Make intr_nesting_level per-process, rather than per-cpu. Setup

interrupt threads to run with it always >= 1, so that malloc can
detect M_WAITOK from "interrupt" context.  This is also necessary
in order to context switch from sched_ithd() directly.

Reviewed By:	peter

54 files changed, 124 insertions, 96 deletions

diff --git a/sys/alpha/alpha/interrupt.c b/sys/alpha/alpha/interrupt.c
index 6311937..c11de3a 100644
--- a/sys/alpha/alpha/interrupt.c
+++ b/sys/alpha/alpha/interrupt.c
@@ -89,15 +89,15 @@ interrupt(a0, a1, a2, framep)
 	unsigned long a0, a1, a2;
 	struct trapframe *framep;
 {
+	struct proc *p;
+
 	/*
 	 * Find our per-cpu globals.
 	 */
 	globalp = (struct globaldata *) alpha_pal_rdval();
-
-	atomic_add_int(PCPU_PTR(intr_nesting_level), 1);
+	p = curproc;
+	atomic_add_int(&p->p_intr_nesting_level, 1);
 	{
-		struct proc *p = curproc;
-		if (!p) p = &proc0;
 		if ((caddr_t) framep < (caddr_t) p->p_addr + 1024) {
 			panic("possible stack overflow\n");
 		}
@@ -116,7 +116,7 @@ interrupt(a0, a1, a2, framep)
 		CTR0(KTR_INTR, "clock interrupt");
 		if (PCPU_GET(cpuno) != hwrpb->rpb_primary_cpu_id) {
 			CTR0(KTR_INTR, "ignoring clock on secondary");
-			atomic_subtract_int(PCPU_PTR(intr_nesting_level), 1);
+			atomic_subtract_int(&p->p_intr_nesting_level, 1);
 			return;
 		}
 			
@@ -152,7 +152,7 @@ interrupt(a0, a1, a2, framep)
 		    a0, a1, a2);
 		/* NOTREACHED */
 	}
-	atomic_subtract_int(PCPU_PTR(intr_nesting_level), 1);
+	atomic_subtract_int(&p->p_intr_nesting_level, 1);
 }
 
 void
diff --git a/sys/alpha/alpha/mp_machdep.c b/sys/alpha/alpha/mp_machdep.c
index 2ece633..e2a2be5 100644
--- a/sys/alpha/alpha/mp_machdep.c
+++ b/sys/alpha/alpha/mp_machdep.c
@@ -1053,7 +1053,7 @@ smp_handle_ipi(struct trapframe *frame)
 			CTR0(KTR_SMP, "IPI_CHECKSTATE");
 			if (frame->tf_regs[FRAME_PS] & ALPHA_PSL_USERMODE)
 				checkstate_cpustate[cpuno] = CHECKSTATE_USER;
-			else if (PCPU_GET(intr_nesting_level) == 1)
+			else if (curproc->p_intr_nesting_level == 1)
 				checkstate_cpustate[cpuno] = CHECKSTATE_SYS;
 			else
 				checkstate_cpustate[cpuno] = CHECKSTATE_INTR;
diff --git a/sys/alpha/include/cpu.h b/sys/alpha/include/cpu.h
index 3da213f..0edb2fc 100644
--- a/sys/alpha/include/cpu.h
+++ b/sys/alpha/include/cpu.h
@@ -65,7 +65,7 @@ struct clockframe {
 #define	CLKF_USERMODE(framep)						\
 	(((framep)->cf_tf.tf_regs[FRAME_PS] & ALPHA_PSL_USERMODE) != 0)
 #define	CLKF_PC(framep)		((framep)->cf_tf.tf_regs[FRAME_PC])
-#define	CLKF_INTR(framep)	(PCPU_GET(intr_nesting_level) >= 2)
+#define	CLKF_INTR(framep)	(curproc->p_intr_nesting_level >= 2)
 
 /*
  * Preempt the current process if in interrupt from user mode,
@@ -92,8 +92,6 @@ struct clockframe {
 #define	aston()		PCPU_SET(astpending, 1)
 
 #ifdef _KERNEL
-extern u_int astpending;
-extern u_int32_t intr_nesting_level;	/* bookeeping only; counts sw intrs */
 extern u_int32_t want_resched;		/* resched() was called */
 #endif
 
diff --git a/sys/alpha/include/globaldata.h b/sys/alpha/include/globaldata.h
index 127dd09..d99c5a9 100644
--- a/sys/alpha/include/globaldata.h
+++ b/sys/alpha/include/globaldata.h
@@ -56,7 +56,6 @@ struct globaldata {
 	u_int64_t	gd_pending_ipis;	/* pending IPI events */
 	u_int32_t	gd_next_asn;		/* next ASN to allocate */
 	u_int32_t	gd_current_asngen;	/* ASN rollover check */
-	u_int32_t	gd_intr_nesting_level;  /* interrupt recursion */
 
 	u_int		gd_astpending;
 	SLIST_ENTRY(globaldata) gd_allcpu;
diff --git a/sys/alpha/include/pcpu.h b/sys/alpha/include/pcpu.h
index 127dd09..d99c5a9 100644
--- a/sys/alpha/include/pcpu.h
+++ b/sys/alpha/include/pcpu.h
@@ -56,7 +56,6 @@ struct globaldata {
 	u_int64_t	gd_pending_ipis;	/* pending IPI events */
 	u_int32_t	gd_next_asn;		/* next ASN to allocate */
 	u_int32_t	gd_current_asngen;	/* ASN rollover check */
-	u_int32_t	gd_intr_nesting_level;  /* interrupt recursion */
 
 	u_int		gd_astpending;
 	SLIST_ENTRY(globaldata) gd_allcpu;
diff --git a/sys/amd64/amd64/apic_vector.S b/sys/amd64/amd64/apic_vector.S
index 4ebf35d..a2cdb47 100644
--- a/sys/amd64/amd64/apic_vector.S
+++ b/sys/amd64/amd64/apic_vector.S
@@ -48,7 +48,8 @@ IDTVEC(vec_name) ;							\
 	movl	$KPSEL,%eax ;						\
 	mov	%ax,%fs ;						\
 	FAKE_MCOUNT(13*4(%esp)) ;					\
-	incb	PCPU(INTR_NESTING_LEVEL) ;				\
+	movl	PCPU(CURPROC),%ebx ;					\
+	incl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	pushl	_intr_unit + (irq_num) * 4 ;				\
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	addl	$4, %esp ;						\
@@ -58,6 +59,7 @@ IDTVEC(vec_name) ;							\
 	movl	_intr_countp + (irq_num) * 4, %eax ;			\
 	lock ; 								\
 	incl	(%eax) ;						\
+	decl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	MEXITCOUNT ;							\
 	jmp	_doreti
 
@@ -150,7 +152,8 @@ IDTVEC(vec_name) ;							\
 	MASK_LEVEL_IRQ(irq_num) ;					\
 	EOI_IRQ(irq_num) ;						\
 0: ;									\
-	incb	PCPU(INTR_NESTING_LEVEL) ;				\
+	movl	PCPU(CURPROC),%ebx ;					\
+	incl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 ;	 								\
   /* entry point used by doreti_unpend for HWIs. */			\
 __CONCAT(Xresume,irq_num): ;						\
@@ -160,6 +163,7 @@ __CONCAT(Xresume,irq_num): ;						\
 	call	_sched_ithd ;						\
 	addl	$4, %esp ;		/* discard the parameter */	\
 ;									\
+	decl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	MEXITCOUNT ;							\
 	jmp	_doreti
 
@@ -301,7 +305,8 @@ _Xcpuast:
 	FAKE_MCOUNT(13*4(%esp))
 
 	orl	$AST_PENDING, PCPU(ASTPENDING)	/* XXX */
-	incb	PCPU(INTR_NESTING_LEVEL)
+	movl	PCPU(CURPROC),%ebx
+	incl	P_INTR_NESTING_LEVEL(%ebx)
 	sti
 	
 	movl	PCPU(CPUID), %eax
@@ -316,6 +321,7 @@ _Xcpuast:
 2:		
 	lock
 	incl	CNAME(cpuast_cnt)
+	decl	P_INTR_NESTING_LEVEL(%ebx)
 	MEXITCOUNT
 	jmp	_doreti
 1:
@@ -323,6 +329,7 @@ _Xcpuast:
 	POP_FRAME
 	iret			
 
+#if 0
 
 /*
  *	 Executed by a CPU when it receives an XFORWARD_IRQ IPI.
@@ -360,7 +367,6 @@ _Xforward_irq:
 	POP_FRAME
 	iret
 
-#if 0
 /*
  * 
  */
diff --git a/sys/amd64/amd64/exception.S b/sys/amd64/amd64/exception.S
index 603b4be..aafbf0b 100644
--- a/sys/amd64/amd64/exception.S
+++ b/sys/amd64/amd64/exception.S
@@ -177,7 +177,6 @@ IDTVEC(fpu)
 	call	_npx_intr
 
 	addl	$4,%esp
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 #else	/* DEV_NPX */
@@ -216,7 +215,6 @@ calltrap:
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -257,7 +255,6 @@ IDTVEC(syscall)
 	cli				/* atomic astpending access */
 	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 /*
@@ -289,7 +286,6 @@ IDTVEC(int0x80_syscall)
 	cli				/* atomic astpending access */
 	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 ENTRY(fork_trampoline)
@@ -323,7 +319,6 @@ ENTRY(fork_trampoline)
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/amd64/amd64/exception.s b/sys/amd64/amd64/exception.s
index 603b4be..aafbf0b 100644
--- a/sys/amd64/amd64/exception.s
+++ b/sys/amd64/amd64/exception.s
@@ -177,7 +177,6 @@ IDTVEC(fpu)
 	call	_npx_intr
 
 	addl	$4,%esp
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 #else	/* DEV_NPX */
@@ -216,7 +215,6 @@ calltrap:
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -257,7 +255,6 @@ IDTVEC(syscall)
 	cli				/* atomic astpending access */
 	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 /*
@@ -289,7 +286,6 @@ IDTVEC(int0x80_syscall)
 	cli				/* atomic astpending access */
 	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 ENTRY(fork_trampoline)
@@ -323,7 +319,6 @@ ENTRY(fork_trampoline)
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/amd64/amd64/genassym.c b/sys/amd64/amd64/genassym.c
index 8b9ca33..c307665 100644
--- a/sys/amd64/amd64/genassym.c
+++ b/sys/amd64/amd64/genassym.c
@@ -80,6 +80,7 @@ ASSYM(P_VMSPACE, offsetof(struct proc, p_vmspace));
 ASSYM(VM_PMAP, offsetof(struct vmspace, vm_pmap));
 ASSYM(PM_ACTIVE, offsetof(struct pmap, pm_active));
 ASSYM(P_ADDR, offsetof(struct proc, p_addr));
+ASSYM(P_INTR_NESTING_LEVEL, offsetof(struct proc, p_intr_nesting_level));
 ASSYM(P_STAT, offsetof(struct proc, p_stat));
 ASSYM(P_WCHAN, offsetof(struct proc, p_wchan));
 
@@ -182,7 +183,6 @@ ASSYM(GD_TSS_GDT, offsetof(struct globaldata, gd_tss_gdt));
 ASSYM(GD_ASTPENDING, offsetof(struct globaldata, gd_astpending));
 ASSYM(AST_PENDING, AST_PENDING);
 ASSYM(AST_RESCHED, AST_RESCHED);
-ASSYM(GD_INTR_NESTING_LEVEL, offsetof(struct globaldata, gd_intr_nesting_level));
 
 #ifdef USER_LDT
 ASSYM(GD_CURRENTLDT, offsetof(struct globaldata, gd_currentldt));
diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index bf560cd..966d3f8 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/amd64/amd64/mptable.c b/sys/amd64/amd64/mptable.c
index bf560cd..966d3f8 100644
--- a/sys/amd64/amd64/mptable.c
+++ b/sys/amd64/amd64/mptable.c
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c
index 6ee3e6d..0170dbe 100644
--- a/sys/amd64/amd64/trap.c
+++ b/sys/amd64/amd64/trap.c
@@ -480,7 +480,7 @@ restart:
 			if (in_vm86call)
 				break;
 
-			if (PCPU_GET(intr_nesting_level) != 0)
+			if (p->p_intr_nesting_level != 0)
 				break;
 
 			/*
@@ -687,7 +687,7 @@ trap_pfault(frame, usermode, eva)
 
 		if (p == NULL ||
 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
-		     (PCPU_GET(intr_nesting_level) != 0 ||
+		     (p->p_intr_nesting_level != 0 ||
 		      PCPU_GET(curpcb) == NULL ||
 		      PCPU_GET(curpcb)->pcb_onfault == NULL))) {
 			trap_fatal(frame, eva);
@@ -751,7 +751,7 @@ trap_pfault(frame, usermode, eva)
 		return (0);
 nogo:
 	if (!usermode) {
-		if (PCPU_GET(intr_nesting_level) == 0 &&
+		if (p->p_intr_nesting_level == 0 &&
 		    PCPU_GET(curpcb) != NULL &&
 		    PCPU_GET(curpcb)->pcb_onfault != NULL) {
 			frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
@@ -858,7 +858,7 @@ trap_pfault(frame, usermode, eva)
 		return (0);
 nogo:
 	if (!usermode) {
-		if (PCPU_GET(intr_nesting_level) == 0 &&
+		if (p->p_intr_nesting_level == 0 &&
 		    PCPU_GET(curpcb) != NULL &&
 		    PCPU_GET(curpcb)->pcb_onfault != NULL) {
 			frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
diff --git a/sys/amd64/include/cpu.h b/sys/amd64/include/cpu.h
index 6c4162b..87c47c5 100644
--- a/sys/amd64/include/cpu.h
+++ b/sys/amd64/include/cpu.h
@@ -62,7 +62,7 @@
 #define	CLKF_USERMODE(framep) \
 	((ISPL((framep)->cf_cs) == SEL_UPL) || (framep->cf_eflags & PSL_VM))
 
-#define CLKF_INTR(framep)	(PCPU_GET(intr_nesting_level) >= 2)
+#define CLKF_INTR(framep)	(curproc->p_intr_nesting_level >= 2)
 #define	CLKF_PC(framep)		((framep)->cf_eip)
 
 /*
diff --git a/sys/amd64/include/mptable.h b/sys/amd64/include/mptable.h
index bf560cd..966d3f8 100644
--- a/sys/amd64/include/mptable.h
+++ b/sys/amd64/include/mptable.h
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/amd64/include/pcpu.h b/sys/amd64/include/pcpu.h
index 149026e..7a7437a6 100644
--- a/sys/amd64/include/pcpu.h
+++ b/sys/amd64/include/pcpu.h
@@ -58,8 +58,6 @@ struct globaldata {
 	struct timeval	gd_switchtime;
 	struct i386tss	gd_common_tss;
 	int		gd_switchticks;
-	u_char		gd_intr_nesting_level;
-	u_char		gd_pad0[3];
 	struct segment_descriptor gd_common_tssd;
 	struct segment_descriptor *gd_tss_gdt;
 	int		gd_currentldt;		/* only used for USER_LDT */
diff --git a/sys/amd64/isa/atpic_vector.S b/sys/amd64/isa/atpic_vector.S
index 635d2ad..0a3456f 100644
--- a/sys/amd64/isa/atpic_vector.S
+++ b/sys/amd64/isa/atpic_vector.S
@@ -61,7 +61,8 @@ IDTVEC(vec_name) ; \
 	mov	$KPSEL,%ax ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -69,6 +70,7 @@ IDTVEC(vec_name) ; \
 	incl	_cnt+V_INTR ;	/* book-keeping can wait */ \
 	movl	_intr_countp + (irq_num) * 4,%eax ; \
 	incl	(%eax) ; \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	jmp	_doreti
 
@@ -102,13 +104,15 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
 	sti ; \
 	call	_sched_ithd ; \
 	addl	$4, %esp ;	/* discard the parameter */ \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	/* We could usually avoid the following jmp by inlining some of */ \
 	/* _doreti, but it's probably better to use less cache. */ \
diff --git a/sys/amd64/isa/icu_vector.S b/sys/amd64/isa/icu_vector.S
index 635d2ad..0a3456f 100644
--- a/sys/amd64/isa/icu_vector.S
+++ b/sys/amd64/isa/icu_vector.S
@@ -61,7 +61,8 @@ IDTVEC(vec_name) ; \
 	mov	$KPSEL,%ax ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -69,6 +70,7 @@ IDTVEC(vec_name) ; \
 	incl	_cnt+V_INTR ;	/* book-keeping can wait */ \
 	movl	_intr_countp + (irq_num) * 4,%eax ; \
 	incl	(%eax) ; \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	jmp	_doreti
 
@@ -102,13 +104,15 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
 	sti ; \
 	call	_sched_ithd ; \
 	addl	$4, %esp ;	/* discard the parameter */ \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	/* We could usually avoid the following jmp by inlining some of */ \
 	/* _doreti, but it's probably better to use less cache. */ \
diff --git a/sys/amd64/isa/icu_vector.s b/sys/amd64/isa/icu_vector.s
index 635d2ad..0a3456f 100644
--- a/sys/amd64/isa/icu_vector.s
+++ b/sys/amd64/isa/icu_vector.s
@@ -61,7 +61,8 @@ IDTVEC(vec_name) ; \
 	mov	$KPSEL,%ax ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -69,6 +70,7 @@ IDTVEC(vec_name) ; \
 	incl	_cnt+V_INTR ;	/* book-keeping can wait */ \
 	movl	_intr_countp + (irq_num) * 4,%eax ; \
 	incl	(%eax) ; \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	jmp	_doreti
 
@@ -102,13 +104,15 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
 	sti ; \
 	call	_sched_ithd ; \
 	addl	$4, %esp ;	/* discard the parameter */ \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	/* We could usually avoid the following jmp by inlining some of */ \
 	/* _doreti, but it's probably better to use less cache. */ \
diff --git a/sys/amd64/isa/intr_machdep.c b/sys/amd64/isa/intr_machdep.c
index 00f5f23..b9328ae 100644
--- a/sys/amd64/isa/intr_machdep.c
+++ b/sys/amd64/isa/intr_machdep.c
@@ -577,6 +577,7 @@ inthand_add(const char *name, int irq, driver_intr_t handler, void *arg,
 				if (errcode)
 					panic("inthand_add: Can't create "
 					      "interrupt thread");
+				p->p_intr_nesting_level = 1;
 				p->p_rtprio.type = RTP_PRIO_ITHREAD;
 				p->p_stat = SWAIT; /* we're idle */
 
diff --git a/sys/amd64/isa/nmi.c b/sys/amd64/isa/nmi.c
index 00f5f23..b9328ae 100644
--- a/sys/amd64/isa/nmi.c
+++ b/sys/amd64/isa/nmi.c
@@ -577,6 +577,7 @@ inthand_add(const char *name, int irq, driver_intr_t handler, void *arg,
 				if (errcode)
 					panic("inthand_add: Can't create "
 					      "interrupt thread");
+				p->p_intr_nesting_level = 1;
 				p->p_rtprio.type = RTP_PRIO_ITHREAD;
 				p->p_stat = SWAIT; /* we're idle */
 
diff --git a/sys/dev/usb/ohci.c b/sys/dev/usb/ohci.c
index c9062a4..c2abcd8 100644
--- a/sys/dev/usb/ohci.c
+++ b/sys/dev/usb/ohci.c
@@ -1889,7 +1889,7 @@ ohci_abort_xfer(xfer, status)
 		timeout(ohci_abort_xfer_end, xfer, hz / USB_FRAMES_PER_SECOND);
 	} else {
 #if defined(DIAGNOSTIC) && defined(__i386__) && defined(__FreeBSD__)
-		KASSERT(PCPU_GET(intr_nesting_level) == 0,
+		KASSERT(curproc->p_intr_nesting_level == 0,
 	        	("ohci_abort_req in interrupt context"));
 #endif
 		usb_delay_ms(opipe->pipe.device->bus, 1);
diff --git a/sys/dev/usb/uhci.c b/sys/dev/usb/uhci.c
index 7098db1..1628268 100644
--- a/sys/dev/usb/uhci.c
+++ b/sys/dev/usb/uhci.c
@@ -1610,7 +1610,7 @@ uhci_abort_xfer(usbd_xfer_handle xfer, usbd_status status)
 		timeout(uhci_abort_xfer_end, xfer, hz / USB_FRAMES_PER_SECOND);
 	} else {
 #if defined(DIAGNOSTIC) && defined(__i386__) && defined(__FreeBSD__)
-		KASSERT(PCPU_GET(intr_nesting_level) == 0,
+		KASSERT(curproc->p_intr_nesting_level == 0,
 	        	("ohci_abort_req in interrupt context"));
 #endif
 		usb_delay_ms(xfer->pipe->device->bus, 1);
diff --git a/sys/dev/usb/usbdi.c b/sys/dev/usb/usbdi.c
index c39a856..6203e8a 100644
--- a/sys/dev/usb/usbdi.c
+++ b/sys/dev/usb/usbdi.c
@@ -931,7 +931,7 @@ usbd_do_request_flags(dev, req, data, flags, actlen)
 
 #ifdef DIAGNOSTIC
 #if defined(__i386__) && defined(__FreeBSD__)
-	KASSERT(PCPU_GET(intr_nesting_level) == 0,
+	KASSERT(curproc->p_intr_nesting_level == 0,
 	       	("usbd_do_request: in interrupt context"));
 #endif
 	if (dev->bus->intr_context) {
diff --git a/sys/dev/vinum/vinumhdr.h b/sys/dev/vinum/vinumhdr.h
index 8a5f70e..2b7d290 100644
--- a/sys/dev/vinum/vinumhdr.h
+++ b/sys/dev/vinum/vinumhdr.h
@@ -87,7 +87,7 @@ void FFree (void *mem, char *, int);
 #define LOCKDRIVE(d) lockdrive (d, __FILE__, __LINE__)
 #else
 #define Malloc(x)  malloc((x), M_DEVBUF, \
-	PCPU_GET(intr_nesting_level) == 0? M_WAITOK: M_NOWAIT)
+	curproc->p_intr_nesting_level == 0? M_WAITOK: M_NOWAIT)
 #define Free(x)    free((x), M_DEVBUF)
 #define LOCKDRIVE(d) lockdrive (d)
 #endif
diff --git a/sys/dev/vinum/vinummemory.c b/sys/dev/vinum/vinummemory.c
index e28f7ae..58eea67 100644
--- a/sys/dev/vinum/vinummemory.c
+++ b/sys/dev/vinum/vinummemory.c
@@ -151,7 +151,7 @@ MMalloc(int size, char *file, int line)
 	return 0;					    /* can't continue */
     }
     /* Wait for malloc if we can */
-    result = malloc(size, M_DEVBUF, PCPU_GET(intr_nesting_level) == 0 ? M_WAITOK : M_NOWAIT);
+    result = malloc(size, M_DEVBUF, curproc->p_intr_nesting_level == 0 ? M_WAITOK : M_NOWAIT);
     if (result == NULL)
 	log(LOG_ERR, "vinum: can't allocate %d bytes from %s:%d\n", size, file, line);
     else {
diff --git a/sys/i386/i386/apic_vector.s b/sys/i386/i386/apic_vector.s
index 4ebf35d..a2cdb47 100644
--- a/sys/i386/i386/apic_vector.s
+++ b/sys/i386/i386/apic_vector.s
@@ -48,7 +48,8 @@ IDTVEC(vec_name) ;							\
 	movl	$KPSEL,%eax ;						\
 	mov	%ax,%fs ;						\
 	FAKE_MCOUNT(13*4(%esp)) ;					\
-	incb	PCPU(INTR_NESTING_LEVEL) ;				\
+	movl	PCPU(CURPROC),%ebx ;					\
+	incl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	pushl	_intr_unit + (irq_num) * 4 ;				\
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	addl	$4, %esp ;						\
@@ -58,6 +59,7 @@ IDTVEC(vec_name) ;							\
 	movl	_intr_countp + (irq_num) * 4, %eax ;			\
 	lock ; 								\
 	incl	(%eax) ;						\
+	decl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	MEXITCOUNT ;							\
 	jmp	_doreti
 
@@ -150,7 +152,8 @@ IDTVEC(vec_name) ;							\
 	MASK_LEVEL_IRQ(irq_num) ;					\
 	EOI_IRQ(irq_num) ;						\
 0: ;									\
-	incb	PCPU(INTR_NESTING_LEVEL) ;				\
+	movl	PCPU(CURPROC),%ebx ;					\
+	incl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 ;	 								\
   /* entry point used by doreti_unpend for HWIs. */			\
 __CONCAT(Xresume,irq_num): ;						\
@@ -160,6 +163,7 @@ __CONCAT(Xresume,irq_num): ;						\
 	call	_sched_ithd ;						\
 	addl	$4, %esp ;		/* discard the parameter */	\
 ;									\
+	decl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	MEXITCOUNT ;							\
 	jmp	_doreti
 
@@ -301,7 +305,8 @@ _Xcpuast:
 	FAKE_MCOUNT(13*4(%esp))
 
 	orl	$AST_PENDING, PCPU(ASTPENDING)	/* XXX */
-	incb	PCPU(INTR_NESTING_LEVEL)
+	movl	PCPU(CURPROC),%ebx
+	incl	P_INTR_NESTING_LEVEL(%ebx)
 	sti
 	
 	movl	PCPU(CPUID), %eax
@@ -316,6 +321,7 @@ _Xcpuast:
 2:		
 	lock
 	incl	CNAME(cpuast_cnt)
+	decl	P_INTR_NESTING_LEVEL(%ebx)
 	MEXITCOUNT
 	jmp	_doreti
 1:
@@ -323,6 +329,7 @@ _Xcpuast:
 	POP_FRAME
 	iret			
 
+#if 0
 
 /*
  *	 Executed by a CPU when it receives an XFORWARD_IRQ IPI.
@@ -360,7 +367,6 @@ _Xforward_irq:
 	POP_FRAME
 	iret
 
-#if 0
 /*
  * 
  */
diff --git a/sys/i386/i386/exception.s b/sys/i386/i386/exception.s
index 603b4be..aafbf0b 100644
--- a/sys/i386/i386/exception.s
+++ b/sys/i386/i386/exception.s
@@ -177,7 +177,6 @@ IDTVEC(fpu)
 	call	_npx_intr
 
 	addl	$4,%esp
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 #else	/* DEV_NPX */
@@ -216,7 +215,6 @@ calltrap:
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -257,7 +255,6 @@ IDTVEC(syscall)
 	cli				/* atomic astpending access */
 	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 /*
@@ -289,7 +286,6 @@ IDTVEC(int0x80_syscall)
 	cli				/* atomic astpending access */
 	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 ENTRY(fork_trampoline)
@@ -323,7 +319,6 @@ ENTRY(fork_trampoline)
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/i386/i386/genassym.c b/sys/i386/i386/genassym.c
index 8b9ca33..c307665 100644
--- a/sys/i386/i386/genassym.c
+++ b/sys/i386/i386/genassym.c
@@ -80,6 +80,7 @@ ASSYM(P_VMSPACE, offsetof(struct proc, p_vmspace));
 ASSYM(VM_PMAP, offsetof(struct vmspace, vm_pmap));
 ASSYM(PM_ACTIVE, offsetof(struct pmap, pm_active));
 ASSYM(P_ADDR, offsetof(struct proc, p_addr));
+ASSYM(P_INTR_NESTING_LEVEL, offsetof(struct proc, p_intr_nesting_level));
 ASSYM(P_STAT, offsetof(struct proc, p_stat));
 ASSYM(P_WCHAN, offsetof(struct proc, p_wchan));
 
@@ -182,7 +183,6 @@ ASSYM(GD_TSS_GDT, offsetof(struct globaldata, gd_tss_gdt));
 ASSYM(GD_ASTPENDING, offsetof(struct globaldata, gd_astpending));
 ASSYM(AST_PENDING, AST_PENDING);
 ASSYM(AST_RESCHED, AST_RESCHED);
-ASSYM(GD_INTR_NESTING_LEVEL, offsetof(struct globaldata, gd_intr_nesting_level));
 
 #ifdef USER_LDT
 ASSYM(GD_CURRENTLDT, offsetof(struct globaldata, gd_currentldt));
diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c
index bf560cd..966d3f8 100644
--- a/sys/i386/i386/mp_machdep.c
+++ b/sys/i386/i386/mp_machdep.c
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/i386/i386/mptable.c b/sys/i386/i386/mptable.c
index bf560cd..966d3f8 100644
--- a/sys/i386/i386/mptable.c
+++ b/sys/i386/i386/mptable.c
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/i386/i386/trap.c b/sys/i386/i386/trap.c
index 6ee3e6d..0170dbe 100644
--- a/sys/i386/i386/trap.c
+++ b/sys/i386/i386/trap.c
@@ -480,7 +480,7 @@ restart:
 			if (in_vm86call)
 				break;
 
-			if (PCPU_GET(intr_nesting_level) != 0)
+			if (p->p_intr_nesting_level != 0)
 				break;
 
 			/*
@@ -687,7 +687,7 @@ trap_pfault(frame, usermode, eva)
 
 		if (p == NULL ||
 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
-		     (PCPU_GET(intr_nesting_level) != 0 ||
+		     (p->p_intr_nesting_level != 0 ||
 		      PCPU_GET(curpcb) == NULL ||
 		      PCPU_GET(curpcb)->pcb_onfault == NULL))) {
 			trap_fatal(frame, eva);
@@ -751,7 +751,7 @@ trap_pfault(frame, usermode, eva)
 		return (0);
 nogo:
 	if (!usermode) {
-		if (PCPU_GET(intr_nesting_level) == 0 &&
+		if (p->p_intr_nesting_level == 0 &&
 		    PCPU_GET(curpcb) != NULL &&
 		    PCPU_GET(curpcb)->pcb_onfault != NULL) {
 			frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
@@ -858,7 +858,7 @@ trap_pfault(frame, usermode, eva)
 		return (0);
 nogo:
 	if (!usermode) {
-		if (PCPU_GET(intr_nesting_level) == 0 &&
+		if (p->p_intr_nesting_level == 0 &&
 		    PCPU_GET(curpcb) != NULL &&
 		    PCPU_GET(curpcb)->pcb_onfault != NULL) {
 			frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
diff --git a/sys/i386/i386/vm86bios.s b/sys/i386/i386/vm86bios.s
index a71dd11..5f3013c 100644
--- a/sys/i386/i386/vm86bios.s
+++ b/sys/i386/i386/vm86bios.s
@@ -131,7 +131,6 @@ ENTRY(vm86_bioscall)
 	/*
 	 * Return via _doreti
 	 */
-	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/i386/include/cpu.h b/sys/i386/include/cpu.h
index 6c4162b..87c47c5 100644
--- a/sys/i386/include/cpu.h
+++ b/sys/i386/include/cpu.h
@@ -62,7 +62,7 @@
 #define	CLKF_USERMODE(framep) \
 	((ISPL((framep)->cf_cs) == SEL_UPL) || (framep->cf_eflags & PSL_VM))
 
-#define CLKF_INTR(framep)	(PCPU_GET(intr_nesting_level) >= 2)
+#define CLKF_INTR(framep)	(curproc->p_intr_nesting_level >= 2)
 #define	CLKF_PC(framep)		((framep)->cf_eip)
 
 /*
diff --git a/sys/i386/include/globaldata.h b/sys/i386/include/globaldata.h
index 149026e..7a7437a6 100644
--- a/sys/i386/include/globaldata.h
+++ b/sys/i386/include/globaldata.h
@@ -58,8 +58,6 @@ struct globaldata {
 	struct timeval	gd_switchtime;
 	struct i386tss	gd_common_tss;
 	int		gd_switchticks;
-	u_char		gd_intr_nesting_level;
-	u_char		gd_pad0[3];
 	struct segment_descriptor gd_common_tssd;
 	struct segment_descriptor *gd_tss_gdt;
 	int		gd_currentldt;		/* only used for USER_LDT */
diff --git a/sys/i386/include/mptable.h b/sys/i386/include/mptable.h
index bf560cd..966d3f8 100644
--- a/sys/i386/include/mptable.h
+++ b/sys/i386/include/mptable.h
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/i386/include/pcpu.h b/sys/i386/include/pcpu.h
index 149026e..7a7437a6 100644
--- a/sys/i386/include/pcpu.h
+++ b/sys/i386/include/pcpu.h
@@ -58,8 +58,6 @@ struct globaldata {
 	struct timeval	gd_switchtime;
 	struct i386tss	gd_common_tss;
 	int		gd_switchticks;
-	u_char		gd_intr_nesting_level;
-	u_char		gd_pad0[3];
 	struct segment_descriptor gd_common_tssd;
 	struct segment_descriptor *gd_tss_gdt;
 	int		gd_currentldt;		/* only used for USER_LDT */
diff --git a/sys/i386/isa/apic_vector.s b/sys/i386/isa/apic_vector.s
index 4ebf35d..a2cdb47 100644
--- a/sys/i386/isa/apic_vector.s
+++ b/sys/i386/isa/apic_vector.s
@@ -48,7 +48,8 @@ IDTVEC(vec_name) ;							\
 	movl	$KPSEL,%eax ;						\
 	mov	%ax,%fs ;						\
 	FAKE_MCOUNT(13*4(%esp)) ;					\
-	incb	PCPU(INTR_NESTING_LEVEL) ;				\
+	movl	PCPU(CURPROC),%ebx ;					\
+	incl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	pushl	_intr_unit + (irq_num) * 4 ;				\
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	addl	$4, %esp ;						\
@@ -58,6 +59,7 @@ IDTVEC(vec_name) ;							\
 	movl	_intr_countp + (irq_num) * 4, %eax ;			\
 	lock ; 								\
 	incl	(%eax) ;						\
+	decl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	MEXITCOUNT ;							\
 	jmp	_doreti
 
@@ -150,7 +152,8 @@ IDTVEC(vec_name) ;							\
 	MASK_LEVEL_IRQ(irq_num) ;					\
 	EOI_IRQ(irq_num) ;						\
 0: ;									\
-	incb	PCPU(INTR_NESTING_LEVEL) ;				\
+	movl	PCPU(CURPROC),%ebx ;					\
+	incl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 ;	 								\
   /* entry point used by doreti_unpend for HWIs. */			\
 __CONCAT(Xresume,irq_num): ;						\
@@ -160,6 +163,7 @@ __CONCAT(Xresume,irq_num): ;						\
 	call	_sched_ithd ;						\
 	addl	$4, %esp ;		/* discard the parameter */	\
 ;									\
+	decl	P_INTR_NESTING_LEVEL(%ebx) ;				\
 	MEXITCOUNT ;							\
 	jmp	_doreti
 
@@ -301,7 +305,8 @@ _Xcpuast:
 	FAKE_MCOUNT(13*4(%esp))
 
 	orl	$AST_PENDING, PCPU(ASTPENDING)	/* XXX */
-	incb	PCPU(INTR_NESTING_LEVEL)
+	movl	PCPU(CURPROC),%ebx
+	incl	P_INTR_NESTING_LEVEL(%ebx)
 	sti
 	
 	movl	PCPU(CPUID), %eax
@@ -316,6 +321,7 @@ _Xcpuast:
 2:		
 	lock
 	incl	CNAME(cpuast_cnt)
+	decl	P_INTR_NESTING_LEVEL(%ebx)
 	MEXITCOUNT
 	jmp	_doreti
 1:
@@ -323,6 +329,7 @@ _Xcpuast:
 	POP_FRAME
 	iret			
 
+#if 0
 
 /*
  *	 Executed by a CPU when it receives an XFORWARD_IRQ IPI.
@@ -360,7 +367,6 @@ _Xforward_irq:
 	POP_FRAME
 	iret
 
-#if 0
 /*
  * 
  */
diff --git a/sys/i386/isa/atpic_vector.s b/sys/i386/isa/atpic_vector.s
index 635d2ad..0a3456f 100644
--- a/sys/i386/isa/atpic_vector.s
+++ b/sys/i386/isa/atpic_vector.s
@@ -61,7 +61,8 @@ IDTVEC(vec_name) ; \
 	mov	$KPSEL,%ax ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -69,6 +70,7 @@ IDTVEC(vec_name) ; \
 	incl	_cnt+V_INTR ;	/* book-keeping can wait */ \
 	movl	_intr_countp + (irq_num) * 4,%eax ; \
 	incl	(%eax) ; \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	jmp	_doreti
 
@@ -102,13 +104,15 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
 	sti ; \
 	call	_sched_ithd ; \
 	addl	$4, %esp ;	/* discard the parameter */ \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	/* We could usually avoid the following jmp by inlining some of */ \
 	/* _doreti, but it's probably better to use less cache. */ \
diff --git a/sys/i386/isa/icu_vector.s b/sys/i386/isa/icu_vector.s
index 635d2ad..0a3456f 100644
--- a/sys/i386/isa/icu_vector.s
+++ b/sys/i386/isa/icu_vector.s
@@ -61,7 +61,8 @@ IDTVEC(vec_name) ; \
 	mov	$KPSEL,%ax ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -69,6 +70,7 @@ IDTVEC(vec_name) ; \
 	incl	_cnt+V_INTR ;	/* book-keeping can wait */ \
 	movl	_intr_countp + (irq_num) * 4,%eax ; \
 	incl	(%eax) ; \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	jmp	_doreti
 
@@ -102,13 +104,15 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	PCPU(INTR_NESTING_LEVEL) ; \
+	movl	PCPU(CURPROC),%ebx ; \
+	incl	P_INTR_NESTING_LEVEL(%ebx) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
 	sti ; \
 	call	_sched_ithd ; \
 	addl	$4, %esp ;	/* discard the parameter */ \
+	decl	P_INTR_NESTING_LEVEL(%ebx) ; \
 	MEXITCOUNT ; \
 	/* We could usually avoid the following jmp by inlining some of */ \
 	/* _doreti, but it's probably better to use less cache. */ \
diff --git a/sys/i386/isa/intr_machdep.c b/sys/i386/isa/intr_machdep.c
index 00f5f23..b9328ae 100644
--- a/sys/i386/isa/intr_machdep.c
+++ b/sys/i386/isa/intr_machdep.c
@@ -577,6 +577,7 @@ inthand_add(const char *name, int irq, driver_intr_t handler, void *arg,
 				if (errcode)
 					panic("inthand_add: Can't create "
 					      "interrupt thread");
+				p->p_intr_nesting_level = 1;
 				p->p_rtprio.type = RTP_PRIO_ITHREAD;
 				p->p_stat = SWAIT; /* we're idle */
 
diff --git a/sys/i386/isa/ipl.s b/sys/i386/isa/ipl.s
index ca874ea..7c41589 100644
--- a/sys/i386/isa/ipl.s
+++ b/sys/i386/isa/ipl.s
@@ -57,8 +57,6 @@
 _doreti:
 	FAKE_MCOUNT(_bintr)		/* init "from" _bintr -> _doreti */
 doreti_next:
-	decb	PCPU(INTR_NESTING_LEVEL)
-
 	/* Check for ASTs that can be handled now. */
 	testl	$AST_PENDING,PCPU(ASTPENDING)
 	je	doreti_exit		/* no AST, exit */
@@ -128,7 +126,6 @@ doreti_ast:
 	sti
 	movl	$T_ASTFLT,TF_TRAPNO(%esp)
 	call	_ast
-	movb	$1,PCPU(INTR_NESTING_LEVEL) /* for doreti_next to decrement */
 	jmp	doreti_next
 
 #ifdef APIC_IO
diff --git a/sys/i386/isa/nmi.c b/sys/i386/isa/nmi.c
index 00f5f23..b9328ae 100644
--- a/sys/i386/isa/nmi.c
+++ b/sys/i386/isa/nmi.c
@@ -577,6 +577,7 @@ inthand_add(const char *name, int irq, driver_intr_t handler, void *arg,
 				if (errcode)
 					panic("inthand_add: Can't create "
 					      "interrupt thread");
+				p->p_intr_nesting_level = 1;
 				p->p_rtprio.type = RTP_PRIO_ITHREAD;
 				p->p_stat = SWAIT; /* we're idle */
 
diff --git a/sys/ia64/ia64/interrupt.c b/sys/ia64/ia64/interrupt.c
index 698b46c..fc2f84a 100644
--- a/sys/ia64/ia64/interrupt.c
+++ b/sys/ia64/ia64/interrupt.c
@@ -78,8 +78,10 @@ static u_int schedclk2;
 void
 interrupt(u_int64_t vector, struct trapframe *framep)
 {
-	atomic_add_int(PCPU_PTR(intr_nesting_level), 1);
+	struct proc *p;
 
+	p = curproc;
+	atomic_add_int(&p->p_intr_nesting_level, 1);
 	switch (vector) {
 	case 240:	/* clock interrupt */
 		CTR0(KTR_INTR, "clock interrupt");
@@ -104,7 +106,7 @@ interrupt(u_int64_t vector, struct trapframe *framep)
 		panic("unexpected interrupt: vec %ld\n", vector);
 		/* NOTREACHED */
 	}
-	atomic_subtract_int(PCPU_PTR(intr_nesting_level), 1);
+	atomic_subtract_int(&p->p_intr_nesting_level, 1);
 }
 
 
diff --git a/sys/ia64/ia64/mp_machdep.c b/sys/ia64/ia64/mp_machdep.c
index 486d0aa..aa42156 100644
--- a/sys/ia64/ia64/mp_machdep.c
+++ b/sys/ia64/ia64/mp_machdep.c
@@ -772,7 +772,7 @@ smp_handle_ipi(struct trapframe *frame)
 			if ((frame->tf_cr_ipsr & IA64_PSR_CPL)
 			    == IA64_PSR_CPL_USER)
 				checkstate_cpustate[cpuno] = CHECKSTATE_USER;
-			else if (PCPU_GET(intr_nesting_level) == 1)
+			else if (curproc->p_intr_nesting_level == 1)
 				checkstate_cpustate[cpuno] = CHECKSTATE_SYS;
 			else
 				checkstate_cpustate[cpuno] = CHECKSTATE_INTR;
diff --git a/sys/ia64/include/cpu.h b/sys/ia64/include/cpu.h
index d581447..ff36f05 100644
--- a/sys/ia64/include/cpu.h
+++ b/sys/ia64/include/cpu.h
@@ -67,7 +67,7 @@ struct clockframe {
 #define	CLKF_BASEPRI(framep)						\
 	(((framep)->cf_tf.tf_cr_ipsr & IA64_PSR_I) == 0)
 #define	CLKF_PC(framep)		((framep)->cf_tf.tf_cr_iip)
-#define	CLKF_INTR(framep)	(PCPU_GET(intr_nesting_level) >= 2)
+#define	CLKF_INTR(framep)	(curproc->p_intr_nesting_level >= 2)
 
 /*
  * Preempt the current process if in interrupt from user mode,
diff --git a/sys/ia64/include/globaldata.h b/sys/ia64/include/globaldata.h
index 6068bd1..91c7fff 100644
--- a/sys/ia64/include/globaldata.h
+++ b/sys/ia64/include/globaldata.h
@@ -56,7 +56,6 @@ struct globaldata {
 	struct pmap	*gd_current_pmap;	/* which pmap is active */
 	u_int32_t	gd_next_asn;		/* next ASN to allocate */
 	u_int32_t	gd_current_asngen;	/* ASN rollover check */
-	u_int32_t	gd_intr_nesting_level;  /* interrupt recursion */
 
 	u_int		gd_astpending;
 	SLIST_ENTRY(globaldata) gd_allcpu;
diff --git a/sys/ia64/include/pcpu.h b/sys/ia64/include/pcpu.h
index 6068bd1..91c7fff 100644
--- a/sys/ia64/include/pcpu.h
+++ b/sys/ia64/include/pcpu.h
@@ -56,7 +56,6 @@ struct globaldata {
 	struct pmap	*gd_current_pmap;	/* which pmap is active */
 	u_int32_t	gd_next_asn;		/* next ASN to allocate */
 	u_int32_t	gd_current_asngen;	/* ASN rollover check */
-	u_int32_t	gd_intr_nesting_level;  /* interrupt recursion */
 
 	u_int		gd_astpending;
 	SLIST_ENTRY(globaldata) gd_allcpu;
diff --git a/sys/kern/kern_fork.c b/sys/kern/kern_fork.c
index 1c8cb38..9dc8ec3 100644
--- a/sys/kern/kern_fork.c
+++ b/sys/kern/kern_fork.c
@@ -351,6 +351,7 @@ again:
 		nextpid = trypid;
 
 	p2 = newproc;
+	p2->p_intr_nesting_level = 0;
 	p2->p_stat = SIDL;			/* protect against others */
 	p2->p_pid = trypid;
 	LIST_INSERT_HEAD(&allproc, p2, p_list);
diff --git a/sys/kern/kern_malloc.c b/sys/kern/kern_malloc.c
index 396f02b..ec33f61 100644
--- a/sys/kern/kern_malloc.c
+++ b/sys/kern/kern_malloc.c
@@ -44,6 +44,7 @@
 #include <sys/mutex.h>
 #include <sys/vmmeter.h>
 #include <sys/lock.h>
+#include <sys/proc.h>
 
 #include <vm/vm.h>
 #include <vm/vm_param.h>
@@ -147,7 +148,7 @@ malloc(size, type, flags)
 
 #if defined(INVARIANTS) && defined(__i386__)
 	if (flags == M_WAITOK)
-		KASSERT(PCPU_GET(intr_nesting_level) == 0,
+		KASSERT(curproc->p_intr_nesting_level == 0,
 		   ("malloc(M_WAITOK) in interrupt context"));
 #endif
 	indx = BUCKETINDX(size);
diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c
index bf560cd..966d3f8 100644
--- a/sys/kern/subr_smp.c
+++ b/sys/kern/subr_smp.c
@@ -581,10 +581,12 @@ mp_enable(u_int boot_addr)
 	/* install an inter-CPU IPI for forcing an additional software trap */
 	setidt(XCPUAST_OFFSET, Xcpuast,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
-	
+
+#if 0	
 	/* install an inter-CPU IPI for interrupt forwarding */
 	setidt(XFORWARD_IRQ_OFFSET, Xforward_irq,
 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
+#endif
 
 	/* install an inter-CPU IPI for CPU stop/restart */
 	setidt(XCPUSTOP_OFFSET, Xcpustop,
diff --git a/sys/kern/subr_trap.c b/sys/kern/subr_trap.c
index 6ee3e6d..0170dbe 100644
--- a/sys/kern/subr_trap.c
+++ b/sys/kern/subr_trap.c
@@ -480,7 +480,7 @@ restart:
 			if (in_vm86call)
 				break;
 
-			if (PCPU_GET(intr_nesting_level) != 0)
+			if (p->p_intr_nesting_level != 0)
 				break;
 
 			/*
@@ -687,7 +687,7 @@ trap_pfault(frame, usermode, eva)
 
 		if (p == NULL ||
 		    (!usermode && va < VM_MAXUSER_ADDRESS &&
-		     (PCPU_GET(intr_nesting_level) != 0 ||
+		     (p->p_intr_nesting_level != 0 ||
 		      PCPU_GET(curpcb) == NULL ||
 		      PCPU_GET(curpcb)->pcb_onfault == NULL))) {
 			trap_fatal(frame, eva);
@@ -751,7 +751,7 @@ trap_pfault(frame, usermode, eva)
 		return (0);
 nogo:
 	if (!usermode) {
-		if (PCPU_GET(intr_nesting_level) == 0 &&
+		if (p->p_intr_nesting_level == 0 &&
 		    PCPU_GET(curpcb) != NULL &&
 		    PCPU_GET(curpcb)->pcb_onfault != NULL) {
 			frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
@@ -858,7 +858,7 @@ trap_pfault(frame, usermode, eva)
 		return (0);
 nogo:
 	if (!usermode) {
-		if (PCPU_GET(intr_nesting_level) == 0 &&
+		if (p->p_intr_nesting_level == 0 &&
 		    PCPU_GET(curpcb) != NULL &&
 		    PCPU_GET(curpcb)->pcb_onfault != NULL) {
 			frame->tf_eip = (int)PCPU_GET(curpcb)->pcb_onfault;
diff --git a/sys/powerpc/include/globaldata.h b/sys/powerpc/include/globaldata.h
index 127dd09..d99c5a9 100644
--- a/sys/powerpc/include/globaldata.h
+++ b/sys/powerpc/include/globaldata.h
@@ -56,7 +56,6 @@ struct globaldata {
 	u_int64_t	gd_pending_ipis;	/* pending IPI events */
 	u_int32_t	gd_next_asn;		/* next ASN to allocate */
 	u_int32_t	gd_current_asngen;	/* ASN rollover check */
-	u_int32_t	gd_intr_nesting_level;  /* interrupt recursion */
 
 	u_int		gd_astpending;
 	SLIST_ENTRY(globaldata) gd_allcpu;
diff --git a/sys/powerpc/include/pcpu.h b/sys/powerpc/include/pcpu.h
index 127dd09..d99c5a9 100644
--- a/sys/powerpc/include/pcpu.h
+++ b/sys/powerpc/include/pcpu.h
@@ -56,7 +56,6 @@ struct globaldata {
 	u_int64_t	gd_pending_ipis;	/* pending IPI events */
 	u_int32_t	gd_next_asn;		/* next ASN to allocate */
 	u_int32_t	gd_current_asngen;	/* ASN rollover check */
-	u_int32_t	gd_intr_nesting_level;  /* interrupt recursion */
 
 	u_int		gd_astpending;
 	SLIST_ENTRY(globaldata) gd_allcpu;
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
index 41f8f6a..ee093ba 100644
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -159,6 +159,7 @@ struct	proc {
 #define	p_rlimit	p_limit->pl_rlimit
 
 	int	p_flag;			/* (c/j) P_* flags. */
+	int	p_intr_nesting_level;	/* (n) Interrupt recursion. */
 	char	p_stat;			/* (j) S* process status. */
 	char	p_pad1[3];