
+	u_int8_t  	product_id[12];
+	u_int32_t	oem_table_pointer;
+	u_int16_t 	oem_table_size;
+	u_int16_t 	entry_count;
+	u_int32_t	apic_address;
+	u_int16_t	ext_len;
+	u_int8_t  	ext_cksum;
+	u_int8_t  	reserved;
+};
+
+struct mpbios_proc {
+	u_int8_t  type;
+	u_int8_t  apic_id;
+	u_int8_t  apic_version;
+	u_int8_t  cpu_flags;
+#define PROCENTRY_FLAG_EN	0x01
+#define PROCENTRY_FLAG_BP	0x02
+	u_long  cpu_signature;
+	u_long  feature_flags;
+	u_long  reserved1;
+	u_long  reserved2;
+};
+
+struct mpbios_bus {
+	u_int8_t  type;
+	u_int8_t  bus_id;
+	char    bus_type[6];
+};
+
+struct mpbios_ioapic {
+	u_int8_t  type;
+	u_int8_t  apic_id;
+	u_int8_t  apic_version;
+	u_int8_t  apic_flags;
+#define IOAPICENTRY_FLAG_EN	0x01
+	void   *apic_address;
+};
+
+struct mpbios_int {
+	u_int8_t  type;
+	u_int8_t  int_type;
+	u_int16_t int_flags;
+	u_int8_t  src_bus_id;
+	u_int8_t  src_bus_irq;
+	u_int8_t  dst_apic_id;
+#define MPS_ALL_APICS	0xff
+	u_int8_t  dst_apic_int;
+};
+
+#endif /* !_I386_MPBIOSREG_H_ */
diff --git a/sys/arch/i386/include/mpbiosvar.h b/sys/arch/i386/include/mpbiosvar.h
new file mode 100644
index 00000000000..b617041b4d9
--- /dev/null
+++ b/sys/arch/i386/include/mpbiosvar.h
@@ -0,0 +1,88 @@
+/*	$OpenBSD: mpbiosvar.h,v 1.2 2004/06/13 21:49:16 niklas Exp $	*/
+/* $NetBSD: mpbiosvar.h,v 1.1.2.3 2000/02/29 13:17:20 sommerfeld Exp $ */
+
+/*-
+ * Copyright (c) 2000 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by RedBack Networks Inc.
+ *
+ * Author: Bill Sommerfeld
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *        This product includes software developed by the NetBSD
+ *        Foundation, Inc. and its contributors.
+ * 4. Neither the name of The NetBSD Foundation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+
+#ifndef _I386_MPBIOSVAR_H_
+#define _I386_MPBIOSVAR_H_
+
+#define MP_TRAMPOLINE  (2 * PAGE_SIZE)
+
+#if !defined(_LOCORE)
+
+#include <machine/mpbiosreg.h>
+
+struct mp_bus 
+{
+	char *mb_name;		/* XXX bus name */
+	int mb_idx;		/* XXX bus index */
+	void (*mb_intr_print) (int);
+	void (*mb_intr_cfg)(const struct mpbios_int *, u_int32_t *);
+	struct mp_intr_map *mb_intrs;
+	u_int32_t mb_data;	/* random bus-specific datum. */
+};
+
+struct mp_intr_map
+{
+	struct mp_intr_map *next;
+	struct mp_bus *bus;
+	int bus_pin;
+	struct ioapic_softc *ioapic;
+	int ioapic_pin;
+	int ioapic_ih;		/* int handle, for apic_intr_est */
+	int type;		/* from mp spec intr record */
+ 	int flags;		/* from mp spec intr record */
+	u_int32_t redir;
+};
+
+#if defined(_KERNEL)
+extern int mp_verbose;
+extern struct mp_bus *mp_busses;
+extern struct mp_intr_map *mp_intrs;
+extern int mp_isa_bus;
+extern int mp_eisa_bus;
+
+void mpbios_scan(struct device *);
+int mpbios_probe(struct device *);
+#endif
+
+#endif
+
+#endif /* !_I386_MPBIOSVAR_H_ */
diff --git a/sys/arch/i386/include/npx.h b/sys/arch/i386/include/npx.h
index d6f906efb46..810c7817bee 100644
--- a/sys/arch/i386/include/npx.h
+++ b/sys/arch/i386/include/npx.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: npx.h,v 1.6 2004/02/01 19:05:23 deraadt Exp $	*/
+/*	$OpenBSD: npx.h,v 1.7 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: npx.h,v 1.11 1994/10/27 04:16:11 cgd Exp $	*/
 
 /*-
@@ -180,5 +180,8 @@ struct	emcsts {
 
 void    process_xmm_to_s87(const struct savexmm *, struct save87 *);
 void    process_s87_to_xmm(const struct save87 *, struct savexmm *);
+struct cpu_info;
+
+void	npxinit(struct cpu_info *);
 
 #endif /* !_I386_NPX_H_ */
diff --git a/sys/arch/i386/include/param.h b/sys/arch/i386/include/param.h
index 0d1dfb68fd9..335d7737cbb 100644
--- a/sys/arch/i386/include/param.h
+++ b/sys/arch/i386/include/param.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: param.h,v 1.27 2004/04/19 22:55:49 deraadt Exp $	*/
+/*	$OpenBSD: param.h,v 1.28 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: param.h,v 1.29 1996/03/04 05:04:26 cgd Exp $	*/
 
 /*-
@@ -99,7 +99,7 @@
 #define	USPACE		(UPAGES * NBPG)	/* total size of u-area */
 
 #ifndef MSGBUFSIZE
-#define MSGBUFSIZE	2*NBPG		/* default message buffer size */
+#define MSGBUFSIZE	4*NBPG		/* default message buffer size */
 #endif
 
 /*
diff --git a/sys/arch/i386/include/pcb.h b/sys/arch/i386/include/pcb.h
index 611ab5996a8..641f8874f17 100644
--- a/sys/arch/i386/include/pcb.h
+++ b/sys/arch/i386/include/pcb.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pcb.h,v 1.11 2004/02/01 19:05:23 deraadt Exp $	*/
+/*	$OpenBSD: pcb.h,v 1.12 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: pcb.h,v 1.21 1996/01/08 13:51:42 mycroft Exp $	*/
 
 /*-
@@ -75,7 +75,8 @@ struct pcb {
 	int	vm86_eflags;		/* virtual eflags for vm86 mode */
 	int	vm86_flagmask;		/* flag mask for vm86 mode */
 	void	*vm86_userp;		/* XXX performance hack */
-	struct pmap *pcb_pmap;		/* back pointer to our pmap */
+	struct  pmap *pcb_pmap;         /* back pointer to our pmap */
+	struct	cpu_info *pcb_fpcpu;	/* cpu holding our fpu state */
 	u_long	pcb_iomap[NIOPORTS/32];	/* I/O bitmap */
 	u_char	pcb_iomap_pad;	/* required; must be 0xff, says intel */
 };
@@ -88,8 +89,4 @@ struct md_coredump {
 	long	md_pad[8];
 };    
 
-#ifdef _KERNEL
-struct pcb *curpcb;		/* our current running pcb */
-#endif
-
 #endif /* _I386_PCB_H_ */
diff --git a/sys/arch/i386/include/pmap.h b/sys/arch/i386/include/pmap.h
index 74013d8824c..a7089d958fd 100644
--- a/sys/arch/i386/include/pmap.h
+++ b/sys/arch/i386/include/pmap.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.h,v 1.34 2004/05/20 09:20:42 kettenis Exp $	*/
+/*	$OpenBSD: pmap.h,v 1.35 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: pmap.h,v 1.44 2000/04/24 17:18:18 thorpej Exp $	*/
 
 /*
@@ -236,6 +236,12 @@
 #define PG_PVLIST	PG_AVAIL2	/* mapping has entry on pvlist */
 #define	PG_X		PG_AVAIL3	/* executable mapping */
 
+/*
+ * Number of PTE's per cache line.  4 byte pte, 32-byte cache line
+ * Used to avoid false sharing of cache lines.
+ */
+#define NPTECL			8
+
 #ifdef _KERNEL
 /*
  * pmap data structures: see pmap.c for details of locking.
@@ -272,6 +278,7 @@ struct pmap {
 	union descriptor *pm_ldt;	/* user-set LDT */
 	int pm_ldt_len;			/* number of LDT entries */
 	int pm_ldt_sel;			/* LDT selector */
+	uint32_t pm_cpus;		/* mask oc CPUs using map */
 };
 
 /* pm_flags */
@@ -330,19 +337,7 @@ struct pv_page {
 };
 
 /*
- * pmap_remove_record: a record of VAs that have been unmapped, used to
- * flush TLB.  If we have more than PMAP_RR_MAX then we stop recording.
- */
-
-#define PMAP_RR_MAX	16	/* max of 16 pages (64K) */
-
-struct pmap_remove_record {
-	int prr_npages;
-	vaddr_t prr_vas[PMAP_RR_MAX];
-};
-
-/*
- * Global kernel variables
+ * global kernel variables
  */
 
 extern pd_entry_t	PTD[];
@@ -394,6 +389,10 @@ int		pmap_exec_fixup(struct vm_map *, struct trapframe *,
 
 vaddr_t reserve_dumppages(vaddr_t); /* XXX: not a pmap fn */
 
+void	pmap_tlb_shootdown(pmap_t, vaddr_t, pt_entry_t, int32_t *);
+void	pmap_tlb_shootnow(int32_t);
+void	pmap_do_tlb_shootdown(struct cpu_info *);
+
 #define PMAP_GROWKERNEL		/* turn on pmap_growkernel interface */
 
 /*
diff --git a/sys/arch/i386/include/proc.h b/sys/arch/i386/include/proc.h
index ff0bc4b1bb3..e5393cd22e4 100644
--- a/sys/arch/i386/include/proc.h
+++ b/sys/arch/i386/include/proc.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: proc.h,v 1.3 2003/06/02 23:27:47 millert Exp $	*/
+/*	$OpenBSD: proc.h,v 1.4 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: proc.h,v 1.10 1995/08/06 05:33:23 mycroft Exp $	*/
 
 /*
@@ -38,6 +38,7 @@
 struct mdproc {
 	struct	trapframe *md_regs;	/* registers on current frame */
 	int	md_flags;		/* machine-dependent flags */
+	int	md_tss_sel;		/* TSS selector */
 };
 
 /* md_flags */
diff --git a/sys/arch/i386/include/segments.h b/sys/arch/i386/include/segments.h
index ac01a2a2e26..3519f0a42ed 100644
--- a/sys/arch/i386/include/segments.h
+++ b/sys/arch/i386/include/segments.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: segments.h,v 1.12 2003/11/16 20:30:06 avsm Exp $	*/
+/*	$OpenBSD: segments.h,v 1.13 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: segments.h,v 1.23 1996/02/01 22:31:03 mycroft Exp $	*/
 
 /*-
@@ -124,7 +124,7 @@ struct region_descriptor {
 #endif
 
 #ifdef _KERNEL
-extern union descriptor gdt[], ldt[];
+extern union descriptor *gdt, ldt[];
 extern struct gate_descriptor idt_region[];
 extern struct gate_descriptor *idt;
 
@@ -132,6 +132,13 @@ void setgate(struct gate_descriptor *, void *, int, int, int, int);
 void setregion(struct region_descriptor *, void *, size_t);
 void setsegment(struct segment_descriptor *, void *, size_t, int, int,
     int, int);
+void unsetgate(struct gate_descriptor *);
+void cpu_init_idt(void);
+
+int idt_vec_alloc(int, int);
+void idt_vec_set(int, void (*)(void));
+void idt_vec_free(int);
+
 #endif /* _KERNEL */
 
 #endif /* !_LOCORE */
@@ -220,7 +227,8 @@ void setsegment(struct segment_descriptor *, void *, size_t, int, int,
 #define	GAPM16CODE_SEL	8	/* 16 bit APM code descriptor */
 #define	GAPMDATA_SEL	9	/* APM data descriptor */
 #define	GICODE_SEL	10	/* Interrupt code descriptor (same as Kernel code) */
-#define	NGDT		11
+#define GCPU_SEL	11	/* per-CPU segment */
+#define	NGDT		12
 
 /*
  * Entries in the Local Descriptor Table (LDT)
diff --git a/sys/arch/i386/include/types.h b/sys/arch/i386/include/types.h
index 0da0bdc06e2..53f8a187136 100644
--- a/sys/arch/i386/include/types.h
+++ b/sys/arch/i386/include/types.h
@@ -1,5 +1,5 @@
 /*	$NetBSD: types.h,v 1.12 1995/12/24 01:08:03 mycroft Exp $	*/
-/*	$OpenBSD: types.h,v 1.13 2003/06/02 23:27:47 millert Exp $	*/
+/*	$OpenBSD: types.h,v 1.14 2004/06/13 21:49:16 niklas Exp $	*/
 
 /*-
  * Copyright (c) 1990 The Regents of the University of California.
@@ -72,5 +72,6 @@ typedef	unsigned long long	 uint64_t;
 typedef int32_t			register_t;
 
 #define __HAVE_NWSCONS
+#define __HAVE_CPUINFO
 
 #endif	/* _MACHTYPES_H_ */
diff --git a/sys/arch/i386/isa/clock.c b/sys/arch/i386/isa/clock.c
index 2d932ad788f..fa18701776d 100644
--- a/sys/arch/i386/isa/clock.c
+++ b/sys/arch/i386/isa/clock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: clock.c,v 1.31 2004/02/27 21:07:49 grange Exp $	*/
+/*	$OpenBSD: clock.c,v 1.32 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: clock.c,v 1.39 1996/05/12 23:11:54 mycroft Exp $	*/
 
 /*-
@@ -331,7 +331,7 @@ gettick()
  * wave' mode counts at 2:1).
  */
 void
-delay(n)
+i8254_delay(n)
 	int n;
 {
 	int limit, tick, otick;
@@ -458,7 +458,7 @@ calibrate_cyclecounter()
 #endif
 
 void
-cpu_initclocks()
+i8254_initclocks()
 {
 	static struct timeout rtcdrain_timeout;
 	stathz = 128;
diff --git a/sys/arch/i386/isa/icu.h b/sys/arch/i386/isa/icu.h
deleted file mode 100644
index 9b68025526b..00000000000
--- a/sys/arch/i386/isa/icu.h
+++ /dev/null
@@ -1,71 +0,0 @@
-/*	$OpenBSD: icu.h,v 1.8 2003/06/02 23:27:47 millert Exp $	*/
-/*	$NetBSD: icu.h,v 1.19 1996/02/01 22:31:21 mycroft Exp $	*/
-
-/*-
- * Copyright (c) 1990 The Regents of the University of California.
- * All rights reserved.
- *
- * This code is derived from software contributed to Berkeley by
- * William Jolitz.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. Neither the name of the University nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- *
- *	@(#)icu.h	5.6 (Berkeley) 5/9/91
- */
-
-/*
- * AT/386 Interrupt Control constants
- * W. Jolitz 8/89
- */
-
-#ifndef	_I386_ISA_ICU_H_
-#define	_I386_ISA_ICU_H_
-
-#ifndef	_LOCORE
-
-/*
- * Interrupt "level" mechanism variables, masks, and macros
- */
-extern	unsigned imen;		/* interrupt mask enable */
-
-#define SET_ICUS()	(outb(IO_ICU1 + 1, imen), outb(IO_ICU2 + 1, imen >> 8))
-
-#endif /* !_LOCORE */
-
-/*
- * Interrupt enable bits -- in order of priority
- */
-#define	IRQ_SLAVE	2
-
-/*
- * Interrupt Control offset into Interrupt descriptor table (IDT)
- * XXX ICU_OFFSET is actually a property of our architecture not of the ICU
- * XXX and therefore ought to use the architecture manifest constant IDTVECOFF
- * XXX for its definition instead.
- */
-#define	ICU_OFFSET	32		/* 0-31 are processor exceptions */
-#define	ICU_LEN		16		/* 32-47 are ISA interrupts */
-
-#endif /* !_I386_ISA_ICU_H_ */
diff --git a/sys/arch/i386/isa/icu.s b/sys/arch/i386/isa/icu.s
index 8bd2af133ca..d3f749b4a05 100644
--- a/sys/arch/i386/isa/icu.s
+++ b/sys/arch/i386/isa/icu.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: icu.s,v 1.20 2003/11/06 21:09:34 mickey Exp $	*/
+/*	$OpenBSD: icu.s,v 1.21 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: icu.s,v 1.45 1996/01/07 03:59:34 mycroft Exp $	*/
 
 /*-
@@ -33,10 +33,11 @@
 #include <net/netisr.h>
 
 	.data
-	.globl	_C_LABEL(imen), _C_LABEL(cpl), _C_LABEL(ipending)
-	.globl	_C_LABEL(astpending), _C_LABEL(netisr)
+	.globl	_C_LABEL(imen),_C_LABEL(ipending),_C_LABEL(netisr)
 _C_LABEL(imen):
 	.long	0xffff		# interrupt mask enable (all off)
+_C_LABEL(ipending):
+	.long	0		# interupts pending
 _C_LABEL(netisr):
 	.long	0		# scheduling bits for network
 
@@ -48,13 +49,13 @@ _C_LABEL(netisr):
 	ALIGN_TEXT
 _C_LABEL(splhigh):
 	movl	$IPL_HIGH,%eax
-	xchgl	%eax,_C_LABEL(cpl)
+	xchgl	%eax,CPL
 	ret
 
 	ALIGN_TEXT
 _C_LABEL(splx):
 	movl	4(%esp),%eax
-	movl	%eax,_C_LABEL(cpl)
+	movl	%eax,CPL
 	testl	%eax,%eax
 	jnz	_C_LABEL(Xspllower)
 	ret
@@ -72,12 +73,12 @@ IDTVEC(spllower)
 	pushl	%ebx
 	pushl	%esi
 	pushl	%edi
-	movl	_C_LABEL(cpl),%ebx	# save priority
+	movl	CPL,%ebx		# save priority
 	movl	$1f,%esi		# address to resume loop at
 1:	movl	%ebx,%eax		# get cpl
 	shrl	$4,%eax			# find its mask.
 	movl	_C_LABEL(iunmask)(,%eax,4),%eax
-	andl	_C_LABEL(ipending),%eax
+	andl	_C_LABEL(ipending),%eax		# any non-masked bits left?
 	jz	2f
 	bsfl	%eax,%eax
 	btrl	%eax,_C_LABEL(ipending)
@@ -98,10 +99,10 @@ IDTVEC(spllower)
  */
 IDTVEC(doreti)
 	popl	%ebx			# get previous priority
-	movl	%ebx,_C_LABEL(cpl)
+	movl	%ebx,CPL
 	movl	$1f,%esi		# address to resume loop at
-1:	movl	%ebx,%eax		# get cpl
-	shrl	$4,%eax			# find its mask
+1:	movl	%ebx,%eax
+	shrl	$4,%eax
 	movl	_C_LABEL(iunmask)(,%eax,4),%eax
 	andl	_C_LABEL(ipending),%eax
 	jz	2f
@@ -111,8 +112,8 @@ IDTVEC(doreti)
 	cli
 	jmp	*_C_LABEL(Xresume)(,%eax,4)
 2:	/* Check for ASTs on exit to user mode. */
+	CHECK_ASTPENDING(%ecx)
 	cli
-	cmpb	$0,_C_LABEL(astpending)
 	je	3f
 	testb   $SEL_RPL,TF_CS(%esp)
 #ifdef VM86
@@ -120,10 +121,12 @@ IDTVEC(doreti)
 	testl	$PSL_VM,TF_EFLAGS(%esp)
 #endif
 	jz	3f
-4:	movb	$0,_C_LABEL(astpending)
+4:	CLEAR_ASTPENDING(%ecx)
 	sti
+	movl	$T_ASTFLT,TF_TRAPNO(%esp)	/* XXX undo later. */
 	/* Pushed T_ASTFLT into tf_trapno on entry. */
 	call	_C_LABEL(trap)
+	cli
 	jmp	2b
 3:	INTRFASTEXIT
 
@@ -137,9 +140,16 @@ IDTVEC(doreti)
 IDTVEC(softtty)
 #if NPCCOM > 0
 	movl	$IPL_SOFTTTY,%eax
-	movl	%eax,_C_LABEL(cpl)
+	movl	%eax,CPL
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
 	call	_C_LABEL(comsoft)
-	movl	%ebx,_C_LABEL(cpl)
+#ifdef MULTIPROCESSOR	
+	call	_C_LABEL(i386_softintunlock)
+#endif
+	movl	%ebx,CPL
 #endif
 	jmp	*%esi
 
@@ -152,18 +162,32 @@ IDTVEC(softtty)
 
 IDTVEC(softnet)
 	movl	$IPL_SOFTNET,%eax
-	movl	%eax,_C_LABEL(cpl)
+	movl	%eax,CPL
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
 	xorl	%edi,%edi
 	xchgl	_C_LABEL(netisr),%edi
 #include <net/netisr_dispatch.h>
-	movl	%ebx,_C_LABEL(cpl)
+#ifdef MULTIPROCESSOR	
+	call	_C_LABEL(i386_softintunlock)
+#endif
+ 	movl	%ebx,CPL
 	jmp	*%esi
 #undef DONETISR
 
 IDTVEC(softclock)
 	movl	$IPL_SOFTCLOCK,%eax
-	movl	%eax,_C_LABEL(cpl)
+	movl	%eax,CPL
+	sti
+#ifdef MULTIPROCESSOR
+	call	_C_LABEL(i386_softintlock)
+#endif
 	call	_C_LABEL(softclock)
-	movl	%ebx,_C_LABEL(cpl)
+#ifdef MULTIPROCESSOR	
+	call	_C_LABEL(i386_softintunlock)
+#endif
+	movl	%ebx,CPL
 	jmp	*%esi
 
diff --git a/sys/arch/i386/isa/isa_machdep.c b/sys/arch/i386/isa/isa_machdep.c
index 0a8b580c6f4..0df077cdb19 100644
--- a/sys/arch/i386/isa/isa_machdep.c
+++ b/sys/arch/i386/isa/isa_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: isa_machdep.c,v 1.48 2003/06/02 23:27:47 millert Exp $	*/
+/*	$OpenBSD: isa_machdep.c,v 1.49 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: isa_machdep.c,v 1.22 1997/06/12 23:57:32 thorpej Exp $	*/
 
 #define ISA_DMA_STATS
@@ -122,18 +122,25 @@
 
 #include <uvm/uvm_extern.h>
 
+#include "ioapic.h"
+
+#if NIOAPIC > 0
+#include <machine/i82093var.h>
+#include <machine/mpbiosvar.h>
+#endif
+
 #define _I386_BUS_DMA_PRIVATE
 #include <machine/bus.h>
 
 #include <machine/intr.h>
 #include <machine/pio.h>
 #include <machine/cpufunc.h>
+#include <machine/i8259.h>
 
 #include <dev/isa/isareg.h>
 #include <dev/isa/isavar.h>
 #include <dev/isa/isadmavar.h>
 #include <i386/isa/isa_machdep.h>
-#include <i386/isa/icu.h>
 
 #include "isadma.h"
 
@@ -250,6 +257,16 @@ isa_defaultirq()
 	outb(IO_ICU2, 0x0a);		/* Read IRR by default. */
 }
 
+void
+isa_nodefaultirq()
+{
+	int i;
+
+	/* icu vectors */
+	for (i = 0; i < ICU_LEN; i++)
+		unsetgate(&idt[ICU_OFFSET + i]);
+}
+
 /*
  * Handle a NMI, possibly a machine check.
  * return true to panic system, false to ignore.
@@ -286,6 +303,9 @@ int intrtype[ICU_LEN], intrmask[ICU_LEN], intrlevel[ICU_LEN];
 int iminlevel[ICU_LEN], imaxlevel[ICU_LEN];
 struct intrhand *intrhand[ICU_LEN];
 
+int imask[NIPL];	/* Bitmask telling what interrupts are blocked. */
+int iunmask[NIPL];	/* Bitmask telling what interrupts are accepted. */
+
 /*
  * Recalculate the interrupt masks from scratch.
  * We could code special registry and deregistry versions of this function that
@@ -295,24 +315,27 @@ struct intrhand *intrhand[ICU_LEN];
 void
 intr_calculatemasks()
 {
-	int irq, level;
+	int irq, level, unusedirqs;
 	struct intrhand *q;
 
 	/* First, figure out which levels each IRQ uses. */
+	unusedirqs = 0xffff;
 	for (irq = 0; irq < ICU_LEN; irq++) {
-		register int levels = 0;
+		int levels = 0;
 		for (q = intrhand[irq]; q; q = q->ih_next)
 			levels |= 1 << IPL(q->ih_level);
 		intrlevel[irq] = levels;
+		if (levels)
+			unusedirqs &= ~(1 << irq);
 	}
 
 	/* Then figure out which IRQs use each level. */
 	for (level = 0; level < NIPL; level++) {
-		register int irqs = 0;
+		int irqs = 0;
 		for (irq = 0; irq < ICU_LEN; irq++)
 			if (intrlevel[irq] & (1 << level))
 				irqs |= 1 << irq;
-		imask[level] = irqs;
+		imask[level] = irqs | unusedirqs;
 	}
 
 	/*
@@ -331,25 +354,40 @@ intr_calculatemasks()
 
 	/* And eventually calculate the complete masks. */
 	for (irq = 0; irq < ICU_LEN; irq++) {
-		register int irqs = 1 << irq;
+		int irqs = 1 << irq;
 		int minlevel = IPL_NONE;
 		int maxlevel = IPL_NONE;
 
-		for (q = intrhand[irq]; q; q = q->ih_next) {
-			irqs |= IMASK(q->ih_level);
-			if (minlevel == IPL_NONE || q->ih_level < minlevel)
-				minlevel = q->ih_level;
-			if (q->ih_level > maxlevel)
-				maxlevel = q->ih_level;
+		if (intrhand[irq] == NULL) {
+			maxlevel = IPL_HIGH;
+			irqs = IMASK(IPL_HIGH);
+		} else {
+			for (q = intrhand[irq]; q; q = q->ih_next) {
+				irqs |= IMASK(q->ih_level);
+				if (minlevel == IPL_NONE ||
+				    q->ih_level < minlevel)
+					minlevel = q->ih_level;
+				if (q->ih_level > maxlevel)
+					maxlevel = q->ih_level;
+			}
 		}
+		if (irqs != IMASK(maxlevel))
+			panic("irq %d level %x mask mismatch: %x vs %x", irq,
+			    maxlevel, irqs, IMASK(maxlevel));
+
 		intrmask[irq] = irqs;
 		iminlevel[irq] = minlevel;
 		imaxlevel[irq] = maxlevel;
+
+#if 0
+		printf("irq %d: level %x, mask 0x%x (%x)\n", irq,
+		    imaxlevel[irq], intrmask[irq], IMASK(imaxlevel[irq]));
+#endif
 	}
 
 	/* Lastly, determine which IRQs are actually in use. */
 	{
-		register int irqs = 0;
+		int irqs = 0;
 		for (irq = 0; irq < ICU_LEN; irq++)
 			if (intrhand[irq])
 				irqs |= 1 << irq;
@@ -496,17 +534,48 @@ isa_intr_establish(ic, irq, type, level, ih_fun, ih_arg, ih_what)
 	struct intrhand **p, *q, *ih;
 	static struct intrhand fakehand = {fakeintr};
 
+#if NIOAPIC > 0
+	struct mp_intr_map *mip;
+
+ 	if (mp_busses != NULL) {
+ 		int mpspec_pin = irq;
+ 		int bus = mp_isa_bus;
+ 		int airq;
+
+ 		for (mip = mp_busses[bus].mb_intrs; mip != NULL; 
+ 		    mip = mip->next) {
+ 			if (mip->bus_pin == mpspec_pin) {
+ 				airq = mip->ioapic_ih | irq;
+ 				break;
+ 			}
+ 		}
+		if (mip == NULL && mp_eisa_bus != -1) {
+			for (mip = mp_busses[mp_eisa_bus].mb_intrs;
+			    mip != NULL; mip=mip->next) {
+				if (mip->bus_pin == mpspec_pin) {
+					airq = mip->ioapic_ih | irq;
+					break;
+				}
+			}
+		}
+ 		if (mip == NULL)
+			printf("isa_intr_establish: no MP mapping found\n");
+ 		else
+			return (apic_intr_establish(airq, type, level, ih_fun,
+			    ih_arg, ih_what));
+ 	}
+#endif
 	/* no point in sleeping unless someone can free memory. */
 	ih = malloc(sizeof *ih, M_DEVBUF, cold ? M_NOWAIT : M_WAITOK);
 	if (ih == NULL) {
 		printf("%s: isa_intr_establish: can't malloc handler info\n",
 		    ih_what);
-		return NULL;
+		return (NULL);
 	}
 
 	if (!LEGAL_IRQ(irq) || type == IST_NONE) {
-		printf("%s: intr_establish: bogus irq or type\n", ih_what);
-		return NULL;
+		printf("%s: isa_intr_establish: bogus irq or type\n", ih_what);
+		return (NULL);
 	}
 	switch (intrtype[irq]) {
 	case IST_NONE:
@@ -521,7 +590,7 @@ isa_intr_establish(ic, irq, type, level, ih_fun, ih_arg, ih_what)
 			/*printf("%s: intr_establish: can't share %s with %s, irq %d\n",
 			    ih_what, isa_intr_typename(intrtype[irq]),
 			    isa_intr_typename(type), irq);*/
-			return NULL;
+			return (NULL);
 		}
 		break;
 	}
@@ -571,8 +640,15 @@ isa_intr_disestablish(ic, arg)
 	int irq = ih->ih_irq;
 	struct intrhand **p, *q;
 
+#if NIOAPIC > 0
+	if (irq & APIC_INT_VIA_APIC) {
+		apic_intr_disestablish(arg);
+		return;
+	}
+#endif
+
 	if (!LEGAL_IRQ(irq))
-		panic("intr_disestablish: bogus irq");
+		panic("intr_disestablish: bogus irq %d", irq);
 
 	/*
 	 * Remove the handler from the chain.
diff --git a/sys/arch/i386/isa/mms.c b/sys/arch/i386/isa/mms.c
index 1b84aecf965..137c18384a5 100644
--- a/sys/arch/i386/isa/mms.c
+++ b/sys/arch/i386/isa/mms.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mms.c,v 1.16 2002/03/14 01:26:33 millert Exp $ */
+/* $OpenBSD: mms.c,v 1.17 2004/06/13 21:49:16 niklas Exp $ */
 /*	$NetBSD: mms.c,v 1.35 2000/01/08 02:57:25 takemura Exp $	*/
 
 /*-
diff --git a/sys/arch/i386/isa/npx.c b/sys/arch/i386/isa/npx.c
index b56a17f3875..4111ba0ab90 100644
--- a/sys/arch/i386/isa/npx.c
+++ b/sys/arch/i386/isa/npx.c
@@ -1,10 +1,10 @@
-/*	$OpenBSD: npx.c,v 1.31 2004/02/01 19:05:21 deraadt Exp $	*/
+/*	$OpenBSD: npx.c,v 1.32 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: npx.c,v 1.57 1996/05/12 23:12:24 mycroft Exp $	*/
 
 #if 0
-#define iprintf(x)	printf x
+#define IPRINTF(x)	printf x
 #else
-#define	iprintf(x)
+#define	IPRINTF(x)
 #endif
 
 /*-
@@ -54,21 +54,16 @@
 
 #include <machine/cpu.h>
 #include <machine/intr.h>
+#include <machine/npx.h>
 #include <machine/pio.h>
 #include <machine/cpufunc.h>
 #include <machine/pcb.h>
 #include <machine/trap.h>
 #include <machine/specialreg.h>
+#include <machine/i8259.h>
 
 #include <dev/isa/isareg.h>
 #include <dev/isa/isavar.h>
-#include <i386/isa/icu.h>
-
-#if 0
-#define IPRINTF(x)      printf x
-#else
-#define IPRINTF(x)
-#endif
 
 /*
  * 387 and 287 Numeric Coprocessor Extension (NPX) Driver.
@@ -104,7 +99,6 @@
 
 int npxintr(void *);
 static int npxprobe1(struct isa_attach_args *);
-static void npxsave1(void);
 
 struct npx_softc {
 	struct device sc_dev;
@@ -129,10 +123,7 @@ enum npx_type {
 	NPX_BROKEN,
 };
 
-struct proc	*npxproc;
-
 static	enum npx_type		npx_type;
-static	int			npx_nointr;
 static	volatile u_int		npx_intrs_while_probing;
 static	volatile u_int		npx_traps_while_probing;
 
@@ -160,15 +151,15 @@ fpu_save(union savefpu *addr)
 }
 
 static int
-npxdna_notset(struct proc *p)
+npxdna_notset(struct cpu_info *ci)
 {
 	panic("npxdna vector not initialized");
 }
 
-int    (*npxdna_func)(struct proc *) = npxdna_notset;
-int    npxdna_s87(struct proc *);
+int    (*npxdna_func)(struct cpu_info *) = npxdna_notset;
+int    npxdna_s87(struct cpu_info *);
 #ifdef I686_CPU
-int    npxdna_xmm(struct proc *);
+int    npxdna_xmm(struct cpu_info *);
 #endif /* I686_CPU */
 void   npxexit(void);
 
@@ -350,6 +341,19 @@ asm (".text\n\t"
 	"popl	%eax\n\t"
 	"ret\n\t");
 
+void
+npxinit(struct cpu_info *ci)
+{
+	lcr0(rcr0() & ~(CR0_EM|CR0_TS));
+	fninit();
+	if (npx586bug1(4195835, 3145727) != 0) {
+		i386_fpu_fdivbug = 1;
+		printf("%s: WARNING: Pentium FDIV bug detected!\n",
+		    ci->ci_dev.dv_xname);
+	}
+	lcr0(rcr0() | (CR0_TS));
+}
+
 /*
  * Attach routine - announce which it is, and wire into system
  */
@@ -379,13 +383,7 @@ npxattach(parent, self, aux)
 		return;
 	}
 
-	lcr0(rcr0() & ~(CR0_EM|CR0_TS));
-	fninit();
-	if (npx586bug1(4195835, 3145727) != 0) {
-		i386_fpu_fdivbug = 1;
-		printf("WARNING: Pentium FDIV bug detected!\n");
-	}
-	lcr0(rcr0() | (CR0_TS));
+	npxinit(&cpu_info_primary);
 	i386_fpu_present = 1;
 
 #ifdef I686_CPU
@@ -415,16 +413,17 @@ int
 npxintr(arg)
 	void *arg;
 {
-	register struct proc *p = npxproc;
+	struct cpu_info *ci = curcpu();
+	struct proc *p = ci->ci_fpcurproc;
 	union savefpu *addr;
 	struct intrframe *frame = arg;
 	int code;
 	union sigval sv;
 
 	uvmexp.traps++;
-	iprintf(("Intr"));
+	IPRINTF(("%s: fp intr\n", ci->ci_dev.dv_xname));
 
-	if (p == 0 || npx_type == NPX_NONE) {
+	if (p == NULL || npx_type == NPX_NONE) {
 		/* XXX no %p in stand/printf.c.  Cast to quiet gcc -Wall. */
 		printf("npxintr: p = %lx, curproc = %lx, npx_type = %d\n",
 		       (u_long) p, (u_long) curproc, npx_type);
@@ -438,11 +437,21 @@ npxintr(arg)
 	 * If we're saving, ignore the interrupt.  The FPU will happily
 	 * generate another one when we restore the state later.
 	 */
-	if (npx_nointr != 0)
+	if (ci->ci_fpsaving)
 		return (1);
+
+#ifdef DIAGNOSTIC
 	/*
-	 * Find the address of npxproc's savefpu.  This is not necessarily
-	 * the one in curpcb.
+	 * At this point, fpcurproc should be curproc.  If it wasn't, the TS
+	 * bit should be set, and we should have gotten a DNA exception.
+	 */
+	if (p != curproc)
+		panic("npxintr: wrong process");
+#endif
+
+	/*
+	 * Find the address of fpcurproc's saved FPU state.  (Given the
+	 * invariant above, this is always the one in curpcb.)
 	 */
 	addr = &p->p_addr->u_pcb.pcb_savefpu;
 	/*
@@ -504,7 +513,7 @@ npxintr(arg)
 		if (addr->sv_87.sv_ex_sw & EN_SW_IE)
 			code = FPE_FLTINV;
 #ifdef notyet
-		else if (addr->sv_ex_sw & EN_SW_DE)
+		else if (addr->sv_87.sv_ex_sw & EN_SW_DE)
 			code = FPE_FLTDEN;
 #endif
 		else if (addr->sv_87.sv_ex_sw & EN_SW_ZE)
@@ -541,115 +550,147 @@ npxintr(arg)
 }
 
 /*
- * Wrapper for fnsave instruction to handle h/w bugs.  If there is an error
- * pending, then fnsave generates a bogus IRQ13 on some systems.  Force any
- * IRQ13 to be handled immediately, and then ignore it.
- *
- * This routine is always called at spl0.  If it might called with the NPX
- * interrupt masked, it would be necessary to forcibly unmask the NPX interrupt
- * so that it could succeed.
- */
-static __inline void
-npxsave1(void)
-{
-	register struct pcb *pcb;
-
-	npx_nointr = 1;
-	pcb = &npxproc->p_addr->u_pcb;
-	fpu_save(&pcb->pcb_savefpu);
-	pcb->pcb_cr0 |= CR0_TS;
-	fwait();
-	npx_nointr = 0;
-}
-
-/*
  * Implement device not available (DNA) exception
  *
- * If the we were the last process to use the FPU, we can simply return.
+ * If we were the last process to use the FPU, we can simply return.
  * Otherwise, we save the previous state, if necessary, and restore our last
  * saved state.
  */
+
+/*
+ * XXX It is unclear if the code below is correct in the multiprocessor
+ * XXX case.  Check the NetBSD sources once again to be sure.
+ */
 #ifdef I686_CPU
 int
-npxdna_xmm(struct proc *p)
+npxdna_xmm(struct cpu_info *ci)
 {
+	struct proc *p;
+	int s;
 
-#ifdef DIAGNOSTIC
-	if (cpl != 0 || npx_nointr != 0)
-	        panic("npxdna: masked");
+	if (ci->ci_fpsaving) {
+		printf("recursive npx trap; cr0=%x\n", rcr0());
+		return (0);
+	}
+
+	s = splipi();		/* lock out IPI's while we clean house.. */
+
+#ifdef MULTIPROCESSOR
+	p = ci->ci_curproc;
+#else
+	p = curproc;
 #endif
 
-	p->p_addr->u_pcb.pcb_cr0 &= ~CR0_TS;
-	clts();
+	IPRINTF(("%s: dna for %lx%s\n", ci->ci_dev.dv_xname, (u_long)p,
+	    (p->p_md.md_flags & MDP_USEDFPU) ? " (used fpu)" : ""));
 
 	/*
+	 * XXX should have a fast-path here when no save/restore is necessary
+	 */
+	/*
 	 * Initialize the FPU state to clear any exceptions.  If someone else
 	 * was using the FPU, save their state (which does an implicit
 	 * initialization).
 	 */
-	npx_nointr = 1;
-	if (npxproc != 0 && npxproc != p) {
-	        IPRINTF(("Save"));
-	        npxsave1();
+	if (ci->ci_fpcurproc != NULL) {
+		IPRINTF(("%s: fp save %lx\n", ci->ci_dev.dv_xname,
+		    (u_long)ci->ci_fpcurproc));
+		npxsave_cpu(ci, 1);
 	} else {
-	        IPRINTF(("Init"));
-	        fninit();
-	        fwait();
+		clts();
+		IPRINTF(("%s: fp init\n", ci->ci_dev.dv_xname));
+		fninit();
+		fwait();
+		stts();
 	}
-	npx_nointr = 0;
-	npxproc = p;
+	splx(s);
+
+	IPRINTF(("%s: done saving\n", ci->ci_dev.dv_xname));
+	KDASSERT(ci->ci_fpcurproc == NULL);
+#ifndef MULTIPROCESSOR
+	KDASSERT(p->p_addr->u_pcb.pcb_fpcpu == NULL);
+#else
+	if (p->p_addr->u_pcb.pcb_fpcpu != NULL)
+		npxsave_proc(p, 1);
+#endif
+	p->p_addr->u_pcb.pcb_cr0 &= ~CR0_TS;
+	clts();
+	s = splipi();
+	ci->ci_fpcurproc = p;
+	p->p_addr->u_pcb.pcb_fpcpu = ci;
+	splx(s);
 
 	if ((p->p_md.md_flags & MDP_USEDFPU) == 0) {
 		fldcw(&p->p_addr->u_pcb.pcb_savefpu.sv_xmm.sv_env.en_cw);
 		p->p_md.md_flags |= MDP_USEDFPU;
-	} else
+	} else {
 		fxrstor(&p->p_addr->u_pcb.pcb_savefpu.sv_xmm);
+	}
 
 	return (1);
 }
 #endif /* I686_CPU */
 
 int
-npxdna_s87(struct proc *p)
+npxdna_s87(struct cpu_info *ci)
 {
-	static u_short control = __INITIAL_NPXCW__;
+	struct proc *p;
+	int s;
+
+	KDASSERT(i386_use_fxsave == 0);
 
-	if (npx_type == NPX_NONE) {
-		iprintf(("Emul"));
+	if (ci->ci_fpsaving) {
+		printf("recursive npx trap; cr0=%x\n", rcr0());
 		return (0);
 	}
 
-#ifdef DIAGNOSTIC
-	if (cpl != IPL_NONE || npx_nointr != 0)
-		panic("npxdna: masked");
+	s = splipi();		/* lock out IPI's while we clean house.. */
+#ifdef MULTIPROCESSOR
+	p = ci->ci_curproc;
+#else
+	p = curproc;
 #endif
 
+	IPRINTF(("%s: dna for %lx%s\n", ci->ci_dev.dv_xname, (u_long)p,
+	    (p->p_md.md_flags & MDP_USEDFPU) ? " (used fpu)" : ""));
+
+	/*
+	 * If someone else was using our FPU, save their state (which does an
+	 * implicit initialization); otherwise, initialize the FPU state to
+	 * clear any exceptions.
+	 */
+	if (ci->ci_fpcurproc != NULL) {
+		IPRINTF(("%s: fp save %lx\n", ci->ci_dev.dv_xname,
+		    (u_long)ci->ci_fpcurproc));
+		npxsave_cpu(ci, 1);
+	} else {
+		clts();
+		IPRINTF(("%s: fp init\n", ci->ci_dev.dv_xname));
+		fninit();
+		fwait();
+		stts();
+	}
+	splx(s);
+
+	IPRINTF(("%s: done saving\n", ci->ci_dev.dv_xname));
+	KDASSERT(ci->ci_fpcurproc == NULL);
+#ifndef MULTIPROCESSOR
+	KDASSERT(p->p_addr->u_pcb.pcb_fpcpu == NULL);
+#else
+	if (p->p_addr->u_pcb.pcb_fpcpu != NULL)
+		npxsave_proc(p, 1);
+#endif
 	p->p_addr->u_pcb.pcb_cr0 &= ~CR0_TS;
 	clts();
+	s = splipi();
+	ci->ci_fpcurproc = p;
+	p->p_addr->u_pcb.pcb_fpcpu = ci;
+	splx(s);
 
 	if ((p->p_md.md_flags & MDP_USEDFPU) == 0) {
+		fldcw(&p->p_addr->u_pcb.pcb_savefpu.sv_87.sv_env.en_cw);
 		p->p_md.md_flags |= MDP_USEDFPU;
-		iprintf(("Init"));
-		if (npxproc != 0 && npxproc != p)
-			npxsave1();
-		else {
-			npx_nointr = 1;
-			fninit();
-			fwait();
-			npx_nointr = 0;
-		}
-		npxproc = p;
-		fldcw(&control);
 	} else {
-		if (npxproc != 0) {
-#ifdef DIAGNOSTIC
-			if (npxproc == p)
-				panic("npxdna: same process");
-#endif
-			iprintf(("Save"));
-			npxsave1();
-		}
-		npxproc = p;
 		/*
 		 * The following frstor may cause an IRQ13 when the state being
 		 * restored has a pending error.  The error will appear to have
@@ -670,37 +711,121 @@ npxdna_s87(struct proc *p)
 }
 
 /*
- * Drop the current FPU state on the floor.
+ * The FNSAVE instruction clears the FPU state.  Rather than reloading the FPU
+ * immediately, we clear fpcurproc and turn on CR0_TS to force a DNA and a
+ * reload of the FPU state the next time we try to use it.  This routine
+ * is only called when forking, core dumping, or debugging, or swapping,
+ * so the lazy reload at worst forces us to trap once per fork(), and at best
+ * saves us a reload once per fork().
  */
 void
-npxdrop()
+npxsave_cpu(struct cpu_info *ci, int save)
 {
+	struct proc *p;
+	int s;
+
+	KDASSERT(ci == curcpu());
+
+	p = ci->ci_fpcurproc;
+	if (p == NULL)
+		return;
+
+	IPRINTF(("%s: fp cpu %s %lx\n", ci->ci_dev.dv_xname,
+	    save ? "save" : "flush", (u_long)p));
+
+	if (save) {
+#ifdef DIAGNOSTIC
+		if (ci->ci_fpsaving != 0)
+			panic("npxsave_cpu: recursive save!");
+#endif
+		 /*
+		  * Set ci->ci_fpsaving, so that any pending exception will be
+		  * thrown away.  (It will be caught again if/when the FPU
+		  * state is restored.)
+		  *
+		  * XXX on i386 and earlier, this routine should always be
+		  * called at spl0; if it might called with the NPX interrupt
+		  * masked, it would be necessary to forcibly unmask the NPX
+		  * interrupt so that it could succeed.
+		  * XXX this is irrelevant on 486 and above (systems
+		  * which report FP failures via traps rather than irq13).
+		  * XXX punting for now..
+		  */
+		clts();
+		ci->ci_fpsaving = 1;
+		fpu_save(&p->p_addr->u_pcb.pcb_savefpu);
+		ci->ci_fpsaving = 0;
+		/* It is unclear if this is needed. */
+		fwait();
+	}
 
+	/*
+	 * We set the TS bit in the saved CR0 for this process, so that it
+	 * will get a DNA exception on any FPU instruction and force a reload.
+	 */
 	stts();
-	npxproc->p_addr->u_pcb.pcb_cr0 |= CR0_TS;
-	npxproc = 0;
+	p->p_addr->u_pcb.pcb_cr0 |= CR0_TS;
+
+	s = splipi();
+	p->p_addr->u_pcb.pcb_fpcpu = NULL;
+	ci->ci_fpcurproc = NULL;
+	splx(s);
 }
 
 /*
- * Save npxproc's FPU state.
- *
- * The FNSAVE instruction clears the FPU state.  Rather than reloading the FPU
- * immediately, we clear npxproc and turn on CR0_TS to force a DNA and a reload
- * of the FPU state the next time we try to use it.  This routine is only
- * called when forking or core dump, so this algorithm at worst forces us to
- * trap once per fork(), and at best saves us a reload once per fork().
+ * Save p's FPU state, which may be on this processor or another processor.
  */
-void
-npxsave()
+ void
+npxsave_proc(struct proc *p, int save)
 {
+	struct cpu_info *ci = curcpu();
+	struct cpu_info *oci;
+
+	KDASSERT(p->p_addr != NULL);
+	KDASSERT(p->p_flag & P_INMEM);
 
+	oci = p->p_addr->u_pcb.pcb_fpcpu;
+	if (oci == NULL)
+		return;
+
+	IPRINTF(("%s: fp proc %s %lx\n", ci->ci_dev.dv_xname,
+	    save ? "save" : "flush", (u_long)p));
+
+#if defined(MULTIPROCESSOR)
+	if (oci == ci) {
+		int s = splipi();
+		npxsave_cpu(ci, save);
+		splx(s);
+	} else {
 #ifdef DIAGNOSTIC
-	if (cpl != IPL_NONE || npx_nointr != 0)
-		panic("npxsave: masked");
+		int spincount;
+#endif
+
+		IPRINTF(("%s: fp ipi to %s %s %lx\n", ci->ci_dev.dv_xname,
+		    oci->ci_dev.dv_xname, save ? "save" : "flush", (u_long)p));
+
+		i386_send_ipi(oci,
+		    save ? I386_IPI_SYNCH_FPU : I386_IPI_FLUSH_FPU);
+
+#ifdef DIAGNOSTIC
+		spincount = 0;
+#endif
+		while (p->p_addr->u_pcb.pcb_fpcpu != NULL)
+#ifdef DIAGNOSTIC
+		{
+			spincount++;
+			if (spincount > 100000000) {
+				panic("fp_save ipi didn't");
+			}
+		}
+#else
+		__splbarrier();		/* XXX replace by generic barrier */
+		;
+#endif
+	}
+#else
+	KASSERT(ci->ci_fpcurproc == p);
+	npxsave_cpu(ci, save);
 #endif
-	iprintf(("Fork"));
-	clts();
-	npxsave1();
-	stts();
-	npxproc = 0;
 }
+
diff --git a/sys/arch/i386/pci/pci_intr_fixup.c b/sys/arch/i386/pci/pci_intr_fixup.c
index 30b47f83807..ea0cfd009f1 100644
--- a/sys/arch/i386/pci/pci_intr_fixup.c
+++ b/sys/arch/i386/pci/pci_intr_fixup.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pci_intr_fixup.c,v 1.32 2004/02/24 19:30:00 markus Exp $	*/
+/*	$OpenBSD: pci_intr_fixup.c,v 1.33 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: pci_intr_fixup.c,v 1.10 2000/08/10 21:18:27 soda Exp $	*/
 
 /*
@@ -100,12 +100,12 @@
 
 #include <machine/bus.h>
 #include <machine/intr.h>
+#include <machine/i8259.h>
 
 #include <dev/pci/pcireg.h>
 #include <dev/pci/pcivar.h>
 #include <dev/pci/pcidevs.h>
 
-#include <i386/isa/icu.h>
 #include <i386/pci/pcibiosvar.h>
 
 struct pciintr_link_map {
diff --git a/sys/arch/i386/pci/pci_machdep.c b/sys/arch/i386/pci/pci_machdep.c
index 1d05bf4a202..4ccb6f0e296 100644
--- a/sys/arch/i386/pci/pci_machdep.c
+++ b/sys/arch/i386/pci/pci_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pci_machdep.c,v 1.24 2003/05/04 08:01:08 deraadt Exp $	*/
+/*	$OpenBSD: pci_machdep.c,v 1.25 2004/06/13 21:49:16 niklas Exp $	*/
 /*	$NetBSD: pci_machdep.c,v 1.28 1997/06/06 23:29:17 thorpej Exp $	*/
 
 /*-
@@ -92,6 +92,7 @@
 #define _I386_BUS_DMA_PRIVATE
 #include <machine/bus.h>
 #include <machine/pio.h>
+#include <machine/i8259.h>
 
 #include "bios.h"
 #if NBIOS > 0
@@ -99,11 +100,17 @@
 extern bios_pciinfo_t *bios_pciinfo;
 #endif
 
-#include <i386/isa/icu.h>
 #include <dev/isa/isavar.h>
 #include <dev/pci/pcivar.h>
 #include <dev/pci/pcireg.h>
 
+#include "ioapic.h"
+
+#if NIOAPIC > 0
+#include <machine/i82093var.h>
+#include <machine/mpbiosvar.h>
+#endif
+
 #include "pcibios.h"
 #if NPCIBIOS > 0
 #include <i386/pci/pcibiosvar.h>
@@ -408,6 +415,11 @@ pci_intr_map(pa, ihp)
 	struct pci_attach_args *pa;
 	pci_intr_handle_t *ihp;
 {
+#if NIOAPIC > 0
+	struct mp_intr_map *mip;
+	int bus, dev, func;
+#endif
+
 #if NPCIBIOS > 0
 	pci_chipset_tag_t pc = pa->pa_pc;
 	pcitag_t intrtag = pa->pa_intrtag;
@@ -459,6 +471,49 @@ pci_intr_map(pa, ihp)
 			line = 9;
 		}
 	}
+#if NIOAPIC > 0
+	pci_decompose_tag (pc, intrtag, &bus, &dev, &func);
+
+	if (mp_busses != NULL) {
+		/*
+		 * Assumes 1:1 mapping between PCI bus numbers and
+		 * the numbers given by the MP bios.
+		 * XXX Is this a valid assumption?
+		 */
+		int mpspec_pin = (dev<<2)|(pin-1);
+
+		for (mip = mp_busses[bus].mb_intrs; mip != NULL; mip=mip->next) {
+			if (mip->bus_pin == mpspec_pin) {
+				ihp->line = mip->ioapic_ih | line;
+				return 0;
+			}
+		}
+		if (mip == NULL && mp_isa_bus != -1) {
+			for (mip = mp_busses[mp_isa_bus].mb_intrs; mip != NULL;
+			    mip=mip->next) {
+				if (mip->bus_pin == line) {
+					ihp->line = mip->ioapic_ih | line;
+					return 0;
+				}
+			}
+		}
+		if (mip == NULL && mp_eisa_bus != -1) {
+			for (mip = mp_busses[mp_eisa_bus].mb_intrs;
+			    mip != NULL; mip=mip->next) {
+				if (mip->bus_pin == line) {
+					ihp->line = mip->ioapic_ih | line;
+					return 0;
+				}
+			}
+		}
+		if (mip == NULL) {
+			printf("pci_intr_map: "
+			    "bus %d dev %d func %d pin %d; line %d\n",
+			    bus, dev, func, pin, line);
+			printf("pci_intr_map: no MP mapping found\n");
+		}
+	}
+#endif
 
 	return 0;
 
@@ -472,14 +527,22 @@ pci_intr_string(pc, ih)
 	pci_chipset_tag_t pc;
 	pci_intr_handle_t ih;
 {
-	static char irqstr[8];		/* 4 + 2 + NULL + sanity */
+	static char irqstr[64];
 
-	if (ih.line == 0 || ih.line >= ICU_LEN || ih.line == 2)
+	if (ih.line == 0 || (ih.line  & 0xff) >= ICU_LEN || ih.line == 2)
 		panic("pci_intr_string: bogus handle 0x%x", ih.line);
 
+#if NIOAPIC > 0
+	if (ih.line & APIC_INT_VIA_APIC) {
+		snprintf(irqstr, sizeof irqstr, "apic %d int %d (irq %d)",
+		     APIC_IRQ_APIC(ih.line), APIC_IRQ_PIN(ih.line),
+		     ih.line & 0xff);
+		return (irqstr);
+	}
+#endif
+
 	snprintf(irqstr, sizeof irqstr, "irq %d", ih.line);
 	return (irqstr);
-	
 }
 
 void *
@@ -492,16 +555,21 @@ pci_intr_establish(pc, ih, level, func, arg, what)
 {
 	void *ret;
 
+#if NIOAPIC > 0
+	if (ih.line != -1 && ih.line & APIC_INT_VIA_APIC)
+		return (apic_intr_establish(ih.line, IST_LEVEL, level, func, 
+		    arg, what));
+#endif
 	if (ih.line == 0 || ih.line >= ICU_LEN || ih.line == 2)
 		panic("pci_intr_establish: bogus handle 0x%x", ih.line);
 
-	ret = isa_intr_establish(NULL, ih.line,
-	    IST_LEVEL, level, func, arg, what);
+	ret = isa_intr_establish(NULL, ih.line, IST_LEVEL, level, func, arg,
+	    what);
 #if NPCIBIOS > 0
 	if (ret)
 		pci_intr_route_link(pc, &ih);
 #endif
-	return ret;
+	return (ret);
 }
 
 void
@@ -510,5 +578,5 @@ pci_intr_disestablish(pc, cookie)
 	void *cookie;
 {
 	/* XXX oh, unroute the pci int link? */
-	return isa_intr_disestablish(NULL, cookie);
+	return (isa_intr_disestablish(NULL, cookie));
 }
diff --git a/sys/arch/m88k/include/cpu.h b/sys/arch/m88k/include/cpu.h
index 213c4a7d672..992d11f0cb5 100644
--- a/sys/arch/m88k/include/cpu.h
+++ b/sys/arch/m88k/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.1 2004/04/26 12:34:05 miod Exp $ */
+/*	$OpenBSD: cpu.h,v 1.2 2004/06/13 21:49:17 niklas Exp $ */
 /*
  * Copyright (c) 1996 Nivas Madhur
  * Copyright (c) 1992, 1993
@@ -103,7 +103,7 @@ extern int	want_ast;
  * or after the current trap/syscall if in system mode.
  */
 extern int	want_resched;		/* resched() was called */
-#define	need_resched()		(want_resched = 1, want_ast = 1)
+#define	need_resched(ci)		(want_resched = 1, want_ast = 1)
 
 /*
  * Give a profiling tick to the current process when the user profiling
diff --git a/sys/arch/m88k/m88k/genassym.cf b/sys/arch/m88k/m88k/genassym.cf
index 9c7329013c3..1f533586cd7 100644
--- a/sys/arch/m88k/m88k/genassym.cf
+++ b/sys/arch/m88k/m88k/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.1 2004/04/29 14:33:27 miod Exp $
+#	$OpenBSD: genassym.cf,v 1.2 2004/06/13 21:49:17 niklas Exp $
 #
 # Copyright (c) 1982, 1990 The Regents of the University of California.
 # All rights reserved.
@@ -28,7 +28,7 @@
 # SUCH DAMAGE.
 #
 #	@(#)genassym.c	7.8 (Berkeley) 5/7/91
-#	$Id: genassym.cf,v 1.1 2004/04/29 14:33:27 miod Exp $
+#	$Id: genassym.cf,v 1.2 2004/06/13 21:49:17 niklas Exp $
 #
 
 include <sys/param.h>
@@ -53,6 +53,7 @@ member	p_stat
 member	p_wchan
 
 export	SRUN
+export	SONPROC
 
 # general constants
 export	UPAGES
diff --git a/sys/arch/m88k/m88k/process.S b/sys/arch/m88k/m88k/process.S
index 9c460e544da..aef28aca221 100644
--- a/sys/arch/m88k/m88k/process.S
+++ b/sys/arch/m88k/m88k/process.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: process.S,v 1.1 2004/04/29 14:33:27 miod Exp $	*/
+/*	$OpenBSD: process.S,v 1.2 2004/06/13 21:49:17 niklas Exp $	*/
 /*
  * Copyright (c) 1996 Nivas Madhur
  * All rights reserved.
@@ -215,6 +215,8 @@ ASLOCAL(Ldoneloop)
 
 	or.u	r11, r0, hi16(_C_LABEL(curproc))
 	st	r9,  r11,lo16(_C_LABEL(curproc))	/* curproc = p */
+	or	r2, r0, SONPROC
+	st.b	r2, r9, P_STAT
 
 	/*  r9 is curproc */
 	st	r0, r9, P_BACK			/* p->p_back = 0 */
diff --git a/sys/arch/mac68k/include/cpu.h b/sys/arch/mac68k/include/cpu.h
index ed14fadfadf..5fdded9e731 100644
--- a/sys/arch/mac68k/include/cpu.h
+++ b/sys/arch/mac68k/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.29 2004/05/20 09:20:42 kettenis Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.30 2004/06/13 21:49:17 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.45 1997/02/10 22:13:40 scottr Exp $	*/
 
 /*
@@ -106,7 +106,7 @@ struct clockframe {
  * or after the current trap/syscall if in system mode.
  */
 extern int want_resched;	/* resched() was called */
-#define	need_resched()	{ want_resched++; aston(); }
+#define	need_resched(ci)	{ want_resched++; aston(); }
 
 /*
  * Give a profiling tick to the current process from the softclock
diff --git a/sys/arch/macppc/macppc/genassym.cf b/sys/arch/macppc/macppc/genassym.cf
index 45d05ddaa5e..a8bce6e5b4f 100644
--- a/sys/arch/macppc/macppc/genassym.cf
+++ b/sys/arch/macppc/macppc/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.9 2003/10/16 05:03:22 deraadt Exp $
+#	$OpenBSD: genassym.cf,v 1.10 2004/06/13 21:49:18 niklas Exp $
 #
 # Copyright (c) 1982, 1990 The Regents of the University of California.
 # All rights reserved.
@@ -71,3 +71,6 @@ struct	proc
 member	p_forw
 member	p_back
 member	p_addr
+member	p_stat
+
+export	SONPROC
diff --git a/sys/arch/macppc/macppc/locore.S b/sys/arch/macppc/macppc/locore.S
index fe21293bff7..4536ffd14d6 100644
--- a/sys/arch/macppc/macppc/locore.S
+++ b/sys/arch/macppc/macppc/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.15 2004/01/03 00:57:06 pvalchev Exp $	*/
+/*	$OpenBSD: locore.S,v 1.16 2004/06/13 21:49:18 niklas Exp $	*/
 /*	$NetBSD: locore.S,v 1.2 1996/10/16 19:33:09 ws Exp $	*/
 
 /*
@@ -247,6 +247,9 @@ _C_LABEL(sw1):
 	lis	%r4,_C_LABEL(curproc)@ha
 	stw	%r31,_C_LABEL(curproc)@l(%r4)	/* record new process */
 
+	li	%r3,SONPROC
+	stb	%r3,P_STAT(%r31)
+
 	mfmsr	%r3
 	ori	%r3,%r3,PSL_EE		/* Now we can interrupt again */
 	mtmsr	%r3
diff --git a/sys/arch/mvme68k/include/cpu.h b/sys/arch/mvme68k/include/cpu.h
index 091cd1aec50..5f3aeb08da3 100644
--- a/sys/arch/mvme68k/include/cpu.h
+++ b/sys/arch/mvme68k/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.17 2004/05/20 09:20:42 kettenis Exp $ */
+/*	$OpenBSD: cpu.h,v 1.18 2004/06/13 21:49:18 niklas Exp $ */
 
 /*
  * Copyright (c) 1995 Theo de Raadt
@@ -127,7 +127,7 @@ struct clockframe {
  * or after the current trap/syscall if in system mode.
  */
 extern int want_resched;
-#define	need_resched()	{ want_resched = 1; aston(); }
+#define	need_resched(ci)	{ want_resched = 1; aston(); }
 
 /*
  * Give a profiling tick to the current process when the user profiling
diff --git a/sys/arch/powerpc/include/cpu.h b/sys/arch/powerpc/include/cpu.h
index f80c925b2a6..9fcae7e00b8 100644
--- a/sys/arch/powerpc/include/cpu.h
+++ b/sys/arch/powerpc/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.18 2004/02/14 15:09:22 grange Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.19 2004/06/13 21:49:19 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 1996/09/30 16:34:21 ws Exp $	*/
 
 /*
@@ -51,7 +51,7 @@ void	delay(unsigned);
 extern volatile int want_resched;
 extern volatile int astpending;
 
-#define	need_resched()		(want_resched = 1, astpending = 1)
+#define	need_resched(ci)		(want_resched = 1, astpending = 1)
 #define	need_proftick(p)	((p)->p_flag |= P_OWEUPC, astpending = 1)
 #define	signotify(p)		(astpending = 1)
 
diff --git a/sys/arch/sparc/include/cpu.h b/sys/arch/sparc/include/cpu.h
index a958a69c803..a1ca691938a 100644
--- a/sys/arch/sparc/include/cpu.h
+++ b/sys/arch/sparc/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.20 2003/06/02 23:27:54 millert Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.21 2004/06/13 21:49:19 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.24 1997/03/15 22:25:15 pk Exp $ */
 
 /*
@@ -129,7 +129,7 @@ extern void	raise(int, int);
  * or after the current trap/syscall if in system mode.
  */
 extern int	want_resched;		/* resched() was called */
-#define	need_resched()		(want_resched = 1, want_ast = 1)
+#define	need_resched(ci)		(want_resched = 1, want_ast = 1)
 extern int	want_ast;
 
 /*
diff --git a/sys/arch/sparc/include/psl.h b/sys/arch/sparc/include/psl.h
index 6bb6785e66b..af175bba370 100644
--- a/sys/arch/sparc/include/psl.h
+++ b/sys/arch/sparc/include/psl.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: psl.h,v 1.18 2003/06/02 23:27:54 millert Exp $	*/
+/*	$OpenBSD: psl.h,v 1.19 2004/06/13 21:49:20 niklas Exp $	*/
 /*	$NetBSD: psl.h,v 1.12 1997/03/10 21:49:11 pk Exp $ */
 
 /*
@@ -95,6 +95,7 @@
  */
 #define IPL_AUHARD	13		/* hard audio interrupts */
 #define IPL_STATCLOCK	14		/* statclock() */
+#define IPL_HIGH	15		/* splhigh() */
 
 #if defined(_KERNEL) && !defined(_LOCORE)
 
diff --git a/sys/arch/sparc/sparc/genassym.cf b/sys/arch/sparc/sparc/genassym.cf
index 3281b9e502d..71660d9efde 100644
--- a/sys/arch/sparc/sparc/genassym.cf
+++ b/sys/arch/sparc/sparc/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.14 2003/06/02 23:27:55 millert Exp $
+#	$OpenBSD: genassym.cf,v 1.15 2004/06/13 21:49:20 niklas Exp $
 #	$NetBSD: genassym.cf,v 1.2 1997/06/28 19:59:04 pk Exp $
 
 #
@@ -84,6 +84,7 @@ member	p_wchan
 member	p_vmspace
 
 export	SRUN
+export	SONPROC
 
 # VM structure fields
 define	VM_PMAP		offsetof(struct vmspace, vm_map.pmap)
diff --git a/sys/arch/sparc/sparc/locore.s b/sys/arch/sparc/sparc/locore.s
index 5dbc64a1ab3..026c076cde2 100644
--- a/sys/arch/sparc/sparc/locore.s
+++ b/sys/arch/sparc/sparc/locore.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.s,v 1.58 2004/06/08 18:06:53 art Exp $	*/
+/*	$OpenBSD: locore.s,v 1.59 2004/06/13 21:49:20 niklas Exp $	*/
 /*	$NetBSD: locore.s,v 1.73 1997/09/13 20:36:48 pk Exp $	*/
 
 /*
@@ -4638,6 +4638,8 @@ Lsw_scan:
 	 * Committed to running process p.
 	 * It may be the same as the one we were running before.
 	 */
+	mov	SONPROC, %o0			! p->p_stat = SONPROC
+	stb	%o0, [%g3 + P_STAT]
 	sethi	%hi(_C_LABEL(want_resched)), %o0
 	st	%g0, [%o0 + %lo(_C_LABEL(want_resched))]	! want_resched = 0;
 	ld	[%g3 + P_ADDR], %g5		! newpcb = p->p_addr;
diff --git a/sys/arch/sparc64/include/cpu.h b/sys/arch/sparc64/include/cpu.h
index 9d73cd48b84..278f7a3031b 100644
--- a/sys/arch/sparc64/include/cpu.h
+++ b/sys/arch/sparc64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.24 2003/07/10 15:26:54 jason Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.25 2004/06/13 21:49:20 niklas Exp $	*/
 /*	$NetBSD: cpu.h,v 1.28 2001/06/14 22:56:58 thorpej Exp $ */
 
 /*
@@ -75,33 +75,7 @@
 #include <machine/reg.h>
 #include <machine/intr.h>
 
-/*#include <sys/sched.h> */
-
-/*
- * CPU states.
- * XXX Not really scheduler state, but no other good place to put
- * it right now, and it really is per-CPU.
- */
-#define CP_USER         0
-#define CP_NICE         1
-#define CP_SYS          2
-#define CP_INTR         3
-#define CP_IDLE         4
-#define CPUSTATES       5
- 
-/*
- * Per-CPU scheduler state.
- */
-struct schedstate_percpu {
-	struct timeval spc_runtime;     /* time curproc started running */
-	__volatile int spc_flags;       /* flags; see below */
-	u_int spc_schedticks;           /* ticks for schedclock() */
-	u_int64_t spc_cp_time[CPUSTATES]; /* CPU state statistics */
-	u_char spc_curpriority;         /* usrpri of curproc */
-	int spc_rrticks;                /* ticks until roundrobin() */
-	int spc_pscnt;			/* prof/stat counter */
-	int spc_psdiv;			/* prof/stat divisor */
-};
+#include <sys/proc.h>
 
 /*
  * The cpu_info structure is part of a 64KB structure mapped both the kernel
@@ -203,7 +177,7 @@ extern	int want_ast;
  * or after the current trap/syscall if in system mode.
  */
 extern	int want_resched;	/* resched() was called */
-#define	need_resched()		(want_resched = 1, want_ast = 1)
+#define	need_resched(ci)	(want_resched = 1, want_ast = 1)
 
 /*
  * Give a profiling tick to the current process when the user profiling
diff --git a/sys/arch/sparc64/sparc64/genassym.cf b/sys/arch/sparc64/sparc64/genassym.cf
index f2c55d313b1..cde690d07d1 100644
--- a/sys/arch/sparc64/sparc64/genassym.cf
+++ b/sys/arch/sparc64/sparc64/genassym.cf
@@ -132,6 +132,7 @@ member	p_pid
 member	P_FPSTATE	p_md.md_fpstate
 
 export	SRUN
+export	SONPROC
 
 # user structure fields
 define	USIZ		sizeof(struct user)
diff --git a/sys/arch/sparc64/sparc64/locore.s b/sys/arch/sparc64/sparc64/locore.s
index 2705d2b30cc..b6fdf4746aa 100644
--- a/sys/arch/sparc64/sparc64/locore.s
+++ b/sys/arch/sparc64/sparc64/locore.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.s,v 1.44 2004/04/23 04:18:17 marc Exp $	*/
+/*	$OpenBSD: locore.s,v 1.45 2004/06/13 21:49:21 niklas Exp $	*/
 /*	$NetBSD: locore.s,v 1.137 2001/08/13 06:10:10 jdolecek Exp $	*/
 
 /*
@@ -5879,10 +5879,8 @@ Lsw_scan:
 	 * p->p_cpu = curcpu();
 	 */
 #endif	/* defined(MULTIPROCESSOR) */
-#ifdef notyet
 	mov	SONPROC, %o0			! p->p_stat = SONPROC
 	stb	%o0, [%l3 + P_STAT]
-#endif	/* notyet */
 	sethi	%hi(_C_LABEL(want_resched)), %o0
 	st	%g0, [%o0 + %lo(_C_LABEL(want_resched))]	! want_resched = 0;
 	ldx	[%l3 + P_ADDR], %l1		! newpcb = p->p_addr;
diff --git a/sys/arch/vax/include/cpu.h b/sys/arch/vax/include/cpu.h
index 03bd14a720d..a0df493ff3e 100644
--- a/sys/arch/vax/include/cpu.h
+++ b/sys/arch/vax/include/cpu.h
@@ -1,4 +1,4 @@
-/*      $OpenBSD: cpu.h,v 1.15 2003/11/10 21:05:06 miod Exp $      */
+/*      $OpenBSD: cpu.h,v 1.16 2004/06/13 21:49:21 niklas Exp $      */
 /*      $NetBSD: cpu.h,v 1.41 1999/10/21 20:01:36 ragge Exp $      */
 
 /*
@@ -88,7 +88,7 @@ extern int bootdev;
  * or after the current trap/syscall if in system mode.
  */
 
-#define need_resched(){ \
+#define need_resched(ci){ \
 	want_resched++; \
 	mtpr(AST_OK,PR_ASTLVL); \
 	}
diff --git a/sys/arch/vax/vax/genassym.cf b/sys/arch/vax/vax/genassym.cf
index 7e2b63bd89b..209fb80ecd1 100644
--- a/sys/arch/vax/vax/genassym.cf
+++ b/sys/arch/vax/vax/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.4 2002/01/23 23:24:40 miod Exp $
+#	$OpenBSD: genassym.cf,v 1.5 2004/06/13 21:49:22 niklas Exp $
 #	$NetBSD: genassym.cf,v 1.10 1999/11/19 22:09:55 ragge Exp $
 #
 # Copyright (c) 1997 Ludd, University of Lule}, Sweden.
@@ -48,6 +48,9 @@ struct	proc
 member	p_priority
 member	p_addr
 member	p_vmspace
+member	p_stat
+
+export	SONPROC
 
 struct	pcb
 member	P0BR
diff --git a/sys/arch/vax/vax/subr.s b/sys/arch/vax/vax/subr.s
index 89d6e993326..0efb63df499 100644
--- a/sys/arch/vax/vax/subr.s
+++ b/sys/arch/vax/vax/subr.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: subr.s,v 1.19 2003/11/07 10:16:45 jmc Exp $     */
+/*	$OpenBSD: subr.s,v 1.20 2004/06/13 21:49:22 niklas Exp $     */
 /*	$NetBSD: subr.s,v 1.32 1999/03/25 00:41:48 mrg Exp $	   */
 
 /*
@@ -306,6 +306,7 @@ noque:	.asciz	"swtch"
 	bbsc	r3,_whichqs,2f		# no, clear bit in whichqs
 2:	clrl	4(r2)			# clear proc backpointer
 	clrl	_want_resched		# we are now changing process
+	movb	$SONPROC,P_STAT(r2)	# p->p_stat = SONPROC
 	movl	r2,_curproc		# set new process running
 	cmpl	r0,r2			# Same process?
 	bneq	1f			# No, continue
diff --git a/sys/compat/linux/linux_sched.c b/sys/compat/linux/linux_sched.c
index 4d47196b37f..c5b1b008db9 100644
--- a/sys/compat/linux/linux_sched.c
+++ b/sys/compat/linux/linux_sched.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: linux_sched.c,v 1.3 2001/11/06 18:41:10 art Exp $	*/
+/*	$OpenBSD: linux_sched.c,v 1.4 2004/06/13 21:49:23 niklas Exp $	*/
 /*	$NetBSD: linux_sched.c,v 1.6 2000/05/28 05:49:05 thorpej Exp $	*/
 
 /*-
@@ -272,7 +272,7 @@ linux_sys_sched_yield(cp, v, retval)
 	void *v;
 	register_t *retval;
 {
-	need_resched();
+	need_resched(curcpu());
 	return (0);
 }
 
diff --git a/sys/ddb/db_command.c b/sys/ddb/db_command.c
index f62fff34850..12477e0b084 100644
--- a/sys/ddb/db_command.c
+++ b/sys/ddb/db_command.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_command.c,v 1.35 2004/04/25 03:21:50 itojun Exp $	*/
+/*	$OpenBSD: db_command.c,v 1.36 2004/06/13 21:49:23 niklas Exp $	*/
 /*	$NetBSD: db_command.c,v 1.20 1996/03/30 22:30:05 christos Exp $	*/
 
 /* 
@@ -60,6 +60,11 @@
 boolean_t	db_cmd_loop_done;
 label_t		*db_recover;
 
+#ifdef MULTIPROCESSOR
+boolean_t	db_switch_cpu;
+long		db_switch_to_cpu;
+#endif
+
 /*
  * if 'ed' style: 'dot' is set at start of last item printed,
  * and '+' points to next line.
@@ -504,6 +509,11 @@ db_command_loop()
 	label_t		*savejmp;
 	extern int	db_output_line;
 
+#ifdef MULTIPROCESSOR
+	db_switch_cpu = 0;
+	db_enter_ddb();
+#endif /* MULTIPROCESSOR */
+
 	/*
 	 * Initialize 'prev' and 'next' to dot.
 	 */
@@ -517,17 +527,35 @@ db_command_loop()
 	(void) setjmp(&db_jmpbuf);
 
 	while (!db_cmd_loop_done) {
+
 		if (db_print_position() != 0)
 			db_printf("\n");
 		db_output_line = 0;
 
+#ifdef MULTIPROCESSOR
+		db_printf("ddb{%ld}> ", (long)cpu_number());
+#else
 		db_printf("ddb> ");
+#endif
 		(void) db_read_line();
 
 		db_command(&db_last_command, db_command_table);
 	}
 
 	db_recover = savejmp;
+
+#ifdef MULTIPROCESSOR
+	if (db_switch_cpu) {
+		db_printf("Moving ddb to cpu %d\n", db_switch_to_cpu);
+		curcpu()->ci_ddb_paused = CI_DDB_STOPPED;
+		db_movetocpu(db_switch_to_cpu);
+		while (curcpu()->ci_ddb_paused == CI_DDB_SHOULDSTOP
+		    || curcpu()->ci_ddb_paused == CI_DDB_STOPPED)
+			;	/* Do nothing */
+	} else {
+		db_leave_ddb();
+	}
+#endif /* MULTIPROCESSOR */
 }
 
 void
diff --git a/sys/dev/isa/aria.c b/sys/dev/isa/aria.c
index 2da86c3bb26..a73d9805f4b 100644
--- a/sys/dev/isa/aria.c
+++ b/sys/dev/isa/aria.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: aria.c,v 1.10 2004/01/09 21:32:23 brad Exp $ */
+/*	$OpenBSD: aria.c,v 1.11 2004/06/13 21:49:24 niklas Exp $ */
 
 /*
  * Copyright (c) 1995, 1996 Roland C. Dowdeswell.  All rights reserved.
@@ -76,7 +76,6 @@
 #include <dev/mulaw.h>
 #include <dev/isa/isavar.h>
 #include <dev/isa/isadmavar.h>
-#include <i386/isa/icu.h>
 
 #include <dev/isa/ariareg.h>
 
diff --git a/sys/dev/isa/gus.c b/sys/dev/isa/gus.c
index 649e7c11d7d..ff9e0ffef32 100644
--- a/sys/dev/isa/gus.c
+++ b/sys/dev/isa/gus.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gus.c,v 1.26 2003/06/08 00:41:47 miod Exp $	*/
+/*	$OpenBSD: gus.c,v 1.27 2004/06/13 21:49:24 niklas Exp $	*/
 /*	$NetBSD: gus.c,v 1.51 1998/01/25 23:48:06 mycroft Exp $	*/
 
 /*-
@@ -119,7 +119,6 @@
 
 #include <dev/isa/isavar.h>
 #include <dev/isa/isadmavar.h>
-#include <i386/isa/icu.h>
 
 #include <dev/ic/ics2101reg.h>
 #include <dev/ic/cs4231reg.h>
diff --git a/sys/dev/isa/gus_isa.c b/sys/dev/isa/gus_isa.c
index 5144a526691..5140bda769b 100644
--- a/sys/dev/isa/gus_isa.c
+++ b/sys/dev/isa/gus_isa.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gus_isa.c,v 1.3 2002/03/14 01:26:56 millert Exp $	*/
+/*	$OpenBSD: gus_isa.c,v 1.4 2004/06/13 21:49:24 niklas Exp $	*/
 /*	$NetBSD: gus.c,v 1.51 1998/01/25 23:48:06 mycroft Exp $	*/
 
 /*-
@@ -119,7 +119,6 @@
 
 #include <dev/isa/isavar.h>
 #include <dev/isa/isadmavar.h>
-#include <i386/isa/icu.h>
 
 #include <dev/ic/ics2101reg.h>
 #include <dev/ic/cs4231reg.h>
diff --git a/sys/dev/isa/gus_isapnp.c b/sys/dev/isa/gus_isapnp.c
index 0700ce4745f..37c5415a643 100644
--- a/sys/dev/isa/gus_isapnp.c
+++ b/sys/dev/isa/gus_isapnp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gus_isapnp.c,v 1.3 2002/03/14 01:26:56 millert Exp $	*/
+/*	$OpenBSD: gus_isapnp.c,v 1.4 2004/06/13 21:49:24 niklas Exp $	*/
 /*	$NetBSD: gus.c,v 1.51 1998/01/25 23:48:06 mycroft Exp $	*/
 
 /*-
@@ -119,7 +119,6 @@
 
 #include <dev/isa/isavar.h>
 #include <dev/isa/isadmavar.h>
-#include <i386/isa/icu.h>
 
 #include <dev/ic/ics2101reg.h>
 #include <dev/ic/cs4231reg.h>
diff --git a/sys/dev/isa/if_hp.c b/sys/dev/isa/if_hp.c
index 2d31185b352..5e05b2296a8 100644
--- a/sys/dev/isa/if_hp.c
+++ b/sys/dev/isa/if_hp.c
@@ -1,4 +1,4 @@
-/*    $OpenBSD: if_hp.c,v 1.11 2004/05/12 06:35:11 tedu Exp $       */
+/*    $OpenBSD: if_hp.c,v 1.12 2004/06/13 21:49:24 niklas Exp $       */
 /*    $NetBSD: if_hp.c,v 1.21 1995/12/24 02:31:31 mycroft Exp $       */
 
 /* XXX THIS DRIVER IS BROKEN.  IT WILL NOT EVEN COMPILE. */
@@ -85,7 +85,6 @@
 
 #include <i386/isa/isa_device.h>	/* XXX BROKEN */
 #include <dev/isa/if_nereg.h>
-#include <i386/isa/icu.h>		/* XXX BROKEN */
 
 int     hpprobe(), hpattach(), hpintr();
 int     hpstart(), hpinit(), ether_output(), hpioctl();
diff --git a/sys/dev/isa/opti.c b/sys/dev/isa/opti.c
index ce2f3b0b00f..e55d4151361 100644
--- a/sys/dev/isa/opti.c
+++ b/sys/dev/isa/opti.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: opti.c,v 1.7 2003/06/02 19:24:22 mickey Exp $	*/
+/*	$OpenBSD: opti.c,v 1.8 2004/06/13 21:49:24 niklas Exp $	*/
 
 /*
  * Copyright (c) 1996 Michael Shalayeff
@@ -41,7 +41,6 @@
 
 #include <machine/pio.h>
 
-#include <i386/isa/icu.h>
 #include <dev/isa/isavar.h>
 
 #include <dev/isa/opti.h>
diff --git a/sys/dev/pci/cy82c693.c b/sys/dev/pci/cy82c693.c
index e50bd84b545..6cf96d3f740 100644
--- a/sys/dev/pci/cy82c693.c
+++ b/sys/dev/pci/cy82c693.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cy82c693.c,v 1.4 2001/02/26 16:21:46 art Exp $	*/
+/*	$OpenBSD: cy82c693.c,v 1.5 2004/06/13 21:49:25 niklas Exp $	*/
 /* $NetBSD: cy82c693.c,v 1.1 2000/06/06 03:07:39 thorpej Exp $ */
 
 /*-
@@ -60,7 +60,7 @@
 static struct cy82c693_handle cyhc_handle;
 static int cyhc_initialized;
 
-struct simplelock cyhc_slock = SLOCK_INITIALIZER;
+struct simplelock cyhc_slock;
 
 #define	CYHC_LOCK(s)							\
 do {									\
@@ -81,6 +81,8 @@ cy82c693_init(bus_space_tag_t iot)
 	int s;
 	int error;
 
+	simple_lock_init(&cyhc_slock);
+
 	CYHC_LOCK(s);
 
 	if (cyhc_initialized) {
diff --git a/sys/kern/init_main.c b/sys/kern/init_main.c
index 55dad5457b4..a8d052d0a05 100644
--- a/sys/kern/init_main.c
+++ b/sys/kern/init_main.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: init_main.c,v 1.115 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: init_main.c,v 1.116 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: init_main.c,v 1.84.4.1 1996/06/02 09:08:06 mrg Exp $	*/
 
 /*
@@ -125,7 +125,6 @@ struct	timeval boottime;
 #ifndef __HAVE_CPUINFO
 struct	timeval runtime;
 #endif
-
 int	ncpus =  1;
 
 #if !defined(NO_PROPOLICE)
@@ -188,6 +187,7 @@ main(framep)
 	int s, i;
 	register_t rval[2];
 	extern struct pdevinit pdevinit[];
+	extern struct SIMPLELOCK kprintf_slock;
 	extern void scheduler_start(void);
 	extern void disk_init(void);
 	extern void endtsleep(void *);
@@ -213,8 +213,13 @@ main(framep)
 	 */
 	config_init();		/* init autoconfiguration data structures */
 	consinit();
+
+	SIMPLE_LOCK_INIT(&kprintf_slock);
+
 	printf("%s\n", copyright);
 
+	KERNEL_LOCK_INIT();
+
 	uvm_init();
 	disk_init();		/* must come before autoconfiguration */
 	tty_init();		/* initialise tty's */
@@ -270,7 +275,7 @@ main(framep)
 	session0.s_leader = p;
 
 	p->p_flag = P_INMEM | P_SYSTEM | P_NOCLDWAIT;
-	p->p_stat = SRUN;
+	p->p_stat = SONPROC;
 	p->p_nice = NZERO;
 	p->p_emul = &emul_native;
 	bcopy("swapper", p->p_comm, sizeof ("swapper"));
@@ -344,6 +349,9 @@ main(framep)
 	/* Start real time and statistics clocks. */
 	initclocks();
 
+	/* Lock the kernel on behalf of proc0. */
+	KERNEL_PROC_LOCK(p);
+
 #ifdef SYSVSHM
 	/* Initialize System V style shared memory. */
 	shminit();
@@ -413,8 +421,6 @@ main(framep)
 	VOP_UNLOCK(rootvnode, 0, p);
 	p->p_fd->fd_rdir = NULL;
 
-	uvm_swap_init();
-
 	/*
 	 * Now can look at time, having had a chance to verify the time
 	 * from the file system.  Reset p->p_rtime as it may have been
@@ -424,10 +430,12 @@ main(framep)
 	p->p_stats->p_start = mono_time = boottime = time;
 	p->p_cpu->ci_schedstate.spc_runtime = time;
 #else
-	p->p_stats->p_start = runtime = mono_time = boottime = time;
+	p->p_stats->p_start = runtime = mono_time = boottime = time;	
 #endif
 	p->p_rtime.tv_sec = p->p_rtime.tv_usec = 0;
 
+	uvm_swap_init();
+
 	/* Create process 1 (init(8)). */
 	if (fork1(p, SIGCHLD, FORK_FORK, NULL, 0, start_init, NULL, rval))
 		panic("fork init");
@@ -465,6 +473,12 @@ main(framep)
 	srandom((u_long)(rtv.tv_sec ^ rtv.tv_usec));
 
 	randompid = 1;
+
+#if defined(MULTIPROCESSOR)
+	/* Boot the secondary processors. */
+	cpu_boot_secondary_processors();
+#endif
+
 	/* The scheduler is an infinite loop. */
 	uvm_scheduler();
 	/* NOTREACHED */
@@ -623,8 +637,10 @@ start_init(arg)
 		 * Now try to exec the program.  If can't for any reason
 		 * other than it doesn't exist, complain.
 		 */
-		if ((error = sys_execve(p, &args, retval)) == 0)
+		if ((error = sys_execve(p, &args, retval)) == 0) {
+			KERNEL_PROC_UNLOCK(p);
 			return;
+		}
 		if (error != ENOENT)
 			printf("exec %s: error %d\n", path, error);
 	}
diff --git a/sys/kern/kern_clock.c b/sys/kern/kern_clock.c
index 8a34e63d16c..5a3df5ecec4 100644
--- a/sys/kern/kern_clock.c
+++ b/sys/kern/kern_clock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_clock.c,v 1.43 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_clock.c,v 1.44 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_clock.c,v 1.34 1996/06/09 04:51:03 briggs Exp $	*/
 
 /*-
@@ -192,9 +192,16 @@ hardclock(struct clockframe *frame)
 	if (stathz == 0)
 		statclock(frame);
 
-#ifdef __HAVE_CPUINFO
+#if defined(__HAVE_CPUINFO)
 	if (--ci->ci_schedstate.spc_rrticks <= 0)
 		roundrobin(ci);
+
+	/*
+	 * If we are not the primary CPU, we're not allowed to do
+	 * any more work.
+	 */
+	if (CPU_IS_PRIMARY(ci) == 0)
+		return;
 #endif
 
 	/*
@@ -420,9 +427,10 @@ statclock(struct clockframe *frame)
 		if (psdiv == 1) {
 			setstatclockrate(stathz);
 		} else {
-			setstatclockrate(profhz);                       
+			setstatclockrate(profhz);			
 		}
 	}
+
 /* XXX Kludgey */
 #define pscnt spc->spc_pscnt
 #define cp_time spc->spc_cp_time
@@ -483,7 +491,7 @@ statclock(struct clockframe *frame)
 	pscnt = psdiv;
 
 #ifdef __HAVE_CPUINFO
-#undef pscnt
+#undef psdiv
 #undef cp_time
 #endif
 
@@ -495,7 +503,8 @@ statclock(struct clockframe *frame)
 		 */
 		if (schedhz == 0) {
 #ifdef __HAVE_CPUINFO
-			if ((++curcpu()->ci_schedstate.spc_schedticks & 3) == 0)
+			if ((++curcpu()->ci_schedstate.spc_schedticks & 3) ==
+			    0)
 				schedclock(p);
 #else
 			if ((++schedclk & 3) == 0)
diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
index 68d77771271..1587724bc08 100644
--- a/sys/kern/kern_exit.c
+++ b/sys/kern/kern_exit.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_exit.c,v 1.50 2004/05/27 20:48:46 tedu Exp $	*/
+/*	$OpenBSD: kern_exit.c,v 1.51 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_exit.c,v 1.39 1996/04/22 01:38:25 christos Exp $	*/
 
 /*
@@ -285,6 +285,9 @@ exit1(p, rv)
 	limfree(p->p_limit);
 	p->p_limit = NULL;
 
+	/* This process no longer needs to hold the kernel lock. */
+	KERNEL_PROC_UNLOCK(p);
+
 	/*
 	 * If emulation has process exit hook, call it now.
 	 */
@@ -319,12 +322,15 @@ void
 exit2(p)
 	struct proc *p;
 {
+	int s;
 
-	simple_lock(&deadproc_slock);
+	SIMPLE_LOCK(&deadproc_slock);
 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
-	simple_unlock(&deadproc_slock);
+	SIMPLE_UNLOCK(&deadproc_slock);
 
 	wakeup(&deadproc);
+
+	SCHED_LOCK(s);
 }
 
 /*
@@ -337,19 +343,22 @@ reaper(void)
 {
 	struct proc *p;
 
+	KERNEL_PROC_UNLOCK(curproc);
+
 	for (;;) {
-		simple_lock(&deadproc_slock);
+		SIMPLE_LOCK(&deadproc_slock);
 		p = LIST_FIRST(&deadproc);
 		if (p == NULL) {
 			/* No work for us; go to sleep until someone exits. */
-			simple_unlock(&deadproc_slock);
+			SIMPLE_UNLOCK(&deadproc_slock);
 			(void) tsleep(&deadproc, PVM, "reaper", 0);
 			continue;
 		}
 
 		/* Remove us from the deadproc list. */
 		LIST_REMOVE(p, p_hash);
-		simple_unlock(&deadproc_slock);
+		SIMPLE_UNLOCK(&deadproc_slock);
+		KERNEL_PROC_LOCK(curproc);
 
 		/*
 		 * Give machine-dependent code a chance to free any
@@ -377,6 +386,9 @@ reaper(void)
 			/* Noone will wait for us. Just zap the process now */
 			proc_zap(p);
 		}
+		/* XXXNJW where should this be with respect to 
+		 * the wakeup() above? */
+		KERNEL_PROC_UNLOCK(curproc);
 	}
 }
 
diff --git a/sys/kern/kern_fork.c b/sys/kern/kern_fork.c
index a103c391634..e33ea08e005 100644
--- a/sys/kern/kern_fork.c
+++ b/sys/kern/kern_fork.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_fork.c,v 1.68 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_fork.c,v 1.69 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_fork.c,v 1.29 1996/02/09 18:59:34 christos Exp $	*/
 
 /*
@@ -204,7 +204,7 @@ fork1(struct proc *p1, int exitsig, int flags, void *stack, size_t stacksize,
 	timeout_set(&p2->p_sleep_to, endtsleep, p2);
 	timeout_set(&p2->p_realit_to, realitexpire, p2);
 
-#ifdef __HAVE_CPUINFO
+#if defined(__HAVE_CPUINFO)
 	p2->p_cpu = NULL;
 #endif
 
@@ -339,12 +339,12 @@ fork1(struct proc *p1, int exitsig, int flags, void *stack, size_t stacksize,
 	/*
 	 * Make child runnable, set start time, and add to run queue.
 	 */
-	s = splstatclock();
+	SCHED_LOCK(s);
 	p2->p_stats->p_start = time;
 	p2->p_acflag = AFORK;
 	p2->p_stat = SRUN;
 	setrunqueue(p2);
-	splx(s);
+	SCHED_UNLOCK(s);
 
 	/*
 	 * Now can be swapped.
@@ -399,3 +399,20 @@ pidtaken(pid_t pid)
 			return (1);
 	return (0);
 }
+
+#if defined(MULTIPROCESSOR)
+/*
+ * XXX This is a slight hack to get newly-formed processes to
+ * XXX acquire the kernel lock as soon as they run.
+ */
+void
+proc_trampoline_mp(void)
+{
+	struct proc *p;
+
+	p = curproc;
+
+	SCHED_ASSERT_UNLOCKED();
+	KERNEL_PROC_LOCK(p);
+}
+#endif
diff --git a/sys/kern/kern_ktrace.c b/sys/kern/kern_ktrace.c
index b811644d403..efb904c589a 100644
--- a/sys/kern/kern_ktrace.c
+++ b/sys/kern/kern_ktrace.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_ktrace.c,v 1.32 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_ktrace.c,v 1.33 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_ktrace.c,v 1.23 1996/02/09 18:59:36 christos Exp $	*/
 
 /*
@@ -37,6 +37,7 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/proc.h>
+#include <sys/sched.h>
 #include <sys/file.h>
 #include <sys/namei.h>
 #include <sys/vnode.h>
diff --git a/sys/kern/kern_lock.c b/sys/kern/kern_lock.c
index d75d09acc94..8ea70f3a097 100644
--- a/sys/kern/kern_lock.c
+++ b/sys/kern/kern_lock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_lock.c,v 1.15 2003/06/02 23:28:05 millert Exp $	*/
+/*	$OpenBSD: kern_lock.c,v 1.16 2004/06/13 21:49:26 niklas Exp $	*/
 
 /* 
  * Copyright (c) 1995
@@ -39,9 +39,20 @@
 #include <sys/proc.h>
 #include <sys/lock.h>
 #include <sys/systm.h>
+#include <sys/sched.h>
 
 #include <machine/cpu.h>
 
+#ifndef spllock
+#define spllock() splhigh()
+#endif
+
+#ifdef MULTIPROCESSOR
+#define CPU_NUMBER() cpu_number()
+#else
+#define CPU_NUMBER() 0
+#endif
+
 void record_stacktrace(int *, int);
 void playback_stacktrace(int *, int);
 
@@ -50,67 +61,243 @@ void playback_stacktrace(int *, int);
  * Locks provide shared/exclusive sychronization.
  */
 
-#if 0
-#ifdef DEBUG
-#define COUNT(p, x) if (p) (p)->p_locks += (x)
+/*
+ * Locking primitives implementation.
+ * Locks provide shared/exclusive synchronization.
+ */
+
+#if defined(LOCKDEBUG) || defined(DIAGNOSTIC) /* { */
+#if defined(MULTIPROCESSOR) /* { */
+#define	COUNT_CPU(cpu_id, x)						\
+	curcpu()->ci_spin_locks += (x)
 #else
-#define COUNT(p, x)
+u_long	spin_locks;
+#define	COUNT_CPU(cpu_id, x)	spin_locks += (x)
+#endif /* MULTIPROCESSOR */ /* } */
+
+#define	COUNT(lkp, p, cpu_id, x)					\
+do {									\
+	if ((lkp)->lk_flags & LK_SPIN)					\
+		COUNT_CPU((cpu_id), (x));				\
+	else								\
+		(p)->p_locks += (x);					\
+} while (/*CONSTCOND*/0)
+#else
+#define COUNT(lkp, p, cpu_id, x)
+#define COUNT_CPU(cpu_id, x)
+#endif /* LOCKDEBUG || DIAGNOSTIC */ /* } */
+
+#ifndef SPINLOCK_SPIN_HOOK		/* from <machine/lock.h> */
+#define	SPINLOCK_SPIN_HOOK		/* nothing */
 #endif
+
+#define	INTERLOCK_ACQUIRE(lkp, flags, s)				\
+do {									\
+	if ((flags) & LK_SPIN)						\
+		s = spllock();						\
+	simple_lock(&(lkp)->lk_interlock);				\
+} while (/*CONSTCOND*/ 0)
+
+#define	INTERLOCK_RELEASE(lkp, flags, s)				\
+do {									\
+	simple_unlock(&(lkp)->lk_interlock);				\
+	if ((flags) & LK_SPIN)						\
+		splx(s);						\
+} while (/*CONSTCOND*/ 0)
+
+#ifdef DDB /* { */
+#ifdef MULTIPROCESSOR
+int simple_lock_debugger = 1;	/* more serious on MP */
+#else
+int simple_lock_debugger = 0;
 #endif
+#define	SLOCK_DEBUGGER()	if (simple_lock_debugger) Debugger()
+#define	SLOCK_TRACE()							\
+	db_stack_trace_print((db_expr_t)__builtin_frame_address(0),	\
+	    TRUE, 65535, "", lock_printf);
+#else
+#define	SLOCK_DEBUGGER()	/* nothing */
+#define	SLOCK_TRACE()		/* nothing */
+#endif /* } */
 
-#define COUNT(p, x)
+#if defined(LOCKDEBUG)
+#if defined(DDB)
+#define	SPINLOCK_SPINCHECK_DEBUGGER	Debugger()
+#else
+#define	SPINLOCK_SPINCHECK_DEBUGGER	/* nothing */
+#endif
+
+#define	SPINLOCK_SPINCHECK_DECL						\
+	/* 32-bits of count -- wrap constitutes a "spinout" */		\
+	uint32_t __spinc = 0
 
-#if NCPUS > 1
+#define	SPINLOCK_SPINCHECK						\
+do {									\
+	if (++__spinc == 0) {						\
+		lock_printf("LK_SPIN spinout, excl %d, share %d\n",	\
+		    lkp->lk_exclusivecount, lkp->lk_sharecount);	\
+		if (lkp->lk_exclusivecount)				\
+			lock_printf("held by CPU %lu\n",		\
+			    (u_long) lkp->lk_cpu);			\
+		if (lkp->lk_lock_file)					\
+			lock_printf("last locked at %s:%d\n",		\
+			    lkp->lk_lock_file, lkp->lk_lock_line);	\
+		if (lkp->lk_unlock_file)				\
+			lock_printf("last unlocked at %s:%d\n",		\
+			    lkp->lk_unlock_file, lkp->lk_unlock_line);	\
+		SLOCK_TRACE();						\
+		SPINLOCK_SPINCHECK_DEBUGGER;				\
+	}								\
+} while (/*CONSTCOND*/ 0)
+#else
+#define	SPINLOCK_SPINCHECK_DECL			/* nothing */
+#define	SPINLOCK_SPINCHECK			/* nothing */
+#endif /* LOCKDEBUG && DDB */
 
 /*
- * For multiprocessor system, try spin lock first.
- *
- * This should be inline expanded below, but we cannot have #if
- * inside a multiline define.
+ * Acquire a resource.
  */
-int lock_wait_time = 100;
-#define PAUSE(lkp, wanted)						\
-		if (lock_wait_time > 0) {				\
-			int i;						\
+#define ACQUIRE(lkp, error, extflags, drain, wanted)			\
+	if ((extflags) & LK_SPIN) {					\
+		int interlocked;					\
+		SPINLOCK_SPINCHECK_DECL;				\
 									\
-			simple_unlock(&lkp->lk_interlock);		\
-			for (i = lock_wait_time; i > 0; i--)		\
-				if (!(wanted))				\
-					break;				\
-			simple_lock(&lkp->lk_interlock);		\
+		if ((drain) == 0)					\
+			(lkp)->lk_waitcount++;				\
+		for (interlocked = 1;;) {				\
+			SPINLOCK_SPINCHECK;				\
+			if (wanted) {					\
+				if (interlocked) {			\
+					INTERLOCK_RELEASE((lkp),	\
+					    LK_SPIN, s);		\
+					interlocked = 0;		\
+				}					\
+				SPINLOCK_SPIN_HOOK;			\
+			} else if (interlocked) {			\
+				break;					\
+			} else {					\
+				INTERLOCK_ACQUIRE((lkp), LK_SPIN, s);	\
+				interlocked = 1;			\
+			}						\
 		}							\
-		if (!(wanted))						\
-			break;
+		if ((drain) == 0)					\
+			(lkp)->lk_waitcount--;				\
+		KASSERT((wanted) == 0);					\
+		error = 0;	/* sanity */				\
+	} else {							\
+		for (error = 0; wanted; ) {				\
+			if ((drain))					\
+				(lkp)->lk_flags |= LK_WAITDRAIN;	\
+			else						\
+				(lkp)->lk_waitcount++;			\
+			/* XXX Cast away volatile. */			\
+			error = ltsleep((drain) ?			\
+			    (void *)&(lkp)->lk_flags :			\
+			    (void *)(lkp), (lkp)->lk_prio,		\
+			    (lkp)->lk_wmesg, (lkp)->lk_timo,		\
+			    &(lkp)->lk_interlock);			\
+			if ((drain) == 0)				\
+				(lkp)->lk_waitcount--;			\
+			if (error)					\
+				break;					\
+			if ((extflags) & LK_SLEEPFAIL) {		\
+				error = ENOLCK;				\
+				break;					\
+			}						\
+		}							\
+	}
 
-#else /* NCPUS == 1 */
+#define	SETHOLDER(lkp, pid, cpu_id)					\
+do {									\
+	if ((lkp)->lk_flags & LK_SPIN)					\
+		(lkp)->lk_cpu = cpu_id;					\
+	else								\
+		(lkp)->lk_lockholder = pid;				\
+} while (/*CONSTCOND*/0)
 
-/*
- * It is an error to spin on a uniprocessor as nothing will ever cause
- * the simple lock to clear while we are executing.
- */
-#define PAUSE(lkp, wanted)
+#define	WEHOLDIT(lkp, pid, cpu_id)					\
+	(((lkp)->lk_flags & LK_SPIN) != 0 ?				\
+	 ((lkp)->lk_cpu == (cpu_id)) :					\
+	 ((lkp)->lk_lockholder == (pid)))
+
+#define	WAKEUP_WAITER(lkp)						\
+do {									\
+	if (((lkp)->lk_flags & LK_SPIN) == 0 && (lkp)->lk_waitcount) {	\
+		/* XXX Cast away volatile. */				\
+		wakeup((void *)(lkp));					\
+	}								\
+} while (/*CONSTCOND*/0)
+
+#if defined(LOCKDEBUG) /* { */
+#if defined(MULTIPROCESSOR) /* { */
+struct simplelock spinlock_list_slock = SIMPLELOCK_INITIALIZER;
 
-#endif /* NCPUS == 1 */
+#define	SPINLOCK_LIST_LOCK()						\
+	__cpu_simple_lock(&spinlock_list_slock.lock_data)
+
+#define	SPINLOCK_LIST_UNLOCK()						\
+	__cpu_simple_unlock(&spinlock_list_slock.lock_data)
+#else
+#define	SPINLOCK_LIST_LOCK()	/* nothing */
 
+#define	SPINLOCK_LIST_UNLOCK()	/* nothing */
+#endif /* MULTIPROCESSOR */ /* } */
+
+TAILQ_HEAD(, lock) spinlock_list =
+    TAILQ_HEAD_INITIALIZER(spinlock_list);
+
+#define	HAVEIT(lkp)							\
+do {									\
+	if ((lkp)->lk_flags & LK_SPIN) {				\
+		int s = spllock();					\
+		SPINLOCK_LIST_LOCK();					\
+		/* XXX Cast away volatile. */				\
+		TAILQ_INSERT_TAIL(&spinlock_list, (struct lock *)(lkp),	\
+		    lk_list);						\
+		SPINLOCK_LIST_UNLOCK();					\
+		splx(s);						\
+	}								\
+} while (/*CONSTCOND*/0)
+
+#define	DONTHAVEIT(lkp)							\
+do {									\
+	if ((lkp)->lk_flags & LK_SPIN) {				\
+		int s = spllock();					\
+		SPINLOCK_LIST_LOCK();					\
+		/* XXX Cast away volatile. */				\
+		TAILQ_REMOVE(&spinlock_list, (struct lock *)(lkp),	\
+		    lk_list);						\
+		SPINLOCK_LIST_UNLOCK();					\
+		splx(s);						\
+	}								\
+} while (/*CONSTCOND*/0)
+#else
+#define	HAVEIT(lkp)		/* nothing */
+
+#define	DONTHAVEIT(lkp)		/* nothing */
+#endif /* LOCKDEBUG */ /* } */
+
+#if defined(LOCKDEBUG)
 /*
- * Acquire a resource.
+ * Lock debug printing routine; can be configured to print to console
+ * or log to syslog.
  */
-#define ACQUIRE(lkp, error, extflags, wanted)				\
-	PAUSE(lkp, wanted);						\
-	for (error = 0; wanted; ) {					\
-		(lkp)->lk_waitcount++;					\
-		simple_unlock(&(lkp)->lk_interlock);			\
-		error = tsleep((void *)lkp, (lkp)->lk_prio,		\
-		    (lkp)->lk_wmesg, (lkp)->lk_timo);			\
-		simple_lock(&(lkp)->lk_interlock);			\
-		(lkp)->lk_waitcount--;					\
-		if (error)						\
-			break;						\
-		if ((extflags) & LK_SLEEPFAIL) {			\
-			error = ENOLCK;					\
-			break;						\
-		}							\
+void
+lock_printf(const char *fmt, ...)
+{
+	char b[150];
+	va_list ap;
+
+	va_start(ap, fmt);
+	if (lock_debug_syslog)
+		vlog(LOG_DEBUG, fmt, ap);
+	else {
+		vsnprintf(b, sizeof(b), fmt, ap);
+		printf_nolog("%s", b);
 	}
+	va_end(ap);
+}
+#endif /* LOCKDEBUG */
 
 /*
  * Initialize a lock; required before use.
@@ -127,10 +314,18 @@ lockinit(lkp, prio, wmesg, timo, flags)
 	bzero(lkp, sizeof(struct lock));
 	simple_lock_init(&lkp->lk_interlock);
 	lkp->lk_flags = flags & LK_EXTFLG_MASK;
-	lkp->lk_prio = prio;
-	lkp->lk_timo = timo;
-	lkp->lk_wmesg = wmesg;
-	lkp->lk_lockholder = LK_NOPROC;
+	if (flags & LK_SPIN)
+		lkp->lk_cpu = LK_NOCPU;
+	else {
+		lkp->lk_lockholder = LK_NOPROC;
+		lkp->lk_prio = prio;
+		lkp->lk_timo = timo;
+	}
+	lkp->lk_wmesg = wmesg;	/* just a name for spin locks */
+#if defined(LOCKDEBUG)
+	lkp->lk_lock_file = NULL;
+	lkp->lk_unlock_file = NULL;
+#endif
 }
 
 /*
@@ -140,14 +335,14 @@ int
 lockstatus(lkp)
 	struct lock *lkp;
 {
-	int lock_type = 0;
+	int s = 0, lock_type = 0;
 
-	simple_lock(&lkp->lk_interlock);
+	INTERLOCK_ACQUIRE(lkp, lkp->lk_flags, s);
 	if (lkp->lk_exclusivecount != 0)
 		lock_type = LK_EXCLUSIVE;
 	else if (lkp->lk_sharecount != 0)
 		lock_type = LK_SHARED;
-	simple_unlock(&lkp->lk_interlock);
+	INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
 	return (lock_type);
 }
 
@@ -168,17 +363,33 @@ lockmgr(lkp, flags, interlkp, p)
 	int error;
 	pid_t pid;
 	int extflags;
+	cpuid_t cpu_id;
+	int s = 0;
 
 	error = 0;
-	if (p)
-		pid = p->p_pid;
-	else
-		pid = LK_KERNPROC;
-	simple_lock(&lkp->lk_interlock);
+
+	INTERLOCK_ACQUIRE(lkp, lkp->lk_flags, s);
 	if (flags & LK_INTERLOCK)
 		simple_unlock(interlkp);
 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
-#ifdef DIAGNOSTIC
+
+#ifdef DIAGNOSTIC /* { */
+	/*
+	 * Don't allow spins on sleep locks and don't allow sleeps
+	 * on spin locks.
+	 */
+	if ((flags ^ lkp->lk_flags) & LK_SPIN)
+		panic("lockmgr: sleep/spin mismatch");
+#endif /* } */
+
+	if (extflags & LK_SPIN) {
+		pid = LK_KERNPROC;
+	} else {
+		/* Process context required. */
+		pid = p->p_pid;
+	}
+	cpu_id = CPU_NUMBER();
+
 	/*
 	 * Once a lock has drained, the LK_DRAINING flag is set and an
 	 * exclusive lock is returned. The only valid operation thereafter
@@ -191,12 +402,14 @@ lockmgr(lkp, flags, interlkp, p)
 	 * the lock by specifying LK_REENABLE.
 	 */
 	if (lkp->lk_flags & (LK_DRAINING|LK_DRAINED)) {
+#ifdef DIAGNOSTIC
 		if (lkp->lk_flags & LK_DRAINED)
 			panic("lockmgr: using decommissioned lock");
 		if ((flags & LK_TYPE_MASK) != LK_RELEASE ||
-		    lkp->lk_lockholder != pid)
+		    WEHOLDIT(lkp, pid, cpu_id) == 0)
 			panic("lockmgr: non-release on draining lock: %d",
 			    flags & LK_TYPE_MASK);
+#endif /* DIAGNOSTIC */
 		lkp->lk_flags &= ~LK_DRAINING;
 		if ((flags & LK_REENABLE) == 0)
 			lkp->lk_flags |= LK_DRAINED;
@@ -208,12 +421,11 @@ lockmgr(lkp, flags, interlkp, p)
 	if ((lkp->lk_flags & (LK_CANRECURSE|LK_RECURSEFAIL)) ==
 	    (LK_CANRECURSE|LK_RECURSEFAIL))
 		panic("lockmgr: make up your mind");
-#endif /* DIAGNOSTIC */
 
 	switch (flags & LK_TYPE_MASK) {
 
 	case LK_SHARED:
-		if (lkp->lk_lockholder != pid) {
+		if (WEHOLDIT(lkp, pid, cpu_id) == 0) {
 			/*
 			 * If just polling, check to see if we will block.
 			 */
@@ -225,12 +437,12 @@ lockmgr(lkp, flags, interlkp, p)
 			/*
 			 * Wait for exclusive locks and upgrades to clear.
 			 */
-			ACQUIRE(lkp, error, extflags, lkp->lk_flags &
+			ACQUIRE(lkp, error, extflags, 0, lkp->lk_flags &
 			    (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE));
 			if (error)
 				break;
 			lkp->lk_sharecount++;
-			COUNT(p, 1);
+			COUNT(lkp, p, cpu_id, 1);
 			break;
 		}
 		/*
@@ -238,18 +450,24 @@ lockmgr(lkp, flags, interlkp, p)
 		 * An alternative would be to fail with EDEADLK.
 		 */
 		lkp->lk_sharecount++;
-		COUNT(p, 1);
+		COUNT(lkp, p, cpu_id, 1);
 		/* fall into downgrade */
 
 	case LK_DOWNGRADE:
-		if (lkp->lk_lockholder != pid || lkp->lk_exclusivecount == 0)
+		if (WEHOLDIT(lkp, pid, cpu_id) == 0 ||
+		    lkp->lk_exclusivecount == 0)
 			panic("lockmgr: not holding exclusive lock");
 		lkp->lk_sharecount += lkp->lk_exclusivecount;
 		lkp->lk_exclusivecount = 0;
+		lkp->lk_recurselevel = 0;
 		lkp->lk_flags &= ~LK_HAVE_EXCL;
-		lkp->lk_lockholder = LK_NOPROC;
-		if (lkp->lk_waitcount)
-			wakeup((void *)lkp);
+		SETHOLDER(lkp, LK_NOPROC, LK_NOCPU);
+#if defined(LOCKDEBUG)
+		lkp->lk_unlock_file = file;
+		lkp->lk_unlock_line = line;
+#endif
+		DONTHAVEIT(lkp);
+		WAKEUP_WAITER(lkp);
 		break;
 
 	case LK_EXCLUPGRADE:
@@ -260,7 +478,7 @@ lockmgr(lkp, flags, interlkp, p)
 		 */
 		if (lkp->lk_flags & LK_WANT_UPGRADE) {
 			lkp->lk_sharecount--;
-			COUNT(p, -1);
+			COUNT(lkp, p, cpu_id, -1);
 			error = EBUSY;
 			break;
 		}
@@ -275,10 +493,10 @@ lockmgr(lkp, flags, interlkp, p)
 		 * after the upgrade). If we return an error, the file
 		 * will always be unlocked.
 		 */
-		if (lkp->lk_lockholder == pid || lkp->lk_sharecount <= 0)
+		if (WEHOLDIT(lkp, pid, cpu_id) || lkp->lk_sharecount <= 0)
 			panic("lockmgr: upgrade exclusive lock");
 		lkp->lk_sharecount--;
-		COUNT(p, -1);
+		COUNT(lkp, p, cpu_id, -1);
 		/*
 		 * If we are just polling, check to see if we will block.
 		 */
@@ -295,16 +513,23 @@ lockmgr(lkp, flags, interlkp, p)
 			 * drop to zero, then take exclusive lock.
 			 */
 			lkp->lk_flags |= LK_WANT_UPGRADE;
-			ACQUIRE(lkp, error, extflags, lkp->lk_sharecount);
+			ACQUIRE(lkp, error, extflags, 0, lkp->lk_sharecount);
 			lkp->lk_flags &= ~LK_WANT_UPGRADE;
 			if (error)
 				break;
 			lkp->lk_flags |= LK_HAVE_EXCL;
-			lkp->lk_lockholder = pid;
+			SETHOLDER(lkp, pid, cpu_id);
+#if defined(LOCKDEBUG)
+			lkp->lk_lock_file = file;
+			lkp->lk_lock_line = line;
+#endif
+			HAVEIT(lkp);
 			if (lkp->lk_exclusivecount != 0)
 				panic("lockmgr: non-zero exclusive count");
 			lkp->lk_exclusivecount = 1;
-			COUNT(p, 1);
+			if (extflags & LK_SETRECURSE)
+				lkp->lk_recurselevel = 1;
+			COUNT(lkp, p, cpu_id, 1);
 			break;
 		}
 		/*
@@ -312,24 +537,28 @@ lockmgr(lkp, flags, interlkp, p)
 		 * lock, awaken upgrade requestor if we are the last shared
 		 * lock, then request an exclusive lock.
 		 */
-		if (lkp->lk_sharecount == 0 && lkp->lk_waitcount)
-			wakeup((void *)lkp);
+		if (lkp->lk_sharecount == 0)
+			WAKEUP_WAITER(lkp);
 		/* fall into exclusive request */
 
 	case LK_EXCLUSIVE:
-		if (lkp->lk_lockholder == pid && pid != LK_KERNPROC) {
+		if (WEHOLDIT(lkp, pid, cpu_id)) {
 			/*
-			 *	Recursive lock.
+			 * Recursive lock.
 			 */
-			if ((extflags & LK_CANRECURSE) == 0) {
+			if ((extflags & LK_CANRECURSE) == 0 &&
+			     lkp->lk_recurselevel == 0) {
 				if (extflags & LK_RECURSEFAIL) {
 					error = EDEADLK;
 					break;
-				}
-				panic("lockmgr: locking against myself");
+				} else
+					panic("lockmgr: locking against myself");
 			}
 			lkp->lk_exclusivecount++;
-			COUNT(p, 1);
+			if (extflags & LK_SETRECURSE &&
+			    lkp->lk_recurselevel == 0)
+				lkp->lk_recurselevel = lkp->lk_exclusivecount;
+			COUNT(lkp, p, cpu_id, 1);
 			break;
 		}
 		/*
@@ -344,7 +573,7 @@ lockmgr(lkp, flags, interlkp, p)
 		/*
 		 * Try to acquire the want_exclusive flag.
 		 */
-		ACQUIRE(lkp, error, extflags, lkp->lk_flags &
+		ACQUIRE(lkp, error, extflags, 0, lkp->lk_flags &
 		    (LK_HAVE_EXCL | LK_WANT_EXCL));
 		if (error)
 			break;
@@ -352,38 +581,62 @@ lockmgr(lkp, flags, interlkp, p)
 		/*
 		 * Wait for shared locks and upgrades to finish.
 		 */
-		ACQUIRE(lkp, error, extflags, lkp->lk_sharecount != 0 ||
+		ACQUIRE(lkp, error, extflags, 0, lkp->lk_sharecount != 0 ||
 		       (lkp->lk_flags & LK_WANT_UPGRADE));
 		lkp->lk_flags &= ~LK_WANT_EXCL;
 		if (error)
 			break;
 		lkp->lk_flags |= LK_HAVE_EXCL;
-		lkp->lk_lockholder = pid;
+		SETHOLDER(lkp, pid, cpu_id);
+#if defined(LOCKDEBUG)
+		lkp->lk_lock_file = file;
+		lkp->lk_lock_line = line;
+#endif
+		HAVEIT(lkp);
 		if (lkp->lk_exclusivecount != 0)
 			panic("lockmgr: non-zero exclusive count");
 		lkp->lk_exclusivecount = 1;
-		COUNT(p, 1);
+		if (extflags & LK_SETRECURSE)
+			lkp->lk_recurselevel = 1;
+		COUNT(lkp, p, cpu_id, 1);
 		break;
 
 	case LK_RELEASE:
 		if (lkp->lk_exclusivecount != 0) {
-			if (pid != lkp->lk_lockholder)
-				panic("lockmgr: pid %d, not %s %d unlocking",
-				    pid, "exclusive lock holder",
-				    lkp->lk_lockholder);
+			if (WEHOLDIT(lkp, pid, cpu_id) == 0) {
+				if (lkp->lk_flags & LK_SPIN) {
+					panic("lockmgr: processor %lu, not "
+					    "exclusive lock holder %lu "
+					    "unlocking", cpu_id, lkp->lk_cpu);
+				} else {
+					panic("lockmgr: pid %d, not "
+					    "exclusive lock holder %d "
+					    "unlocking", pid,
+					    lkp->lk_lockholder);
+				}
+			}
+			if (lkp->lk_exclusivecount == lkp->lk_recurselevel)
+				lkp->lk_recurselevel = 0;
 			lkp->lk_exclusivecount--;
-			COUNT(p, -1);
+			COUNT(lkp, p, cpu_id, -1);
 			if (lkp->lk_exclusivecount == 0) {
 				lkp->lk_flags &= ~LK_HAVE_EXCL;
-				lkp->lk_lockholder = LK_NOPROC;
+				SETHOLDER(lkp, LK_NOPROC, LK_NOCPU);
+#if defined(LOCKDEBUG)
+				lkp->lk_unlock_file = file;
+				lkp->lk_unlock_line = line;
+#endif
+				DONTHAVEIT(lkp);
 			}
 		} else if (lkp->lk_sharecount != 0) {
 			lkp->lk_sharecount--;
-			COUNT(p, -1);
-		} else
-			panic("lockmgr: LK_RELEASE of unlocked lock");
-		if (lkp->lk_waitcount)
-			wakeup((void *)lkp);
+			COUNT(lkp, p, cpu_id, -1);
+		}
+#ifdef DIAGNOSTIC
+		else
+			panic("lockmgr: release of unlocked lock!");
+#endif
+		WAKEUP_WAITER(lkp);
 		break;
 
 	case LK_DRAIN:
@@ -393,7 +646,7 @@ lockmgr(lkp, flags, interlkp, p)
 		 * check for holding a shared lock, but at least we can
 		 * check for an exclusive one.
 		 */
-		if (lkp->lk_lockholder == pid)
+		if (WEHOLDIT(lkp, pid, cpu_id))
 			panic("lockmgr: draining against myself");
 		/*
 		 * If we are just polling, check to see if we will sleep.
@@ -404,66 +657,228 @@ lockmgr(lkp, flags, interlkp, p)
 			error = EBUSY;
 			break;
 		}
-		PAUSE(lkp, ((lkp->lk_flags &
+		ACQUIRE(lkp, error, extflags, 1,
+		    ((lkp->lk_flags &
 		     (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) ||
-		     lkp->lk_sharecount != 0 || lkp->lk_waitcount != 0));
-		for (error = 0; ((lkp->lk_flags &
-		     (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) ||
-		     lkp->lk_sharecount != 0 || lkp->lk_waitcount != 0); ) {
-			lkp->lk_flags |= LK_WAITDRAIN;
-			simple_unlock(&lkp->lk_interlock);
-			if ((error = tsleep((void *)&lkp->lk_flags, lkp->lk_prio,
-			    lkp->lk_wmesg, lkp->lk_timo)) != 0)
-				return (error);
-			if ((extflags) & LK_SLEEPFAIL)
-				return (ENOLCK);
-			simple_lock(&lkp->lk_interlock);
-		}
+		     lkp->lk_sharecount != 0 ||
+		     lkp->lk_waitcount != 0));
+		if (error)
+			break;
 		lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
-		lkp->lk_lockholder = pid;
+		SETHOLDER(lkp, pid, cpu_id);
+#if defined(LOCKDEBUG)
+		lkp->lk_lock_file = file;
+		lkp->lk_lock_line = line;
+#endif
+		HAVEIT(lkp);
 		lkp->lk_exclusivecount = 1;
-		COUNT(p, 1);
+		/* XXX unlikely that we'd want this */
+		if (extflags & LK_SETRECURSE)
+			lkp->lk_recurselevel = 1;
+		COUNT(lkp, p, cpu_id, 1);
 		break;
 
 	default:
-		simple_unlock(&lkp->lk_interlock);
+		INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
 		panic("lockmgr: unknown locktype request %d",
 		    flags & LK_TYPE_MASK);
 		/* NOTREACHED */
 	}
-	if ((lkp->lk_flags & LK_WAITDRAIN) && ((lkp->lk_flags &
-	     (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) == 0 &&
-	     lkp->lk_sharecount == 0 && lkp->lk_waitcount == 0)) {
+	if ((lkp->lk_flags & (LK_WAITDRAIN | LK_SPIN)) == LK_WAITDRAIN &&
+	    ((lkp->lk_flags &
+	    (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) == 0 &&
+	    lkp->lk_sharecount == 0 && lkp->lk_waitcount == 0)) {
 		lkp->lk_flags &= ~LK_WAITDRAIN;
 		wakeup((void *)&lkp->lk_flags);
 	}
-	simple_unlock(&lkp->lk_interlock);
+	INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
 	return (error);
 }
 
 /*
+ * For a recursive spinlock held one or more times by the current CPU,
+ * release all N locks, and return N.
+ * Intended for use in mi_switch() shortly before context switching.
+ */
+
+#ifdef notyet
+int
+#if defined(LOCKDEBUG)
+_spinlock_release_all(__volatile struct lock *lkp, const char *file, int line)
+#else
+spinlock_release_all(__volatile struct lock *lkp)
+#endif
+{
+	int s, count;
+	cpuid_t cpu_id;
+
+	KASSERT(lkp->lk_flags & LK_SPIN);
+	
+	INTERLOCK_ACQUIRE(lkp, LK_SPIN, s);
+
+	cpu_id = CPU_NUMBER();
+	count = lkp->lk_exclusivecount;
+	
+	if (count != 0) {
+#ifdef DIAGNOSTIC		
+		if (WEHOLDIT(lkp, 0, cpu_id) == 0) {
+			panic("spinlock_release_all: processor %lu, not "
+			    "exclusive lock holder %lu "
+			    "unlocking", (long)cpu_id, lkp->lk_cpu);
+		}
+#endif
+		lkp->lk_recurselevel = 0;
+		lkp->lk_exclusivecount = 0;
+		COUNT_CPU(cpu_id, -count);
+		lkp->lk_flags &= ~LK_HAVE_EXCL;
+		SETHOLDER(lkp, LK_NOPROC, LK_NOCPU);
+#if defined(LOCKDEBUG)
+		lkp->lk_unlock_file = file;
+		lkp->lk_unlock_line = line;
+#endif
+		DONTHAVEIT(lkp);
+	}
+#ifdef DIAGNOSTIC
+	else if (lkp->lk_sharecount != 0)
+		panic("spinlock_release_all: release of shared lock!");
+	else
+		panic("spinlock_release_all: release of unlocked lock!");
+#endif
+	INTERLOCK_RELEASE(lkp, LK_SPIN, s);	
+
+	return (count);
+}
+#endif
+
+/*
+ * For a recursive spinlock held one or more times by the current CPU,
+ * release all N locks, and return N.
+ * Intended for use in mi_switch() right after resuming execution.
+ */
+
+#ifdef notyet
+void
+#if defined(LOCKDEBUG)
+_spinlock_acquire_count(__volatile struct lock *lkp, int count,
+    const char *file, int line)
+#else
+spinlock_acquire_count(__volatile struct lock *lkp, int count)
+#endif
+{
+	int s, error;
+	cpuid_t cpu_id;
+	
+	KASSERT(lkp->lk_flags & LK_SPIN);
+	
+	INTERLOCK_ACQUIRE(lkp, LK_SPIN, s);
+
+	cpu_id = CPU_NUMBER();
+
+#ifdef DIAGNOSTIC
+	if (WEHOLDIT(lkp, LK_NOPROC, cpu_id))
+		panic("spinlock_acquire_count: processor %lu already holds lock", (long)cpu_id);
+#endif
+	/*
+	 * Try to acquire the want_exclusive flag.
+	 */
+	ACQUIRE(lkp, error, LK_SPIN, 0, lkp->lk_flags &
+	    (LK_HAVE_EXCL | LK_WANT_EXCL));
+	lkp->lk_flags |= LK_WANT_EXCL;
+	/*
+	 * Wait for shared locks and upgrades to finish.
+	 */
+	ACQUIRE(lkp, error, LK_SPIN, 0, lkp->lk_sharecount != 0 ||
+	    (lkp->lk_flags & LK_WANT_UPGRADE));
+	lkp->lk_flags &= ~LK_WANT_EXCL;
+	lkp->lk_flags |= LK_HAVE_EXCL;
+	SETHOLDER(lkp, LK_NOPROC, cpu_id);
+#if defined(LOCKDEBUG)
+	lkp->lk_lock_file = file;
+	lkp->lk_lock_line = line;
+#endif
+	HAVEIT(lkp);
+	if (lkp->lk_exclusivecount != 0)
+		panic("lockmgr: non-zero exclusive count");
+	lkp->lk_exclusivecount = count;
+	lkp->lk_recurselevel = 1;
+	COUNT_CPU(cpu_id, count);
+
+	INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);	
+}
+#endif
+
+/*
  * Print out information about state of a lock. Used by VOP_PRINT
  * routines to display ststus about contained locks.
  */
 void
 lockmgr_printinfo(lkp)
-	struct lock *lkp;
+	__volatile struct lock *lkp;
 {
 
 	if (lkp->lk_sharecount)
 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
 		    lkp->lk_sharecount);
-	else if (lkp->lk_flags & LK_HAVE_EXCL)
-		printf(" lock type %s: EXCL (count %d) by pid %d",
-		    lkp->lk_wmesg, lkp->lk_exclusivecount, lkp->lk_lockholder);
-	if (lkp->lk_waitcount > 0)
+	else if (lkp->lk_flags & LK_HAVE_EXCL) {
+		printf(" lock type %s: EXCL (count %d) by ",
+		    lkp->lk_wmesg, lkp->lk_exclusivecount);
+		if (lkp->lk_flags & LK_SPIN)
+			printf("processor %lu", lkp->lk_cpu);
+		else
+			printf("pid %d", lkp->lk_lockholder);
+	} else
+		printf(" not locked");
+	if ((lkp->lk_flags & LK_SPIN) == 0 && lkp->lk_waitcount > 0)
 		printf(" with %d pending", lkp->lk_waitcount);
 }
 
 #if defined(LOCKDEBUG)
+TAILQ_HEAD(, simplelock) simplelock_list =
+    TAILQ_HEAD_INITIALIZER(simplelock_list);
+
+#if defined(MULTIPROCESSOR) /* { */
+struct simplelock simplelock_list_slock = SIMPLELOCK_INITIALIZER;
+
+#define	SLOCK_LIST_LOCK()						\
+	__cpu_simple_lock(&simplelock_list_slock.lock_data)
+
+#define	SLOCK_LIST_UNLOCK()						\
+	__cpu_simple_unlock(&simplelock_list_slock.lock_data)
+
+#define	SLOCK_COUNT(x)							\
+	curcpu()->ci_simple_locks += (x)
+#else
+u_long simple_locks;
+
+#define	SLOCK_LIST_LOCK()	/* nothing */
+
+#define	SLOCK_LIST_UNLOCK()	/* nothing */
 
-int lockdebug_print = 0;
-int lockdebug_debugger = 0;
+#define	SLOCK_COUNT(x)		simple_locks += (x)
+#endif /* MULTIPROCESSOR */ /* } */
+
+#ifdef MULTIPROCESSOR
+#define SLOCK_MP()		lock_printf("on cpu %ld\n", 		\
+				    (u_long) cpu_number())
+#else
+#define SLOCK_MP()		/* nothing */
+#endif
+
+#define	SLOCK_WHERE(str, alp, id, l)					\
+do {									\
+	lock_printf("\n");						\
+	lock_printf(str);						\
+	lock_printf("lock: %p, currently at: %s:%d\n", (alp), (id), (l)); \
+	SLOCK_MP();							\
+	if ((alp)->lock_file != NULL)					\
+		lock_printf("last locked: %s:%d\n", (alp)->lock_file,	\
+		    (alp)->lock_line);					\
+	if ((alp)->unlock_file != NULL)					\
+		lock_printf("last unlocked: %s:%d\n", (alp)->unlock_file, \
+		    (alp)->unlock_line);				\
+	SLOCK_TRACE()							\
+	SLOCK_DEBUGGER();						\
+} while (/*CONSTCOND*/0)
 
 /*
  * Simple lock functions so that the debugger can see from whence
@@ -474,7 +889,16 @@ simple_lock_init(lkp)
 	struct simplelock *lkp;
 {
 
-	lkp->lock_data = SLOCK_UNLOCKED;
+#if defined(MULTIPROCESSOR) /* { */
+	__cpu_simple_lock_init(&alp->lock_data);
+#else
+	alp->lock_data = __SIMPLELOCK_UNLOCKED;
+#endif /* } */
+	alp->lock_file = NULL;
+	alp->lock_line = 0;
+	alp->unlock_file = NULL;
+	alp->unlock_line = 0;
+	alp->lock_holder = LK_NOCPU;
 }
 
 void
@@ -483,16 +907,80 @@ _simple_lock(lkp, id, l)
 	const char *id;
 	int l;
 {
+	cpuid_t cpu_id = CPU_NUMBER();
+	int s;
+
+	s = spllock();
+
+	/*
+	 * MULTIPROCESSOR case: This is `safe' since if it's not us, we
+	 * don't take any action, and just fall into the normal spin case.
+	 */
+	if (alp->lock_data == __SIMPLELOCK_LOCKED) {
+#if defined(MULTIPROCESSOR) /* { */
+		if (alp->lock_holder == cpu_id) {
+			SLOCK_WHERE("simple_lock: locking against myself\n",
+			    alp, id, l);
+			goto out;
+		}
+#else
+		SLOCK_WHERE("simple_lock: lock held\n", alp, id, l);
+		goto out;
+#endif /* MULTIPROCESSOR */ /* } */
+	}
+
+#if defined(MULTIPROCESSOR) /* { */
+	/* Acquire the lock before modifying any fields. */
+	splx(s);
+	__cpu_simple_lock(&alp->lock_data);
+	s = spllock();
+#else
+	alp->lock_data = __SIMPLELOCK_LOCKED;
+#endif /* } */
 
-	if (lkp->lock_data == SLOCK_LOCKED) {
-		if (lockdebug_print)
-			printf("%s:%d simple_lock: lock held...\n", id, l);
-		if (lockdebug_debugger)
-			Debugger();
+	if (alp->lock_holder != LK_NOCPU) {
+		SLOCK_WHERE("simple_lock: uninitialized lock\n",
+		    alp, id, l);
 	}
-	lkp->lock_data = SLOCK_LOCKED;
+	alp->lock_file = id;
+	alp->lock_line = l;
+	alp->lock_holder = cpu_id;
+
+	SLOCK_LIST_LOCK();
+	/* XXX Cast away volatile */
+	TAILQ_INSERT_TAIL(&simplelock_list, (struct simplelock *)alp, list);
+	SLOCK_LIST_UNLOCK();
+
+	SLOCK_COUNT(1);
+
+ out:
+	splx(s);
 }
 
+int
+_simple_lock_held(__volatile struct simplelock *alp)
+{
+	cpuid_t cpu_id = CPU_NUMBER();
+	int s, locked = 0;
+
+	s = spllock();
+
+#if defined(MULTIPROCESSOR)
+	if (__cpu_simple_lock_try(&alp->lock_data) == 0)
+		locked = (alp->lock_holder == cpu_id);
+	else
+		__cpu_simple_unlock(&alp->lock_data);
+#else
+	if (alp->lock_data == __SIMPLELOCK_LOCKED) {
+		locked = 1;
+		KASSERT(alp->lock_holder == cpu_id);
+	}
+#endif
+
+	splx(s);
+
+	return (locked);
+}
 
 int
 _simple_lock_try(lkp, id, l)
@@ -500,14 +988,50 @@ _simple_lock_try(lkp, id, l)
 	const char *id;
 	int l;
 {
+	cpuid_t cpu_id = CPU_NUMBER();
+	int s, rv = 0;
+
+	s = spllock();
 
-	if (lkp->lock_data == SLOCK_LOCKED) {
-		if (lockdebug_print)
-			printf("%s:%d simple_lock: lock held...\n", id, l);
-		if (lockdebug_debugger)
-			Debugger();
+	/*
+	 * MULTIPROCESSOR case: This is `safe' since if it's not us, we
+	 * don't take any action.
+	 */
+#if defined(MULTIPROCESSOR) /* { */
+	if ((rv = __cpu_simple_lock_try(&alp->lock_data)) == 0) {
+		if (alp->lock_holder == cpu_id)
+			SLOCK_WHERE("simple_lock_try: locking against myself\n",
+			    alp, id, l);
+		goto out;
 	}
-	return lkp->lock_data = SLOCK_LOCKED;
+#else
+	if (alp->lock_data == __SIMPLELOCK_LOCKED) {
+		SLOCK_WHERE("simple_lock_try: lock held\n", alp, id, l);
+		goto out;
+	}
+	alp->lock_data = __SIMPLELOCK_LOCKED;
+#endif /* MULTIPROCESSOR */ /* } */
+
+	/*
+	 * At this point, we have acquired the lock.
+	 */
+
+	rv = 1;
+
+	alp->lock_file = id;
+	alp->lock_line = l;
+	alp->lock_holder = cpu_id;
+
+	SLOCK_LIST_LOCK();
+	/* XXX Cast away volatile. */
+	TAILQ_INSERT_TAIL(&simplelock_list, (struct simplelock *)alp, list);
+	SLOCK_LIST_UNLOCK();
+
+	SLOCK_COUNT(1);
+
+ out:
+	splx(s);
+	return (rv);
 }
 
 void
@@ -516,30 +1040,239 @@ _simple_unlock(lkp, id, l)
 	const char *id;
 	int l;
 {
+	int s;
 
-	if (lkp->lock_data == SLOCK_UNLOCKED) {
-		if (lockdebug_print)
-			printf("%s:%d simple_unlock: lock not held...\n",
-			       id, l);
-		if (lockdebug_debugger)
-			Debugger();
+	s = spllock();
+
+	/*
+	 * MULTIPROCESSOR case: This is `safe' because we think we hold
+	 * the lock, and if we don't, we don't take any action.
+	 */
+	if (alp->lock_data == __SIMPLELOCK_UNLOCKED) {
+		SLOCK_WHERE("simple_unlock: lock not held\n",
+		    alp, id, l);
+		goto out;
 	}
-	lkp->lock_data = SLOCK_UNLOCKED;
+
+	SLOCK_LIST_LOCK();
+	TAILQ_REMOVE(&simplelock_list, alp, list);
+	SLOCK_LIST_UNLOCK();
+
+	SLOCK_COUNT(-1);
+
+	alp->list.tqe_next = NULL;	/* sanity */
+	alp->list.tqe_prev = NULL;	/* sanity */
+
+	alp->unlock_file = id;
+	alp->unlock_line = l;
+
+#if defined(MULTIPROCESSOR) /* { */
+	alp->lock_holder = LK_NOCPU;
+	/* Now that we've modified all fields, release the lock. */
+	__cpu_simple_unlock(&alp->lock_data);
+#else
+	alp->lock_data = __SIMPLELOCK_UNLOCKED;
+	KASSERT(alp->lock_holder == CPU_NUMBER());
+	alp->lock_holder = LK_NOCPU;
+#endif /* } */
+
+ out:
+	splx(s);
 }
 
 void
-_simple_lock_assert(lkp, state, id, l)
-	__volatile struct simplelock *lkp;
-	int state;
-	const char *id;
-	int l;
+simple_lock_dump(void)
+{
+	struct simplelock *alp;
+	int s;
+
+	s = spllock();
+	SLOCK_LIST_LOCK();
+	lock_printf("all simple locks:\n");
+	TAILQ_FOREACH(alp, &simplelock_list, list) {
+		lock_printf("%p CPU %lu %s:%d\n", alp, alp->lock_holder,
+		    alp->lock_file, alp->lock_line);
+	}
+	SLOCK_LIST_UNLOCK();
+	splx(s);
+}
+
+void
+simple_lock_freecheck(void *start, void *end)
+{
+	struct simplelock *alp;
+	int s;
+
+	s = spllock();
+	SLOCK_LIST_LOCK();
+	TAILQ_FOREACH(alp, &simplelock_list, list) {
+		if ((void *)alp >= start && (void *)alp < end) {
+			lock_printf("freeing simple_lock %p CPU %lu %s:%d\n",
+			    alp, alp->lock_holder, alp->lock_file,
+			    alp->lock_line);
+			SLOCK_DEBUGGER();
+		}
+	}
+	SLOCK_LIST_UNLOCK();
+	splx(s);
+ }
+
+/*
+ * We must be holding exactly one lock: the sched_lock.
+ */
+
+#ifdef notyet
+void
+simple_lock_switchcheck(void)
+{
+
+	simple_lock_only_held(&sched_lock, "switching");
+}
+#endif
+
+void
+simple_lock_only_held(volatile struct simplelock *lp, const char *where)
 {
-	if (lkp->lock_data != state) {
-		if (lockdebug_print)
-			printf("%s:%d simple_lock_assert: wrong state: %d",
-			       id, l, lkp->lock_data);
-		if (lockdebug_debugger)
-			Debugger();
+	struct simplelock *alp;
+	cpuid_t cpu_id = CPU_NUMBER();
+	int s;
+
+	if (lp) {
+		LOCK_ASSERT(simple_lock_held(lp));
+	}
+	s = spllock();
+	SLOCK_LIST_LOCK();
+	TAILQ_FOREACH(alp, &simplelock_list, list) {
+		if (alp == lp)
+			continue;
+		if (alp->lock_holder == cpu_id)
+			break;
+	}
+	SLOCK_LIST_UNLOCK();
+	splx(s);
+
+	if (alp != NULL) {
+		lock_printf("\n%s with held simple_lock %p "
+		    "CPU %lu %s:%d\n",
+		    where, alp, alp->lock_holder, alp->lock_file,
+		    alp->lock_line);
+		SLOCK_TRACE();
+		SLOCK_DEBUGGER();
 	}
 }
 #endif /* LOCKDEBUG */
+
+#if defined(MULTIPROCESSOR)
+/*
+ * Functions for manipulating the kernel_lock.  We put them here
+ * so that they show up in profiles.
+ */
+
+/*
+ * XXX Instead of using struct lock for the kernel lock and thus requiring us
+ * XXX to implement simplelocks, causing all sorts of fine-grained locks all
+ * XXX over our tree getting activated consuming both time and potentially
+ * XXX introducing locking protocol bugs.
+ */
+#ifdef notyet
+
+struct lock kernel_lock; 
+
+void
+_kernel_lock_init(void)
+{
+	spinlockinit(&kernel_lock, "klock", 0);
+}
+
+/*
+ * Acquire/release the kernel lock.  Intended for use in the scheduler
+ * and the lower half of the kernel.
+ */
+void
+_kernel_lock(int flag)
+{
+	SCHED_ASSERT_UNLOCKED();
+	spinlockmgr(&kernel_lock, flag, 0);
+}
+
+void
+_kernel_unlock(void)
+{
+	spinlockmgr(&kernel_lock, LK_RELEASE, 0);
+}
+
+/*
+ * Acquire/release the kernel_lock on behalf of a process.  Intended for
+ * use in the top half of the kernel.
+ */
+void
+_kernel_proc_lock(struct proc *p)
+{
+	SCHED_ASSERT_UNLOCKED();
+	spinlockmgr(&kernel_lock, LK_EXCLUSIVE, 0);
+	p->p_flag |= P_BIGLOCK;
+}
+
+void
+_kernel_proc_unlock(struct proc *p)
+{
+	p->p_flag &= ~P_BIGLOCK;
+	spinlockmgr(&kernel_lock, LK_RELEASE, 0);
+}
+
+#else
+
+struct __mp_lock kernel_lock; 
+
+void
+_kernel_lock_init(void)
+{
+	__mp_lock_init(&kernel_lock);
+}
+
+/*
+ * Acquire/release the kernel lock.  Intended for use in the scheduler
+ * and the lower half of the kernel.
+ */
+
+/* XXX The flag should go, all callers want equal behaviour. */
+void
+_kernel_lock(int flag)
+{
+	SCHED_ASSERT_UNLOCKED();
+	__mp_lock(&kernel_lock);
+}
+
+void
+_kernel_unlock(void)
+{
+	__mp_unlock(&kernel_lock);
+}
+
+/*
+ * Acquire/release the kernel_lock on behalf of a process.  Intended for
+ * use in the top half of the kernel.
+ */
+void
+_kernel_proc_lock(struct proc *p)
+{
+	SCHED_ASSERT_UNLOCKED();
+	__mp_lock(&kernel_lock);
+	p->p_flag |= P_BIGLOCK;
+}
+
+void
+_kernel_proc_unlock(struct proc *p)
+{
+	p->p_flag &= ~P_BIGLOCK;
+	__mp_unlock(&kernel_lock);
+}
+
+#endif
+
+#ifdef MP_LOCKDEBUG
+/* CPU-dependent timing, needs this to be settable from ddb. */
+int __mp_lock_spinout = 200000000;
+#endif
+
+#endif /* MULTIPROCESSOR */
diff --git a/sys/kern/kern_proc.c b/sys/kern/kern_proc.c
index 84519c2b60e..aa7ec306c56 100644
--- a/sys/kern/kern_proc.c
+++ b/sys/kern/kern_proc.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_proc.c,v 1.18 2004/01/29 17:19:42 millert Exp $	*/
+/*	$OpenBSD: kern_proc.c,v 1.19 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_proc.c,v 1.14 1996/02/09 18:59:41 christos Exp $	*/
 
 /*
@@ -85,7 +85,7 @@ struct pool pcred_pool;
  * proclist.  Processes on this proclist are also on zombproc;
  * we use the p_hash member to linkup to deadproc.
  */
-struct simplelock deadproc_slock;
+struct SIMPLELOCK deadproc_slock;
 struct proclist deadproc;		/* dead, but not yet undead */
 
 static void orphanpg(struct pgrp *);
@@ -104,7 +104,7 @@ procinit()
 	LIST_INIT(&zombproc);
 
 	LIST_INIT(&deadproc);
-	simple_lock_init(&deadproc_slock);
+	SIMPLE_LOCK_INIT(&deadproc_slock);
 
 	pidhashtbl = hashinit(maxproc / 4, M_PROC, M_WAITOK, &pidhash);
 	pgrphashtbl = hashinit(maxproc / 4, M_PROC, M_WAITOK, &pgrphash);
diff --git a/sys/kern/kern_resource.c b/sys/kern/kern_resource.c
index 1e868518989..6d7af1fd136 100644
--- a/sys/kern/kern_resource.c
+++ b/sys/kern/kern_resource.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_resource.c,v 1.26 2003/12/11 23:02:30 millert Exp $	*/
+/*	$OpenBSD: kern_resource.c,v 1.27 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_resource.c,v 1.38 1996/10/23 07:19:38 matthias Exp $	*/
 
 /*-
@@ -44,6 +44,7 @@
 #include <sys/resourcevar.h>
 #include <sys/pool.h>
 #include <sys/proc.h>
+#include <sys/sched.h>
 
 #include <sys/mount.h>
 #include <sys/syscallargs.h>
@@ -184,6 +185,7 @@ donice(curp, chgp, n)
 	register int n;
 {
 	register struct pcred *pcred = curp->p_cred;
+	int s;
 
 	if (pcred->pc_ucred->cr_uid && pcred->p_ruid &&
 	    pcred->pc_ucred->cr_uid != chgp->p_ucred->cr_uid &&
@@ -197,7 +199,9 @@ donice(curp, chgp, n)
 	if (n < chgp->p_nice && suser(curp, 0))
 		return (EACCES);
 	chgp->p_nice = n;
+	SCHED_LOCK(s);
 	(void)resetpriority(chgp);
+	SCHED_UNLOCK(s);
 	return (0);
 }
 
diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c
index 494f6878d80..0913d2b2a1f 100644
--- a/sys/kern/kern_sig.c
+++ b/sys/kern/kern_sig.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sig.c,v 1.70 2004/04/06 17:24:11 mickey Exp $	*/
+/*	$OpenBSD: kern_sig.c,v 1.71 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $	*/
 
 /*
@@ -62,6 +62,7 @@
 #include <sys/malloc.h>
 #include <sys/pool.h>
 #include <sys/ptrace.h>
+#include <sys/sched.h>
 
 #include <sys/mount.h>
 #include <sys/syscallargs.h>
@@ -805,19 +806,30 @@ trapsignal(p, signum, code, type, sigval)
  *     regardless of the signal action (eg, blocked or ignored).
  *
  * Other ignored signals are discarded immediately.
+ *
+ * XXXSMP: Invoked as psignal() or sched_psignal().
  */
 void
-psignal(p, signum)
+psignal1(p, signum, dolock)
 	register struct proc *p;
 	register int signum;
+	int dolock;		/* XXXSMP: works, but icky */
 {
 	register int s, prop;
 	register sig_t action;
 	int mask;
 
+#ifdef DIAGNOSTIC
 	if ((u_int)signum >= NSIG || signum == 0)
 		panic("psignal signal number");
 
+	/* XXXSMP: works, but icky */
+	if (dolock)
+		SCHED_ASSERT_UNLOCKED();
+	else
+		SCHED_ASSERT_LOCKED();
+#endif
+
 	/* Ignore signal if we are exiting */
 	if (p->p_flag & P_WEXIT)
 		return;
@@ -879,7 +891,10 @@ psignal(p, signum)
 	 */
 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
 		return;
-	s = splhigh();
+	/* XXXSMP: works, but icky */
+	if (dolock)
+		SCHED_LOCK(s);
+
 	switch (p->p_stat) {
 
 	case SSLEEP:
@@ -921,7 +936,11 @@ psignal(p, signum)
 			p->p_siglist &= ~mask;
 			p->p_xstat = signum;
 			if ((p->p_pptr->p_flag & P_NOCLDSTOP) == 0)
-				psignal(p->p_pptr, SIGCHLD);
+				/*
+				 * XXXSMP: recursive call; don't lock
+				 * the second time around.
+				 */
+				sched_psignal(p->p_pptr, SIGCHLD);
 			proc_stop(p);
 			goto out;
 		}
@@ -1009,7 +1028,9 @@ runfast:
 run:
 	setrunnable(p);
 out:
-	splx(s);
+	/* XXXSMP: works, but icky */
+	if (dolock)
+		SCHED_UNLOCK(s);
 }
 
 /*
@@ -1054,7 +1075,7 @@ issignal(struct proc *p)
 			 */
 			p->p_xstat = signum;
 
-			s = splstatclock();	/* protect mi_switch */
+			SCHED_LOCK(s);	/* protect mi_switch */
 			if (p->p_flag & P_FSTRACE) {
 #ifdef	PROCFS
 				/* procfs debugging */
@@ -1070,6 +1091,7 @@ issignal(struct proc *p)
 				proc_stop(p);
 				mi_switch();
 			}
+			SCHED_ASSERT_UNLOCKED();
 			splx(s);
 
 			/*
@@ -1130,8 +1152,9 @@ issignal(struct proc *p)
 				if ((p->p_pptr->p_flag & P_NOCLDSTOP) == 0)
 					psignal(p->p_pptr, SIGCHLD);
 				proc_stop(p);
-				s = splstatclock();
+				SCHED_LOCK(s);
 				mi_switch();
+				SCHED_ASSERT_UNLOCKED();
 				splx(s);
 				break;
 			} else if (prop & SA_IGNORE) {
@@ -1179,6 +1202,9 @@ void
 proc_stop(p)
 	struct proc *p;
 {
+#ifdef MULTIPROCESSOR
+	SCHED_ASSERT_LOCKED();
+#endif
 
 	p->p_stat = SSTOP;
 	p->p_flag &= ~P_WAITED;
@@ -1205,6 +1231,9 @@ postsig(signum)
 	if (signum == 0)
 		panic("postsig");
 #endif
+
+	KERNEL_PROC_LOCK(p);
+
 	mask = sigmask(signum);
 	p->p_siglist &= ~mask;
 	action = ps->ps_sigact[signum];
@@ -1254,7 +1283,11 @@ postsig(signum)
 		 * mask from before the sigpause is what we want
 		 * restored after the signal processing is completed.
 		 */
+#ifdef MULTIPROCESSOR
+		s = splsched();
+#else
 		s = splhigh();
+#endif
 		if (ps->ps_flags & SAS_OLDMASK) {
 			returnmask = ps->ps_oldmask;
 			ps->ps_flags &= ~SAS_OLDMASK;
@@ -1279,6 +1312,8 @@ postsig(signum)
 		(*p->p_emul->e_sendsig)(action, signum, returnmask, code,
 		    type, sigval);
 	}
+
+	KERNEL_PROC_UNLOCK(p);
 }
 
 /*
@@ -1308,7 +1343,6 @@ sigexit(p, signum)
 	register struct proc *p;
 	int signum;
 {
-
 	/* Mark process as going away */
 	p->p_flag |= P_WEXIT;
 
diff --git a/sys/kern/kern_subr.c b/sys/kern/kern_subr.c
index 52432e05522..32e659af713 100644
--- a/sys/kern/kern_subr.c
+++ b/sys/kern/kern_subr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_subr.c,v 1.27 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_subr.c,v 1.28 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_subr.c,v 1.15 1996/04/09 17:21:56 ragge Exp $	*/
 
 /*
@@ -40,6 +40,7 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/proc.h>
+#include <sys/sched.h>
 #include <sys/malloc.h>
 #include <sys/queue.h>
 #include <sys/kernel.h>
diff --git a/sys/kern/kern_synch.c b/sys/kern/kern_synch.c
index baba311f668..ab7ca8f7ed3 100644
--- a/sys/kern/kern_synch.c
+++ b/sys/kern/kern_synch.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_synch.c,v 1.55 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_synch.c,v 1.56 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_synch.c,v 1.37 1996/04/22 01:38:37 christos Exp $	*/
 
 /*-
@@ -59,15 +59,19 @@ u_char	curpriority;		/* usrpri of curproc */
 #endif
 int	lbolt;			/* once a second sleep address */
 #ifdef __HAVE_CPUINFO
-int	rrticks_init;		/* # of harclock ticks per roundrobin */
+int	rrticks_init;		/* # of hardclock ticks per roundrobin() */
 #endif
 
 int whichqs;			/* Bit mask summary of non-empty Q's. */
 struct prochd qs[NQS];
 
+struct SIMPLELOCK sched_lock;
+
 void scheduler_start(void);
 
-#ifndef __HAVE_CPUINFO
+#ifdef __HAVE_CPUINFO
+void roundrobin(struct cpu_info *);
+#else
 void roundrobin(void *);
 #endif
 void schedcpu(void *);
@@ -85,11 +89,13 @@ scheduler_start()
 	/*
 	 * We avoid polluting the global namespace by keeping the scheduler
 	 * timeouts static in this function.
-	 * We setup the timeouts here and kick roundrobin and schedcpu once to
+	 * We setup the timeouts here and kick schedcpu and roundrobin once to
 	 * make them do their job.
 	 */
 
-	timeout_set(&roundrobin_to, roundrobin, &roundrobin_to);
+#ifndef __HAVE_CPUINFO
+	timeout_set(&roundrobin_to, schedcpu, &roundrobin_to);
+#endif
 	timeout_set(&schedcpu_to, schedcpu, &schedcpu_to);
 
 #ifdef __HAVE_CPUINFO
@@ -103,6 +109,7 @@ scheduler_start()
 /*
  * Force switch among equal priority processes every 100ms.
  */
+/* ARGSUSED */
 #ifdef __HAVE_CPUINFO
 void
 roundrobin(struct cpu_info *ci)
@@ -122,7 +129,7 @@ roundrobin(struct cpu_info *ci)
 			 */
 			spc->spc_schedflags |= SPCF_SHOULDYIELD;
 		} else {
-			 spc->spc_schedflags |= SPCF_SEENRR;
+			spc->spc_schedflags |= SPCF_SEENRR;
 		}
 		splx(s);
 	}
@@ -130,7 +137,6 @@ roundrobin(struct cpu_info *ci)
 	need_resched(curcpu());
 }
 #else
-/* ARGSUSED */
 void
 roundrobin(void *arg)
 {
@@ -152,7 +158,8 @@ roundrobin(void *arg)
 		}
 		splx(s);
 	}
-	need_resched();
+
+	need_resched(0);
 	timeout_add(to, hz / 10);
 }
 #endif
@@ -298,6 +305,8 @@ schedcpu(arg)
 		p->p_cpticks = 0;
 		newcpu = (u_int) decay_cpu(loadfac, p->p_estcpu);
 		p->p_estcpu = newcpu;
+		splx(s);
+		SCHED_LOCK(s);
 		resetpriority(p);
 		if (p->p_priority >= PUSER) {
 			if ((p != curproc) &&
@@ -310,7 +319,7 @@ schedcpu(arg)
 			} else
 				p->p_priority = p->p_usrpri;
 		}
-		splx(s);
+		SCHED_UNLOCK(s);
 	}
 	uvm_meter();
 	wakeup((caddr_t)&lbolt);
@@ -329,6 +338,8 @@ updatepri(p)
 	register unsigned int newcpu = p->p_estcpu;
 	register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
 
+	SCHED_ASSERT_LOCKED();
+
 	if (p->p_slptime > 5 * loadfac)
 		p->p_estcpu = 0;
 	else {
@@ -392,11 +403,6 @@ ltsleep(ident, priority, wmesg, timo, interlock)
 	int catch = priority & PCATCH;
 	int relock = (priority & PNORELOCK) == 0;
 
-#ifdef KTRACE
-	if (KTRPOINT(p, KTR_CSW))
-		ktrcsw(p, 1, 0);
-#endif
-	s = splhigh();
 	if (cold || panicstr) {
 		/*
 		 * After a panic, or during autoconfiguration,
@@ -404,16 +410,26 @@ ltsleep(ident, priority, wmesg, timo, interlock)
 		 * don't run any other procs or panic below,
 		 * in case this is the idle process and already asleep.
 		 */
+		s = splhigh();
 		splx(safepri);
 		splx(s);
 		if (interlock != NULL && relock == 0)
 			simple_unlock(interlock);
 		return (0);
 	}
+
+#ifdef KTRACE
+	if (KTRPOINT(p, KTR_CSW))
+		ktrcsw(p, 1, 0);
+#endif
+
+	SCHED_LOCK(s);
+
 #ifdef DIAGNOSTIC
-	if (ident == NULL || p->p_stat != SRUN || p->p_back)
+	if (ident == NULL || p->p_stat != SONPROC || p->p_back != NULL)
 		panic("tsleep");
 #endif
+
 	p->p_wchan = ident;
 	p->p_wmesg = wmesg;
 	p->p_slptime = 0;
@@ -452,29 +468,39 @@ ltsleep(ident, priority, wmesg, timo, interlock)
 		if ((sig = CURSIG(p)) != 0) {
 			if (p->p_wchan)
 				unsleep(p);
-			p->p_stat = SRUN;
+			p->p_stat = SONPROC;
+			SCHED_UNLOCK(s);
 			goto resume;
 		}
 		if (p->p_wchan == 0) {
 			catch = 0;
+			SCHED_UNLOCK(s);
 			goto resume;
 		}
 	} else
 		sig = 0;
 	p->p_stat = SSLEEP;
 	p->p_stats->p_ru.ru_nvcsw++;
+	SCHED_ASSERT_LOCKED();
 	mi_switch();
 #ifdef	DDB
 	/* handy breakpoint location after process "wakes" */
 	__asm(".globl bpendtsleep\nbpendtsleep:");
 #endif
+
+	SCHED_ASSERT_UNLOCKED();
+	/*
+	 * Note! this splx belongs to the SCHED_LOCK(s) above, mi_switch
+	 * releases the scheduler lock, but does not lower the spl.
+	 */
+	splx(s);
+
 resume:
 #ifdef __HAVE_CPUINFO
 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
 #else
 	curpriority = p->p_usrpri;
 #endif
-	splx(s);
 	p->p_flag &= ~P_SINTR;
 	if (p->p_flag & P_TIMEOUT) {
 		p->p_flag &= ~P_TIMEOUT;
@@ -504,6 +530,7 @@ resume:
 	if (KTRPOINT(p, KTR_CSW))
 		ktrcsw(p, 0, 0);
 #endif
+
 	if (interlock != NULL && relock)
 		simple_lock(interlock);
 	return (0);
@@ -523,7 +550,7 @@ endtsleep(arg)
 	int s;
 
 	p = (struct proc *)arg;
-	s = splhigh();
+	SCHED_LOCK(s);
 	if (p->p_wchan) {
 		if (p->p_stat == SSLEEP)
 			setrunnable(p);
@@ -531,75 +558,7 @@ endtsleep(arg)
 			unsleep(p);
 		p->p_flag |= P_TIMEOUT;
 	}
-	splx(s);
-}
-
-/*
- * Short-term, non-interruptable sleep.
- */
-void
-sleep(ident, priority)
-	void *ident;
-	int priority;
-{
-	register struct proc *p = curproc;
-	register struct slpque *qp;
-	register int s;
-
-#ifdef DIAGNOSTIC
-	if (priority > PZERO) {
-		printf("sleep called with priority %d > PZERO, wchan: %p\n",
-		    priority, ident);
-		panic("old sleep");
-	}
-#endif
-	s = splhigh();
-	if (cold || panicstr) {
-		/*
-		 * After a panic, or during autoconfiguration,
-		 * just give interrupts a chance, then just return;
-		 * don't run any other procs or panic below,
-		 * in case this is the idle process and already asleep.
-		 */
-		splx(safepri);
-		splx(s);
-		return;
-	}
-#ifdef DIAGNOSTIC
-	if (ident == NULL || p->p_stat != SRUN || p->p_back)
-		panic("sleep");
-#endif
-	p->p_wchan = ident;
-	p->p_wmesg = NULL;
-	p->p_slptime = 0;
-	p->p_priority = priority;
-	qp = &slpque[LOOKUP(ident)];
-	if (qp->sq_head == 0)
-		qp->sq_head = p;
-	else
-		*qp->sq_tailp = p;
-	*(qp->sq_tailp = &p->p_forw) = 0;
-	p->p_stat = SSLEEP;
-	p->p_stats->p_ru.ru_nvcsw++;
-#ifdef KTRACE
-	if (KTRPOINT(p, KTR_CSW))
-		ktrcsw(p, 1, 0);
-#endif
-	mi_switch();
-#ifdef	DDB
-	/* handy breakpoint location after process "wakes" */
-	__asm(".globl bpendsleep\nbpendsleep:");
-#endif
-#ifdef KTRACE
-	if (KTRPOINT(p, KTR_CSW))
-		ktrcsw(p, 0, 0);
-#endif
-#ifdef __HAVE_CPUINFO
-	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
-#else
-	curpriority = p->p_usrpri;
-#endif
-	splx(s);
+	SCHED_UNLOCK(s);
 }
 
 /*
@@ -611,9 +570,15 @@ unsleep(p)
 {
 	register struct slpque *qp;
 	register struct proc **hp;
+#if 0
 	int s;
 
-	s = splhigh();
+	/*
+	 * XXX we cannot do recursive SCHED_LOCKing yet.  All callers lock
+	 * anyhow.
+	 */
+	SCHED_LOCK(s);
+#endif
 	if (p->p_wchan) {
 		hp = &(qp = &slpque[LOOKUP(p->p_wchan)])->sq_head;
 		while (*hp != p)
@@ -623,9 +588,25 @@ unsleep(p)
 			qp->sq_tailp = hp;
 		p->p_wchan = 0;
 	}
-	splx(s);
+#if 0
+	SCHED_UNLOCK(s);
+#endif
+}
+
+#if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
+void
+sched_unlock_idle(void)
+{
+	SIMPLE_UNLOCK(&sched_lock);
 }
 
+void
+sched_lock_idle(void)
+{
+	SIMPLE_LOCK(&sched_lock);
+}
+#endif /* MULTIPROCESSOR || LOCKDEBUG */
+
 /*
  * Make all processes sleeping on the specified identifier runnable.
  */
@@ -638,7 +619,7 @@ wakeup_n(ident, n)
 	struct proc *p, **q;
 	int s;
 
-	s = splhigh();
+	SCHED_LOCK(s);
 	qp = &slpque[LOOKUP(ident)];
 restart:
 	for (q = &qp->sq_head; (p = *q) != NULL; ) {
@@ -662,15 +643,19 @@ restart:
 				/*
 				 * Since curpriority is a user priority,
 				 * p->p_priority is always better than
-				 * curpriority.
+				 * curpriority on the last CPU on
+				 * which it ran.
+				 *
+				 * XXXSMP See affinity comment in
+				 * resched_proc().
 				 */
-
 				if ((p->p_flag & P_INMEM) != 0) {
 					setrunqueue(p);
 #ifdef __HAVE_CPUINFO
+					KASSERT(p->p_cpu != NULL);
 					need_resched(p->p_cpu);
 #else
-					need_resched();
+					need_resched(0);
 #endif
 				} else {
 					wakeup((caddr_t)&proc0);
@@ -685,7 +670,7 @@ restart:
 		} else
 			q = &p->p_forw;
 	}
-	splx(s);
+	SCHED_UNLOCK(s);
 }
 
 void
@@ -705,11 +690,12 @@ yield()
 	struct proc *p = curproc;
 	int s;
 
-	s = splstatclock();
+	SCHED_LOCK(s);
 	p->p_priority = p->p_usrpri;
 	setrunqueue(p);
 	p->p_stats->p_ru.ru_nvcsw++;
 	mi_switch();
+	SCHED_ASSERT_UNLOCKED();
 	splx(s);
 }
 
@@ -732,11 +718,13 @@ preempt(newp)
 	if (newp != NULL)
 		panic("preempt: cpu_preempt not yet implemented");
 
-	s = splstatclock();
+	SCHED_LOCK(s);
 	p->p_priority = p->p_usrpri;
+	p->p_stat = SRUN;
 	setrunqueue(p);
 	p->p_stats->p_ru.ru_nivcsw++;
 	mi_switch();
+	SCHED_ASSERT_UNLOCKED();
 	splx(s);
 }
 
@@ -750,11 +738,28 @@ mi_switch()
 	struct proc *p = curproc;	/* XXX */
 	struct rlimit *rlim;
 	struct timeval tv;
+#if defined(MULTIPROCESSOR)
+	int hold_count;
+#endif
 #ifdef __HAVE_CPUINFO
 	struct schedstate_percpu *spc = &p->p_cpu->ci_schedstate;
 #endif
 
-	splassert(IPL_STATCLOCK);
+	SCHED_ASSERT_LOCKED();
+
+#if defined(MULTIPROCESSOR)
+	/*
+	 * Release the kernel_lock, as we are about to yield the CPU.
+	 * The scheduler lock is still held until cpu_switch()
+	 * selects a new process and removes it from the run queue.
+	 */
+	if (p->p_flag & P_BIGLOCK)
+#ifdef notyet
+		hold_count = spinlock_release_all(&kernel_lock);
+#else
+		hold_count = __mp_release_all(&kernel_lock);
+#endif
+#endif
 
 	/*
 	 * Compute the amount of time during which the current
@@ -765,19 +770,19 @@ mi_switch()
 	if (timercmp(&tv, &spc->spc_runtime, <)) {
 #if 0
 		printf("time is not monotonic! "
-		    "tv=%ld.%06ld, runtime=%ld.%06ld\n",
+		    "tv=%lu.%06lu, runtime=%lu.%06lu\n",
 		    tv.tv_sec, tv.tv_usec, spc->spc_runtime.tv_sec,
 		    spc->spc_runtime.tv_usec);
 #endif
 	} else {
-		timersub(&tv, &spc->runtime, &tv);
+		timersub(&tv, &spc->spc_runtime, &tv);
 		timeradd(&p->p_rtime, &tv, &p->p_rtime);
 	}
 #else
 	if (timercmp(&tv, &runtime, <)) {
 #if 0
 		printf("time is not monotonic! "
-		    "tv=%ld.%06ld, runtime=%ld.%06ld\n",
+		    "tv=%lu.%06lu, runtime=%lu.%06lu\n",
 		    tv.tv_sec, tv.tv_usec, runtime.tv_sec, runtime.tv_usec);
 #endif
 	} else {
@@ -817,12 +822,38 @@ mi_switch()
 	uvmexp.swtch++;
 	cpu_switch(p);
 
+	/*
+	 * Make sure that MD code released the scheduler lock before
+	 * resuming us.
+	 */
+	SCHED_ASSERT_UNLOCKED();
+
+	/*
+	 * We're running again; record our new start time.  We might
+	 * be running on a new CPU now, so don't use the cache'd
+	 * schedstate_percpu pointer.
+	 */
 #ifdef __HAVE_CPUINFO
-	/* p->p_cpu might have changed in cpu_switch() */
+	KDASSERT(p->p_cpu != NULL);
+	KDASSERT(p->p_cpu == curcpu());
 	microtime(&p->p_cpu->ci_schedstate.spc_runtime);
 #else
 	microtime(&runtime);
 #endif
+
+#if defined(MULTIPROCESSOR)
+	/*
+	 * Reacquire the kernel_lock now.  We do this after we've
+	 * released the scheduler lock to avoid deadlock, and before
+	 * we reacquire the interlock.
+	 */
+	if (p->p_flag & P_BIGLOCK)
+#ifdef notyet
+		spinlock_acquire_count(&kernel_lock, hold_count);
+#else
+		__mp_acquire_count(&kernel_lock, hold_count);
+#endif
+#endif
 }
 
 /*
@@ -836,6 +867,7 @@ rqinit()
 
 	for (i = 0; i < NQS; i++)
 		qs[i].ph_link = qs[i].ph_rlink = (struct proc *)&qs[i];
+	SIMPLE_LOCK_INIT(&sched_lock);
 }
 
 static __inline void
@@ -845,13 +877,35 @@ resched_proc(struct proc *p, u_char pri)
 	struct cpu_info *ci;
 #endif
 
+	/*
+	 * XXXSMP
+	 * Since p->p_cpu persists across a context switch,
+	 * this gives us *very weak* processor affinity, in
+	 * that we notify the CPU on which the process last
+	 * ran that it should try to switch.
+	 *
+	 * This does not guarantee that the process will run on
+	 * that processor next, because another processor might
+	 * grab it the next time it performs a context switch.
+	 *
+	 * This also does not handle the case where its last
+	 * CPU is running a higher-priority process, but every
+	 * other CPU is running a lower-priority process.  There
+	 * are ways to handle this situation, but they're not
+	 * currently very pretty, and we also need to weigh the
+	 * cost of moving a process from one CPU to another.
+	 *
+	 * XXXSMP
+	 * There is also the issue of locking the other CPU's
+	 * sched state, which we currently do not do.
+	 */
 #ifdef __HAVE_CPUINFO
 	ci = (p->p_cpu != NULL) ? p->p_cpu : curcpu();
 	if (pri < ci->ci_schedstate.spc_curpriority)
 		need_resched(ci);
 #else
 	if (pri < curpriority)
-		need_resched();
+		need_resched(0);
 #endif
 }
 
@@ -864,12 +918,12 @@ void
 setrunnable(p)
 	register struct proc *p;
 {
-	register int s;
+	SCHED_ASSERT_LOCKED();
 
-	s = splhigh();
 	switch (p->p_stat) {
 	case 0:
 	case SRUN:
+	case SONPROC:
 	case SZOMB:
 	case SDEAD:
 	default:
@@ -890,7 +944,6 @@ setrunnable(p)
 	p->p_stat = SRUN;
 	if (p->p_flag & P_INMEM)
 		setrunqueue(p);
-	splx(s);
 	if (p->p_slptime > 1)
 		updatepri(p);
 	p->p_slptime = 0;
@@ -911,6 +964,8 @@ resetpriority(p)
 {
 	register unsigned int newpriority;
 
+	SCHED_ASSERT_LOCKED();
+
 	newpriority = PUSER + p->p_estcpu + NICE_WEIGHT * (p->p_nice - NZERO);
 	newpriority = min(newpriority, MAXPRI);
 	p->p_usrpri = newpriority;
@@ -936,8 +991,12 @@ void
 schedclock(p)
 	struct proc *p;
 {
+	int s;
+
 	p->p_estcpu = ESTCPULIM(p->p_estcpu + 1);
+	SCHED_LOCK(s);
 	resetpriority(p);
+	SCHED_UNLOCK(s);
 	if (p->p_priority >= PUSER)
 		p->p_priority = p->p_usrpri;
 }
diff --git a/sys/kern/kern_sysctl.c b/sys/kern/kern_sysctl.c
index a517ba2951c..603a354b76b 100644
--- a/sys/kern/kern_sysctl.c
+++ b/sys/kern/kern_sysctl.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sysctl.c,v 1.111 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_sysctl.c,v 1.112 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_sysctl.c,v 1.17 1996/05/20 17:49:05 mrg Exp $	*/
 
 /*-
@@ -429,6 +429,20 @@ kern_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
 		return (sysctl_malloc(name + 1, namelen - 1, oldp, oldlenp,
 		    newp, newlen, p));
 	case KERN_CPTIME:
+#ifdef MULTIPROCESSOR
+	{
+		CPU_INFO_ITERATOR cii;
+		struct cpu_info *ci;
+		int i;
+
+		bzero(cp_time, sizeof(cp_time));
+
+		for (CPU_INFO_FOREACH(cii, ci)) {
+			for (i = 0; i < CPUSTATES; i++)
+				cp_time[i] += ci->ci_schedstate.spc_cp_time[i];
+		}
+	}
+#endif
 		return (sysctl_rdstruct(oldp, oldlenp, newp, &cp_time,
 		    sizeof(cp_time)));
 	case KERN_NCHSTATS:
@@ -1317,6 +1331,11 @@ fill_kproc2(struct proc *p, struct kinfo_proc2 *ki)
 			 &p->p_stats->p_cru.ru_stime, &ut);
 		ki->p_uctime_sec = ut.tv_sec;
 		ki->p_uctime_usec = ut.tv_usec;
+		ki->p_cpuid = KI_NOCPU;
+#ifdef MULTIPROCESSOR
+		if (p->p_cpu != NULL)
+			ki->p_cpuid = p->p_cpu->ci_cpuid;
+#endif
 		PRELE(p);
 	}
 }
diff --git a/sys/kern/kern_time.c b/sys/kern/kern_time.c
index f40928b2824..1ed5f182d36 100644
--- a/sys/kern/kern_time.c
+++ b/sys/kern/kern_time.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_time.c,v 1.40 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kern_time.c,v 1.41 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: kern_time.c,v 1.20 1996/02/18 11:57:06 fvdl Exp $	*/
 
 /*
@@ -99,7 +99,7 @@ settime(struct timeval *tv)
 	timersub(tv, &time, &delta);
 	time = *tv;
 	timeradd(&boottime, &delta, &boottime);
-#ifdef __HAVE_CURCPU
+#ifdef __HAVE_CPUINFO
 	/*
 	 * XXXSMP
 	 * This is wrong.  We should traverse a list of all
diff --git a/sys/kern/subr_pool.c b/sys/kern/subr_pool.c
index e9114b34990..db01a8a868c 100644
--- a/sys/kern/subr_pool.c
+++ b/sys/kern/subr_pool.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: subr_pool.c,v 1.41 2004/06/02 22:17:22 tedu Exp $	*/
+/*	$OpenBSD: subr_pool.c,v 1.42 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: subr_pool.c,v 1.61 2001/09/26 07:14:56 chs Exp $	*/
 
 /*-
@@ -54,7 +54,6 @@
 /*
  * XXX - for now.
  */
-#define SIMPLELOCK_INITIALIZER { SLOCK_UNLOCKED }
 #ifdef LOCKDEBUG
 #define simple_lock_freecheck(a, s) do { /* nothing */ } while (0)
 #define simple_lock_only_held(lkp, str) do { /* nothing */ } while (0)
@@ -86,7 +85,7 @@ int pool_inactive_time = 10;
 static struct pool	*drainpp;
 
 /* This spin lock protects both pool_head and drainpp. */
-struct simplelock pool_head_slock = SIMPLELOCK_INITIALIZER;
+struct simplelock pool_head_slock;
 
 struct pool_item_header {
 	/* Page headers */
@@ -529,6 +528,8 @@ pool_init(struct pool *pp, size_t size, u_int align, u_int ioff, int flags,
 		    0, "pcgpool", NULL);
 	}
 
+	simple_lock_init(&pool_head_slock);
+
 	/* Insert this into the list of all pools. */
 	simple_lock(&pool_head_slock);
 	TAILQ_INSERT_TAIL(&pool_head, pp, pr_poollist);
@@ -2062,9 +2063,9 @@ pool_allocator_drain(struct pool_allocator *pa, struct pool *org, int need)
 		TAILQ_INSERT_TAIL(&pa->pa_list, pp, pr_alloc_list);
 		if (pp == org)
 			continue;
-		simple_unlock(&pa->pa_list);
-		freed = pool_reclaim(pp)
-		simple_lock(&pa->pa_list);
+		simple_unlock(&pa->pa_slock);
+		freed = pool_reclaim(pp);
+		simple_lock(&pa->pa_slock);
 	} while ((pp = TAILQ_FIRST(&pa->pa_list)) != start && (freed < need));
 
 	if (!freed) {
diff --git a/sys/kern/subr_prf.c b/sys/kern/subr_prf.c
index 906ce028770..a843c2f1dc2 100644
--- a/sys/kern/subr_prf.c
+++ b/sys/kern/subr_prf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: subr_prf.c,v 1.58 2004/01/03 14:08:53 espie Exp $	*/
+/*	$OpenBSD: subr_prf.c,v 1.59 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: subr_prf.c,v 1.45 1997/10/24 18:14:25 chuck Exp $	*/
 
 /*-
@@ -47,6 +47,7 @@
 #include <sys/ioctl.h>
 #include <sys/vnode.h>
 #include <sys/file.h>
+#include <sys/simplelock.h>
 #include <sys/tty.h>
 #include <sys/tprintf.h>
 #include <sys/syslog.h>
@@ -97,6 +98,50 @@ extern int uvm_doswapencrypt;
 int	 kprintf(const char *, int, void *, char *, va_list);
 void	 kputchar(int, int, struct tty *);
 
+#ifdef MULTIPROCESSOR
+
+#ifdef notdef
+
+struct simplelock kprintf_slock;
+
+#define KPRINTF_MUTEX_ENTER(s)						\
+do {									\
+	(s) = splhigh();						\
+	simple_lock(&kprintf_slock);					\
+} while (/*CONSTCOND*/0)
+
+#define KPRINTF_MUTEX_EXIT(s)						\
+do {									\
+	simple_unlock(&kprintf_slock);					\
+	splx((s));							\
+} while (/*CONSTCOND*/0)
+
+#else
+
+struct __mp_lock kprintf_slock;
+
+#define KPRINTF_MUTEX_ENTER(s)						\
+do {									\
+	(s) = splhigh();						\
+	__mp_lock(&kprintf_slock);					\
+} while (/*CONSTCOND*/0)
+
+#define KPRINTF_MUTEX_EXIT(s)						\
+do {									\
+	__mp_unlock(&kprintf_slock);					\
+	splx((s));							\
+} while (/*CONSTCOND*/0)
+
+#endif
+
+#else
+
+struct simplelock kprintf_slock;
+#define KPRINTF_MUTEX_ENTER(s) (s) = splhigh()
+#define KPRINTF_MUTEX_EXIT(s) splx((s))
+
+#endif /* MULTIPROCESSOR */
+
 /*
  * globals
  */
@@ -506,6 +551,9 @@ printf(const char *fmt, ...)
 {
 	va_list ap;
 	int savintr, retval;
+	int s;
+
+	KPRINTF_MUTEX_ENTER(s);
 
 	savintr = consintr;		/* disable interrupts */
 	consintr = 0;
@@ -515,6 +563,9 @@ printf(const char *fmt, ...)
 	if (!panicstr)
 		logwakeup();
 	consintr = savintr;		/* reenable interrupts */
+
+	KPRINTF_MUTEX_EXIT(s);
+
 	return(retval);
 }
 
diff --git a/sys/kern/sys_generic.c b/sys/kern/sys_generic.c
index a6fdd09c95e..c512b9a40dd 100644
--- a/sys/kern/sys_generic.c
+++ b/sys/kern/sys_generic.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sys_generic.c,v 1.47 2003/12/10 23:10:08 millert Exp $	*/
+/*	$OpenBSD: sys_generic.c,v 1.48 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: sys_generic.c,v 1.24 1996/03/29 00:25:32 cgd Exp $	*/
 
 /*
@@ -55,6 +55,7 @@
 #ifdef KTRACE
 #include <sys/ktrace.h>
 #endif
+#include <sys/sched.h>
 
 #include <sys/mount.h>
 #include <sys/syscallargs.h>
@@ -852,7 +853,7 @@ selwakeup(sip)
 	p = pfind(sip->si_selpid);
 	sip->si_selpid = 0;
 	if (p != NULL) {
-		s = splhigh();
+		SCHED_LOCK(s);
 		if (p->p_wchan == (caddr_t)&selwait) {
 			if (p->p_stat == SSLEEP)
 				setrunnable(p);
@@ -860,7 +861,7 @@ selwakeup(sip)
 				unsleep(p);
 		} else if (p->p_flag & P_SELECT)
 			p->p_flag &= ~P_SELECT;
-		splx(s);
+		SCHED_UNLOCK(s);
 	}
 }
 
diff --git a/sys/kern/sys_process.c b/sys/kern/sys_process.c
index 37664becb7d..21180cda141 100644
--- a/sys/kern/sys_process.c
+++ b/sys/kern/sys_process.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sys_process.c,v 1.27 2004/02/08 00:04:21 deraadt Exp $	*/
+/*	$OpenBSD: sys_process.c,v 1.28 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: sys_process.c,v 1.55 1996/05/15 06:17:47 tls Exp $	*/
 
 /*-
@@ -57,6 +57,7 @@
 #include <sys/ptrace.h>
 #include <sys/uio.h>
 #include <sys/user.h>
+#include <sys/sched.h>
 
 #include <sys/mount.h>
 #include <sys/syscallargs.h>
@@ -91,6 +92,7 @@ sys_ptrace(p, v, retval)
 #endif
 	int error, write;
 	int temp;
+	int s;
 
 	/* "A foolish consistency..." XXX */
 	if (SCARG(uap, req) == PT_TRACE_ME)
@@ -353,7 +355,9 @@ sys_ptrace(p, v, retval)
 		/* Finally, deliver the requested signal (or none). */
 		if (t->p_stat == SSTOP) {
 			t->p_xstat = SCARG(uap, data);
+			SCHED_LOCK(s);
 			setrunnable(t);
+			SCHED_UNLOCK(s);
 		} else {
 			if (SCARG(uap, data) != 0)
 				psignal(t, SCARG(uap, data));
diff --git a/sys/kern/tty.c b/sys/kern/tty.c
index 0aca0394f7a..6dec67d20fe 100644
--- a/sys/kern/tty.c
+++ b/sys/kern/tty.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tty.c,v 1.64 2004/03/19 19:03:07 deraadt Exp $	*/
+/*	$OpenBSD: tty.c,v 1.65 2004/06/13 21:49:26 niklas Exp $	*/
 /*	$NetBSD: tty.c,v 1.68.4.2 1996/06/06 16:04:52 thorpej Exp $	*/
 
 /*-
@@ -2098,7 +2098,8 @@ ttyinfo(tp)
 				pick = p;
 
 		ttyprintf(tp, " cmd: %s %d [%s] ", pick->p_comm, pick->p_pid,
-		    pick->p_stat == SRUN ? "running" :
+		    pick->p_stat == SONPROC ? "running" :
+		    pick->p_stat == SRUN ? "runnable" :
 		    pick->p_wmesg ? pick->p_wmesg : "iowait");
 
 		calcru(pick, &utime, &stime, NULL);
diff --git a/sys/net/netisr_dispatch.h b/sys/net/netisr_dispatch.h
index 822577ad843..96493b324db 100644
--- a/sys/net/netisr_dispatch.h
+++ b/sys/net/netisr_dispatch.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: netisr_dispatch.h,v 1.3 2001/06/09 06:16:38 angelos Exp $	*/
+/*	$OpenBSD: netisr_dispatch.h,v 1.4 2004/06/13 21:49:27 niklas Exp $	*/
 /* $NetBSD: netisr_dispatch.h,v 1.2 2000/07/02 04:40:47 cgd Exp $ */
 
 /*
@@ -16,20 +16,23 @@
  * }
  */
 
-#ifndef _NET_NETISR_DISPATCH_H_
-#define _NET_NETISR_DISPATCH_H_
-
 #ifndef _NET_NETISR_H_
 #error <net/netisr.h> must be included before <net/netisr_dispatch.h>
 #endif
 
+#ifndef _NET_NETISR_DISPATCH_H_
+#define _NET_NETISR_DISPATCH_H_
+#include "ether.h"
+#include "ppp.h"
+#include "bridge.h"
+#endif
+
 /*
  * When adding functions to this list, be sure to add headers to provide
  * their prototypes in <net/netisr.h> (if necessary).
  */
 
 #ifdef INET
-#include "ether.h"
 #if NETHER > 0
 	DONETISR(NETISR_ARP,arpintr);
 #endif
@@ -59,12 +62,9 @@
 #ifdef NATM
 	DONETISR(NETISR_NATM,natmintr);
 #endif
-#include "ppp.h"
 #if NPPP > 0
 	DONETISR(NETISR_PPP,pppintr);
 #endif
-#include "bridge.h"
 #if NBRIDGE > 0
         DONETISR(NETISR_BRIDGE,bridgeintr);
 #endif
-#endif /* _NET_NETISR_DISPATCH_H_ */
diff --git a/sys/sys/kernel.h b/sys/sys/kernel.h
index b3c92ca019b..c2bc3965541 100644
--- a/sys/sys/kernel.h
+++ b/sys/sys/kernel.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kernel.h,v 1.9 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: kernel.h,v 1.10 2004/06/13 21:49:28 niklas Exp $	*/
 /*	$NetBSD: kernel.h,v 1.11 1995/03/03 01:24:16 cgd Exp $	*/
 
 /*-
@@ -49,7 +49,7 @@ extern int domainnamelen;
 /* 1.2 */
 extern volatile struct timeval mono_time;
 extern struct timeval boottime;
-#ifndef __HAVE_CURCPU
+#ifndef __HAVE_CPUINFO
 extern struct timeval runtime;
 #endif
 extern volatile struct timeval time;
diff --git a/sys/sys/lock.h b/sys/sys/lock.h
index e5d1899d35a..5da1c1bfd73 100644
--- a/sys/sys/lock.h
+++ b/sys/sys/lock.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: lock.h,v 1.11 2003/06/02 23:28:21 millert Exp $	*/
+/*	$OpenBSD: lock.h,v 1.12 2004/06/13 21:49:28 niklas Exp $	*/
 
 /* 
  * Copyright (c) 1995
@@ -56,11 +56,51 @@ struct lock {
 	int	lk_sharecount;		/* # of accepted shared locks */
 	int	lk_waitcount;		/* # of processes sleeping for lock */
 	short	lk_exclusivecount;	/* # of recursive exclusive locks */
-	short	lk_prio;		/* priority at which to sleep */
+	short	lk_recurselevel;	/* lvl above which recursion ok */
+
+	/*
+	 * This is the sleep message for sleep locks, and a simple name
+	 * for spin locks.
+	 */
 	char	*lk_wmesg;		/* resource sleeping (for tsleep) */
-	int	lk_timo;		/* maximum sleep time (for tsleep) */
-	pid_t	lk_lockholder;		/* pid of exclusive lock holder */
+
+	union {
+		struct {
+			/* pid of exclusive lock holder */
+			pid_t lk_sleep_lockholder;
+
+			/* priority at which to sleep */
+			int lk_sleep_prio;
+
+			/* maximum sleep time (for tsleep) */
+			int lk_sleep_timo;
+		} lk_un_sleep;
+		struct {
+			/* CPU ID of exclusive lock holder */
+			cpuid_t lk_spin_cpu;
+#if defined(LOCKDEBUG)
+			TAILQ_ENTRY(lock) lk_spin_list;
+#endif
+		} lk_un_spin;
+	} lk_un;
+
+#define	lk_lockholder	lk_un.lk_un_sleep.lk_sleep_lockholder
+#define	lk_prio		lk_un.lk_un_sleep.lk_sleep_prio
+#define	lk_timo		lk_un.lk_un_sleep.lk_sleep_timo
+
+#define	lk_cpu		lk_un.lk_un_spin.lk_spin_cpu
+#if defined(LOCKDEBUG)
+#define	lk_list		lk_un.lk_un_spin.lk_spin_list
+#endif
+
+#if defined(LOCKDEBUG)
+	const char *lk_lock_file;
+	const char *lk_unlock_file;
+	int lk_lock_line;
+	int lk_unlock_line;
+#endif
 };
+
 /*
  * Lock request types:
  *   LK_SHARED - get one of many possible shared locks. If a process
@@ -109,12 +149,14 @@ struct lock {
  * or passed in as arguments to the lock manager. The LK_REENABLE flag may be
  * set only at the release of a lock obtained by drain.
  */
-#define LK_EXTFLG_MASK	0x00000770	/* mask of external flags */
+#define LK_EXTFLG_MASK	0x00700070	/* mask of external flags */
 #define LK_NOWAIT	0x00000010	/* do not sleep to await lock */
 #define LK_SLEEPFAIL	0x00000020	/* sleep, then return failure */
 #define LK_CANRECURSE	0x00000040	/* allow recursive exclusive lock */
 #define LK_REENABLE	0x00000080	/* lock is be reenabled after drain */
-#define LK_RECURSEFAIL	0x00000100	/* fail if recursive exclusive lock */
+#define LK_SETRECURSE	0x00100000	/* other locks while we have it OK */
+#define LK_RECURSEFAIL	0x00200000	/* fail if recursive exclusive lock */
+#define LK_SPIN		0x00400000	/* lock spins instead of sleeps */
 /*
  * Internal lock flags.
  *
@@ -131,9 +173,9 @@ struct lock {
  *
  * Non-persistent external flags.
  */
-#define LK_INTERLOCK	0x00100000	/* unlock passed simple lock after
+#define LK_INTERLOCK	0x00010000	/* unlock passed simple lock after
 					   getting lk_interlock */
-#define LK_RETRY	0x00200000	/* vn_lock: retry until locked */
+#define LK_RETRY	0x00020000	/* vn_lock: retry until locked */
 
 /*
  * Lock return status.
@@ -157,6 +199,7 @@ struct lock {
  */
 #define LK_KERNPROC ((pid_t) -2)
 #define LK_NOPROC ((pid_t) -1)
+#define LK_NOCPU ((cpuid_t) -1)
 
 struct proc;
 
@@ -164,14 +207,69 @@ void	lockinit(struct lock *, int prio, char *wmesg, int timo,
 			int flags);
 int	lockmgr(__volatile struct lock *, u_int flags,
 			struct simplelock *, struct proc *p);
-void    lockmgr_printinfo(struct lock *);
+void	lockmgr_printinfo(__volatile struct lock *);
 int	lockstatus(struct lock *);
 
+#if (0 && defined(MULTIPROCESSOR)) || defined(LOCKDEBUG)
+#define spinlockinit(lkp, name, flags)					\
+	lockinit((lkp), 0, (name), 0, (flags) | LK_SPIN)
+#define spinlockmgr(lkp, flags, intrlk)					\
+	lockmgr((lkp), (flags) | LK_SPIN, (intrlk), curproc)
+#else
+#define spinlockinit(lkp, name, flags)	(void)(lkp)
+#define spinlockmgr(lkp, flags, intrlk)	(0)
+#endif
+
+#if defined(LOCKDEBUG)
+int	_spinlock_release_all(__volatile struct lock *, const char *, int);
+void	_spinlock_acquire_count(__volatile struct lock *, int, const char *,
+	    int);
+
+#define	spinlock_release_all(l)	_spinlock_release_all((l), __FILE__, __LINE__)
+#define	spinlock_acquire_count(l, c) _spinlock_acquire_count((l), (c),	\
+					__FILE__, __LINE__)
+#else
+int	spinlock_release_all(__volatile struct lock *);
+void	spinlock_acquire_count(__volatile struct lock *, int);
+#endif
+
 #ifdef LOCKDEBUG
 #define LOCK_ASSERT(x)	KASSERT(x)
 #else
 #define LOCK_ASSERT(x)	/* nothing */
 #endif
 
-#endif /* !_LOCK_H_ */
+#if defined(MULTIPROCESSOR)
+/*
+ * XXX Instead of using struct lock for the kernel lock and thus requiring us
+ * XXX to implement simplelocks, causing all sorts of fine-grained locks all
+ * XXX over our tree getting activated consuming both time and potentially
+ * XXX introducing locking protocol bugs.
+ */
+#ifdef notyet
+
+extern struct lock kernel_lock;
+
+/*
+ * XXX Simplelock macros used at "trusted" places.
+ */
+#define SIMPLELOCK		simplelock
+#define SIMPLE_LOCK_INIT	simple_lock_init
+#define SIMPLE_LOCK		simple_lock
+#define SIMPLE_UNLOCK		simple_unlock
+
+#endif
 
+#else
+
+/*
+ * XXX Simplelock macros used at "trusted" places.
+ */
+#define SIMPLELOCK		simplelock
+#define SIMPLE_LOCK_INIT	simple_lock_init
+#define SIMPLE_LOCK		simple_lock
+#define SIMPLE_UNLOCK		simple_unlock
+
+#endif
+
+#endif /* !_LOCK_H_ */
diff --git a/sys/sys/mplock.h b/sys/sys/mplock.h
new file mode 100644
index 00000000000..73f3c0b1e40
--- /dev/null
+++ b/sys/sys/mplock.h
@@ -0,0 +1,169 @@
+/*
+ * Copyright (c) 2004 Niklas Hallqvist.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _MPLOCK_H_
+#define _MPLOCK_H_
+
+/*
+ * Really simple spinlock implementation with recursive capabilities.
+ * Correctness is paramount, no fancyness allowed.
+ */
+
+struct __mp_lock {
+	__cpu_simple_lock_t mpl_lock;
+	cpuid_t	mpl_cpu;
+	int	mpl_count;
+};
+
+static __inline void __mp_lock_init(struct __mp_lock *);
+static __inline void __mp_lock(struct __mp_lock *);
+static __inline void __mp_unlock(struct __mp_lock *);
+static __inline int __mp_release_all(struct __mp_lock *);
+static __inline void __mp_acquire_count(struct __mp_lock *, int);
+static __inline int __mp_lock_held(struct __mp_lock *);
+
+/*
+ * XXX Simplelocks macros used at "trusted" places.
+ */
+#define SIMPLELOCK		__mp_lock
+#define SIMPLE_LOCK_INIT	__mp_lock_init
+#define SIMPLE_LOCK		__mp_lock
+#define SIMPLE_UNLOCK		__mp_unlock
+
+static __inline void
+__mp_lock_init(struct __mp_lock *lock)
+{
+	__cpu_simple_lock_init(&lock->mpl_lock);
+	lock->mpl_cpu = LK_NOCPU;
+	lock->mpl_count = 0;
+}
+
+#if defined(MP_LOCKDEBUG)
+#ifndef DDB
+#error "MP_LOCKDEBUG requires DDB"
+#endif
+
+extern void Debugger(void);
+extern int db_printf(const char *, ...)
+    __attribute__((__format__(__kprintf__,1,2)));
+
+/* CPU-dependent timing, needs this to be settable from ddb. */
+extern int __mp_lock_spinout;
+#endif
+
+static __inline void
+__mp_lock(struct __mp_lock *lock)
+{
+	int s = spllock();
+
+	if (lock->mpl_cpu != cpu_number()) {
+#ifndef MP_LOCKDEBUG
+		__cpu_simple_lock(&lock->mpl_lock);
+#else
+		{
+			int got_it;
+			do {
+				int ticks = __mp_lock_spinout;
+
+				do {
+					got_it = __cpu_simple_lock_try(
+					    &lock->mpl_lock);
+				} while (!got_it && ticks-- > 0);
+				if (!got_it) {
+ 					db_printf(
+					    "__mp_lock(0x%x): lock spun out",
+					    lock);
+					Debugger();
+				}
+			} while (!got_it);
+		}
+#endif
+		lock->mpl_cpu = cpu_number();
+	}
+	lock->mpl_count++;
+	splx(s);
+}
+
+static __inline void
+__mp_unlock(struct __mp_lock *lock)
+{
+	int s = spllock();
+
+#ifdef MP_LOCKDEBUG
+	if (lock->mpl_count == 0 || lock->mpl_cpu == LK_NOCPU) {
+		db_printf("__mp_unlock(0x%x): releasing not locked lock\n",
+		    lock);
+		Debugger();
+	}
+#endif
+
+	if (--lock->mpl_count == 0) {
+		lock->mpl_cpu = LK_NOCPU;
+		__cpu_simple_unlock(&lock->mpl_lock);
+	}
+	splx(s);
+}
+
+static __inline int
+__mp_release_all(struct __mp_lock *lock) {
+	int s = spllock();
+	int rv = lock->mpl_count;
+
+#ifdef MP_LOCKDEBUG
+	if (lock->mpl_count == 0 || lock->mpl_cpu == LK_NOCPU) {
+		db_printf(
+		    "__mp_release_all(0x%x): releasing not locked lock\n",
+		    lock);
+		Debugger();
+	}
+#endif
+
+	lock->mpl_cpu = LK_NOCPU;
+	lock->mpl_count = 0;
+	__cpu_simple_unlock(&lock->mpl_lock);
+	splx(s);
+	return (rv);
+}
+
+static __inline void
+__mp_acquire_count(struct __mp_lock *lock, int count) {
+	int s = spllock();
+
+	__cpu_simple_lock(&lock->mpl_lock);
+	lock->mpl_cpu = cpu_number();
+	lock->mpl_count = count;
+	splx(s);
+}
+
+static __inline int
+__mp_lock_held(struct __mp_lock *lock) {
+	return lock->mpl_count;
+}
+
+extern struct __mp_lock kernel_lock;
+
+/* XXX Should really be in proc.h but then __mp_lock is not defined. */
+extern struct SIMPLELOCK deadproc_slock;
+
+#endif /* !_MPLOCK_H */
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
index c06b14eda43..9f64ed24839 100644
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: proc.h,v 1.70 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: proc.h,v 1.71 2004/06/13 21:49:28 niklas Exp $	*/
 /*	$NetBSD: proc.h,v 1.44 1996/04/22 01:23:21 christos Exp $	*/
 
 /*-
@@ -46,7 +46,6 @@
 #include <sys/timeout.h>		/* For struct timeout. */
 #include <sys/event.h>			/* For struct klist */
 
-#ifdef __HAVE_CPUINFO
 /*
  * CPU states.
  * XXX Not really scheduler state, but no other good place to put
@@ -60,24 +59,25 @@
 #define CPUSTATES	5
 
 /*
- * Per-CPU scheduler state. XXX - this should be in sys/sched.h
+ * Per-CPU scheduler state.
  */
 struct schedstate_percpu {
 	struct timeval spc_runtime;	/* time curproc started running */
 	__volatile int spc_schedflags;	/* flags; see below */
 	u_int spc_schedticks;		/* ticks for schedclock() */
 	u_int64_t spc_cp_time[CPUSTATES]; /* CPU state statistics */
-	u_char spc_curpriority;         /* usrpri of curproc */
+	u_char spc_curpriority;		/* usrpri of curproc */
 	int spc_rrticks;		/* ticks until roundrobin() */
 	int spc_pscnt;			/* prof/stat counter */
-	int spc_psdiv;			/* prof/stat divisor */ 
+	int spc_psdiv;			/* prof/stat divisor */	
 };
 
 /* spc_flags */
-#define SPCF_SEENRR		0x0001	/* process has seen roundrobin() */
-#define SPCF_SHOULDYIELD	0x0002	/* process should yield the CPU */
-#define SPCF_SWITCHCLEAR	(SPCF_SEENRR|SPCF_SHOULDYIELD)
+#define SPCF_SEENRR             0x0001  /* process has seen roundrobin() */
+#define SPCF_SHOULDYIELD        0x0002  /* process should yield the CPU */
+#define SPCF_SWITCHCLEAR        (SPCF_SEENRR|SPCF_SHOULDYIELD)
 
+#ifdef __HAVE_CPUINFO
 /*
  * These are the fields we require in struct cpu_info that we get from
  * curcpu():
@@ -208,7 +208,7 @@ struct	proc {
 	u_int	p_swtime;	 /* Time swapped in or out. */
 	u_int	p_slptime;	 /* Time since last blocked. */
 #ifdef __HAVE_CPUINFO
-	struct cpu_info * __volatile p_cpu;
+	struct	cpu_info * __volatile p_cpu; /* CPU we're running on. */
 #else
 	int	p_schedflags;	 /* PSCHED_* flags */
 #endif
@@ -264,6 +264,7 @@ struct	proc {
 	u_short	p_xstat;	/* Exit status for wait; also stop signal. */
 	u_short	p_acflag;	/* Accounting flags. */
 	struct	rusage *p_ru;	/* Exit information. XXX */
+	int	p_locks;       	/* DEBUG: lockmgr count of held locks */
 };
 
 #define	p_session	p_pgrp->pg_session
@@ -276,6 +277,7 @@ struct	proc {
 #define	SSTOP	4		/* Process debugging or suspension. */
 #define	SZOMB	5		/* Awaiting collection by parent. */
 #define SDEAD	6		/* Process is almost a zombie. */
+#define	SONPROC	7		/* Process is currently on a CPU. */
 
 #define P_ZOMBIE(p)	((p)->p_stat == SZOMB || (p)->p_stat == SDEAD)
 
@@ -311,12 +313,13 @@ struct	proc {
 #define P_SYSTRACE	0x400000	/* Process system call tracing active*/
 #define P_CONTINUED	0x800000	/* Proc has continued from a stopped state. */
 #define P_SWAPIN	0x1000000	/* Swapping in right now */
+#define P_BIGLOCK	0x2000000	/* Process needs kernel "big lock" to run */
 
 #define	P_BITS \
     ("\20\01ADVLOCK\02CTTY\03INMEM\04NOCLDSTOP\05PPWAIT\06PROFIL\07SELECT" \
      "\010SINTR\011SUGID\012SYSTEM\013TIMEOUT\014TRACED\015WAITED\016WEXIT" \
      "\017EXEC\020PWEUPC\021FSTRACE\022SSTEP\023SUGIDEXEC\024NOCLDWAIT" \
-     "\025NOZOMBIE\026INEXEC\027SYSTRACE\030CONTINUED")
+     "\025NOZOMBIE\026INEXEC\027SYSTRACE\030CONTINUED\031SWAPIN\032BIGLOCK")
 
 /* Macro to compute the exit signal to be delivered. */
 #define P_EXITSIG(p) \
@@ -404,7 +407,6 @@ extern struct proclist allproc;		/* List of all processes. */
 extern struct proclist zombproc;	/* List of zombie processes. */
 
 extern struct proclist deadproc;	/* List of dead processes. */
-extern struct simplelock deadproc_slock;
 
 extern struct proc *initproc;		/* Process slots for init, pager. */
 extern struct proc *syncerproc;		/* filesystem syncer daemon */
@@ -448,7 +450,6 @@ void	setrunnable(struct proc *);
 #if !defined(setrunqueue)
 void	setrunqueue(struct proc *);
 #endif
-void	sleep(void *chan, int pri);
 void	uvm_swapin(struct proc *);  /* XXX: uvm_extern.h? */
 int	ltsleep(void *chan, int pri, const char *wmesg, int timo,
 	    volatile struct simplelock *);
@@ -476,5 +477,9 @@ void	child_return(void *);
 
 int	proc_cansugid(struct proc *);
 void	proc_zap(struct proc *);
+
+#if defined(MULTIPROCESSOR)
+void	proc_trampoline_mp(void);	/* XXX */
+#endif
 #endif	/* _KERNEL */
 #endif	/* !_SYS_PROC_H_ */
diff --git a/sys/sys/sched.h b/sys/sys/sched.h
index 08b1fb49904..50625e84149 100644
--- a/sys/sys/sched.h
+++ b/sys/sys/sched.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sched.h,v 1.5 2004/06/09 20:18:28 art Exp $	*/
+/*	$OpenBSD: sched.h,v 1.6 2004/06/13 21:49:28 niklas Exp $	*/
 /* $NetBSD: sched.h,v 1.2 1999/02/28 18:14:58 ross Exp $ */
 
 /*-
@@ -86,10 +86,11 @@
 #define NICE_WEIGHT 2			/* priorities per nice level */
 #define	ESTCPULIM(e) min((e), NICE_WEIGHT * PRIO_MAX - PPQ)
 
-extern int	schedhz;			/* ideally: 16 */
+extern int schedhz;			/* ideally: 16 */
+extern int rrticks_init;		/* ticks per roundrobin() */
 
 #ifdef	_SYS_PROC_H_
-void schedclock(struct proc *p);
+void schedclock(struct proc *);
 #ifdef __HAVE_CPUINFO
 void roundrobin(struct cpu_info *);
 #endif
@@ -118,5 +119,75 @@ scheduler_wait_hook(parent, child)
 	parent->p_estcpu = ESTCPULIM(parent->p_estcpu + child->p_estcpu);
 }
 #endif	/* _SYS_PROC_H_ */
+
+#ifndef splsched
+#define splsched() splhigh()
+#endif
+#ifndef IPL_SCHED
+#define IPL_SCHED IPL_HIGH
+#endif
+
+#if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
+#include <sys/lock.h>
+
+/*
+ * XXX Instead of using struct lock for the kernel lock and thus requiring us
+ * XXX to implement simplelocks, causing all sorts of fine-grained locks all
+ * XXX over our tree getting activated consuming both time and potentially
+ * XXX introducing locking protocol bugs.
+ */
+#ifdef notyet
+
+extern struct simplelock sched_lock;
+
+#define	SCHED_ASSERT_LOCKED()	LOCK_ASSERT(simple_lock_held(&sched_lock))
+#define	SCHED_ASSERT_UNLOCKED()	LOCK_ASSERT(simple_lock_held(&sched_lock) == 0)
+
+#define	SCHED_LOCK(s)							\
+do {									\
+	s = splsched();							\
+	simple_lock(&sched_lock);					\
+} while (/* CONSTCOND */ 0)
+
+#define	SCHED_UNLOCK(s)							\
+do {									\
+	simple_unlock(&sched_lock);					\
+	splx(s);							\
+} while (/* CONSTCOND */ 0)
+
+#else
+
+extern struct __mp_lock sched_lock;
+
+#define	SCHED_ASSERT_LOCKED()	LOCK_ASSERT(__mp_lock_held(&sched_lock))
+#define	SCHED_ASSERT_UNLOCKED()	LOCK_ASSERT(__mp_lock_held(&sched_lock) == 0)
+
+#define	SCHED_LOCK(s)							\
+do {									\
+	s = splsched();							\
+	__mp_lock(&sched_lock);						\
+} while (/* CONSTCOND */ 0)
+
+#define	SCHED_UNLOCK(s)							\
+do {									\
+	__mp_unlock(&sched_lock);					\
+	splx(s);							\
+} while (/* CONSTCOND */ 0)
+
+#endif
+
+void	sched_lock_idle(void);
+void	sched_unlock_idle(void);
+
+#else /* ! MULTIPROCESSOR || LOCKDEBUG */
+
+#define	SCHED_ASSERT_LOCKED()		splassert(IPL_SCHED);
+#define	SCHED_ASSERT_UNLOCKED()		/* nothing */
+
+#define	SCHED_LOCK(s)			s = splsched()
+#define	SCHED_UNLOCK(s)			splx(s)
+
+#endif /* MULTIPROCESSOR || LOCKDEBUG */
+
 #endif	/* _KERNEL */
 #endif	/* _SYS_SCHED_H_ */
diff --git a/sys/sys/signalvar.h b/sys/sys/signalvar.h
index cac591c71d5..9db5681a21d 100644
--- a/sys/sys/signalvar.h
+++ b/sys/sys/signalvar.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: signalvar.h,v 1.11 2004/04/06 17:24:12 mickey Exp $	*/
+/*	$OpenBSD: signalvar.h,v 1.12 2004/06/13 21:49:28 niklas Exp $	*/
 /*	$NetBSD: signalvar.h,v 1.17 1996/04/22 01:23:31 christos Exp $	*/
 
 /*
@@ -159,7 +159,9 @@ void	csignal(pid_t pgid, int signum, uid_t uid, uid_t euid);
 int	issignal(struct proc *p);
 void	pgsignal(struct pgrp *pgrp, int sig, int checkctty);
 void	postsig(int sig);
-void	psignal(struct proc *p, int sig);
+void	psignal1(struct proc *p, int sig, int dolock);
+#define	psignal(p, sig)		psignal1((p), (sig), 1)
+#define	sched_psignal(p, sig)	psignal1((p), (sig), 0)
 void	siginit(struct proc *p);
 void	trapsignal(struct proc *p, int sig, u_long code, int type,
 	    union sigval val);
diff --git a/sys/sys/simplelock.h b/sys/sys/simplelock.h
index cfc6fdd8ed0..fbd39ef4d4e 100644
--- a/sys/sys/simplelock.h
+++ b/sys/sys/simplelock.h
@@ -1,7 +1,12 @@
-/*	$OpenBSD: simplelock.h,v 1.10 2002/03/14 01:27:14 millert Exp $	*/
+/*	$OpenBSD: simplelock.h,v 1.11 2004/06/13 21:49:28 niklas Exp $	*/
 
 #ifndef _SIMPLELOCK_H_
 #define _SIMPLELOCK_H_
+
+#ifdef MULTIPROCESSOR
+#include <machine/lock.h>
+#endif
+
 /*
  * A simple spin lock.
  *
@@ -12,28 +17,26 @@
  * of these locks while a process is sleeping.
  */
 struct simplelock {
+#ifdef MULTIPROCESSOR
+	__cpu_simple_lock_t lock_data;
+#else
 	int	lock_data;
+#endif
 };
 
 #ifdef _KERNEL
 
-#ifndef NCPUS
-#define NCPUS 1
-#endif
-
-#define SLOCK_LOCKED 1
-#define SLOCK_UNLOCKED 0
-
-#define SLOCK_INITIALIZER { SLOCK_UNLOCKED }
-
 /*
  * We can't debug locks when we use them in real life.
  */
-#if (NCPUS != 1) && defined(LOCKDEBUG)
+#if defined(MULTIPROCESSOR) && defined(LOCKDEBUG)
 #undef LOCKDEBUG
 #endif
 
-#if NCPUS == 1
+#if !defined(MULTIPROCESSOR) || 1
+
+#define SLOCK_LOCKED 1
+#define SLOCK_UNLOCKED 0
 
 #ifndef LOCKDEBUG
 
@@ -42,11 +45,7 @@ struct simplelock {
 #define	simple_unlock(lkp)
 #define simple_lock_assert(lkp)
 
-static __inline void simple_lock_init(struct simplelock *);
-
-static __inline void
-simple_lock_init(lkp)
-	struct simplelock *lkp;
+static __inline void simple_lock_init(struct simplelock *lkp)
 {
 
 	lkp->lock_data = SLOCK_UNLOCKED;
@@ -67,7 +66,7 @@ void simple_lock_init(struct simplelock *);
 
 #endif /* !defined(LOCKDEBUG) */
 
-#else  /* NCPUS >  1 */
+#else  /* MULTIPROCESSOR */
 
 /*
  * The simple-lock routines are the primitives out of which the lock
@@ -78,31 +77,26 @@ void simple_lock_init(struct simplelock *);
  * only be used for exclusive locks.
  */
 
-static __inline void
-simple_lock(lkp)
-	__volatile struct simplelock *lkp;
+static __inline void simple_lock_init(struct simplelock *lkp)
 {
-
-	while (test_and_set(&lkp->lock_data))
-		continue;
+	__cpu_simple_lock_init(&lkp->lock_data);
 }
 
-static __inline int
-simple_lock_try(lkp)
-	__volatile struct simplelock *lkp;
+static __inline void simple_lock(__volatile struct simplelock *lkp)
 {
-
-	return (!test_and_set(&lkp->lock_data))
+	__cpu_simple_lock(&lkp->lock_data);
 }
 
-static __inline void
-simple_unlock(lkp)
-	__volatile struct simplelock *lkp;
+static __inline int simple_lock_try(__volatile struct simplelock *lkp)
 {
+	return (__cpu_simple_lock_try(&lkp->lock_data));
+}
 
-	lkp->lock_data = 0;
+static __inline void simple_unlock(__volatile struct simplelock *lkp)
+{
+	__cpu_simple_unlock(&lkp->lock_data);
 }
-#endif /* NCPUS > 1 */
+#endif /* MULTIPROCESSOR */
 
 #endif /* _KERNEL */
 
diff --git a/sys/sys/systm.h b/sys/sys/systm.h
index 40f921050c3..b0bf65341a1 100644
--- a/sys/sys/systm.h
+++ b/sys/sys/systm.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: systm.h,v 1.61 2004/06/08 18:09:31 marc Exp $	*/
+/*	$OpenBSD: systm.h,v 1.62 2004/06/13 21:49:28 niklas Exp $	*/
 /*	$NetBSD: systm.h,v 1.50 1996/06/09 04:55:09 briggs Exp $	*/
 
 /*-
@@ -85,7 +85,11 @@ extern int nchrdev;		/* number of entries in cdevsw */
 
 extern int selwait;		/* select timeout address */
 
+#ifdef MULTIPROCESSOR
+#define curpriority (curcpu()->ci_schedstate.spc_curpriority)
+#else
 extern u_char curpriority;	/* priority of current process */
+#endif
 
 extern int maxmem;		/* max memory per process */
 extern int physmem;		/* physical memory */
@@ -294,4 +298,27 @@ int	read_symtab_from_file(struct proc *,struct vnode *,const char *);
 void	user_config(void);
 #endif
 
+#if defined(MULTIPROCESSOR)
+void	_kernel_lock_init(void);
+void	_kernel_lock(int);
+void	_kernel_unlock(void);
+void	_kernel_proc_lock(struct proc *);
+void	_kernel_proc_unlock(struct proc *);
+
+#define	KERNEL_LOCK_INIT()		_kernel_lock_init()
+#define	KERNEL_LOCK(flag)		_kernel_lock((flag))
+#define	KERNEL_UNLOCK()			_kernel_unlock()
+#define	KERNEL_PROC_LOCK(p)		_kernel_proc_lock((p))
+#define	KERNEL_PROC_UNLOCK(p)		_kernel_proc_unlock((p))
+
+#else /* ! MULTIPROCESSOR */
+
+#define	KERNEL_LOCK_INIT()		/* nothing */
+#define	KERNEL_LOCK(flag)		/* nothing */
+#define	KERNEL_UNLOCK()			/* nothing */
+#define	KERNEL_PROC_LOCK(p)		/* nothing */
+#define	KERNEL_PROC_UNLOCK(p)		/* nothing */
+
+#endif /* MULTIPROCESSOR */
+
 #endif /* __SYSTM_H__ */
diff --git a/sys/uvm/uvm_meter.c b/sys/uvm/uvm_meter.c
index 5e2a1ed6ef9..15bc3cd8651 100644
--- a/sys/uvm/uvm_meter.c
+++ b/sys/uvm/uvm_meter.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_meter.c,v 1.18 2004/05/17 13:08:42 pedro Exp $	*/
+/*	$OpenBSD: uvm_meter.c,v 1.19 2004/06/13 21:49:29 niklas Exp $	*/
 /*	$NetBSD: uvm_meter.c,v 1.21 2001/07/14 06:36:03 matt Exp $	*/
 
 /*
@@ -110,6 +110,7 @@ uvm_loadav(avg)
 		/* fall through */
 		case SRUN:
 		case SIDL:
+		case SONPROC:
 			nrun++;
 		}
 	}
@@ -263,6 +264,7 @@ uvm_total(totalp)
 
 		case SRUN:
 		case SIDL:
+		case SONPROC:
 			if (p->p_flag & P_INMEM)
 				totalp->t_rq++;
 			else