Overhaul the per-CPU support a bit:

- The MI portions of struct globaldata have been consolidated into a MI
  struct pcpu.  The MD per-CPU data are specified via a macro defined in
  machine/pcpu.h.  A macro was chosen over a struct mdpcpu so that the
  interface would be cleaner (PCPU_GET(my_md_field) vs.
  PCPU_GET(md.md_my_md_field)).
- All references to globaldata are changed to pcpu instead.  In a UP kernel,
  this data was stored as global variables which is where the original name
  came from.  In an SMP world this data is per-CPU and ideally private to each
  CPU outside of the context of debuggers.  This also included combining
  machine/globaldata.h and machine/globals.h into machine/pcpu.h.
- The pointer to the thread using the FPU on i386 was renamed from
  npxthread to fpcurthread to be identical with other architectures.
- Make the show pcpu ddb command MI with a MD callout to display MD
  fields.
- The globaldata_register() function was renamed to pcpu_init() and now
  init's MI fields of a struct pcpu in addition to registering it with
  the internal array and list.
- A pcpu_destroy() function was added to remove a struct pcpu from the
  internal array and list.

Tested on:	alpha, i386
Reviewed by:	peter, jake

19 files changed, 235 insertions, 227 deletions

diff --git a/sys/amd64/amd64/bios.c b/sys/amd64/amd64/bios.c
index cc887e7..6eb90b2 100644
--- a/sys/amd64/amd64/bios.c
+++ b/sys/amd64/amd64/bios.c
@@ -36,9 +36,9 @@
 #include <sys/kernel.h>
 #include <sys/malloc.h>
 #include <sys/bus.h>
+#include <sys/pcpu.h>
 #include <vm/vm.h>
 #include <vm/pmap.h>
-#include <machine/globals.h>
 #include <machine/md_var.h>
 #include <machine/segments.h>
 #include <machine/stdarg.h>
diff --git a/sys/amd64/amd64/cpu_switch.S b/sys/amd64/amd64/cpu_switch.S
index 14e4495..ae967c4 100644
--- a/sys/amd64/amd64/cpu_switch.S
+++ b/sys/amd64/amd64/cpu_switch.S
@@ -123,7 +123,7 @@ ENTRY(cpu_switch)
 
 #ifdef DEV_NPX
 	/* have we used fp, and need a save? */
-	cmpl	%ecx,PCPU(NPXTHREAD)
+	cmpl	%ecx,PCPU(FPCURTHREAD)
 	jne	1f
 	addl	$PCB_SAVEFPU,%edx		/* h/w bugs make saving complicated */
 	pushl	%edx
@@ -337,20 +337,20 @@ ENTRY(savectx)
 
 #ifdef DEV_NPX
 	/*
-	 * If npxthread == NULL, then the npx h/w state is irrelevant and the
+	 * If fpcurthread == NULL, then the npx h/w state is irrelevant and the
 	 * state had better already be in the pcb.  This is true for forks
 	 * but not for dumps (the old book-keeping with FP flags in the pcb
 	 * always lost for dumps because the dump pcb has 0 flags).
 	 *
-	 * If npxthread != NULL, then we have to save the npx h/w state to
-	 * npxthread's pcb and copy it to the requested pcb, or save to the
+	 * If fpcurthread != NULL, then we have to save the npx h/w state to
+	 * fpcurthread's pcb and copy it to the requested pcb, or save to the
 	 * requested pcb and reload.  Copying is easier because we would
 	 * have to handle h/w bugs for reloading.  We used to lose the
 	 * parent's npx state for forks by forgetting to reload.
 	 */
 	pushfl
 	cli
-	movl	PCPU(NPXTHREAD),%eax
+	movl	PCPU(FPCURTHREAD),%eax
 	testl	%eax,%eax
 	je	1f
 
diff --git a/sys/amd64/amd64/db_interface.c b/sys/amd64/amd64/db_interface.c
index 7e78088..9708cad 100644
--- a/sys/amd64/amd64/db_interface.c
+++ b/sys/amd64/amd64/db_interface.c
@@ -332,48 +332,9 @@ Debugger(msg)
 	}
 }
 
-DB_SHOW_COMMAND(pcpu, db_show_pcpu)
+void
+db_show_mdpcpu(struct pcpu *pc)
 {
-	struct globaldata *gd;
-#ifdef SMP
-	int id;
-
-	if (have_addr)
-		id = ((addr >> 4) % 16) * 10 + (addr % 16);
-	else
-		id = PCPU_GET(cpuid);
-	gd = globaldata_find(id);
-	if (gd == NULL) {
-		db_printf("CPU %d not found\n", id);
-		return;
-	}
-#else
-	gd = GLOBALDATA;
-#endif
-	db_printf("cpuid    = %d\n", gd->gd_cpuid);
-	db_printf("curthread  = ");
-	if (gd->gd_curthread != NULL)
-		db_printf("%p: pid %d \"%s\"\n", gd->gd_curthread,
-		    gd->gd_curthread->td_proc->p_pid, gd->gd_curthread->td_proc->p_comm);
-	else
-		db_printf("none\n");
-	db_printf("curpcb   = %p\n", gd->gd_curpcb);
-	db_printf("npxthread  = ");
-	if (gd->gd_npxthread != NULL)
-		db_printf("%p: pid %d \"%s\"\n", gd->gd_npxthread,
-		    gd->gd_npxthread->td_proc->p_pid, gd->gd_npxthread->td_proc->p_comm);
-	else
-		db_printf("none\n");
-	db_printf("idlethread = ");
-	if (gd->gd_idlethread != NULL)
-		db_printf("%p: pid %d \"%s\"\n", gd->gd_idlethread,
-		    gd->gd_idlethread->td_proc->p_pid,
-		    gd->gd_idlethread->td_proc->p_comm);
-	else
-		db_printf("none\n");
-		
-#ifdef WITNESS
-	db_printf("spin locks held:\n");
-	witness_list_locks(&gd->gd_spinlocks);
-#endif
+
+	db_printf("currentldt   = 0x%x\n", pc->pc_currentldt);
 }
diff --git a/sys/amd64/amd64/fpu.c b/sys/amd64/amd64/fpu.c
index 1238d67..ff8c4cf 100644
--- a/sys/amd64/amd64/fpu.c
+++ b/sys/amd64/amd64/fpu.c
@@ -237,7 +237,7 @@ npx_intr(dummy)
 #endif
 
 	/*
-	 * npxthread is normally non-null here.  In that case, schedule an
+	 * fpcurthread is normally non-null here.  In that case, schedule an
 	 * AST to finish the exception handling in the correct context
 	 * (this interrupt may occur after the thread has entered the
 	 * kernel via a syscall or an interrupt).  Otherwise, the npx
@@ -248,7 +248,7 @@ npx_intr(dummy)
 	 * that caused it and it will repeat.  We will eventually (usually
 	 * soon) win the race to handle the interrupt properly.
 	 */
-	td = PCPU_GET(npxthread);
+	td = PCPU_GET(fpcurthread);
 	if (td != NULL) {
 		td->td_pcb->pcb_flags |= PCB_NPXTRAP;
 		mtx_lock_spin(&sched_lock);
@@ -513,7 +513,7 @@ npxinit(control)
 	/*
 	 * fninit has the same h/w bugs as fnsave.  Use the detoxified
 	 * fnsave to throw away any junk in the fpu.  npxsave() initializes
-	 * the fpu and sets npxthread = NULL as important side effects.
+	 * the fpu and sets fpcurthread = NULL as important side effects.
 	 */
 	savecrit = critical_enter();
 	npxsave(&dummy);
@@ -540,7 +540,7 @@ npxexit(td)
 	critical_t savecrit;
 
 	savecrit = critical_enter();
-	if (td == PCPU_GET(npxthread))
+	if (td == PCPU_GET(fpcurthread))
 		npxsave(&PCPU_GET(curpcb)->pcb_save);
 	critical_exit(savecrit);
 #ifdef NPX_DEBUG
@@ -758,8 +758,8 @@ npxtrap()
 	u_long *exstat;
 
 	if (!npx_exists) {
-		printf("npxtrap: npxthread = %p, curthread = %p, npx_exists = %d\n",
-		       PCPU_GET(npxthread), curthread, npx_exists);
+		printf("npxtrap: fpcurthread = %p, curthread = %p, npx_exists = %d\n",
+		       PCPU_GET(fpcurthread), curthread, npx_exists);
 		panic("npxtrap from nowhere");
 	}
 	savecrit = critical_enter();
@@ -769,7 +769,7 @@ npxtrap()
 	 * state to memory.  Fetch the relevant parts of the state from
 	 * wherever they are.
 	 */
-	if (PCPU_GET(npxthread) != curthread) {
+	if (PCPU_GET(fpcurthread) != curthread) {
 		control = GET_FPU_CW(curthread);
 		status = GET_FPU_SW(curthread);
 	} else {
@@ -779,7 +779,7 @@ npxtrap()
 
 	exstat = GET_FPU_EXSW_PTR(curthread->td_pcb);
 	*exstat = status;
-	if (PCPU_GET(npxthread) != curthread)
+	if (PCPU_GET(fpcurthread) != curthread)
 		GET_FPU_SW(curthread) &= ~0x80bf;
 	else
 		fnclex();
@@ -790,7 +790,7 @@ npxtrap()
 /*
  * Implement device not available (DNA) exception
  *
- * It would be better to switch FP context here (if curthread != npxthread)
+ * It would be better to switch FP context here (if curthread != fpcurthread)
  * and not necessarily for every context switch, but it is too hard to
  * access foreign pcb's.
  */
@@ -802,9 +802,9 @@ npxdna()
 
 	if (!npx_exists)
 		return (0);
-	if (PCPU_GET(npxthread) != NULL) {
-		printf("npxdna: npxthread = %p, curthread = %p\n",
-		       PCPU_GET(npxthread), curthread);
+	if (PCPU_GET(fpcurthread) != NULL) {
+		printf("npxdna: fpcurthread = %p, curthread = %p\n",
+		       PCPU_GET(fpcurthread), curthread);
 		panic("npxdna");
 	}
 	s = critical_enter();
@@ -812,7 +812,7 @@ npxdna()
 	/*
 	 * Record new context early in case frstor causes an IRQ13.
 	 */
-	PCPU_SET(npxthread, curthread);
+	PCPU_SET(fpcurthread, curthread);
 
 	exstat = GET_FPU_EXSW_PTR(PCPU_GET(curpcb));
 	*exstat = 0;
@@ -844,13 +844,13 @@ npxdna()
  * after the process has entered the kernel.  It may even be delivered after
  * the fnsave here completes.  A spurious IRQ13 for the fnsave is handled in
  * the same way as a very-late-arriving non-spurious IRQ13 from user mode:
- * it is normally ignored at first because we set npxthread to NULL; it is
+ * it is normally ignored at first because we set fpcurthread to NULL; it is
  * normally retriggered in npxdna() after return to user mode.
  *
  * npxsave() must be called with interrupts disabled, so that it clears
- * npxthread atomically with saving the state.  We require callers to do the
+ * fpcurthread atomically with saving the state.  We require callers to do the
  * disabling, since most callers need to disable interrupts anyway to call
- * npxsave() atomically with checking npxthread.
+ * npxsave() atomically with checking fpcurthread.
  *
  * A previous version of npxsave() went to great lengths to excecute fnsave
  * with interrupts enabled in case executing it froze the CPU.  This case
@@ -866,7 +866,7 @@ npxsave(addr)
 	fpusave(addr);
 
 	start_emulating();
-	PCPU_SET(npxthread, NULL);
+	PCPU_SET(fpcurthread, NULL);
 }
 
 static void
diff --git a/sys/amd64/amd64/genassym.c b/sys/amd64/amd64/genassym.c
index 9f41665..59f3147 100644
--- a/sys/amd64/amd64/genassym.c
+++ b/sys/amd64/amd64/genassym.c
@@ -51,10 +51,6 @@
 #include <sys/socket.h>
 #include <sys/resourcevar.h>
 #include <sys/user.h>
-/* XXX */
-#ifdef KTR_PERCPU
-#include <sys/ktr.h>
-#endif
 #include <machine/bootinfo.h>
 #include <machine/tss.h>
 #include <sys/vmmeter.h>
@@ -75,7 +71,6 @@
 #endif
 #include <machine/cpu.h>
 #include <machine/sigframe.h>
-#include <machine/globaldata.h>
 #include <machine/vm86.h>
 #include <machine/proc.h>
 
@@ -177,27 +172,19 @@ ASSYM(BI_SIZE, offsetof(struct bootinfo, bi_size));
 ASSYM(BI_SYMTAB, offsetof(struct bootinfo, bi_symtab));
 ASSYM(BI_ESYMTAB, offsetof(struct bootinfo, bi_esymtab));
 ASSYM(BI_KERNEND, offsetof(struct bootinfo, bi_kernend));
-ASSYM(GD_SIZEOF, sizeof(struct globaldata));
-ASSYM(GD_PRVSPACE, offsetof(struct globaldata, gd_prvspace));
-ASSYM(GD_CURTHREAD, offsetof(struct globaldata, gd_curthread));
-ASSYM(GD_NPXTHREAD, offsetof(struct globaldata, gd_npxthread));
-ASSYM(GD_IDLETHREAD, offsetof(struct globaldata, gd_idlethread));
-ASSYM(GD_CURPCB, offsetof(struct globaldata, gd_curpcb));
-ASSYM(GD_COMMON_TSS, offsetof(struct globaldata, gd_common_tss));
-ASSYM(GD_SWITCHTIME, offsetof(struct globaldata, gd_switchtime));
-ASSYM(GD_SWITCHTICKS, offsetof(struct globaldata, gd_switchticks));
-ASSYM(GD_COMMON_TSSD, offsetof(struct globaldata, gd_common_tssd));
-ASSYM(GD_TSS_GDT, offsetof(struct globaldata, gd_tss_gdt));
-ASSYM(GD_CURRENTLDT, offsetof(struct globaldata, gd_currentldt));
-
-/* XXX */
-#ifdef KTR_PERCPU
-ASSYM(GD_KTR_IDX, offsetof(struct globaldata, gd_ktr_idx));
-ASSYM(GD_KTR_BUF, offsetof(struct globaldata, gd_ktr_buf));
-ASSYM(GD_KTR_BUF_DATA, offsetof(struct globaldata, gd_ktr_buf_data));
-#endif
-
-ASSYM(GD_CPUID, offsetof(struct globaldata, gd_cpuid));
+ASSYM(PC_SIZEOF, sizeof(struct pcpu));
+ASSYM(PC_PRVSPACE, offsetof(struct pcpu, pc_prvspace));
+ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
+ASSYM(PC_FPCURTHREAD, offsetof(struct pcpu, pc_fpcurthread));
+ASSYM(PC_IDLETHREAD, offsetof(struct pcpu, pc_idlethread));
+ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
+ASSYM(PC_COMMON_TSS, offsetof(struct pcpu, pc_common_tss));
+ASSYM(PC_SWITCHTIME, offsetof(struct pcpu, pc_switchtime));
+ASSYM(PC_SWITCHTICKS, offsetof(struct pcpu, pc_switchticks));
+ASSYM(PC_COMMON_TSSD, offsetof(struct pcpu, pc_common_tssd));
+ASSYM(PC_TSS_GDT, offsetof(struct pcpu, pc_tss_gdt));
+ASSYM(PC_CURRENTLDT, offsetof(struct pcpu, pc_currentldt));
+ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
 
 #ifdef SMP
 ASSYM(LA_VER, offsetof(struct LAPIC, version));
diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c
index bdc0980..c2ef5dc 100644
--- a/sys/amd64/amd64/machdep.c
+++ b/sys/amd64/amd64/machdep.c
@@ -104,7 +104,6 @@
 #include <machine/pc/bios.h>
 #include <machine/pcb_ext.h>		/* pcb.h included via sys/user.h */
 #include <machine/proc.h>
-#include <machine/globals.h>
 #ifdef PERFMON
 #include <machine/perfmon.h>
 #endif
@@ -207,7 +206,7 @@ struct kva_md_info kmi;
 
 static struct trapframe proc0_tf;
 #ifndef SMP
-static struct globaldata __globaldata;
+static struct pcpu __pcpu;
 #endif
 
 struct mtx sched_lock;
@@ -262,7 +261,6 @@ cpu_startup(dummy)
 	bufinit();
 	vm_pager_bufferinit();
 
-	globaldata_register(GLOBALDATA);
 #ifndef SMP
 	/* For SMP, we delay the cpu_setregs() until after SMP startup. */
 	cpu_setregs();
@@ -1670,6 +1668,7 @@ init386(first)
 	/* table descriptors - used to load tables by microp */
 	struct region_descriptor r_gdt, r_idt;
 #endif
+	struct pcpu *pc;
 
 	proc_linkup(&proc0);
 	proc0.p_uarea = proc0uarea;
@@ -1706,20 +1705,16 @@ init386(first)
 	gdt_segs[GCODE_SEL].ssd_limit = atop(0 - 1);
 	gdt_segs[GDATA_SEL].ssd_limit = atop(0 - 1);
 #ifdef SMP
+	pc = &SMP_prvspace[0];
 	gdt_segs[GPRIV_SEL].ssd_limit =
 		atop(sizeof(struct privatespace) - 1);
-	gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[0];
-	gdt_segs[GPROC0_SEL].ssd_base =
-		(int) &SMP_prvspace[0].globaldata.gd_common_tss;
-	SMP_prvspace[0].globaldata.gd_prvspace = &SMP_prvspace[0].globaldata;
 #else
+	pc = &__pcpu;
 	gdt_segs[GPRIV_SEL].ssd_limit =
-		atop(sizeof(struct globaldata) - 1);
-	gdt_segs[GPRIV_SEL].ssd_base = (int) &__globaldata;
-	gdt_segs[GPROC0_SEL].ssd_base =
-		(int) &__globaldata.gd_common_tss;
-	__globaldata.gd_prvspace = &__globaldata;
+		atop(sizeof(struct pcpu) - 1);
 #endif
+	gdt_segs[GPRIV_SEL].ssd_base = (int) pc;
+	gdt_segs[GPROC0_SEL].ssd_base = (int) &pc->pc_common_tss;
 
 	for (x = 0; x < NGDT; x++) {
 #ifdef BDE_DEBUGGER
@@ -1734,10 +1729,11 @@ init386(first)
 	r_gdt.rd_base =  (int) gdt;
 	lgdt(&r_gdt);
 
-	/* setup curproc so that mutexes work */
+	pcpu_init(pc, 0, sizeof(struct pcpu));
+	PCPU_SET(prvspace, pc);
 
+	/* setup curproc so that mutexes work */
 	PCPU_SET(curthread, thread0);
-	PCPU_SET(spinlocks, NULL);
 
 	LIST_INIT(&thread0->td_contested);
 
@@ -1907,6 +1903,11 @@ init386(first)
 	thread0->td_frame = &proc0_tf;
 }
 
+void
+cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
+{
+}
+
 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
 static void f00f_hack(void *unused);
 SYSINIT(f00f_hack, SI_SUB_INTRINSIC, SI_ORDER_FIRST, f00f_hack, NULL);
diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 13144b9..9d86dae 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -71,7 +71,6 @@
 #include <machine/smptests.h>	/** TEST_DEFAULT_CONFIG, TEST_TEST1 */
 #include <machine/tss.h>
 #include <machine/specialreg.h>
-#include <machine/globaldata.h>
 #include <machine/privatespace.h>
 
 #if defined(APIC_IO)
@@ -477,9 +476,9 @@ init_secondary(void)
 
 	gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid];
 	gdt_segs[GPROC0_SEL].ssd_base =
-		(int) &SMP_prvspace[myid].globaldata.gd_common_tss;
-	SMP_prvspace[myid].globaldata.gd_prvspace =
-		&SMP_prvspace[myid].globaldata;
+		(int) &SMP_prvspace[myid].pcpu.pc_common_tss;
+	SMP_prvspace[myid].pcpu.pc_prvspace =
+		&SMP_prvspace[myid].pcpu;
 
 	for (x = 0; x < NGDT; x++) {
 		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
@@ -1915,7 +1914,7 @@ start_all_aps(u_int boot_addr)
 	int     x, i, pg;
 	u_char  mpbiosreason;
 	u_long  mpbioswarmvec;
-	struct globaldata *gd;
+	struct pcpu *pc;
 	char *stack;
 	uintptr_t kptbase;
 
@@ -1955,10 +1954,10 @@ start_all_aps(u_int boot_addr)
 		pg = x * i386_btop(sizeof(struct privatespace));
 
 		/* allocate a new private data page */
-		gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE);
+		pc = (struct pcpu *)kmem_alloc(kernel_map, PAGE_SIZE);
 
 		/* wire it into the private page table page */
-		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd));
+		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(pc));
 
 		/* allocate and set up an idle stack data page */
 		stack = (char *)kmem_alloc(kernel_map, KSTACK_PAGES * PAGE_SIZE); /* XXXKSE */
@@ -1967,8 +1966,7 @@ start_all_aps(u_int boot_addr)
 			    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
 
 		/* prime data page for it to use */
-		gd->gd_cpuid = x;
-		globaldata_register(gd);
+		pcpu_init(pc, x, sizeof(struct pcpu));
 
 		/* setup a vector to our boot code */
 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
diff --git a/sys/amd64/amd64/mptable.c b/sys/amd64/amd64/mptable.c
index 13144b9..9d86dae 100644
--- a/sys/amd64/amd64/mptable.c
+++ b/sys/amd64/amd64/mptable.c
@@ -71,7 +71,6 @@
 #include <machine/smptests.h>	/** TEST_DEFAULT_CONFIG, TEST_TEST1 */
 #include <machine/tss.h>
 #include <machine/specialreg.h>
-#include <machine/globaldata.h>
 #include <machine/privatespace.h>
 
 #if defined(APIC_IO)
@@ -477,9 +476,9 @@ init_secondary(void)
 
 	gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid];
 	gdt_segs[GPROC0_SEL].ssd_base =
-		(int) &SMP_prvspace[myid].globaldata.gd_common_tss;
-	SMP_prvspace[myid].globaldata.gd_prvspace =
-		&SMP_prvspace[myid].globaldata;
+		(int) &SMP_prvspace[myid].pcpu.pc_common_tss;
+	SMP_prvspace[myid].pcpu.pc_prvspace =
+		&SMP_prvspace[myid].pcpu;
 
 	for (x = 0; x < NGDT; x++) {
 		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
@@ -1915,7 +1914,7 @@ start_all_aps(u_int boot_addr)
 	int     x, i, pg;
 	u_char  mpbiosreason;
 	u_long  mpbioswarmvec;
-	struct globaldata *gd;
+	struct pcpu *pc;
 	char *stack;
 	uintptr_t kptbase;
 
@@ -1955,10 +1954,10 @@ start_all_aps(u_int boot_addr)
 		pg = x * i386_btop(sizeof(struct privatespace));
 
 		/* allocate a new private data page */
-		gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE);
+		pc = (struct pcpu *)kmem_alloc(kernel_map, PAGE_SIZE);
 
 		/* wire it into the private page table page */
-		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd));
+		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(pc));
 
 		/* allocate and set up an idle stack data page */
 		stack = (char *)kmem_alloc(kernel_map, KSTACK_PAGES * PAGE_SIZE); /* XXXKSE */
@@ -1967,8 +1966,7 @@ start_all_aps(u_int boot_addr)
 			    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
 
 		/* prime data page for it to use */
-		gd->gd_cpuid = x;
-		globaldata_register(gd);
+		pcpu_init(pc, x, sizeof(struct pcpu));
 
 		/* setup a vector to our boot code */
 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c
index 789c491..d7a95a4 100644
--- a/sys/amd64/amd64/pmap.c
+++ b/sys/amd64/amd64/pmap.c
@@ -105,7 +105,6 @@
 #include <machine/apic.h>
 #include <machine/segments.h>
 #include <machine/tss.h>
-#include <machine/globaldata.h>
 #endif /* SMP || APIC_IO */
 
 #define PMAP_KEEP_PDIRS
diff --git a/sys/amd64/amd64/support.S b/sys/amd64/amd64/support.S
index 81181cc..bc58672 100644
--- a/sys/amd64/amd64/support.S
+++ b/sys/amd64/amd64/support.S
@@ -216,8 +216,8 @@ ENTRY(i586_bzero)
 	 * complicated since we avoid it if possible at all levels.  We
 	 * want to localize the complications even when that increases them.
 	 * Here the extra work involves preserving CR0_TS in TS.
-	 * `npxthread != NULL' is supposed to be the condition that all the
-	 * FPU resources belong to an application, but npxthread and CR0_TS
+	 * `fpcurthread != NULL' is supposed to be the condition that all the
+	 * FPU resources belong to an application, but fpcurthread and CR0_TS
 	 * aren't set atomically enough for this condition to work in
 	 * interrupt handlers.
 	 *
@@ -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,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bz1
 
 	/*
@@ -303,7 +303,7 @@ fpureg_i586_bzero_loop:
 	cmpl	$8,%ecx
 	jae	fpureg_i586_bzero_loop
 
-	cmpl	$0,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bz3
 
 	/* XXX check that the condition for cases 1-2 stayed false. */
@@ -517,7 +517,7 @@ ENTRY(i586_bcopy)
 
 	sarb	$1,kernel_fpu_lock
 	jc	small_i586_bcopy
-	cmpl	$0,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bc1
 
 	/* XXX turn off handling of cases 1-2, as above. */
@@ -593,7 +593,7 @@ large_i586_bcopy_loop:
 	cmpl	$64,%ecx
 	jae	4b
 
-	cmpl	$0,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bc2
 
 	/* XXX check that the condition for cases 1-2 stayed false. */
@@ -991,14 +991,14 @@ ENTRY(fastmove)
 	/* XXX grab FPU context atomically. */
 	cli
 
-/* if (npxthread != NULL) { */
-	cmpl	$0,PCPU(NPXTHREAD)
+/* if (fpcurthread != NULL) { */
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	6f
 /*    fnsave(&curpcb->pcb_savefpu); */
 	movl	PCPU(CURPCB),%eax
 	fnsave	PCB_SAVEFPU(%eax)
-/*   NPXTHREAD = NULL; */
-	movl	$0,PCPU(NPXTHREAD)
+/*   FPCURTHREAD = NULL; */
+	movl	$0,PCPU(FPCURTHREAD)
 /* } */
 6:
 /* now we own the FPU. */
@@ -1026,9 +1026,9 @@ ENTRY(fastmove)
 	movl	-4(%ebp),%edi
 /* stop_emulating(); */
 	clts
-/* npxthread = curthread; */
+/* fpcurthread = curthread; */
 	movl	PCPU(CURTHREAD),%eax
-	movl	%eax,PCPU(NPXTHREAD)
+	movl	%eax,PCPU(FPCURTHREAD)
 	movl	PCPU(CURPCB),%eax
 
 	/* XXX end of atomic FPU context grab. */
@@ -1113,8 +1113,8 @@ fastmove_loop:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-/* npxthread = NULL; */
-	movl	$0,PCPU(NPXTHREAD)
+/* fpcurthread = NULL; */
+	movl	$0,PCPU(FPCURTHREAD)
 
 	/* XXX end of atomic FPU context ungrab. */
 	sti
@@ -1154,7 +1154,7 @@ fastmove_fault:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-	movl	$0,PCPU(NPXTHREAD)
+	movl	$0,PCPU(FPCURTHREAD)
 
 	/* XXX end of atomic FPU context ungrab. */
 	sti
diff --git a/sys/amd64/amd64/support.s b/sys/amd64/amd64/support.s
index 81181cc..bc58672 100644
--- a/sys/amd64/amd64/support.s
+++ b/sys/amd64/amd64/support.s
@@ -216,8 +216,8 @@ ENTRY(i586_bzero)
 	 * complicated since we avoid it if possible at all levels.  We
 	 * want to localize the complications even when that increases them.
 	 * Here the extra work involves preserving CR0_TS in TS.
-	 * `npxthread != NULL' is supposed to be the condition that all the
-	 * FPU resources belong to an application, but npxthread and CR0_TS
+	 * `fpcurthread != NULL' is supposed to be the condition that all the
+	 * FPU resources belong to an application, but fpcurthread and CR0_TS
 	 * aren't set atomically enough for this condition to work in
 	 * interrupt handlers.
 	 *
@@ -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,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bz1
 
 	/*
@@ -303,7 +303,7 @@ fpureg_i586_bzero_loop:
 	cmpl	$8,%ecx
 	jae	fpureg_i586_bzero_loop
 
-	cmpl	$0,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bz3
 
 	/* XXX check that the condition for cases 1-2 stayed false. */
@@ -517,7 +517,7 @@ ENTRY(i586_bcopy)
 
 	sarb	$1,kernel_fpu_lock
 	jc	small_i586_bcopy
-	cmpl	$0,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bc1
 
 	/* XXX turn off handling of cases 1-2, as above. */
@@ -593,7 +593,7 @@ large_i586_bcopy_loop:
 	cmpl	$64,%ecx
 	jae	4b
 
-	cmpl	$0,PCPU(NPXTHREAD)
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	i586_bc2
 
 	/* XXX check that the condition for cases 1-2 stayed false. */
@@ -991,14 +991,14 @@ ENTRY(fastmove)
 	/* XXX grab FPU context atomically. */
 	cli
 
-/* if (npxthread != NULL) { */
-	cmpl	$0,PCPU(NPXTHREAD)
+/* if (fpcurthread != NULL) { */
+	cmpl	$0,PCPU(FPCURTHREAD)
 	je	6f
 /*    fnsave(&curpcb->pcb_savefpu); */
 	movl	PCPU(CURPCB),%eax
 	fnsave	PCB_SAVEFPU(%eax)
-/*   NPXTHREAD = NULL; */
-	movl	$0,PCPU(NPXTHREAD)
+/*   FPCURTHREAD = NULL; */
+	movl	$0,PCPU(FPCURTHREAD)
 /* } */
 6:
 /* now we own the FPU. */
@@ -1026,9 +1026,9 @@ ENTRY(fastmove)
 	movl	-4(%ebp),%edi
 /* stop_emulating(); */
 	clts
-/* npxthread = curthread; */
+/* fpcurthread = curthread; */
 	movl	PCPU(CURTHREAD),%eax
-	movl	%eax,PCPU(NPXTHREAD)
+	movl	%eax,PCPU(FPCURTHREAD)
 	movl	PCPU(CURPCB),%eax
 
 	/* XXX end of atomic FPU context grab. */
@@ -1113,8 +1113,8 @@ fastmove_loop:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-/* npxthread = NULL; */
-	movl	$0,PCPU(NPXTHREAD)
+/* fpcurthread = NULL; */
+	movl	$0,PCPU(FPCURTHREAD)
 
 	/* XXX end of atomic FPU context ungrab. */
 	sti
@@ -1154,7 +1154,7 @@ fastmove_fault:
 	smsw	%ax
 	orb	$CR0_TS,%al
 	lmsw	%ax
-	movl	$0,PCPU(NPXTHREAD)
+	movl	$0,PCPU(FPCURTHREAD)
 
 	/* XXX end of atomic FPU context ungrab. */
 	sti
diff --git a/sys/amd64/amd64/swtch.s b/sys/amd64/amd64/swtch.s
index 14e4495..ae967c4 100644
--- a/sys/amd64/amd64/swtch.s
+++ b/sys/amd64/amd64/swtch.s
@@ -123,7 +123,7 @@ ENTRY(cpu_switch)
 
 #ifdef DEV_NPX
 	/* have we used fp, and need a save? */
-	cmpl	%ecx,PCPU(NPXTHREAD)
+	cmpl	%ecx,PCPU(FPCURTHREAD)
 	jne	1f
 	addl	$PCB_SAVEFPU,%edx		/* h/w bugs make saving complicated */
 	pushl	%edx
@@ -337,20 +337,20 @@ ENTRY(savectx)
 
 #ifdef DEV_NPX
 	/*
-	 * If npxthread == NULL, then the npx h/w state is irrelevant and the
+	 * If fpcurthread == NULL, then the npx h/w state is irrelevant and the
 	 * state had better already be in the pcb.  This is true for forks
 	 * but not for dumps (the old book-keeping with FP flags in the pcb
 	 * always lost for dumps because the dump pcb has 0 flags).
 	 *
-	 * If npxthread != NULL, then we have to save the npx h/w state to
-	 * npxthread's pcb and copy it to the requested pcb, or save to the
+	 * If fpcurthread != NULL, then we have to save the npx h/w state to
+	 * fpcurthread's pcb and copy it to the requested pcb, or save to the
 	 * requested pcb and reload.  Copying is easier because we would
 	 * have to handle h/w bugs for reloading.  We used to lose the
 	 * parent's npx state for forks by forgetting to reload.
 	 */
 	pushfl
 	cli
-	movl	PCPU(NPXTHREAD),%eax
+	movl	PCPU(FPCURTHREAD),%eax
 	testl	%eax,%eax
 	je	1f
 
diff --git a/sys/amd64/amd64/vm_machdep.c b/sys/amd64/amd64/vm_machdep.c
index 59898e3..322b5e7 100644
--- a/sys/amd64/amd64/vm_machdep.c
+++ b/sys/amd64/amd64/vm_machdep.c
@@ -154,7 +154,7 @@ cpu_fork(td1, p2, flags)
 	if (td1 == curthread)
 		td1->td_pcb->pcb_gs = rgs();
 	savecrit = critical_enter();
-	if (PCPU_GET(npxthread) == td1)
+	if (PCPU_GET(fpcurthread) == td1)
 		npxsave(&td1->td_pcb->pcb_save);
 	critical_exit(savecrit);
 #endif
diff --git a/sys/amd64/include/asmacros.h b/sys/amd64/include/asmacros.h
index ec43cda..325e3d6 100644
--- a/sys/amd64/include/asmacros.h
+++ b/sys/amd64/include/asmacros.h
@@ -64,9 +64,9 @@
 #define NON_GPROF_RET		.byte 0xc3	/* opcode for `ret' */
 
 #ifdef LOCORE
-#define	PCPU(member)	%fs:GD_ ## member
-#define	PCPU_ADDR(member, reg)	movl %fs:GD_PRVSPACE,reg; \
-			addl $GD_ ## member,reg
+#define	PCPU(member)	%fs:PC_ ## member
+#define	PCPU_ADDR(member, reg)	movl %fs:PC_PRVSPACE,reg; \
+			addl $PC_ ## member,reg
 #endif
 
 #ifdef GPROF
diff --git a/sys/amd64/include/cpu.h b/sys/amd64/include/cpu.h
index a7783a0..10b9499 100644
--- a/sys/amd64/include/cpu.h
+++ b/sys/amd64/include/cpu.h
@@ -46,7 +46,6 @@
 #include <machine/psl.h>
 #include <machine/frame.h>
 #include <machine/segments.h>
-#include <machine/globals.h>
 
 /*
  * definitions of cpu-dependent requirements
diff --git a/sys/amd64/include/mptable.h b/sys/amd64/include/mptable.h
index 13144b9..9d86dae 100644
--- a/sys/amd64/include/mptable.h
+++ b/sys/amd64/include/mptable.h
@@ -71,7 +71,6 @@
 #include <machine/smptests.h>	/** TEST_DEFAULT_CONFIG, TEST_TEST1 */
 #include <machine/tss.h>
 #include <machine/specialreg.h>
-#include <machine/globaldata.h>
 #include <machine/privatespace.h>
 
 #if defined(APIC_IO)
@@ -477,9 +476,9 @@ init_secondary(void)
 
 	gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid];
 	gdt_segs[GPROC0_SEL].ssd_base =
-		(int) &SMP_prvspace[myid].globaldata.gd_common_tss;
-	SMP_prvspace[myid].globaldata.gd_prvspace =
-		&SMP_prvspace[myid].globaldata;
+		(int) &SMP_prvspace[myid].pcpu.pc_common_tss;
+	SMP_prvspace[myid].pcpu.pc_prvspace =
+		&SMP_prvspace[myid].pcpu;
 
 	for (x = 0; x < NGDT; x++) {
 		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
@@ -1915,7 +1914,7 @@ start_all_aps(u_int boot_addr)
 	int     x, i, pg;
 	u_char  mpbiosreason;
 	u_long  mpbioswarmvec;
-	struct globaldata *gd;
+	struct pcpu *pc;
 	char *stack;
 	uintptr_t kptbase;
 
@@ -1955,10 +1954,10 @@ start_all_aps(u_int boot_addr)
 		pg = x * i386_btop(sizeof(struct privatespace));
 
 		/* allocate a new private data page */
-		gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE);
+		pc = (struct pcpu *)kmem_alloc(kernel_map, PAGE_SIZE);
 
 		/* wire it into the private page table page */
-		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd));
+		SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(pc));
 
 		/* allocate and set up an idle stack data page */
 		stack = (char *)kmem_alloc(kernel_map, KSTACK_PAGES * PAGE_SIZE); /* XXXKSE */
@@ -1967,8 +1966,7 @@ start_all_aps(u_int boot_addr)
 			    (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack));
 
 		/* prime data page for it to use */
-		gd->gd_cpuid = x;
-		globaldata_register(gd);
+		pcpu_init(pc, x, sizeof(struct pcpu));
 
 		/* setup a vector to our boot code */
 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
diff --git a/sys/amd64/include/pcpu.h b/sys/amd64/include/pcpu.h
index 46c26c4..5fecb98 100644
--- a/sys/amd64/include/pcpu.h
+++ b/sys/amd64/include/pcpu.h
@@ -26,54 +26,122 @@
  * $FreeBSD$
  */
 
-#ifndef _MACHINE_GLOBALDATA_H_
-#define _MACHINE_GLOBALDATA_H_
+#ifndef _MACHINE_PCPU_H_
+#define _MACHINE_PCPU_H_
 
 #ifdef _KERNEL
 
+#ifndef	__GNUC__
+#error	gcc is required to use this file
+#endif
+
 #include <machine/segments.h>
 #include <machine/tss.h>
 
-/* XXX */
-#ifdef KTR_PERCPU
-#include <sys/ktr.h>
-#endif
-
 /*
- * This structure maps out the global data that needs to be kept on a
- * per-cpu basis.  genassym uses this to generate offsets for the assembler
- * code, which also provides external symbols so that C can get at them as
- * though they were really globals.
- *
  * The SMP parts are setup in pmap.c and locore.s for the BSP, and
  * mp_machdep.c sets up the data for the AP's to "see" when they awake.
  * The reason for doing it via a struct is so that an array of pointers
  * to each CPU's data can be set up for things like "check curproc on all
  * other processors"
  */
-struct globaldata {
-	struct	globaldata *gd_prvspace;	/* Self-reference */
-	struct	thread *gd_curthread;
-	struct	thread *gd_npxthread;
-	struct	pcb *gd_curpcb;
-	struct	thread *gd_idlethread;
-	struct	timeval gd_switchtime;
-	struct	i386tss gd_common_tss;
-	int	gd_switchticks;
-	struct	segment_descriptor gd_common_tssd;
-	struct	segment_descriptor *gd_tss_gdt;
-	int	gd_currentldt;
-	u_int	gd_cpuid;
-	u_int	gd_other_cpus;
-	SLIST_ENTRY(globaldata) gd_allcpu;
-	struct	lock_list_entry *gd_spinlocks;
-#ifdef KTR_PERCPU
-	int	gd_ktr_idx;			/* Index into trace table */
-	char	*gd_ktr_buf;
-	char	gd_ktr_buf_data[KTR_SIZE];
-#endif
-};
+#define	PCPU_MD_FIELDS							\
+	struct	pcpu *pc_prvspace;		/* Self-reference */	\
+	struct	i386tss pc_common_tss;					\
+	struct	segment_descriptor pc_common_tssd;			\
+	struct	segment_descriptor *pc_tss_gdt;				\
+	int	pc_currentldt
+
+/*
+ * Evaluates to the byte offset of the per-cpu variable name.
+ */
+#define	__pcpu_offset(name)						\
+	__offsetof(struct pcpu, name)
+
+/*
+ * Evaluates to the type of the per-cpu variable name.
+ */
+#define	__pcpu_type(name)						\
+	__typeof(((struct pcpu *)0)->name)
+
+/*
+ * Evaluates to the address of the per-cpu variable name.
+ */
+#define	__PCPU_PTR(name) ({						\
+	__pcpu_type(name) *__p;						\
+									\
+	__asm __volatile("movl %%fs:%1,%0; addl %2,%0"			\
+	    : "=r" (__p)						\
+	    : "m" (*(struct pcpu *)(__pcpu_offset(pc_prvspace))),	\
+	      "i" (__pcpu_offset(name)));				\
+									\
+	__p;								\
+})
+
+/*
+ * Evaluates to the value of the per-cpu variable name.
+ */
+#define	__PCPU_GET(name) ({						\
+	__pcpu_type(name) __result;					\
+									\
+	if (sizeof(__result) == 1) {					\
+		u_char __b;						\
+		__asm __volatile("movb %%fs:%1,%0"			\
+		    : "=r" (__b)					\
+		    : "m" (*(u_char *)(__pcpu_offset(name))));		\
+		__result = *(__pcpu_type(name) *)&__b;			\
+	} else if (sizeof(__result) == 2) {				\
+		u_short __w;						\
+		__asm __volatile("movw %%fs:%1,%0"			\
+		    : "=r" (__w)					\
+		    : "m" (*(u_short *)(__pcpu_offset(name))));		\
+		__result = *(__pcpu_type(name) *)&__w;			\
+	} else if (sizeof(__result) == 4) {				\
+		u_int __i;						\
+		__asm __volatile("movl %%fs:%1,%0"			\
+		    : "=r" (__i)					\
+		    : "m" (*(u_int *)(__pcpu_offset(name))));		\
+		__result = *(__pcpu_type(name) *)&__i;			\
+	} else {							\
+		__result = *__PCPU_PTR(name);				\
+	}								\
+									\
+	__result;							\
+})
+
+/*
+ * Sets the value of the per-cpu variable name to value val.
+ */
+#define	__PCPU_SET(name, val) ({					\
+	__pcpu_type(name) __val = (val);				\
+									\
+	if (sizeof(__val) == 1) {					\
+		u_char __b;						\
+		__b = *(u_char *)&__val;				\
+		__asm __volatile("movb %1,%%fs:%0"			\
+		    : "=m" (*(u_char *)(__pcpu_offset(name)))		\
+		    : "r" (__b));					\
+	} else if (sizeof(__val) == 2) {				\
+		u_short __w;						\
+		__w = *(u_short *)&__val;				\
+		__asm __volatile("movw %1,%%fs:%0"			\
+		    : "=m" (*(u_short *)(__pcpu_offset(name)))		\
+		    : "r" (__w));					\
+	} else if (sizeof(__val) == 4) {				\
+		u_int __i;						\
+		__i = *(u_int *)&__val;					\
+		__asm __volatile("movl %1,%%fs:%0"			\
+		    : "=m" (*(u_int *)(__pcpu_offset(name)))		\
+		    : "r" (__i));					\
+	} else {							\
+		*__PCPU_PTR(name) = __val;				\
+	}								\
+})
+
+#define	PCPU_GET(member)	__PCPU_GET(pc_ ## member)
+#define	PCPU_PTR(member)	__PCPU_PTR(pc_ ## member)
+#define	PCPU_SET(member, val)	__PCPU_SET(pc_ ## member, val)
 
 #endif	/* _KERNEL */
 
-#endif	/* ! _MACHINE_GLOBALDATA_H_ */
+#endif	/* ! _MACHINE_PCPU_H_ */
diff --git a/sys/amd64/include/proc.h b/sys/amd64/include/proc.h
index 911f6b8..5b50824 100644
--- a/sys/amd64/include/proc.h
+++ b/sys/amd64/include/proc.h
@@ -37,7 +37,6 @@
 #ifndef _MACHINE_PROC_H_
 #define	_MACHINE_PROC_H_
 
-#include <machine/globals.h>
 #include <machine/segments.h>
 
 struct proc_ldt {
diff --git a/sys/amd64/isa/npx.c b/sys/amd64/isa/npx.c
index 1238d67..ff8c4cf 100644
--- a/sys/amd64/isa/npx.c
+++ b/sys/amd64/isa/npx.c
@@ -237,7 +237,7 @@ npx_intr(dummy)
 #endif
 
 	/*
-	 * npxthread is normally non-null here.  In that case, schedule an
+	 * fpcurthread is normally non-null here.  In that case, schedule an
 	 * AST to finish the exception handling in the correct context
 	 * (this interrupt may occur after the thread has entered the
 	 * kernel via a syscall or an interrupt).  Otherwise, the npx
@@ -248,7 +248,7 @@ npx_intr(dummy)
 	 * that caused it and it will repeat.  We will eventually (usually
 	 * soon) win the race to handle the interrupt properly.
 	 */
-	td = PCPU_GET(npxthread);
+	td = PCPU_GET(fpcurthread);
 	if (td != NULL) {
 		td->td_pcb->pcb_flags |= PCB_NPXTRAP;
 		mtx_lock_spin(&sched_lock);
@@ -513,7 +513,7 @@ npxinit(control)
 	/*
 	 * fninit has the same h/w bugs as fnsave.  Use the detoxified
 	 * fnsave to throw away any junk in the fpu.  npxsave() initializes
-	 * the fpu and sets npxthread = NULL as important side effects.
+	 * the fpu and sets fpcurthread = NULL as important side effects.
 	 */
 	savecrit = critical_enter();
 	npxsave(&dummy);
@@ -540,7 +540,7 @@ npxexit(td)
 	critical_t savecrit;
 
 	savecrit = critical_enter();
-	if (td == PCPU_GET(npxthread))
+	if (td == PCPU_GET(fpcurthread))
 		npxsave(&PCPU_GET(curpcb)->pcb_save);
 	critical_exit(savecrit);
 #ifdef NPX_DEBUG
@@ -758,8 +758,8 @@ npxtrap()
 	u_long *exstat;
 
 	if (!npx_exists) {
-		printf("npxtrap: npxthread = %p, curthread = %p, npx_exists = %d\n",
-		       PCPU_GET(npxthread), curthread, npx_exists);
+		printf("npxtrap: fpcurthread = %p, curthread = %p, npx_exists = %d\n",
+		       PCPU_GET(fpcurthread), curthread, npx_exists);
 		panic("npxtrap from nowhere");
 	}
 	savecrit = critical_enter();
@@ -769,7 +769,7 @@ npxtrap()
 	 * state to memory.  Fetch the relevant parts of the state from
 	 * wherever they are.
 	 */
-	if (PCPU_GET(npxthread) != curthread) {
+	if (PCPU_GET(fpcurthread) != curthread) {
 		control = GET_FPU_CW(curthread);
 		status = GET_FPU_SW(curthread);
 	} else {
@@ -779,7 +779,7 @@ npxtrap()
 
 	exstat = GET_FPU_EXSW_PTR(curthread->td_pcb);
 	*exstat = status;
-	if (PCPU_GET(npxthread) != curthread)
+	if (PCPU_GET(fpcurthread) != curthread)
 		GET_FPU_SW(curthread) &= ~0x80bf;
 	else
 		fnclex();
@@ -790,7 +790,7 @@ npxtrap()
 /*
  * Implement device not available (DNA) exception
  *
- * It would be better to switch FP context here (if curthread != npxthread)
+ * It would be better to switch FP context here (if curthread != fpcurthread)
  * and not necessarily for every context switch, but it is too hard to
  * access foreign pcb's.
  */
@@ -802,9 +802,9 @@ npxdna()
 
 	if (!npx_exists)
 		return (0);
-	if (PCPU_GET(npxthread) != NULL) {
-		printf("npxdna: npxthread = %p, curthread = %p\n",
-		       PCPU_GET(npxthread), curthread);
+	if (PCPU_GET(fpcurthread) != NULL) {
+		printf("npxdna: fpcurthread = %p, curthread = %p\n",
+		       PCPU_GET(fpcurthread), curthread);
 		panic("npxdna");
 	}
 	s = critical_enter();
@@ -812,7 +812,7 @@ npxdna()
 	/*
 	 * Record new context early in case frstor causes an IRQ13.
 	 */
-	PCPU_SET(npxthread, curthread);
+	PCPU_SET(fpcurthread, curthread);
 
 	exstat = GET_FPU_EXSW_PTR(PCPU_GET(curpcb));
 	*exstat = 0;
@@ -844,13 +844,13 @@ npxdna()
  * after the process has entered the kernel.  It may even be delivered after
  * the fnsave here completes.  A spurious IRQ13 for the fnsave is handled in
  * the same way as a very-late-arriving non-spurious IRQ13 from user mode:
- * it is normally ignored at first because we set npxthread to NULL; it is
+ * it is normally ignored at first because we set fpcurthread to NULL; it is
  * normally retriggered in npxdna() after return to user mode.
  *
  * npxsave() must be called with interrupts disabled, so that it clears
- * npxthread atomically with saving the state.  We require callers to do the
+ * fpcurthread atomically with saving the state.  We require callers to do the
  * disabling, since most callers need to disable interrupts anyway to call
- * npxsave() atomically with checking npxthread.
+ * npxsave() atomically with checking fpcurthread.
  *
  * A previous version of npxsave() went to great lengths to excecute fnsave
  * with interrupts enabled in case executing it froze the CPU.  This case
@@ -866,7 +866,7 @@ npxsave(addr)
 	fpusave(addr);
 
 	start_emulating();
-	PCPU_SET(npxthread, NULL);
+	PCPU_SET(fpcurthread, NULL);
 }
 
 static void