Switch from TSS-per-process to TSS-per-CPU, placing the TSS right

next to the cpu's GDT, also making the double-fault stack per-CPU,
leaving it at the top of the page of the CPU's idle process.  Inline
pmap_activate() and pmap_deactivate() into the asm cpu_switchto
routine, adding a check for the new pmap already being marked as
active on the CPU.  Garbage collect the hasn't-been-used-in-years
GDT update IPI.

Tested by many; ok mikeb@, kettenis@

18 files changed, 121 insertions, 351 deletions

diff --git a/sys/arch/amd64/amd64/acpi_machdep.c b/sys/arch/amd64/amd64/acpi_machdep.c
index 1d6a29833c3..b521cf00ade 100644
--- a/sys/arch/amd64/amd64/acpi_machdep.c
+++ b/sys/arch/amd64/amd64/acpi_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: acpi_machdep.c,v 1.46 2010/10/06 16:37:29 deraadt Exp $	*/
+/*	$OpenBSD: acpi_machdep.c,v 1.47 2010/11/13 04:16:42 guenther Exp $	*/
 /*
  * Copyright (c) 2005 Thorsten Lockert <tholo@sigmasoft.com>
  *
@@ -303,7 +303,7 @@ acpi_resume_machdep(void)
 		p = ci->ci_schedstate.spc_idleproc;
 		pcb = &p->p_addr->u_pcb;
 
-		tf = (struct trapframe *)pcb->pcb_tss.tss_rsp0 - 1;
+		tf = (struct trapframe *)pcb->pcb_kstack - 1;
 		sf = (struct switchframe *)tf - 1;
 		sf->sf_r12 = (u_int64_t)sched_idle;
 		sf->sf_r13 = (u_int64_t)ci;
diff --git a/sys/arch/amd64/amd64/cpu.c b/sys/arch/amd64/amd64/cpu.c
index 6219a90cdea..0a29134d946 100644
--- a/sys/arch/amd64/amd64/cpu.c
+++ b/sys/arch/amd64/amd64/cpu.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.c,v 1.37 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: cpu.c,v 1.38 2010/11/13 04:16:42 guenther Exp $	*/
 /* $NetBSD: cpu.c,v 1.1 2003/04/26 18:39:26 fvdl Exp $ */
 
 /*-
@@ -280,9 +280,8 @@ cpu_attach(struct device *parent, struct device *self, void *aux)
 	pcb = ci->ci_idle_pcb = (struct pcb *) kstack;
 	memset(pcb, 0, USPACE);
 
-	pcb->pcb_tss.tss_rsp0 = kstack + USPACE - 16;
+	pcb->pcb_kstack = kstack + USPACE - 16;
 	pcb->pcb_rbp = pcb->pcb_rsp = kstack + USPACE - 16;
-	pcb->pcb_tss.tss_ist[0] = kstack + PAGE_SIZE - 16;
 	pcb->pcb_pmap = pmap_kernel();
 	pcb->pcb_cr0 = rcr0();
 	pcb->pcb_cr3 = pcb->pcb_pmap->pm_pdirpa;
diff --git a/sys/arch/amd64/amd64/gdt.c b/sys/arch/amd64/amd64/gdt.c
index aec3012d6dc..df975b5758a 100644
--- a/sys/arch/amd64/amd64/gdt.c
+++ b/sys/arch/amd64/amd64/gdt.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gdt.c,v 1.17 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: gdt.c,v 1.18 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: gdt.c,v 1.1 2003/04/26 18:39:28 fvdl Exp $	*/
 
 /*-
@@ -33,8 +33,6 @@
 /*
  * Modified to deal with variable-length entries for amd64 by
  * fvdl@wasabisystems.com, may 2001
- * XXX this file should be shared with the i386 code, the difference
- * can be hidden in macros.
  */
 
 #include <sys/param.h>
@@ -45,90 +43,9 @@
 #include <uvm/uvm.h>
 
 #include <machine/gdt.h>
+#include <machine/tss.h>
 #include <machine/pcb.h>
 
-int gdt_next;		/* next available slot for sweeping */
-int gdt_free;		/* next free slot; terminated with GNULL_SEL */
-
-struct mutex gdt_lock_store = MUTEX_INITIALIZER(IPL_HIGH);
-
-void gdt_init(void);
-int gdt_get_slot(void);
-void gdt_put_slot(int);
-
-/*
- * Lock and unlock the GDT.
- */
-#define gdt_lock()	(mtx_enter(&gdt_lock_store))
-#define gdt_unlock()	(mtx_leave(&gdt_lock_store))
-
-void
-set_mem_gdt(struct mem_segment_descriptor *sd, void *base, size_t limit,
-    int type, int dpl, int gran, int def32, int is64)
-{
-	CPU_INFO_ITERATOR cii;
-	struct cpu_info *ci;
-	int off;
-
-        set_mem_segment(sd, base, limit, type, dpl, gran, def32, is64);
-	off = (char *)sd - gdtstore;
-        CPU_INFO_FOREACH(cii, ci) {
-                if (ci->ci_gdt != NULL)
-			*(struct mem_segment_descriptor *)(ci->ci_gdt + off) =
-			    *sd;
-        }
-}
-
-void
-set_sys_gdt(struct sys_segment_descriptor *sd, void *base, size_t limit,
-    int type, int dpl, int gran)
-{
-	CPU_INFO_ITERATOR cii;
-	struct cpu_info *ci;
-	int off;
-
-        set_sys_segment(sd, base, limit, type, dpl, gran);
-	off = (char *)sd - gdtstore;
-        CPU_INFO_FOREACH(cii, ci) {
-                if (ci->ci_gdt != NULL)
-			*(struct sys_segment_descriptor *)(ci->ci_gdt + off) =
-			    *sd;
-        }
-}
-
-
-/*
- * Initialize the GDT.
- */
-void
-gdt_init(void)
-{
-	char *old_gdt;
-	struct vm_page *pg;
-	vaddr_t va;
-	struct cpu_info *ci = &cpu_info_primary;
-
-	gdt_next = 0;
-	gdt_free = GNULL_SEL;
-
-	old_gdt = gdtstore;
-	gdtstore = (char *)uvm_km_valloc(kernel_map, MAXGDTSIZ);
-	for (va = (vaddr_t)gdtstore; va < (vaddr_t)gdtstore + MAXGDTSIZ;
-	    va += PAGE_SIZE) {
-		pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_ZERO);
-		if (pg == NULL) {
-			panic("gdt_init: no pages");
-		}
-		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg),
-		    VM_PROT_READ | VM_PROT_WRITE);
-	}
-	bcopy(old_gdt, gdtstore, DYNSEL_START);
-	ci->ci_gdt = gdtstore;
-
-	gdt_init_cpu(ci);
-}
-
-#ifdef MULTIPROCESSOR
 /*
  * Allocate shadow GDT for a slave cpu.
  */
@@ -138,10 +55,13 @@ gdt_alloc_cpu(struct cpu_info *ci)
 	struct vm_page *pg;
 	vaddr_t va;
 
-	ci->ci_gdt = (char *)uvm_km_valloc(kernel_map, MAXGDTSIZ);
+	ci->ci_gdt = (char *)uvm_km_valloc(kernel_map,
+	    GDT_SIZE + sizeof(*ci->ci_tss));
+	ci->ci_tss = (void *)(ci->ci_gdt + GDT_SIZE);
 	uvm_map_pageable(kernel_map, (vaddr_t)ci->ci_gdt,
-            (vaddr_t)ci->ci_gdt + MAXGDTSIZ, FALSE, FALSE);
-	for (va = (vaddr_t)ci->ci_gdt; va < (vaddr_t)ci->ci_gdt + MAXGDTSIZ;
+            (vaddr_t)ci->ci_gdt + GDT_SIZE, FALSE, FALSE);
+	for (va = (vaddr_t)ci->ci_gdt;
+	    va < (vaddr_t)ci->ci_gdt + GDT_SIZE + sizeof(*ci->ci_tss);
 	    va += PAGE_SIZE) {
 		pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_ZERO);
 		if (pg == NULL)
@@ -149,10 +69,9 @@ gdt_alloc_cpu(struct cpu_info *ci)
 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg),
 		    VM_PROT_READ | VM_PROT_WRITE);
 	}
-	bzero(ci->ci_gdt, MAXGDTSIZ);
-	bcopy(gdtstore, ci->ci_gdt, MAXGDTSIZ);
+	bcopy(gdtstore, ci->ci_gdt, GDT_SIZE);
+	bzero(ci->ci_tss, sizeof(*ci->ci_tss));
 }
-#endif	/* MULTIPROCESSOR */
 
 
 /*
@@ -164,107 +83,10 @@ gdt_init_cpu(struct cpu_info *ci)
 {
 	struct region_descriptor region;
 
-	setregion(&region, ci->ci_gdt, (u_int16_t)(MAXGDTSIZ - 1));
-	lgdt(&region);
-}
-
-#ifdef MULTIPROCESSOR
-
-void
-gdt_reload_cpu(struct cpu_info *ci)
-{
-	struct region_descriptor region;
+	set_sys_segment(GDT_ADDR_SYS(ci->ci_gdt, GPROC0_SEL), ci->ci_tss,
+	    sizeof (struct x86_64_tss)-1, SDT_SYS386TSS, SEL_KPL, 0);
 
-	setregion(&region, ci->ci_gdt, MAXGDTSIZ - 1);
+	setregion(&region, ci->ci_gdt, GDT_SIZE - 1);
 	lgdt(&region);
+	ltr(GSYSSEL(GPROC0_SEL, SEL_KPL));
 }
-#endif
-
-/*
- * Allocate a GDT slot as follows:
- * 1) If there are entries on the free list, use those.
- * 2) If there are fewer than MAXGDTSIZ entries in use, there are free slots
- *    near the end that we can sweep through.
- */
-int
-gdt_get_slot(void)
-{
-	int slot;
-	struct sys_segment_descriptor *gdt;
-
-	gdt = (struct sys_segment_descriptor *)&gdtstore[DYNSEL_START];
-
-	gdt_lock();
-
-	if (gdt_free != GNULL_SEL) {
-		slot = gdt_free;
-		gdt_free = gdt[slot].sd_xx3;	/* XXXfvdl res. field abuse */
-	} else {
-		if (gdt_next >= MAXGDTSIZ)
-			panic("gdt_get_slot: out of GDT descriptors");
-		slot = gdt_next++;
-	}
-
-	gdt_unlock();
-	return (slot);
-}
-
-/*
- * Deallocate a GDT slot, putting it on the free list.
- */
-void
-gdt_put_slot(int slot)
-{
-	struct sys_segment_descriptor *gdt;
-
-	gdt = (struct sys_segment_descriptor *)&gdtstore[DYNSEL_START];
-
-	gdt_lock();
-
-	gdt[slot].sd_type = SDT_SYSNULL;
-	gdt[slot].sd_xx3 = gdt_free;
-	gdt_free = slot;
-
-	gdt_unlock();
-}
-
-int
-tss_alloc(struct pcb *pcb)
-{
-	int slot;
-	struct sys_segment_descriptor *gdt;
-
-	gdt = (struct sys_segment_descriptor *)&gdtstore[DYNSEL_START];
-
-	slot = gdt_get_slot();
-#if 0
-	printf("tss_alloc: slot %d addr %p\n", slot, &gdt[slot]);
-#endif
-	set_sys_gdt(&gdt[slot], &pcb->pcb_tss, sizeof (struct x86_64_tss)-1,
-	    SDT_SYS386TSS, SEL_KPL, 0);
-#if 0
-	printf("lolimit %lx lobase %lx type %lx dpl %lx p %lx hilimit %lx\n"
-	       "xx1 %lx gran %lx hibase %lx xx2 %lx zero %lx xx3 %lx pad %lx\n",
-		(unsigned long)gdt[slot].sd_lolimit,
-		(unsigned long)gdt[slot].sd_lobase,
-		(unsigned long)gdt[slot].sd_type,
-		(unsigned long)gdt[slot].sd_dpl,
-		(unsigned long)gdt[slot].sd_p,
-		(unsigned long)gdt[slot].sd_hilimit,
-		(unsigned long)gdt[slot].sd_xx1,
-		(unsigned long)gdt[slot].sd_gran,
-		(unsigned long)gdt[slot].sd_hibase,
-		(unsigned long)gdt[slot].sd_xx2,
-		(unsigned long)gdt[slot].sd_zero,
-		(unsigned long)gdt[slot].sd_xx3);
-#endif
-	return GDYNSEL(slot, SEL_KPL);
-}
-
-void
-tss_free(int sel)
-{
-
-	gdt_put_slot(IDXDYNSEL(sel));
-}
-
diff --git a/sys/arch/amd64/amd64/genassym.cf b/sys/arch/amd64/amd64/genassym.cf
index f82dabe9c33..b882b230349 100644
--- a/sys/arch/amd64/amd64/genassym.cf
+++ b/sys/arch/amd64/amd64/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.24 2010/10/26 05:49:10 guenther Exp $
+#	$OpenBSD: genassym.cf,v 1.25 2010/11/13 04:16:42 guenther Exp $
 #	Written by Artur Grabowski art@openbsd.org, Public Domain
 
 include <sys/param.h>
@@ -15,6 +15,7 @@ include <machine/pte.h>
 include <machine/vmparam.h>
 include <machine/intr.h>
 include <machine/pic.h>
+include <machine/tss.h>
 include <machine/i82093var.h>
 include <machine/mutex.h>
 
@@ -36,7 +37,6 @@ member	p_addr
 member	p_priority
 member	p_stat
 member	p_wchan
-member	P_MD_TSS_SEL	p_md.md_tss_sel
 member	P_MD_REGS	p_md.md_regs
 member	P_MD_FLAGS	p_md.md_flags
 member	P_MD_ASTPENDING	p_md.md_astpending
@@ -82,22 +82,29 @@ define  FRAMESIZE               sizeof(struct trapframe)
 
 struct pcb
 member	pcb_cr3
-member	pcb_rbp
 member	pcb_rsp
-member	PCB_RSP0	pcb_tss.tss_rsp0
-member	pcb_cr0
+member	pcb_rbp
+member	pcb_kstack
 member	pcb_onfault
 member	pcb_fpcpu
+member	pcb_pmap
+member	pcb_cr0
+
+struct pmap
+member	pm_cpus
+
+struct x86_64_tss
+member	tss_rsp0
 
 struct	cpu_info
 member	CPU_INFO_SCRATCH	ci_scratch
 member  CPU_INFO_SELF		ci_self
+member  CPU_INFO_CPUID		ci_cpuid
 member  CPU_INFO_APICID		ci_apicid
 member  CPU_INFO_RESCHED	ci_want_resched
 member  CPU_INFO_CURPROC	ci_curproc
 member  CPU_INFO_CURPCB		ci_curpcb
 member  CPU_INFO_IDLE_PCB	ci_idle_pcb
-member  CPU_INFO_IDLE_TSS_SEL	ci_idle_tss_sel
 member	CPU_INFO_ILEVEL		ci_ilevel
 member	CPU_INFO_IDEPTH		ci_idepth
 member	CPU_INFO_ISOURCES	ci_isources
@@ -107,6 +114,7 @@ ifdef DIAGNOSTIC
 member	CPU_INFO_MUTEX_LEVEL	ci_mutex_level
 endif
 member	CPU_INFO_GDT		ci_gdt
+member	CPU_INFO_TSS		ci_tss
 
 struct	intrsource
 member	is_recurse
diff --git a/sys/arch/amd64/amd64/ipifuncs.c b/sys/arch/amd64/amd64/ipifuncs.c
index fdf6373cc9f..6ab5a412390 100644
--- a/sys/arch/amd64/amd64/ipifuncs.c
+++ b/sys/arch/amd64/amd64/ipifuncs.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ipifuncs.c,v 1.14 2010/10/02 23:14:32 deraadt Exp $	*/
+/*	$OpenBSD: ipifuncs.c,v 1.15 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: ipifuncs.c,v 1.1 2003/04/26 18:39:28 fvdl Exp $ */
 
 /*-
@@ -54,7 +54,6 @@
 #include <machine/i82489reg.h>
 #include <machine/i82489var.h>
 #include <machine/mtrr.h>
-#include <machine/gdt.h>
 #include <machine/fpu.h>
 
 #include <ddb/db_output.h>
@@ -80,13 +79,12 @@ void (*ipifunc[X86_NIPI])(struct cpu_info *) =
 	x86_64_ipi_synch_fpu,
 	NULL,
 	x86_64_ipi_reload_mtrr,
-	gdt_reload_cpu,
+	x86_setperf_ipi,
 #ifdef DDB
 	x86_ipi_db,
 #else
 	NULL,
 #endif
-	x86_setperf_ipi,
 };
 
 void
diff --git a/sys/arch/amd64/amd64/locore.S b/sys/arch/amd64/amd64/locore.S
index e5270186815..556b51a35a9 100644
--- a/sys/arch/amd64/amd64/locore.S
+++ b/sys/arch/amd64/amd64/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.42 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: locore.S,v 1.43 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: locore.S,v 1.13 2004/03/25 18:33:17 drochner Exp $	*/
 
 /*
@@ -741,6 +741,8 @@ ENTRY(cpu_switchto)
 	movb	$SONPROC,P_STAT(%r12)	# p->p_stat = SONPROC
 	SET_CURPROC(%r12,%rcx)
 
+	movl	CPUVAR(CPUID),%edi
+
 	/* If old proc exited, don't bother. */
 	testq	%r13,%r13
 	jz	switch_exited
@@ -752,13 +754,16 @@ ENTRY(cpu_switchto)
 	 *   %rax, %rcx - scratch
 	 *   %r13 - old proc, then old pcb
 	 *   %r12 - new proc
+	 *   %edi - cpuid
 	 */
 
-	movq	%r13,%rdi
-	call	pmap_deactivate
-
 	movq	P_ADDR(%r13),%r13
 
+	/* clear the old pmap's bit for the cpu */
+	movq	PCB_PMAP(%r13),%rcx
+	lock
+	btrl	%edi,PM_CPUS(%rcx)
+
 	/* Save stack pointers. */
 	movq	%rsp,PCB_RSP(%r13)
 	movq	%rbp,PCB_RBP(%r13)
@@ -781,30 +786,28 @@ switch_exited:
 	movq	PCB_RSP(%r13),%rsp
 	movq	PCB_RBP(%r13),%rbp
 
-#if 0
+	movq	CPUVAR(TSS),%rcx
+	movq	PCB_KSTACK(%r13),%rdx
+	movq	%rdx,TSS_RSP0(%rcx)
+
+	movq	PCB_CR3(%r13),%rax
+	movq	%rax,%cr3
+
 	/* Don't bother with the rest if switching to a system process. */
 	testl	$P_SYSTEM,P_FLAG(%r12)
 	jnz	switch_restored
-#endif
 
-	/* Load TSS info. */
-#ifdef MULTIPROCESSOR
-	movq    CPUVAR(GDT),%rax
-#else   
-	movq	_C_LABEL(gdtstore)(%rip),%rax
+	/* set the new pmap's bit for the cpu */
+	movl	CPUVAR(CPUID),%edi
+	movq	PCB_PMAP(%r13),%rcx
+	movl	PM_CPUS(%rcx),%eax
+	lock
+	btsl	%edi,PM_CPUS(%rcx)
+#ifdef DIAGNOSTIC
+	jc	_C_LABEL(switch_pmcpu_set)
 #endif
-	movl	P_MD_TSS_SEL(%r12),%edx
-
-	/* Switch TSS. Reset "task busy" flag before */
-	andl	$~0x0200,4(%rax,%rdx, 1)
-	ltr	%dx
-
-	movq	%r12,%rdi
-	call	_C_LABEL(pmap_activate)
 
-#if 0
 switch_restored:
-#endif
 	/* Restore cr0 (including FPU state). */
 	movl	PCB_CR0(%r13),%ecx
 #ifdef MULTIPROCESSOR
@@ -876,6 +879,11 @@ NENTRY(switch_error3)
 	call	_C_LABEL(panic)
 	/* NOTREACHED */
 1:	.asciz	"cpu_switch 3"
+NENTRY(switch_pmcpu_set)
+	movabsq	$1f,%rdi
+	call	_C_LABEL(panic)
+	/* NOTREACHED */
+1:	.asciz	"activate already active pmap"
 #endif /* DIAGNOSTIC */
 
 /*
@@ -900,7 +908,7 @@ IDTVEC(syscall)
 	swapgs
 	movq	%r15,CPUVAR(SCRATCH)
 	movq	CPUVAR(CURPCB),%r15
-	movq	PCB_RSP0(%r15),%r15
+	movq	PCB_KSTACK(%r15),%r15
 	xchgq	%r15,%rsp
 	sti
 
diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index 0572a2b3cae..b5eaae0e9b7 100644
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.126 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.127 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: machdep.c,v 1.3 2003/05/07 22:58:18 fvdl Exp $	*/
 
 /*-
@@ -111,6 +111,7 @@
 #include <machine/mpbiosvar.h>
 #include <machine/reg.h>
 #include <machine/kcore.h>
+#include <machine/tss.h>
 
 #include <dev/isa/isareg.h>
 #include <machine/isa_machdep.h>
@@ -360,30 +361,25 @@ cpu_startup(void)
 }
 
 /*
- * Set up proc0's TSS
+ * Set up proc0's PCB and the cpu's TSS.
  */
 void
 x86_64_proc0_tss_ldt_init(void)
 {
 	struct pcb *pcb;
-	int x;
-
-	gdt_init();
 
 	cpu_info_primary.ci_curpcb = pcb = &proc0.p_addr->u_pcb;
-
-	pcb->pcb_tss.tss_iobase =
-	    (u_int16_t)((caddr_t)pcb->pcb_iomap - (caddr_t)&pcb->pcb_tss);
-	for (x = 0; x < sizeof(pcb->pcb_iomap) / 4; x++)
-		pcb->pcb_iomap[x] = 0xffffffff;
-
 	pcb->pcb_cr0 = rcr0();
-	pcb->pcb_tss.tss_rsp0 = (u_int64_t)proc0.p_addr + USPACE - 16;
-	pcb->pcb_tss.tss_ist[0] = (u_int64_t)proc0.p_addr + PAGE_SIZE;
-	proc0.p_md.md_regs = (struct trapframe *)pcb->pcb_tss.tss_rsp0 - 1;
-	proc0.p_md.md_tss_sel = tss_alloc(pcb);
+	pcb->pcb_kstack = (u_int64_t)proc0.p_addr + USPACE - 16;
+	proc0.p_md.md_regs = (struct trapframe *)pcb->pcb_kstack - 1;
 
-	ltr(proc0.p_md.md_tss_sel);
+	/* an empty iomap, by setting its offset to the TSS limit */
+	cpu_info_primary.ci_tss->tss_iobase = sizeof(struct x86_64_tss);
+	cpu_info_primary.ci_tss->tss_rsp0 = pcb->pcb_kstack;
+	cpu_info_primary.ci_tss->tss_ist[0] =
+	    (u_int64_t)proc0.p_addr + PAGE_SIZE - 16;
+
+	ltr(GSYSSEL(GPROC0_SEL, SEL_KPL));
 	lldt(0);
 }
 
@@ -395,17 +391,13 @@ x86_64_proc0_tss_ldt_init(void)
 void    
 x86_64_init_pcb_tss_ldt(struct cpu_info *ci)   
 {        
-	int x;      
 	struct pcb *pcb = ci->ci_idle_pcb;
  
-	pcb->pcb_tss.tss_iobase =
-	    (u_int16_t)((caddr_t)pcb->pcb_iomap - (caddr_t)&pcb->pcb_tss);
-	for (x = 0; x < sizeof(pcb->pcb_iomap) / 4; x++)
-		pcb->pcb_iomap[x] = 0xffffffff;
+	ci->ci_tss->tss_iobase = sizeof(*ci->ci_tss);
+	ci->ci_tss->tss_rsp0 = pcb->pcb_kstack;
+	ci->ci_tss->tss_ist[0] = pcb->pcb_kstack - USPACE + PAGE_SIZE;
 
 	pcb->pcb_cr0 = rcr0();
-        
-        ci->ci_idle_tss_sel = tss_alloc(pcb);
 }       
 #endif	/* MULTIPROCESSOR */
 
@@ -1471,7 +1463,8 @@ init_x86_64(paddr_t first_avail)
 #endif
 
 	idt = (struct gate_descriptor *)idt_vaddr;
-	gdtstore = (char *)(idt + NIDT);
+	cpu_info_primary.ci_tss = (void *)(idt + NIDT);
+	gdtstore = (void *)(cpu_info_primary.ci_tss + 1);
 
 	/* make gdt gates and memory segments */
 	set_mem_segment(GDT_ADDR_MEM(gdtstore, GCODE_SEL), 0,
@@ -1489,6 +1482,10 @@ init_x86_64(paddr_t first_avail)
 	set_mem_segment(GDT_ADDR_MEM(gdtstore, GUCODE_SEL), 0,
 	    atop(VM_MAXUSER_ADDRESS) - 1, SDT_MEMERA, SEL_UPL, 1, 0, 1);
 
+	set_sys_segment(GDT_ADDR_SYS(gdtstore, GPROC0_SEL),
+	    cpu_info_primary.ci_tss, sizeof (struct x86_64_tss)-1,
+	    SDT_SYS386TSS, SEL_KPL, 0);
+
 	/* exceptions */
 	for (x = 0; x < 32; x++) {
 		ist = (x == 8) ? 1 : 0;
@@ -1503,7 +1500,7 @@ init_x86_64(paddr_t first_avail)
 	    GSEL(GCODE_SEL, SEL_KPL));
 	idt_allocmap[128] = 1;
 
-	setregion(&region, gdtstore, DYNSEL_START - 1);
+	setregion(&region, gdtstore, GDT_SIZE - 1);
 	lgdt(&region);
 
 	cpu_init_idt();
@@ -1528,9 +1525,6 @@ init_x86_64(paddr_t first_avail)
 	}
 #endif
 
-        /* Make sure maxproc is sane */ 
-        if (maxproc > cpu_maxproc())
-                maxproc = cpu_maxproc();
 }
 
 #ifdef KGDB
@@ -1704,15 +1698,6 @@ idt_vec_free(int vec)
 	simple_unlock(&idt_lock);
 }
 
-/*
- * Number of processes is limited by number of available GDT slots.
- */
-int
-cpu_maxproc(void)
-{
-	return (MAXGDTSIZ - DYNSEL_START) / 16;
-}
-
 #ifdef DIAGNOSTIC
 void
 splassert_check(int wantipl, const char *func)
diff --git a/sys/arch/amd64/amd64/mptramp.S b/sys/arch/amd64/amd64/mptramp.S
index 98d9f861f2d..aece9c33443 100644
--- a/sys/arch/amd64/amd64/mptramp.S
+++ b/sys/arch/amd64/amd64/mptramp.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mptramp.S,v 1.6 2010/04/01 19:47:59 kettenis Exp $	*/
+/*	$OpenBSD: mptramp.S,v 1.7 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: mptramp.S,v 1.1 2003/04/26 18:39:30 fvdl Exp $	*/
 
 /*-
@@ -83,7 +83,6 @@
 #include <machine/segments.h>
 #include <machine/mpbiosvar.h>
 #include <machine/i82489reg.h>
-#include <machine/gdt.h>
 
 #define _RELOC(x)       ((x) - KERNBASE)
 #define RELOC(x)        _RELOC(_C_LABEL(x))
@@ -224,7 +223,7 @@ _C_LABEL(cpu_spinup_trampoline_end):	#end of code copied to MP_TRAMPOLINE
 	movq	PCB_RBP(%rsi),%rbp
 
 	movq	CPU_INFO_GDT(%rdi),%rax
-	movw	$(MAXGDTSIZ-1),-10(%rsp)
+	movw	$(GDT_SIZE-1),-10(%rsp)
 	movq	%rax,-8(%rsp)
 	lgdt	-10(%rsp)
 
diff --git a/sys/arch/amd64/amd64/pmap.c b/sys/arch/amd64/amd64/pmap.c
index a49ca83dc68..8a754946bc3 100644
--- a/sys/arch/amd64/amd64/pmap.c
+++ b/sys/arch/amd64/amd64/pmap.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.c,v 1.57 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: pmap.c,v 1.58 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: pmap.c,v 1.3 2003/05/08 18:13:13 thorpej Exp $	*/
 
 /*
@@ -1118,7 +1118,7 @@ pmap_reference(struct pmap *pmap)
 /*
  * pmap_activate: activate a process' pmap (fill in %cr3)
  *
- * => called from cpu_switch()
+ * => called from cpu_fork() and when switching pmaps during exec
  * => if p is the curproc, then load it into the MMU
  */
 
diff --git a/sys/arch/amd64/amd64/vm_machdep.c b/sys/arch/amd64/amd64/vm_machdep.c
index 6237a0e9080..19a669d5514 100644
--- a/sys/arch/amd64/amd64/vm_machdep.c
+++ b/sys/arch/amd64/amd64/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.23 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.24 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.1 2003/04/26 18:39:33 fvdl Exp $	*/
 
 /*-
@@ -111,15 +111,13 @@ cpu_fork(struct proc *p1, struct proc *p2, void *stack, size_t stacksize,
 	 */
 	pmap_activate(p2);
 
-	/* Fix up the TSS. */
-	pcb->pcb_tss.tss_rsp0 = (u_int64_t)p2->p_addr + USPACE - 16;
-	pcb->pcb_tss.tss_ist[0] = (u_int64_t)p2->p_addr + PAGE_SIZE - 16;
-	p2->p_md.md_tss_sel = tss_alloc(pcb);
+	/* Record where this process's kernel stack is */
+	pcb->pcb_kstack = (u_int64_t)p2->p_addr + USPACE - 16;
 
 	/*
 	 * Copy the trapframe.
 	 */
-	p2->p_md.md_regs = tf = (struct trapframe *)pcb->pcb_tss.tss_rsp0 - 1;
+	p2->p_md.md_regs = tf = (struct trapframe *)pcb->pcb_kstack - 1;
 	*tf = *p1->p_md.md_regs;
 
 	setredzone(p2);
@@ -160,7 +158,6 @@ cpu_exit(struct proc *p)
 		mtrr_clean(p);
 
 	pmap_deactivate(p);
-	tss_free(p->p_md.md_tss_sel);
 	sched_exit(p);
 }
 
diff --git a/sys/arch/amd64/conf/files.amd64 b/sys/arch/amd64/conf/files.amd64
index 519b24be01e..5d5931aa8c2 100644
--- a/sys/arch/amd64/conf/files.amd64
+++ b/sys/arch/amd64/conf/files.amd64
@@ -1,4 +1,4 @@
-#	$OpenBSD: files.amd64,v 1.58 2010/09/07 16:22:48 mikeb Exp $
+#	$OpenBSD: files.amd64,v 1.59 2010/11/13 04:16:42 guenther Exp $
 
 maxpartitions 16
 maxusers 2 16 128
@@ -6,7 +6,7 @@ maxusers 2 16 128
 file	arch/amd64/amd64/autoconf.c
 file	arch/amd64/amd64/conf.c
 file	arch/amd64/amd64/disksubr.c		disk
-file	arch/amd64/amd64/gdt.c
+file	arch/amd64/amd64/gdt.c			multiprocessor
 file	arch/amd64/amd64/machdep.c
 file	arch/amd64/amd64/identcpu.c
 file	arch/amd64/amd64/via.c
diff --git a/sys/arch/amd64/include/cpu.h b/sys/arch/amd64/include/cpu.h
index b2a1ab3cd7d..ad8df981d52 100644
--- a/sys/arch/amd64/include/cpu.h
+++ b/sys/arch/amd64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.58 2010/10/02 23:13:27 deraadt Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.59 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 2003/04/26 18:39:39 fvdl Exp $	*/
 
 /*-
@@ -43,7 +43,6 @@
  */
 #include <machine/frame.h>
 #include <machine/segments.h>
-#include <machine/tss.h>
 #include <machine/intrdefs.h>
 #include <machine/cacheinfo.h>
 
@@ -59,6 +58,7 @@
 
 #ifdef _KERNEL
 
+struct x86_64_tss;
 struct cpu_info {
 	struct device *ci_dev;
 	struct cpu_info *ci_self;
@@ -79,7 +79,6 @@ struct cpu_info {
 
 	struct pcb *ci_curpcb;
 	struct pcb *ci_idle_pcb;
-	int ci_idle_tss_sel;
 
 	struct intrsource *ci_isources[MAX_INTR_SOURCES];
 	u_int32_t	ci_ipending;
@@ -110,6 +109,7 @@ struct cpu_info {
 
 	struct x86_cache_info ci_cinfo[CAI_COUNT];
 
+	struct	x86_64_tss *ci_tss;
 	char		*ci_gdt;
 
 	volatile int	ci_ddb_paused;
@@ -125,10 +125,6 @@ struct cpu_info {
 #define CI_SETPERF_INTRANSIT	2
 #define CI_SETPERF_DONE		3
 
-	struct x86_64_tss	ci_doubleflt_tss;
-
-	char *ci_doubleflt_stack;
-
 	struct ksensordev	ci_sensordev;
 	struct ksensor		ci_sensor;
 };
@@ -272,7 +268,6 @@ void cpu_probe_features(struct cpu_info *);
 
 /* machdep.c */
 void	dumpconf(void);
-int	cpu_maxproc(void);
 void	cpu_reset(void);
 void	x86_64_proc0_tss_ldt_init(void);
 void	x86_64_bufinit(void);
diff --git a/sys/arch/amd64/include/gdt.h b/sys/arch/amd64/include/gdt.h
index b52bc98e3eb..65a116e8bc1 100644
--- a/sys/arch/amd64/include/gdt.h
+++ b/sys/arch/amd64/include/gdt.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: gdt.h,v 1.4 2009/02/03 11:24:19 mikeb Exp $	*/
+/*	$OpenBSD: gdt.h,v 1.5 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: gdt.h,v 1.1 2003/04/26 18:39:40 fvdl Exp $	*/
 
 /*-
@@ -30,29 +30,5 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef _LOCORE
-struct proc;
-struct pmap;
-
-void gdt_init(void);
 void gdt_init_cpu(struct cpu_info *);
-void gdt_reload_cpu(struct cpu_info *);
 void gdt_alloc_cpu(struct cpu_info *);
-
-int tss_alloc(struct pcb *);
-void tss_free(int);
-
-void ldt_alloc(struct pmap *, char *, size_t);
-void ldt_free(struct pmap *);
-
-void set_mem_gdt(struct mem_segment_descriptor *, void *, size_t,
-		 int, int, int, int, int);
-void set_sys_gdt(struct sys_segment_descriptor *, void *, size_t, int, int,
-		 int);
-#endif
-
-/*
- * Maximum GDT size.  It cannot exceed 65536 since the selector field of
- * a descriptor is just 16 bits, and used as free list link.
- */
-#define MAXGDTSIZ       65536
diff --git a/sys/arch/amd64/include/intrdefs.h b/sys/arch/amd64/include/intrdefs.h
index c61cc1f1c45..96a196ff7cd 100644
--- a/sys/arch/amd64/include/intrdefs.h
+++ b/sys/arch/amd64/include/intrdefs.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: intrdefs.h,v 1.7 2010/05/22 21:31:05 deraadt Exp $	*/
+/*	$OpenBSD: intrdefs.h,v 1.8 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: intrdefs.h,v 1.2 2003/05/04 22:01:56 fvdl Exp $	*/
 
 #ifndef _AMD64_INTRDEFS_H
@@ -75,16 +75,14 @@
 #define X86_IPI_SYNCH_FPU		0x00000008
 #define X86_IPI_TLB			0x00000010
 #define X86_IPI_MTRR			0x00000020
-#define X86_IPI_GDT			0x00000040
+#define X86_IPI_SETPERF			0x00000040
 #define X86_IPI_DDB			0x00000080
-#define X86_IPI_SETPERF			0x00000100
 
-#define X86_NIPI			9
+#define X86_NIPI			8
 
 #define X86_IPI_NAMES { "halt IPI", "nop IPI", "FPU flush IPI", \
 			 "FPU synch IPI", "TLB shootdown IPI", \
-			 "MTRR update IPI", "GDT update IPI", "ddb IPI", \
-			 "setperf IPI"}
+			 "MTRR update IPI", "setperf IPI", "ddb IPI" }
 
 #define IREENT_MAGIC	0x18041969
 
diff --git a/sys/arch/amd64/include/pcb.h b/sys/arch/amd64/include/pcb.h
index db8b0cab91a..8151dc666d4 100644
--- a/sys/arch/amd64/include/pcb.h
+++ b/sys/arch/amd64/include/pcb.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pcb.h,v 1.6 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: pcb.h,v 1.7 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: pcb.h,v 1.1 2003/04/26 18:39:45 fvdl Exp $	*/
 
 /*-
@@ -73,36 +73,23 @@
 
 #include <sys/signal.h>
 
-#include <machine/segments.h>
 #include <machine/tss.h>
 #include <machine/fpu.h>
-#include <machine/sysarch.h>
-
-#define	NIOPORTS	1024		/* # of ports we allow to be mapped */
 
 /*
  * Please note that the pcb_savefpu field in struct below must be
  * on a 16-byte boundary.
  */
 struct pcb {
-	/*
-	 * XXXfvdl
-	 * It's overkill to have a TSS here, as it's only needed
-	 * for compatibility processes who use an I/O permission map.
-	 * The pcb fields below are not in the TSS anymore (and there's
-	 * not enough room in the TSS to store them all)
-	 * Should just make this a pointer and allocate.
-	 */
-	struct	x86_64_tss pcb_tss;
-	u_int64_t pcb_cr3;
-	u_int64_t pcb_rsp;
-	u_int64_t pcb_rbp;
 	struct	savefpu pcb_savefpu;	/* floating point state */
-	int	pcb_cr0;		/* saved image of CR0 */
+	u_int64_t	pcb_cr3;
+	u_int64_t	pcb_rsp;
+	u_int64_t	pcb_rbp;
+	u_int64_t	pcb_kstack;	/* kernel stack address */
 	caddr_t	pcb_onfault;		/* copyin/out fault recovery */
-	struct cpu_info *pcb_fpcpu;	/* cpu holding our fp state. */
-	unsigned pcb_iomap[NIOPORTS/32];	/* I/O bitmap */
-	struct pmap *pcb_pmap;		/* back pointer to our pmap */
+	struct	cpu_info *pcb_fpcpu;	/* cpu holding our fp state. */
+	struct	pmap *pcb_pmap;		/* back pointer to our pmap */
+	int	pcb_cr0;		/* saved image of CR0 */
 };
 
 /*    
diff --git a/sys/arch/amd64/include/proc.h b/sys/arch/amd64/include/proc.h
index fb662a6ecd7..46ca006f6b4 100644
--- a/sys/arch/amd64/include/proc.h
+++ b/sys/arch/amd64/include/proc.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: proc.h,v 1.3 2005/12/13 00:18:19 jsg Exp $	*/
+/*	$OpenBSD: proc.h,v 1.4 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: proc.h,v 1.1 2003/04/26 18:39:46 fvdl Exp $	*/
 
 /*
@@ -42,7 +42,6 @@
  */
 struct mdproc {
 	struct	trapframe *md_regs;	/* registers on current frame */
-	int	md_tss_sel;		/* TSS selector */
 	int	md_flags;
 	__volatile int md_astpending;
 };
diff --git a/sys/arch/amd64/include/segments.h b/sys/arch/amd64/include/segments.h
index 79f36ed301f..87650c8388a 100644
--- a/sys/arch/amd64/include/segments.h
+++ b/sys/arch/amd64/include/segments.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: segments.h,v 1.7 2010/10/26 05:49:10 guenther Exp $	*/
+/*	$OpenBSD: segments.h,v 1.8 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: segments.h,v 1.1 2003/04/26 18:39:47 fvdl Exp $	*/
 
 /*-
@@ -61,16 +61,14 @@
 #define	ISLDT(s)	((s) & SEL_LDT)	/* is it local or global */
 #define	SEL_LDT		4		/* local descriptor table */	
 
-/* Dynamically allocated TSSs start (byte offset) */
 #define SYSSEL_START	(NGDT_MEM << 3)
-#define DYNSEL_START	(SYSSEL_START + (NGDT_SYS << 4))
+#define GDT_SIZE	(SYSSEL_START + (NGDT_SYS << 4))
 
 /*
  * These define the index not from the start of the GDT, but from
  * the part of the GDT that they're allocated from.
  * First NGDT_MEM entries are 8-byte descriptors for CS and DS.
- *
- * The rest is 16-byte descriptors for TSSs
+ * Next NGDT_SYS entries are 16-byte descriptors defining TSSs.
  */
 
 #define	IDXSEL(s)	(((s) >> 3) & 0x1fff)
@@ -249,7 +247,7 @@ void cpu_init_idt(void);
  * The code and data descriptors must come first. There
  * are NGDT_MEM of them.
  *
- * Then come the predefined LDT (and possibly TSS) descriptors.
+ * Then comes the predefined TSS descriptor.
  * There are NGDT_SYS of them.
  *
  * The particular order of the UCODE32, UDATA, and UCODE descriptors is 
@@ -263,7 +261,8 @@ void cpu_init_idt(void);
 #define	GUCODE_SEL	5	/* User code descriptor */
 #define NGDT_MEM 6
 
-#define NGDT_SYS	0
+#define	GPROC0_SEL	0	/* common TSS */
+#define NGDT_SYS	1
 
 #define GDT_SYS_OFFSET	(NGDT_MEM << 3)
 
diff --git a/sys/arch/amd64/include/tss.h b/sys/arch/amd64/include/tss.h
index c674e0952dd..22b44bca8b2 100644
--- a/sys/arch/amd64/include/tss.h
+++ b/sys/arch/amd64/include/tss.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tss.h,v 1.1 2004/01/28 01:39:39 mickey Exp $	*/
+/*	$OpenBSD: tss.h,v 1.2 2010/11/13 04:16:42 guenther Exp $	*/
 /*	$NetBSD: tss.h,v 1.1 2003/04/26 18:39:49 fvdl Exp $	*/
 
 /*
@@ -49,7 +49,7 @@ struct x86_64_tss {
 	u_int32_t	tss_reserved1;
 	u_int64_t	tss_rsp0;
 	u_int64_t	tss_rsp1;
-	u_int64_t	tss_rsp3;
+	u_int64_t	tss_rsp2;
 	u_int32_t	tss_reserved2;
 	u_int32_t	tss_reserved3;
 	u_int64_t	tss_ist[7];