diff --git a/sys/amd64/amd64/apic_vector.S b/sys/amd64/amd64/apic_vector.S
index 995cf213fb25..a23ccff7b67d 100644
--- a/sys/amd64/amd64/apic_vector.S
+++ b/sys/amd64/amd64/apic_vector.S
@@ -48,7 +48,7 @@ IDTVEC(vec_name) ;							\
 	movl	$KPSEL,%eax ;						\
 	mov	%ax,%fs ;						\
 	FAKE_MCOUNT(13*4(%esp)) ;					\
-	incb	_intr_nesting_level ;					\
+	incb	PCPU(INTR_NESTING_LEVEL) ;				\
 	pushl	_intr_unit + (irq_num) * 4 ;				\
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	addl	$4, %esp ;						\
@@ -136,7 +136,7 @@ log_intr_event:
 	addl	$4, %esp
 	movl	CNAME(apic_itrace_debugbuffer_idx), %ecx
 	andl	$32767, %ecx
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	shll	$8,	%eax
 	orl	8(%esp), %eax
 	movw	%ax,	CNAME(apic_itrace_debugbuffer)(,%ecx,2)
@@ -218,7 +218,7 @@ IDTVEC(vec_name) ;							\
 	MASK_LEVEL_IRQ(irq_num) ;					\
 	EOI_IRQ(irq_num) ;						\
 0: ;									\
-	incb	_intr_nesting_level ;					\
+	incb	PCPU(INTR_NESTING_LEVEL) ;				\
 ;	 								\
   /* entry point used by doreti_unpend for HWIs. */			\
 __CONCAT(Xresume,irq_num): ;						\
@@ -263,7 +263,7 @@ _Xinvltlb:
 	pushl	%fs
 	movl	$KPSEL, %eax
 	mov	%ax, %fs
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	popl	%fs
 	ss
 	incl	_xhits(,%eax,4)
@@ -321,10 +321,11 @@ _Xcpucheckstate:
 	jne	1f
 	incl	%ebx			/* system or interrupt */
 1:	
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	movl	%ebx, _checkstate_cpustate(,%eax,4)
-	movl	_curproc, %ebx
+	movl	PCPU(CURPROC), %ebx
 	movl	%ebx, _checkstate_curproc(,%eax,4)
+
 	movl	16(%esp), %ebx
 	movl	%ebx, _checkstate_pc(,%eax,4)
 
@@ -358,7 +359,7 @@ _Xcpuast:
 	movl	$KPSEL, %eax
 	mov	%ax, %fs
 
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	lock				/* checkstate_need_ast &= ~(1<<id) */
 	btrl	%eax, _checkstate_need_ast
 	movl	$0, lapic_eoi		/* End Of Interrupt to APIC */
@@ -369,17 +370,17 @@ _Xcpuast:
 
 	FAKE_MCOUNT(13*4(%esp))
 
-	orl	$AST_PENDING, _astpending	/* XXX */
-	incb	_intr_nesting_level
+	orl	$AST_PENDING, PCPU(ASTPENDING)	/* XXX */
+	incb	PCPU(INTR_NESTING_LEVEL)
 	sti
 	
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	lock	
 	btrl	%eax, _checkstate_pending_ast
 	lock	
 	btrl	%eax, CNAME(resched_cpus)
 	jnc	2f
-	orl	$AST_PENDING+AST_RESCHED, _astpending
+	orl	$AST_PENDING+AST_RESCHED, PCPU(ASTPENDING)
 	lock
 	incl	CNAME(want_resched_cnt)
 2:		
@@ -414,10 +415,10 @@ _Xforward_irq:
 
 	lock
 	incl	CNAME(forward_irq_hitcnt)
-	cmpb	$4, _intr_nesting_level
+	cmpb	$4, PCPU(INTR_NESTING_LEVEL)
 	jae	1f
 	
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	sti
 	
 	MEXITCOUNT
@@ -499,7 +500,7 @@ _Xcpustop:
 
 	movl	$0, lapic_eoi		/* End Of Interrupt to APIC */
 
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	imull	$PCB_SIZE, %eax
 	leal	CNAME(stoppcbs)(%eax), %eax
 	pushl	%eax
@@ -507,7 +508,7 @@ _Xcpustop:
 	addl	$4, %esp
 	
 		
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 
 	lock
 	btsl	%eax, _stopped_cpus	/* stopped_cpus |= (1<<id) */
diff --git a/sys/amd64/amd64/cpu_switch.S b/sys/amd64/amd64/cpu_switch.S
index 2d315fc5f787..c2cfa5e49136 100644
--- a/sys/amd64/amd64/cpu_switch.S
+++ b/sys/amd64/amd64/cpu_switch.S
@@ -82,7 +82,7 @@ ENTRY(cpu_throw)
 ENTRY(cpu_switch)
 	
 	/* switch to new process. first, save context as needed */
-	movl	_curproc,%ecx
+	movl	PCPU(CURPROC),%ecx
 
 	/* if no process to save, don't bother */
 	testl	%ecx,%ecx
@@ -95,7 +95,7 @@ ENTRY(cpu_switch)
 #endif /* SMP */
 	movl	P_VMSPACE(%ecx), %edx
 #ifdef SMP
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 #else
 	xorl	%eax, %eax
 #endif /* SMP */
@@ -142,7 +142,7 @@ ENTRY(cpu_switch)
 
 #if NNPX > 0
 	/* have we used fp, and need a save? */
-	cmpl	%ecx,_npxproc
+	cmpl	%ecx,PCPU(NPXPROC)
 	jne	1f
 	addl	$PCB_SAVEFPU,%edx		/* h/w bugs make saving complicated */
 	pushl	%edx
@@ -158,10 +158,10 @@ sw1:
 	/* Stop scheduling if smp_active goes zero and we are not BSP */
 	cmpl	$0,_smp_active
 	jne	1f
-	cmpl	$0,_cpuid
+	cmpl	$0,PCPU(CPUID)
 	je	1f
 
-	movl	_idleproc, %eax
+	movl	PCPU(IDLEPROC), %eax
 	jmp	sw1b
 1:
 #endif
@@ -181,7 +181,7 @@ sw1b:
 	movl	%eax,%ecx
 
 	xorl	%eax,%eax
-	andl	$~AST_RESCHED,_astpending
+	andl	$~AST_RESCHED,PCPU(ASTPENDING)
 
 #ifdef	INVARIANTS
 	cmpb	$SRUN,P_STAT(%ecx)
@@ -206,7 +206,7 @@ sw1b:
 4:
 
 #ifdef SMP
-	movl	_cpuid, %esi
+	movl	PCPU(CPUID), %esi
 #else
 	xorl	%esi, %esi
 #endif
@@ -220,19 +220,19 @@ sw1b:
 	/* update common_tss.tss_esp0 pointer */
 	movl	%edx, %ebx			/* pcb */
 	addl	$(UPAGES * PAGE_SIZE - 16), %ebx
-	movl	%ebx, _common_tss + TSS_ESP0
+	movl	%ebx, PCPU(COMMON_TSS) + TSS_ESP0
 
 	btrl	%esi, _private_tss
 	jae	3f
 #ifdef SMP
-	movl	$gd_common_tssd, %edi
+	movl	$GD_COMMON_TSSD, %edi
 	addl	%fs:0, %edi
 #else
-	movl	$_common_tssd, %edi
+	movl	$PCPU(COMMON_TSSD), %edi
 #endif
 2:
 	/* move correct tss descriptor into GDT slot, then reload tr */
-	movl	_tss_gdt, %ebx			/* entry in GDT */
+	movl	PCPU(TSS_GDT), %ebx		/* entry in GDT */
 	movl	0(%edi), %eax
 	movl	%eax, 0(%ebx)
 	movl	4(%edi), %eax
@@ -242,7 +242,7 @@ sw1b:
 3:
 	movl	P_VMSPACE(%ecx), %ebx
 #ifdef SMP
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 #else
 	xorl	%eax, %eax
 #endif
@@ -265,11 +265,11 @@ sw1b:
 	andl	$~APIC_TPR_PRIO, lapic_tpr
 #endif /** CHEAP_TPR */
 #endif /** GRAB_LOPRIO */
-	movl	_cpuid,%eax
+	movl	PCPU(CPUID),%eax
 	movb	%al, P_ONCPU(%ecx)
 #endif /* SMP */
-	movl	%edx, _curpcb
-	movl	%ecx, _curproc			/* into next process */
+	movl	%edx, PCPU(CURPCB)
+	movl	%ecx, PCPU(CURPROC)		/* into next process */
 
 #ifdef SMP
 	/* XXX FIXME: we should be restoring the local APIC TPR */
@@ -279,10 +279,10 @@ sw1b:
 	cmpl	$0, PCB_USERLDT(%edx)
 	jnz	1f
 	movl	__default_ldt,%eax
-	cmpl	_currentldt,%eax
+	cmpl	PCPU(CURRENTLDT),%eax
 	je	2f
 	lldt	__default_ldt
-	movl	%eax,_currentldt
+	movl	%eax,PCPU(CURRENTLDT)
 	jmp	2f
 1:	pushl	%edx
 	call	_set_user_ldt
@@ -320,7 +320,7 @@ cpu_switch_load_gs:
 	movl	PCB_SCHEDNEST(%edx),%eax
 	movl	%eax,_sched_lock+MTX_RECURSE
 
-	movl	_curproc,%eax
+	movl	PCPU(CURPROC),%eax
 	movl	%eax,_sched_lock+MTX_LOCK
 
 	ret
@@ -376,7 +376,7 @@ ENTRY(savectx)
 	 * have to handle h/w bugs for reloading.  We used to lose the
 	 * parent's npx state for forks by forgetting to reload.
 	 */
-	movl	_npxproc,%eax
+	movl	PCPU(NPXPROC),%eax
 	testl	%eax,%eax
 	je	1f
 
diff --git a/sys/amd64/amd64/exception.S b/sys/amd64/amd64/exception.S
index 99584a6d6cf7..6ba916f5f58b 100644
--- a/sys/amd64/amd64/exception.S
+++ b/sys/amd64/amd64/exception.S
@@ -184,7 +184,7 @@ IDTVEC(fpu)
 	call	__mtx_exit_giant_def
 
 	addl	$4,%esp
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 #else	/* NNPX > 0 */
@@ -223,7 +223,7 @@ calltrap:
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -262,9 +262,9 @@ IDTVEC(syscall)
 	call	_syscall2
 	MEXITCOUNT
 	cli				/* atomic astpending access */
-	cmpl    $0,_astpending		/* AST pending? */
+	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 /*
@@ -294,9 +294,9 @@ IDTVEC(int0x80_syscall)
 	call	_syscall2
 	MEXITCOUNT
 	cli				/* atomic astpending access */
-	cmpl    $0,_astpending		/* AST pending? */
+	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 ENTRY(fork_trampoline)
@@ -306,15 +306,15 @@ ENTRY(fork_trampoline)
 					   are enabled */
 
 #ifdef SMP
-	cmpl	$0,_switchtime
+	cmpl	$0,PCPU(SWITCHTIME)
 	jne	1f
-	movl	$gd_switchtime,%eax
+	movl	$GD_SWITCHTIME,%eax
 	addl	%fs:0,%eax
 	pushl	%eax
 	call	_microuptime
 	popl	%edx
 	movl	_ticks,%eax
-	movl	%eax,_switchticks
+	movl	%eax,PCPU(SWITCHTICKS)
 1:
 #endif
 
@@ -331,7 +331,7 @@ ENTRY(fork_trampoline)
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/amd64/amd64/exception.s b/sys/amd64/amd64/exception.s
index 99584a6d6cf7..6ba916f5f58b 100644
--- a/sys/amd64/amd64/exception.s
+++ b/sys/amd64/amd64/exception.s
@@ -184,7 +184,7 @@ IDTVEC(fpu)
 	call	__mtx_exit_giant_def
 
 	addl	$4,%esp
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 #else	/* NNPX > 0 */
@@ -223,7 +223,7 @@ calltrap:
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -262,9 +262,9 @@ IDTVEC(syscall)
 	call	_syscall2
 	MEXITCOUNT
 	cli				/* atomic astpending access */
-	cmpl    $0,_astpending		/* AST pending? */
+	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 /*
@@ -294,9 +294,9 @@ IDTVEC(int0x80_syscall)
 	call	_syscall2
 	MEXITCOUNT
 	cli				/* atomic astpending access */
-	cmpl    $0,_astpending		/* AST pending? */
+	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 ENTRY(fork_trampoline)
@@ -306,15 +306,15 @@ ENTRY(fork_trampoline)
 					   are enabled */
 
 #ifdef SMP
-	cmpl	$0,_switchtime
+	cmpl	$0,PCPU(SWITCHTIME)
 	jne	1f
-	movl	$gd_switchtime,%eax
+	movl	$GD_SWITCHTIME,%eax
 	addl	%fs:0,%eax
 	pushl	%eax
 	call	_microuptime
 	popl	%edx
 	movl	_ticks,%eax
-	movl	%eax,_switchticks
+	movl	%eax,PCPU(SWITCHTICKS)
 1:
 #endif
 
@@ -331,7 +331,7 @@ ENTRY(fork_trampoline)
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/amd64/amd64/support.S b/sys/amd64/amd64/support.S
index cb3a14404acb..e240de4b324c 100644
--- a/sys/amd64/amd64/support.S
+++ b/sys/amd64/amd64/support.S
@@ -241,7 +241,7 @@ ENTRY(i586_bzero)
 	 * method.  CR0_TS must be preserved although it is very likely to
 	 * always end up as clear.
 	 */
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bz1
 	cmpl	$256+184,%ecx		/* empirical; not quite 2*108 more */
 	jb	intreg_i586_bzero
@@ -293,7 +293,7 @@ fpureg_i586_bzero_loop:
 	cmpl	$8,%ecx
 	jae	fpureg_i586_bzero_loop
 
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bz3
 	frstor	0(%esp)
 	addl	$108,%esp
@@ -501,7 +501,7 @@ ENTRY(i586_bcopy)
 
 	sarb	$1,kernel_fpu_lock
 	jc	small_i586_bcopy
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bc1
 	smsw	%dx
 	clts
@@ -572,7 +572,7 @@ large_i586_bcopy_loop:
 	cmpl	$64,%ecx
 	jae	4b
 
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bc2
 	frstor	0(%esp)
 	addl	$108,%esp
@@ -670,7 +670,7 @@ ENTRY(copyout)
 	jmp	*_copyout_vector
 
 ENTRY(generic_copyout)
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyout_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -781,7 +781,7 @@ done_copyout:
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	%eax,PCB_ONFAULT(%edx)
 	ret
 
@@ -790,7 +790,7 @@ copyout_fault:
 	popl	%ebx
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -800,7 +800,7 @@ ENTRY(i586_copyout)
 	/*
 	 * Duplicated from generic_copyout.  Could be done a bit better.
 	 */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyout_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -857,7 +857,7 @@ ENTRY(copyin)
 	jmp	*_copyin_vector
 
 ENTRY(generic_copyin)
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyin_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -895,7 +895,7 @@ done_copyin:
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	%eax,PCB_ONFAULT(%edx)
 	ret
 
@@ -903,7 +903,7 @@ done_copyin:
 copyin_fault:
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -913,7 +913,7 @@ ENTRY(i586_copyin)
 	/*
 	 * Duplicated from generic_copyin.  Could be done a bit better.
 	 */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyin_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -967,13 +967,13 @@ ENTRY(fastmove)
 	jnz	fastmove_tail
 
 /* if (npxproc != NULL) { */
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	6f
 /*    fnsave(&curpcb->pcb_savefpu); */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	fnsave	PCB_SAVEFPU(%eax)
 /*   npxproc = NULL; */
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 /* } */
 6:
 /* now we own the FPU. */
@@ -990,7 +990,7 @@ ENTRY(fastmove)
 	movl	%esi,-8(%ebp)
 	movl	%edi,-4(%ebp)
 	movl	%esp,%edi
-	movl	_curpcb,%esi
+	movl	PCPU(CURPCB),%esi
 	addl	$PCB_SAVEFPU,%esi
 	cld
 	movl	$PCB_SAVEFPU_SIZE>>2,%ecx
@@ -1002,9 +1002,9 @@ ENTRY(fastmove)
 /* stop_emulating(); */
 	clts
 /* npxproc = curproc; */
-	movl	_curproc,%eax
-	movl	%eax,_npxproc
-	movl	_curpcb,%eax
+	movl	PCPU(CURPROC),%eax
+	movl	%eax,PCPU(NPXPROC)
+	movl	PCPU(CURPCB),%eax
 	movl	$fastmove_fault,PCB_ONFAULT(%eax)
 4:
 	movl	%ecx,-12(%ebp)
@@ -1066,7 +1066,7 @@ fastmove_loop:
 	movl	%ecx,-12(%ebp)
 	movl	%esi,-8(%ebp)
 	movl	%edi,-4(%ebp)
-	movl	_curpcb,%edi
+	movl	PCPU(CURPCB),%edi
 	addl	$PCB_SAVEFPU,%edi
 	movl	%esp,%esi
 	cld
@@ -1082,11 +1082,11 @@ fastmove_loop:
 	orb	$CR0_TS,%al
 	lmsw	%ax
 /* npxproc = NULL; */
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 
 	ALIGN_TEXT
 fastmove_tail:
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$fastmove_tail_fault,PCB_ONFAULT(%eax)
 
 	movb	%cl,%al
@@ -1105,7 +1105,7 @@ fastmove_tail:
 
 	ALIGN_TEXT
 fastmove_fault:
-	movl	_curpcb,%edi
+	movl	PCPU(CURPCB),%edi
 	addl	$PCB_SAVEFPU,%edi
 	movl	%esp,%esi
 	cld
@@ -1116,7 +1116,7 @@ fastmove_fault:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 
 fastmove_tail_fault:
 	movl	%ebp,%esp
@@ -1125,7 +1125,7 @@ fastmove_tail_fault:
 	popl	%ebx
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -1137,7 +1137,7 @@ fastmove_tail_fault:
  *	Fetch a byte (sword, word) from user memory
  */
 ENTRY(fuword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx			/* from */
 
@@ -1163,7 +1163,7 @@ ENTRY(fuswintr)
  * fusword - MP SAFE
  */
 ENTRY(fusword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1178,7 +1178,7 @@ ENTRY(fusword)
  * fubyte - MP SAFE
  */
 ENTRY(fubyte)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1191,7 +1191,7 @@ ENTRY(fubyte)
 
 	ALIGN_TEXT
 fusufault:
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	xorl	%eax,%eax
 	movl	%eax,PCB_ONFAULT(%ecx)
 	decl	%eax
@@ -1203,7 +1203,7 @@ fusufault:
  *	Write a byte (word, longword) to user memory
  */
 ENTRY(suword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1247,7 +1247,7 @@ ENTRY(suword)
 	movl	8(%esp),%eax
 	movl	%eax,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1255,7 +1255,7 @@ ENTRY(suword)
  * susword - MP SAFE (if not I386_CPU)
  */
 ENTRY(susword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1299,7 +1299,7 @@ ENTRY(susword)
 	movw	8(%esp),%ax
 	movw	%ax,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx			/* restore trashed register */
+	movl	PCPU(CURPCB),%ecx		/* restore trashed register */
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1308,7 +1308,7 @@ ENTRY(susword)
  */
 ALTENTRY(suibyte)
 ENTRY(subyte)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1351,7 +1351,7 @@ ENTRY(subyte)
 	movb	8(%esp),%al
 	movb	%al,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx			/* restore trashed register */
+	movl	PCPU(CURPCB),%ecx		/* restore trashed register */
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1366,7 +1366,7 @@ ENTRY(subyte)
 ENTRY(copyinstr)
 	pushl	%esi
 	pushl	%edi
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$cpystrflt,PCB_ONFAULT(%ecx)
 
 	movl	12(%esp),%esi			/* %esi = from */
@@ -1414,7 +1414,7 @@ cpystrflt:
 
 cpystrflt_x:
 	/* set *lencopied and return %eax */
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$0,PCB_ONFAULT(%ecx)
 	movl	20(%esp),%ecx
 	subl	%edx,%ecx
diff --git a/sys/amd64/amd64/support.s b/sys/amd64/amd64/support.s
index cb3a14404acb..e240de4b324c 100644
--- a/sys/amd64/amd64/support.s
+++ b/sys/amd64/amd64/support.s
@@ -241,7 +241,7 @@ ENTRY(i586_bzero)
 	 * method.  CR0_TS must be preserved although it is very likely to
 	 * always end up as clear.
 	 */
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bz1
 	cmpl	$256+184,%ecx		/* empirical; not quite 2*108 more */
 	jb	intreg_i586_bzero
@@ -293,7 +293,7 @@ fpureg_i586_bzero_loop:
 	cmpl	$8,%ecx
 	jae	fpureg_i586_bzero_loop
 
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bz3
 	frstor	0(%esp)
 	addl	$108,%esp
@@ -501,7 +501,7 @@ ENTRY(i586_bcopy)
 
 	sarb	$1,kernel_fpu_lock
 	jc	small_i586_bcopy
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bc1
 	smsw	%dx
 	clts
@@ -572,7 +572,7 @@ large_i586_bcopy_loop:
 	cmpl	$64,%ecx
 	jae	4b
 
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bc2
 	frstor	0(%esp)
 	addl	$108,%esp
@@ -670,7 +670,7 @@ ENTRY(copyout)
 	jmp	*_copyout_vector
 
 ENTRY(generic_copyout)
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyout_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -781,7 +781,7 @@ done_copyout:
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	%eax,PCB_ONFAULT(%edx)
 	ret
 
@@ -790,7 +790,7 @@ copyout_fault:
 	popl	%ebx
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -800,7 +800,7 @@ ENTRY(i586_copyout)
 	/*
 	 * Duplicated from generic_copyout.  Could be done a bit better.
 	 */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyout_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -857,7 +857,7 @@ ENTRY(copyin)
 	jmp	*_copyin_vector
 
 ENTRY(generic_copyin)
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyin_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -895,7 +895,7 @@ done_copyin:
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	%eax,PCB_ONFAULT(%edx)
 	ret
 
@@ -903,7 +903,7 @@ done_copyin:
 copyin_fault:
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -913,7 +913,7 @@ ENTRY(i586_copyin)
 	/*
 	 * Duplicated from generic_copyin.  Could be done a bit better.
 	 */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyin_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -967,13 +967,13 @@ ENTRY(fastmove)
 	jnz	fastmove_tail
 
 /* if (npxproc != NULL) { */
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	6f
 /*    fnsave(&curpcb->pcb_savefpu); */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	fnsave	PCB_SAVEFPU(%eax)
 /*   npxproc = NULL; */
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 /* } */
 6:
 /* now we own the FPU. */
@@ -990,7 +990,7 @@ ENTRY(fastmove)
 	movl	%esi,-8(%ebp)
 	movl	%edi,-4(%ebp)
 	movl	%esp,%edi
-	movl	_curpcb,%esi
+	movl	PCPU(CURPCB),%esi
 	addl	$PCB_SAVEFPU,%esi
 	cld
 	movl	$PCB_SAVEFPU_SIZE>>2,%ecx
@@ -1002,9 +1002,9 @@ ENTRY(fastmove)
 /* stop_emulating(); */
 	clts
 /* npxproc = curproc; */
-	movl	_curproc,%eax
-	movl	%eax,_npxproc
-	movl	_curpcb,%eax
+	movl	PCPU(CURPROC),%eax
+	movl	%eax,PCPU(NPXPROC)
+	movl	PCPU(CURPCB),%eax
 	movl	$fastmove_fault,PCB_ONFAULT(%eax)
 4:
 	movl	%ecx,-12(%ebp)
@@ -1066,7 +1066,7 @@ fastmove_loop:
 	movl	%ecx,-12(%ebp)
 	movl	%esi,-8(%ebp)
 	movl	%edi,-4(%ebp)
-	movl	_curpcb,%edi
+	movl	PCPU(CURPCB),%edi
 	addl	$PCB_SAVEFPU,%edi
 	movl	%esp,%esi
 	cld
@@ -1082,11 +1082,11 @@ fastmove_loop:
 	orb	$CR0_TS,%al
 	lmsw	%ax
 /* npxproc = NULL; */
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 
 	ALIGN_TEXT
 fastmove_tail:
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$fastmove_tail_fault,PCB_ONFAULT(%eax)
 
 	movb	%cl,%al
@@ -1105,7 +1105,7 @@ fastmove_tail:
 
 	ALIGN_TEXT
 fastmove_fault:
-	movl	_curpcb,%edi
+	movl	PCPU(CURPCB),%edi
 	addl	$PCB_SAVEFPU,%edi
 	movl	%esp,%esi
 	cld
@@ -1116,7 +1116,7 @@ fastmove_fault:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 
 fastmove_tail_fault:
 	movl	%ebp,%esp
@@ -1125,7 +1125,7 @@ fastmove_tail_fault:
 	popl	%ebx
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -1137,7 +1137,7 @@ fastmove_tail_fault:
  *	Fetch a byte (sword, word) from user memory
  */
 ENTRY(fuword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx			/* from */
 
@@ -1163,7 +1163,7 @@ ENTRY(fuswintr)
  * fusword - MP SAFE
  */
 ENTRY(fusword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1178,7 +1178,7 @@ ENTRY(fusword)
  * fubyte - MP SAFE
  */
 ENTRY(fubyte)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1191,7 +1191,7 @@ ENTRY(fubyte)
 
 	ALIGN_TEXT
 fusufault:
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	xorl	%eax,%eax
 	movl	%eax,PCB_ONFAULT(%ecx)
 	decl	%eax
@@ -1203,7 +1203,7 @@ fusufault:
  *	Write a byte (word, longword) to user memory
  */
 ENTRY(suword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1247,7 +1247,7 @@ ENTRY(suword)
 	movl	8(%esp),%eax
 	movl	%eax,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1255,7 +1255,7 @@ ENTRY(suword)
  * susword - MP SAFE (if not I386_CPU)
  */
 ENTRY(susword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1299,7 +1299,7 @@ ENTRY(susword)
 	movw	8(%esp),%ax
 	movw	%ax,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx			/* restore trashed register */
+	movl	PCPU(CURPCB),%ecx		/* restore trashed register */
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1308,7 +1308,7 @@ ENTRY(susword)
  */
 ALTENTRY(suibyte)
 ENTRY(subyte)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1351,7 +1351,7 @@ ENTRY(subyte)
 	movb	8(%esp),%al
 	movb	%al,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx			/* restore trashed register */
+	movl	PCPU(CURPCB),%ecx		/* restore trashed register */
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1366,7 +1366,7 @@ ENTRY(subyte)
 ENTRY(copyinstr)
 	pushl	%esi
 	pushl	%edi
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$cpystrflt,PCB_ONFAULT(%ecx)
 
 	movl	12(%esp),%esi			/* %esi = from */
@@ -1414,7 +1414,7 @@ cpystrflt:
 
 cpystrflt_x:
 	/* set *lencopied and return %eax */
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$0,PCB_ONFAULT(%ecx)
 	movl	20(%esp),%ecx
 	subl	%edx,%ecx
diff --git a/sys/amd64/amd64/swtch.s b/sys/amd64/amd64/swtch.s
index 2d315fc5f787..c2cfa5e49136 100644
--- a/sys/amd64/amd64/swtch.s
+++ b/sys/amd64/amd64/swtch.s
@@ -82,7 +82,7 @@ ENTRY(cpu_throw)
 ENTRY(cpu_switch)
 	
 	/* switch to new process. first, save context as needed */
-	movl	_curproc,%ecx
+	movl	PCPU(CURPROC),%ecx
 
 	/* if no process to save, don't bother */
 	testl	%ecx,%ecx
@@ -95,7 +95,7 @@ ENTRY(cpu_switch)
 #endif /* SMP */
 	movl	P_VMSPACE(%ecx), %edx
 #ifdef SMP
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 #else
 	xorl	%eax, %eax
 #endif /* SMP */
@@ -142,7 +142,7 @@ ENTRY(cpu_switch)
 
 #if NNPX > 0
 	/* have we used fp, and need a save? */
-	cmpl	%ecx,_npxproc
+	cmpl	%ecx,PCPU(NPXPROC)
 	jne	1f
 	addl	$PCB_SAVEFPU,%edx		/* h/w bugs make saving complicated */
 	pushl	%edx
@@ -158,10 +158,10 @@ sw1:
 	/* Stop scheduling if smp_active goes zero and we are not BSP */
 	cmpl	$0,_smp_active
 	jne	1f
-	cmpl	$0,_cpuid
+	cmpl	$0,PCPU(CPUID)
 	je	1f
 
-	movl	_idleproc, %eax
+	movl	PCPU(IDLEPROC), %eax
 	jmp	sw1b
 1:
 #endif
@@ -181,7 +181,7 @@ sw1b:
 	movl	%eax,%ecx
 
 	xorl	%eax,%eax
-	andl	$~AST_RESCHED,_astpending
+	andl	$~AST_RESCHED,PCPU(ASTPENDING)
 
 #ifdef	INVARIANTS
 	cmpb	$SRUN,P_STAT(%ecx)
@@ -206,7 +206,7 @@ sw1b:
 4:
 
 #ifdef SMP
-	movl	_cpuid, %esi
+	movl	PCPU(CPUID), %esi
 #else
 	xorl	%esi, %esi
 #endif
@@ -220,19 +220,19 @@ sw1b:
 	/* update common_tss.tss_esp0 pointer */
 	movl	%edx, %ebx			/* pcb */
 	addl	$(UPAGES * PAGE_SIZE - 16), %ebx
-	movl	%ebx, _common_tss + TSS_ESP0
+	movl	%ebx, PCPU(COMMON_TSS) + TSS_ESP0
 
 	btrl	%esi, _private_tss
 	jae	3f
 #ifdef SMP
-	movl	$gd_common_tssd, %edi
+	movl	$GD_COMMON_TSSD, %edi
 	addl	%fs:0, %edi
 #else
-	movl	$_common_tssd, %edi
+	movl	$PCPU(COMMON_TSSD), %edi
 #endif
 2:
 	/* move correct tss descriptor into GDT slot, then reload tr */
-	movl	_tss_gdt, %ebx			/* entry in GDT */
+	movl	PCPU(TSS_GDT), %ebx		/* entry in GDT */
 	movl	0(%edi), %eax
 	movl	%eax, 0(%ebx)
 	movl	4(%edi), %eax
@@ -242,7 +242,7 @@ sw1b:
 3:
 	movl	P_VMSPACE(%ecx), %ebx
 #ifdef SMP
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 #else
 	xorl	%eax, %eax
 #endif
@@ -265,11 +265,11 @@ sw1b:
 	andl	$~APIC_TPR_PRIO, lapic_tpr
 #endif /** CHEAP_TPR */
 #endif /** GRAB_LOPRIO */
-	movl	_cpuid,%eax
+	movl	PCPU(CPUID),%eax
 	movb	%al, P_ONCPU(%ecx)
 #endif /* SMP */
-	movl	%edx, _curpcb
-	movl	%ecx, _curproc			/* into next process */
+	movl	%edx, PCPU(CURPCB)
+	movl	%ecx, PCPU(CURPROC)		/* into next process */
 
 #ifdef SMP
 	/* XXX FIXME: we should be restoring the local APIC TPR */
@@ -279,10 +279,10 @@ sw1b:
 	cmpl	$0, PCB_USERLDT(%edx)
 	jnz	1f
 	movl	__default_ldt,%eax
-	cmpl	_currentldt,%eax
+	cmpl	PCPU(CURRENTLDT),%eax
 	je	2f
 	lldt	__default_ldt
-	movl	%eax,_currentldt
+	movl	%eax,PCPU(CURRENTLDT)
 	jmp	2f
 1:	pushl	%edx
 	call	_set_user_ldt
@@ -320,7 +320,7 @@ cpu_switch_load_gs:
 	movl	PCB_SCHEDNEST(%edx),%eax
 	movl	%eax,_sched_lock+MTX_RECURSE
 
-	movl	_curproc,%eax
+	movl	PCPU(CURPROC),%eax
 	movl	%eax,_sched_lock+MTX_LOCK
 
 	ret
@@ -376,7 +376,7 @@ ENTRY(savectx)
 	 * have to handle h/w bugs for reloading.  We used to lose the
 	 * parent's npx state for forks by forgetting to reload.
 	 */
-	movl	_npxproc,%eax
+	movl	PCPU(NPXPROC),%eax
 	testl	%eax,%eax
 	je	1f
 
diff --git a/sys/amd64/include/asmacros.h b/sys/amd64/include/asmacros.h
index 92b9e1a5d700..c2d89b83c446 100644
--- a/sys/amd64/include/asmacros.h
+++ b/sys/amd64/include/asmacros.h
@@ -69,6 +69,12 @@
 #define NON_GPROF_ENTRY(name)	GEN_ENTRY(name)
 #define NON_GPROF_RET		.byte 0xc3	/* opcode for `ret' */
 
+#ifdef SMP
+#define	PCPU(member)	%fs:GD_ ## member
+#else
+#define	PCPU(member)	CNAME(globaldata) + GD_ ## member
+#endif
+
 #ifdef GPROF
 /*
  * __mcount is like [.]mcount except that doesn't require its caller to set
diff --git a/sys/amd64/isa/atpic_vector.S b/sys/amd64/isa/atpic_vector.S
index f8104016619f..e28c37598678 100644
--- a/sys/amd64/isa/atpic_vector.S
+++ b/sys/amd64/isa/atpic_vector.S
@@ -60,7 +60,7 @@ IDTVEC(vec_name) ; \
 	mov	%ax,%es ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -100,7 +100,7 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
diff --git a/sys/amd64/isa/icu_vector.S b/sys/amd64/isa/icu_vector.S
index f8104016619f..e28c37598678 100644
--- a/sys/amd64/isa/icu_vector.S
+++ b/sys/amd64/isa/icu_vector.S
@@ -60,7 +60,7 @@ IDTVEC(vec_name) ; \
 	mov	%ax,%es ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -100,7 +100,7 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
diff --git a/sys/amd64/isa/icu_vector.s b/sys/amd64/isa/icu_vector.s
index f8104016619f..e28c37598678 100644
--- a/sys/amd64/isa/icu_vector.s
+++ b/sys/amd64/isa/icu_vector.s
@@ -60,7 +60,7 @@ IDTVEC(vec_name) ; \
 	mov	%ax,%es ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -100,7 +100,7 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
diff --git a/sys/i386/i386/apic_vector.s b/sys/i386/i386/apic_vector.s
index 995cf213fb25..a23ccff7b67d 100644
--- a/sys/i386/i386/apic_vector.s
+++ b/sys/i386/i386/apic_vector.s
@@ -48,7 +48,7 @@ IDTVEC(vec_name) ;							\
 	movl	$KPSEL,%eax ;						\
 	mov	%ax,%fs ;						\
 	FAKE_MCOUNT(13*4(%esp)) ;					\
-	incb	_intr_nesting_level ;					\
+	incb	PCPU(INTR_NESTING_LEVEL) ;				\
 	pushl	_intr_unit + (irq_num) * 4 ;				\
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	addl	$4, %esp ;						\
@@ -136,7 +136,7 @@ log_intr_event:
 	addl	$4, %esp
 	movl	CNAME(apic_itrace_debugbuffer_idx), %ecx
 	andl	$32767, %ecx
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	shll	$8,	%eax
 	orl	8(%esp), %eax
 	movw	%ax,	CNAME(apic_itrace_debugbuffer)(,%ecx,2)
@@ -218,7 +218,7 @@ IDTVEC(vec_name) ;							\
 	MASK_LEVEL_IRQ(irq_num) ;					\
 	EOI_IRQ(irq_num) ;						\
 0: ;									\
-	incb	_intr_nesting_level ;					\
+	incb	PCPU(INTR_NESTING_LEVEL) ;				\
 ;	 								\
   /* entry point used by doreti_unpend for HWIs. */			\
 __CONCAT(Xresume,irq_num): ;						\
@@ -263,7 +263,7 @@ _Xinvltlb:
 	pushl	%fs
 	movl	$KPSEL, %eax
 	mov	%ax, %fs
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	popl	%fs
 	ss
 	incl	_xhits(,%eax,4)
@@ -321,10 +321,11 @@ _Xcpucheckstate:
 	jne	1f
 	incl	%ebx			/* system or interrupt */
 1:	
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	movl	%ebx, _checkstate_cpustate(,%eax,4)
-	movl	_curproc, %ebx
+	movl	PCPU(CURPROC), %ebx
 	movl	%ebx, _checkstate_curproc(,%eax,4)
+
 	movl	16(%esp), %ebx
 	movl	%ebx, _checkstate_pc(,%eax,4)
 
@@ -358,7 +359,7 @@ _Xcpuast:
 	movl	$KPSEL, %eax
 	mov	%ax, %fs
 
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	lock				/* checkstate_need_ast &= ~(1<<id) */
 	btrl	%eax, _checkstate_need_ast
 	movl	$0, lapic_eoi		/* End Of Interrupt to APIC */
@@ -369,17 +370,17 @@ _Xcpuast:
 
 	FAKE_MCOUNT(13*4(%esp))
 
-	orl	$AST_PENDING, _astpending	/* XXX */
-	incb	_intr_nesting_level
+	orl	$AST_PENDING, PCPU(ASTPENDING)	/* XXX */
+	incb	PCPU(INTR_NESTING_LEVEL)
 	sti
 	
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	lock	
 	btrl	%eax, _checkstate_pending_ast
 	lock	
 	btrl	%eax, CNAME(resched_cpus)
 	jnc	2f
-	orl	$AST_PENDING+AST_RESCHED, _astpending
+	orl	$AST_PENDING+AST_RESCHED, PCPU(ASTPENDING)
 	lock
 	incl	CNAME(want_resched_cnt)
 2:		
@@ -414,10 +415,10 @@ _Xforward_irq:
 
 	lock
 	incl	CNAME(forward_irq_hitcnt)
-	cmpb	$4, _intr_nesting_level
+	cmpb	$4, PCPU(INTR_NESTING_LEVEL)
 	jae	1f
 	
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	sti
 	
 	MEXITCOUNT
@@ -499,7 +500,7 @@ _Xcpustop:
 
 	movl	$0, lapic_eoi		/* End Of Interrupt to APIC */
 
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	imull	$PCB_SIZE, %eax
 	leal	CNAME(stoppcbs)(%eax), %eax
 	pushl	%eax
@@ -507,7 +508,7 @@ _Xcpustop:
 	addl	$4, %esp
 	
 		
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 
 	lock
 	btsl	%eax, _stopped_cpus	/* stopped_cpus |= (1<<id) */
diff --git a/sys/i386/i386/exception.s b/sys/i386/i386/exception.s
index 99584a6d6cf7..6ba916f5f58b 100644
--- a/sys/i386/i386/exception.s
+++ b/sys/i386/i386/exception.s
@@ -184,7 +184,7 @@ IDTVEC(fpu)
 	call	__mtx_exit_giant_def
 
 	addl	$4,%esp
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 #else	/* NNPX > 0 */
@@ -223,7 +223,7 @@ calltrap:
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -262,9 +262,9 @@ IDTVEC(syscall)
 	call	_syscall2
 	MEXITCOUNT
 	cli				/* atomic astpending access */
-	cmpl    $0,_astpending		/* AST pending? */
+	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 /*
@@ -294,9 +294,9 @@ IDTVEC(int0x80_syscall)
 	call	_syscall2
 	MEXITCOUNT
 	cli				/* atomic astpending access */
-	cmpl    $0,_astpending		/* AST pending? */
+	cmpl    $0,PCPU(ASTPENDING)	/* AST pending? */
 	je	doreti_syscall_ret	/* no, get out of here */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	jmp	_doreti
 
 ENTRY(fork_trampoline)
@@ -306,15 +306,15 @@ ENTRY(fork_trampoline)
 					   are enabled */
 
 #ifdef SMP
-	cmpl	$0,_switchtime
+	cmpl	$0,PCPU(SWITCHTIME)
 	jne	1f
-	movl	$gd_switchtime,%eax
+	movl	$GD_SWITCHTIME,%eax
 	addl	%fs:0,%eax
 	pushl	%eax
 	call	_microuptime
 	popl	%edx
 	movl	_ticks,%eax
-	movl	%eax,_switchticks
+	movl	%eax,PCPU(SWITCHTICKS)
 1:
 #endif
 
@@ -331,7 +331,7 @@ ENTRY(fork_trampoline)
 	/*
 	 * Return via _doreti to handle ASTs.
 	 */
-	movb	$1,_intr_nesting_level
+	movb	$1,PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
diff --git a/sys/i386/i386/simplelock.s b/sys/i386/i386/simplelock.s
index 378cf85389e9..ecfb562dd5a2 100644
--- a/sys/i386/i386/simplelock.s
+++ b/sys/i386/i386/simplelock.s
@@ -35,6 +35,8 @@
 	
 #include <machine/smptests.h>			/** FAST_HI */
 
+#include "assym.s"
+
 /*
  * The following impliments the primitives described in i386/i386/param.h
  * necessary for the Lite2 lock manager system.
@@ -105,7 +107,7 @@ gotit:
 ENTRY(s_lock)
 	movl	4(%esp), %edx		/* get the address of the lock */
 setlock:
-	movl	_cpu_lockid, %ecx	/* add cpu id portion */
+	movl	PCPU(CPU_LOCKID), %ecx	/* add cpu id portion */
 	incl	%ecx			/* add lock portion */
 	movl	$0, %eax
 	lock
@@ -113,7 +115,7 @@ setlock:
 	jz	gotit			/* it was clear, return */
 	pushl	%eax			/* save what we xchanged */
 	decl	%eax			/* remove lock portion */
-	cmpl	_cpu_lockid, %eax	/* do we hold it? */
+	cmpl	PCPU(CPU_LOCKID), %eax	/* do we hold it? */
 	je	bad_slock		/* yes, thats not good... */
 	addl	$4, %esp		/* clear the stack */
 wait:
@@ -127,7 +129,7 @@ gotit:
 bad_slock:
 	/* %eax (current lock) is already on the stack */
 	pushl	%edx
-	pushl	_cpuid
+	pushl	PCPU(CPUID)
 	pushl	$bsl1
 	call	_panic
 
@@ -160,7 +162,7 @@ ENTRY(s_lock_try)
 
 ENTRY(s_lock_try)
 	movl	4(%esp), %edx		/* get the address of the lock */
-	movl	_cpu_lockid, %ecx	/* add cpu id portion */
+	movl	PCPU(CPU_LOCKID), %ecx	/* add cpu id portion */
 	incl	%ecx			/* add lock portion */
 
 	xorl	%eax, %eax
@@ -227,7 +229,7 @@ swait:
 	jne	swait			/* still set... */
 	jmp	ssetlock		/* empty again, try once more */
 sgotit:
-	popl	_ss_eflags		/* save the old eflags */
+	popl	PCPU(SS_EFLAGS)		/* save the old eflags */
 	ret
 
 #else /* SL_DEBUG */
@@ -235,7 +237,7 @@ sgotit:
 ENTRY(ss_lock)
 	movl	4(%esp), %edx		/* get the address of the lock */
 ssetlock:
-	movl	_cpu_lockid, %ecx	/* add cpu id portion */
+	movl	PCPU(CPU_LOCKID), %ecx	/* add cpu id portion */
 	incl	%ecx			/* add lock portion */
 	pushfl
 	cli
@@ -245,7 +247,7 @@ ssetlock:
 	jz	sgotit			/* it was clear, return */
 	pushl	%eax			/* save what we xchanged */
 	decl	%eax			/* remove lock portion */
-	cmpl	_cpu_lockid, %eax	/* do we hold it? */
+	cmpl	PCPU(CPU_LOCKID), %eax	/* do we hold it? */
 	je	sbad_slock		/* yes, thats not good... */
 	addl	$4, %esp		/* clear the stack */
 	popfl
@@ -254,7 +256,7 @@ swait:
 	jne	swait			/* still set... */
 	jmp	ssetlock		/* empty again, try once more */
 sgotit:
-	popl	_ss_eflags		/* save the old task priority */
+	popl	PCPU(SS_EFLAGS)		/* save the old task priority */
 sgotit2:
 	ret
 
@@ -262,7 +264,7 @@ sgotit2:
 sbad_slock:
 	/* %eax (current lock) is already on the stack */
 	pushl	%edx
-	pushl	_cpuid
+	pushl	PCPU(CPUID)
 	pushl	$sbsl1
 	call	_panic
 
@@ -276,7 +278,7 @@ sbsl1:	.asciz	"rsslock: cpu: %d, addr: 0x%08x, lock: 0x%08x"
 ENTRY(ss_unlock)
 	movl	4(%esp), %eax		/* get the address of the lock */
 	movl	$0, (%eax)		/* clear the simple lock */
-	testl	$PSL_I, _ss_eflags
+	testl	$PSL_I, PCPU(SS_EFLAGS)
 	jz	ss_unlock2
 	sti
 ss_unlock2:	
diff --git a/sys/i386/i386/support.s b/sys/i386/i386/support.s
index cb3a14404acb..e240de4b324c 100644
--- a/sys/i386/i386/support.s
+++ b/sys/i386/i386/support.s
@@ -241,7 +241,7 @@ ENTRY(i586_bzero)
 	 * method.  CR0_TS must be preserved although it is very likely to
 	 * always end up as clear.
 	 */
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bz1
 	cmpl	$256+184,%ecx		/* empirical; not quite 2*108 more */
 	jb	intreg_i586_bzero
@@ -293,7 +293,7 @@ fpureg_i586_bzero_loop:
 	cmpl	$8,%ecx
 	jae	fpureg_i586_bzero_loop
 
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bz3
 	frstor	0(%esp)
 	addl	$108,%esp
@@ -501,7 +501,7 @@ ENTRY(i586_bcopy)
 
 	sarb	$1,kernel_fpu_lock
 	jc	small_i586_bcopy
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bc1
 	smsw	%dx
 	clts
@@ -572,7 +572,7 @@ large_i586_bcopy_loop:
 	cmpl	$64,%ecx
 	jae	4b
 
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	i586_bc2
 	frstor	0(%esp)
 	addl	$108,%esp
@@ -670,7 +670,7 @@ ENTRY(copyout)
 	jmp	*_copyout_vector
 
 ENTRY(generic_copyout)
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyout_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -781,7 +781,7 @@ done_copyout:
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	%eax,PCB_ONFAULT(%edx)
 	ret
 
@@ -790,7 +790,7 @@ copyout_fault:
 	popl	%ebx
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -800,7 +800,7 @@ ENTRY(i586_copyout)
 	/*
 	 * Duplicated from generic_copyout.  Could be done a bit better.
 	 */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyout_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -857,7 +857,7 @@ ENTRY(copyin)
 	jmp	*_copyin_vector
 
 ENTRY(generic_copyin)
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyin_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -895,7 +895,7 @@ done_copyin:
 	popl	%edi
 	popl	%esi
 	xorl	%eax,%eax
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	%eax,PCB_ONFAULT(%edx)
 	ret
 
@@ -903,7 +903,7 @@ done_copyin:
 copyin_fault:
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -913,7 +913,7 @@ ENTRY(i586_copyin)
 	/*
 	 * Duplicated from generic_copyin.  Could be done a bit better.
 	 */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$copyin_fault,PCB_ONFAULT(%eax)
 	pushl	%esi
 	pushl	%edi
@@ -967,13 +967,13 @@ ENTRY(fastmove)
 	jnz	fastmove_tail
 
 /* if (npxproc != NULL) { */
-	cmpl	$0,_npxproc
+	cmpl	$0,PCPU(NPXPROC)
 	je	6f
 /*    fnsave(&curpcb->pcb_savefpu); */
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	fnsave	PCB_SAVEFPU(%eax)
 /*   npxproc = NULL; */
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 /* } */
 6:
 /* now we own the FPU. */
@@ -990,7 +990,7 @@ ENTRY(fastmove)
 	movl	%esi,-8(%ebp)
 	movl	%edi,-4(%ebp)
 	movl	%esp,%edi
-	movl	_curpcb,%esi
+	movl	PCPU(CURPCB),%esi
 	addl	$PCB_SAVEFPU,%esi
 	cld
 	movl	$PCB_SAVEFPU_SIZE>>2,%ecx
@@ -1002,9 +1002,9 @@ ENTRY(fastmove)
 /* stop_emulating(); */
 	clts
 /* npxproc = curproc; */
-	movl	_curproc,%eax
-	movl	%eax,_npxproc
-	movl	_curpcb,%eax
+	movl	PCPU(CURPROC),%eax
+	movl	%eax,PCPU(NPXPROC)
+	movl	PCPU(CURPCB),%eax
 	movl	$fastmove_fault,PCB_ONFAULT(%eax)
 4:
 	movl	%ecx,-12(%ebp)
@@ -1066,7 +1066,7 @@ fastmove_loop:
 	movl	%ecx,-12(%ebp)
 	movl	%esi,-8(%ebp)
 	movl	%edi,-4(%ebp)
-	movl	_curpcb,%edi
+	movl	PCPU(CURPCB),%edi
 	addl	$PCB_SAVEFPU,%edi
 	movl	%esp,%esi
 	cld
@@ -1082,11 +1082,11 @@ fastmove_loop:
 	orb	$CR0_TS,%al
 	lmsw	%ax
 /* npxproc = NULL; */
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 
 	ALIGN_TEXT
 fastmove_tail:
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	movl	$fastmove_tail_fault,PCB_ONFAULT(%eax)
 
 	movb	%cl,%al
@@ -1105,7 +1105,7 @@ fastmove_tail:
 
 	ALIGN_TEXT
 fastmove_fault:
-	movl	_curpcb,%edi
+	movl	PCPU(CURPCB),%edi
 	addl	$PCB_SAVEFPU,%edi
 	movl	%esp,%esi
 	cld
@@ -1116,7 +1116,7 @@ fastmove_fault:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-	movl	$0,_npxproc
+	movl	$0,PCPU(NPXPROC)
 
 fastmove_tail_fault:
 	movl	%ebp,%esp
@@ -1125,7 +1125,7 @@ fastmove_tail_fault:
 	popl	%ebx
 	popl	%edi
 	popl	%esi
-	movl	_curpcb,%edx
+	movl	PCPU(CURPCB),%edx
 	movl	$0,PCB_ONFAULT(%edx)
 	movl	$EFAULT,%eax
 	ret
@@ -1137,7 +1137,7 @@ fastmove_tail_fault:
  *	Fetch a byte (sword, word) from user memory
  */
 ENTRY(fuword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx			/* from */
 
@@ -1163,7 +1163,7 @@ ENTRY(fuswintr)
  * fusword - MP SAFE
  */
 ENTRY(fusword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1178,7 +1178,7 @@ ENTRY(fusword)
  * fubyte - MP SAFE
  */
 ENTRY(fubyte)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1191,7 +1191,7 @@ ENTRY(fubyte)
 
 	ALIGN_TEXT
 fusufault:
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	xorl	%eax,%eax
 	movl	%eax,PCB_ONFAULT(%ecx)
 	decl	%eax
@@ -1203,7 +1203,7 @@ fusufault:
  *	Write a byte (word, longword) to user memory
  */
 ENTRY(suword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1247,7 +1247,7 @@ ENTRY(suword)
 	movl	8(%esp),%eax
 	movl	%eax,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1255,7 +1255,7 @@ ENTRY(suword)
  * susword - MP SAFE (if not I386_CPU)
  */
 ENTRY(susword)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1299,7 +1299,7 @@ ENTRY(susword)
 	movw	8(%esp),%ax
 	movw	%ax,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx			/* restore trashed register */
+	movl	PCPU(CURPCB),%ecx		/* restore trashed register */
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1308,7 +1308,7 @@ ENTRY(susword)
  */
 ALTENTRY(suibyte)
 ENTRY(subyte)
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$fusufault,PCB_ONFAULT(%ecx)
 	movl	4(%esp),%edx
 
@@ -1351,7 +1351,7 @@ ENTRY(subyte)
 	movb	8(%esp),%al
 	movb	%al,(%edx)
 	xorl	%eax,%eax
-	movl	_curpcb,%ecx			/* restore trashed register */
+	movl	PCPU(CURPCB),%ecx		/* restore trashed register */
 	movl	%eax,PCB_ONFAULT(%ecx)
 	ret
 
@@ -1366,7 +1366,7 @@ ENTRY(subyte)
 ENTRY(copyinstr)
 	pushl	%esi
 	pushl	%edi
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$cpystrflt,PCB_ONFAULT(%ecx)
 
 	movl	12(%esp),%esi			/* %esi = from */
@@ -1414,7 +1414,7 @@ cpystrflt:
 
 cpystrflt_x:
 	/* set *lencopied and return %eax */
-	movl	_curpcb,%ecx
+	movl	PCPU(CURPCB),%ecx
 	movl	$0,PCB_ONFAULT(%ecx)
 	movl	20(%esp),%ecx
 	subl	%edx,%ecx
diff --git a/sys/i386/i386/swtch.s b/sys/i386/i386/swtch.s
index 2d315fc5f787..c2cfa5e49136 100644
--- a/sys/i386/i386/swtch.s
+++ b/sys/i386/i386/swtch.s
@@ -82,7 +82,7 @@ ENTRY(cpu_throw)
 ENTRY(cpu_switch)
 	
 	/* switch to new process. first, save context as needed */
-	movl	_curproc,%ecx
+	movl	PCPU(CURPROC),%ecx
 
 	/* if no process to save, don't bother */
 	testl	%ecx,%ecx
@@ -95,7 +95,7 @@ ENTRY(cpu_switch)
 #endif /* SMP */
 	movl	P_VMSPACE(%ecx), %edx
 #ifdef SMP
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 #else
 	xorl	%eax, %eax
 #endif /* SMP */
@@ -142,7 +142,7 @@ ENTRY(cpu_switch)
 
 #if NNPX > 0
 	/* have we used fp, and need a save? */
-	cmpl	%ecx,_npxproc
+	cmpl	%ecx,PCPU(NPXPROC)
 	jne	1f
 	addl	$PCB_SAVEFPU,%edx		/* h/w bugs make saving complicated */
 	pushl	%edx
@@ -158,10 +158,10 @@ sw1:
 	/* Stop scheduling if smp_active goes zero and we are not BSP */
 	cmpl	$0,_smp_active
 	jne	1f
-	cmpl	$0,_cpuid
+	cmpl	$0,PCPU(CPUID)
 	je	1f
 
-	movl	_idleproc, %eax
+	movl	PCPU(IDLEPROC), %eax
 	jmp	sw1b
 1:
 #endif
@@ -181,7 +181,7 @@ sw1b:
 	movl	%eax,%ecx
 
 	xorl	%eax,%eax
-	andl	$~AST_RESCHED,_astpending
+	andl	$~AST_RESCHED,PCPU(ASTPENDING)
 
 #ifdef	INVARIANTS
 	cmpb	$SRUN,P_STAT(%ecx)
@@ -206,7 +206,7 @@ sw1b:
 4:
 
 #ifdef SMP
-	movl	_cpuid, %esi
+	movl	PCPU(CPUID), %esi
 #else
 	xorl	%esi, %esi
 #endif
@@ -220,19 +220,19 @@ sw1b:
 	/* update common_tss.tss_esp0 pointer */
 	movl	%edx, %ebx			/* pcb */
 	addl	$(UPAGES * PAGE_SIZE - 16), %ebx
-	movl	%ebx, _common_tss + TSS_ESP0
+	movl	%ebx, PCPU(COMMON_TSS) + TSS_ESP0
 
 	btrl	%esi, _private_tss
 	jae	3f
 #ifdef SMP
-	movl	$gd_common_tssd, %edi
+	movl	$GD_COMMON_TSSD, %edi
 	addl	%fs:0, %edi
 #else
-	movl	$_common_tssd, %edi
+	movl	$PCPU(COMMON_TSSD), %edi
 #endif
 2:
 	/* move correct tss descriptor into GDT slot, then reload tr */
-	movl	_tss_gdt, %ebx			/* entry in GDT */
+	movl	PCPU(TSS_GDT), %ebx		/* entry in GDT */
 	movl	0(%edi), %eax
 	movl	%eax, 0(%ebx)
 	movl	4(%edi), %eax
@@ -242,7 +242,7 @@ sw1b:
 3:
 	movl	P_VMSPACE(%ecx), %ebx
 #ifdef SMP
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 #else
 	xorl	%eax, %eax
 #endif
@@ -265,11 +265,11 @@ sw1b:
 	andl	$~APIC_TPR_PRIO, lapic_tpr
 #endif /** CHEAP_TPR */
 #endif /** GRAB_LOPRIO */
-	movl	_cpuid,%eax
+	movl	PCPU(CPUID),%eax
 	movb	%al, P_ONCPU(%ecx)
 #endif /* SMP */
-	movl	%edx, _curpcb
-	movl	%ecx, _curproc			/* into next process */
+	movl	%edx, PCPU(CURPCB)
+	movl	%ecx, PCPU(CURPROC)		/* into next process */
 
 #ifdef SMP
 	/* XXX FIXME: we should be restoring the local APIC TPR */
@@ -279,10 +279,10 @@ sw1b:
 	cmpl	$0, PCB_USERLDT(%edx)
 	jnz	1f
 	movl	__default_ldt,%eax
-	cmpl	_currentldt,%eax
+	cmpl	PCPU(CURRENTLDT),%eax
 	je	2f
 	lldt	__default_ldt
-	movl	%eax,_currentldt
+	movl	%eax,PCPU(CURRENTLDT)
 	jmp	2f
 1:	pushl	%edx
 	call	_set_user_ldt
@@ -320,7 +320,7 @@ cpu_switch_load_gs:
 	movl	PCB_SCHEDNEST(%edx),%eax
 	movl	%eax,_sched_lock+MTX_RECURSE
 
-	movl	_curproc,%eax
+	movl	PCPU(CURPROC),%eax
 	movl	%eax,_sched_lock+MTX_LOCK
 
 	ret
@@ -376,7 +376,7 @@ ENTRY(savectx)
 	 * have to handle h/w bugs for reloading.  We used to lose the
 	 * parent's npx state for forks by forgetting to reload.
 	 */
-	movl	_npxproc,%eax
+	movl	PCPU(NPXPROC),%eax
 	testl	%eax,%eax
 	je	1f
 
diff --git a/sys/i386/i386/vm86bios.s b/sys/i386/i386/vm86bios.s
index 95988ef3310c..be227e7ecc36 100644
--- a/sys/i386/i386/vm86bios.s
+++ b/sys/i386/i386/vm86bios.s
@@ -63,8 +63,8 @@ ENTRY(vm86_bioscall)
 	pushl	%gs
 
 #if NNPX > 0
-	movl	_curproc,%ecx
-	cmpl	%ecx,_npxproc		/* do we need to save fp? */
+	movl	PCPU(CURPROC),%ecx
+	cmpl	%ecx,PCPU(NPXPROC)	/* do we need to save fp? */
 	jne	1f
 	testl	%ecx,%ecx
 	je 	1f			/* no curproc/npxproc */
@@ -86,11 +86,11 @@ ENTRY(vm86_bioscall)
 	rep
 	movsl				/* copy frame to new stack */
 
-	movl	_curpcb,%eax
+	movl	PCPU(CURPCB),%eax
 	pushl	%eax			/* save curpcb */
-	movl	%edx,_curpcb		/* set curpcb to vm86pcb */
+	movl	%edx,PCPU(CURPCB)	/* set curpcb to vm86pcb */
 
-	movl	_tss_gdt,%ebx		/* entry in GDT */
+	movl	PCPU(TSS_GDT),%ebx	/* entry in GDT */
 	movl	0(%ebx),%eax
 	movl	%eax,SCR_TSS0(%edx)	/* save first word */
 	movl	4(%ebx),%eax
@@ -129,7 +129,7 @@ ENTRY(vm86_bioscall)
 	/*
 	 * Return via _doreti
 	 */
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	MEXITCOUNT
 	jmp	_doreti
 
@@ -156,7 +156,7 @@ ENTRY(vm86_biosret)
 
 	movl	$0,_in_vm86call		/* reset trapflag */
 
-	movl	_tss_gdt,%ebx		/* entry in GDT */
+	movl	PCPU(TSS_GDT),%ebx		/* entry in GDT */
 	movl	SCR_TSS0(%edx),%eax
 	movl	%eax,0(%ebx)		/* restore first word */
 	movl	SCR_TSS1(%edx),%eax
@@ -164,7 +164,7 @@ ENTRY(vm86_biosret)
 	movl	$GPROC0_SEL*8,%esi	/* GSEL(entry, SEL_KPL) */
 	ltr	%si
 	
-	popl	_curpcb			/* restore curpcb/curproc */
+	popl	PCPU(CURPCB)		/* restore curpcb/curproc */
 	movl	SCR_ARGFRAME(%edx),%edx	/* original stack frame */
 	movl	TF_TRAPNO(%edx),%eax	/* return (trapno) */
 
diff --git a/sys/i386/include/asmacros.h b/sys/i386/include/asmacros.h
index 92b9e1a5d700..c2d89b83c446 100644
--- a/sys/i386/include/asmacros.h
+++ b/sys/i386/include/asmacros.h
@@ -69,6 +69,12 @@
 #define NON_GPROF_ENTRY(name)	GEN_ENTRY(name)
 #define NON_GPROF_RET		.byte 0xc3	/* opcode for `ret' */
 
+#ifdef SMP
+#define	PCPU(member)	%fs:GD_ ## member
+#else
+#define	PCPU(member)	CNAME(globaldata) + GD_ ## member
+#endif
+
 #ifdef GPROF
 /*
  * __mcount is like [.]mcount except that doesn't require its caller to set
diff --git a/sys/i386/isa/apic_vector.s b/sys/i386/isa/apic_vector.s
index 995cf213fb25..a23ccff7b67d 100644
--- a/sys/i386/isa/apic_vector.s
+++ b/sys/i386/isa/apic_vector.s
@@ -48,7 +48,7 @@ IDTVEC(vec_name) ;							\
 	movl	$KPSEL,%eax ;						\
 	mov	%ax,%fs ;						\
 	FAKE_MCOUNT(13*4(%esp)) ;					\
-	incb	_intr_nesting_level ;					\
+	incb	PCPU(INTR_NESTING_LEVEL) ;				\
 	pushl	_intr_unit + (irq_num) * 4 ;				\
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	addl	$4, %esp ;						\
@@ -136,7 +136,7 @@ log_intr_event:
 	addl	$4, %esp
 	movl	CNAME(apic_itrace_debugbuffer_idx), %ecx
 	andl	$32767, %ecx
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	shll	$8,	%eax
 	orl	8(%esp), %eax
 	movw	%ax,	CNAME(apic_itrace_debugbuffer)(,%ecx,2)
@@ -218,7 +218,7 @@ IDTVEC(vec_name) ;							\
 	MASK_LEVEL_IRQ(irq_num) ;					\
 	EOI_IRQ(irq_num) ;						\
 0: ;									\
-	incb	_intr_nesting_level ;					\
+	incb	PCPU(INTR_NESTING_LEVEL) ;				\
 ;	 								\
   /* entry point used by doreti_unpend for HWIs. */			\
 __CONCAT(Xresume,irq_num): ;						\
@@ -263,7 +263,7 @@ _Xinvltlb:
 	pushl	%fs
 	movl	$KPSEL, %eax
 	mov	%ax, %fs
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	popl	%fs
 	ss
 	incl	_xhits(,%eax,4)
@@ -321,10 +321,11 @@ _Xcpucheckstate:
 	jne	1f
 	incl	%ebx			/* system or interrupt */
 1:	
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	movl	%ebx, _checkstate_cpustate(,%eax,4)
-	movl	_curproc, %ebx
+	movl	PCPU(CURPROC), %ebx
 	movl	%ebx, _checkstate_curproc(,%eax,4)
+
 	movl	16(%esp), %ebx
 	movl	%ebx, _checkstate_pc(,%eax,4)
 
@@ -358,7 +359,7 @@ _Xcpuast:
 	movl	$KPSEL, %eax
 	mov	%ax, %fs
 
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	lock				/* checkstate_need_ast &= ~(1<<id) */
 	btrl	%eax, _checkstate_need_ast
 	movl	$0, lapic_eoi		/* End Of Interrupt to APIC */
@@ -369,17 +370,17 @@ _Xcpuast:
 
 	FAKE_MCOUNT(13*4(%esp))
 
-	orl	$AST_PENDING, _astpending	/* XXX */
-	incb	_intr_nesting_level
+	orl	$AST_PENDING, PCPU(ASTPENDING)	/* XXX */
+	incb	PCPU(INTR_NESTING_LEVEL)
 	sti
 	
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	lock	
 	btrl	%eax, _checkstate_pending_ast
 	lock	
 	btrl	%eax, CNAME(resched_cpus)
 	jnc	2f
-	orl	$AST_PENDING+AST_RESCHED, _astpending
+	orl	$AST_PENDING+AST_RESCHED, PCPU(ASTPENDING)
 	lock
 	incl	CNAME(want_resched_cnt)
 2:		
@@ -414,10 +415,10 @@ _Xforward_irq:
 
 	lock
 	incl	CNAME(forward_irq_hitcnt)
-	cmpb	$4, _intr_nesting_level
+	cmpb	$4, PCPU(INTR_NESTING_LEVEL)
 	jae	1f
 	
-	incb	_intr_nesting_level
+	incb	PCPU(INTR_NESTING_LEVEL)
 	sti
 	
 	MEXITCOUNT
@@ -499,7 +500,7 @@ _Xcpustop:
 
 	movl	$0, lapic_eoi		/* End Of Interrupt to APIC */
 
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 	imull	$PCB_SIZE, %eax
 	leal	CNAME(stoppcbs)(%eax), %eax
 	pushl	%eax
@@ -507,7 +508,7 @@ _Xcpustop:
 	addl	$4, %esp
 	
 		
-	movl	_cpuid, %eax
+	movl	PCPU(CPUID), %eax
 
 	lock
 	btsl	%eax, _stopped_cpus	/* stopped_cpus |= (1<<id) */
diff --git a/sys/i386/isa/atpic_vector.s b/sys/i386/isa/atpic_vector.s
index f8104016619f..e28c37598678 100644
--- a/sys/i386/isa/atpic_vector.s
+++ b/sys/i386/isa/atpic_vector.s
@@ -60,7 +60,7 @@ IDTVEC(vec_name) ; \
 	mov	%ax,%es ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -100,7 +100,7 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
diff --git a/sys/i386/isa/icu_vector.s b/sys/i386/isa/icu_vector.s
index f8104016619f..e28c37598678 100644
--- a/sys/i386/isa/icu_vector.s
+++ b/sys/i386/isa/icu_vector.s
@@ -60,7 +60,7 @@ IDTVEC(vec_name) ; \
 	mov	%ax,%es ; \
 	mov	%ax,%fs ; \
 	FAKE_MCOUNT((12+ACTUALLY_PUSHED)*4(%esp)) ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 	pushl	_intr_unit + (irq_num) * 4 ; \
 	call	*_intr_handler + (irq_num) * 4 ; /* do the work ASAP */ \
 	enable_icus ;		/* (re)enable ASAP (helps edge trigger?) */ \
@@ -100,7 +100,7 @@ IDTVEC(vec_name) ; \
 	movb	%al,_imen + IRQ_BYTE(irq_num) ; \
 	outb	%al,$icu+ICU_IMR_OFFSET ; \
 	enable_icus ; \
-	incb	_intr_nesting_level ; \
+	incb	PCPU(INTR_NESTING_LEVEL) ; \
 __CONCAT(Xresume,irq_num): ; \
 	FAKE_MCOUNT(13*4(%esp)) ;	/* XXX late to avoid double count */ \
 	pushl	$irq_num; 	/* pass the IRQ */ \
diff --git a/sys/i386/isa/ipl.s b/sys/i386/isa/ipl.s
index a079d2c5a43b..3f8fc9692c7b 100644
--- a/sys/i386/isa/ipl.s
+++ b/sys/i386/isa/ipl.s
@@ -80,10 +80,10 @@ _softtty_imask:	.long	0
 _doreti:
 	FAKE_MCOUNT(_bintr)		/* init "from" _bintr -> _doreti */
 doreti_next:
-	decb	_intr_nesting_level
+	decb	PCPU(INTR_NESTING_LEVEL)
 
 	/* Check for ASTs that can be handled now. */
-	testl	$AST_PENDING,_astpending
+	testl	$AST_PENDING,PCPU(ASTPENDING)
 	je	doreti_exit		/* no AST, exit */
 	testb	$SEL_RPL_MASK,TF_CS(%esp)  /* are we in user mode? */
 	jne	doreti_ast		/* yes, do it now. */
@@ -147,11 +147,11 @@ doreti_popl_fs_fault:
 
 	ALIGN_TEXT
 doreti_ast:
-	andl	$~AST_PENDING,_astpending
+	andl	$~AST_PENDING,PCPU(ASTPENDING)
 	sti
 	movl	$T_ASTFLT,TF_TRAPNO(%esp)
 	call	_ast
-	movb	$1,_intr_nesting_level	/* for doreti_next to decrement */
+	movb	$1,PCPU(INTR_NESTING_LEVEL) /* for doreti_next to decrement */
 	jmp	doreti_next
 
 #ifdef APIC_IO