diff --git a/sys/amd64/amd64/cpu_switch.S b/sys/amd64/amd64/cpu_switch.S
index a0b11f8..99d6716 100644
--- a/sys/amd64/amd64/cpu_switch.S
+++ b/sys/amd64/amd64/cpu_switch.S
@@ -266,7 +266,7 @@ store_seg:
 	movl	%es,PCB_ES(%r8)
 	movl	%fs,PCB_FS(%r8)
 	jmp	done_store_seg
-2:	movq	PCB_GS32P(%r8),%rax
+2:	movq	PCPU(GS32P),%rax
 	movq	(%rax),%rax
 	movq	%rax,PCB_GS32SD(%r8)
 	jmp	1b
@@ -283,7 +283,7 @@ load_seg:
 	movl	PCB_FS(%r8),%fs
 	jmp	done_load_seg
 	/* Restore userland %gs while preserving kernel gsbase */
-2:	movq	PCB_GS32P(%r8),%rax
+2:	movq	PCPU(GS32P),%rax
 	movq	PCB_GS32SD(%r8),%rcx
 	movq	%rcx,(%rax)
 	jmp	1b
diff --git a/sys/amd64/amd64/genassym.c b/sys/amd64/amd64/genassym.c
index fec0380..f749dd6 100644
--- a/sys/amd64/amd64/genassym.c
+++ b/sys/amd64/amd64/genassym.c
@@ -147,7 +147,6 @@ ASSYM(PCB_FLAGS, offsetof(struct pcb, pcb_flags));
 ASSYM(PCB_SAVEFPU, offsetof(struct pcb, pcb_save));
 ASSYM(PCB_SAVEFPU_SIZE, sizeof(struct savefpu));
 ASSYM(PCB_ONFAULT, offsetof(struct pcb, pcb_onfault));
-ASSYM(PCB_GS32P, offsetof(struct pcb, pcb_gs32p));
 ASSYM(PCB_GS32SD, offsetof(struct pcb, pcb_gs32sd));
 
 ASSYM(PCB_SIZE, sizeof(struct pcb));
@@ -199,6 +198,7 @@ ASSYM(PC_SCRATCH_RSP, offsetof(struct pcpu, pc_scratch_rsp));
 ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
 ASSYM(PC_TSSP, offsetof(struct pcpu, pc_tssp));
 ASSYM(PC_RSP0, offsetof(struct pcpu, pc_rsp0));
+ASSYM(PC_GS32P, offsetof(struct pcpu, pc_gs32p));
  
 ASSYM(LA_VER, offsetof(struct LAPIC, version));
 ASSYM(LA_TPR, offsetof(struct LAPIC, tpr));
diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c
index f3c41f7..988b039 100644
--- a/sys/amd64/amd64/machdep.c
+++ b/sys/amd64/amd64/machdep.c
@@ -803,7 +803,7 @@ cpu_setregs(void)
  * Initialize segments & interrupt table
  */
 
-struct user_segment_descriptor gdt[NGDT * MAXCPU];/* global descriptor table */
+struct user_segment_descriptor gdt[NGDT * MAXCPU];/* global descriptor tables */
 static struct gate_descriptor idt0[NIDT];
 struct gate_descriptor *idt = &idt0[0];	/* interrupt descriptor table */
 
@@ -870,7 +870,7 @@ struct soft_segment_descriptor gdt_segs[] = {
 /* GPROC0_SEL	6 Proc 0 Tss Descriptor */
 {
 	0x0,			/* segment base address */
-	sizeof(struct amd64tss)-1,/* length - all address space */
+	sizeof(struct amd64tss)-1,/* length */
 	SDT_SYSTSS,		/* segment type */
 	SEL_KPL,		/* segment descriptor priority level */
 	1,			/* segment descriptor present */
@@ -1347,6 +1347,7 @@ hammer_time(u_int64_t modulep, u_int64_t physfree)
 	PCPU_SET(curthread, &thread0);
 	PCPU_SET(curpcb, thread0.td_pcb);
 	PCPU_SET(tssp, &common_tss[0]);
+	PCPU_SET(gs32p, &gdt[GUGS32_SEL]);
 
 	/*
 	 * Initialize mutexes.
diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 304deaa..2de63e3 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -436,7 +436,8 @@ init_secondary(void)
 {
 	struct pcpu *pc;
 	u_int64_t msr, cr0;
-	int cpu, gsel_tss;
+	int cpu, gsel_tss, x;
+	struct region_descriptor ap_gdt;
 
 	/* Set by the startup code for us to use */
 	cpu = bootAP;
@@ -447,11 +448,17 @@ init_secondary(void)
 	common_tss[cpu].tss_iobase = sizeof(struct amd64tss);
 	common_tss[cpu].tss_ist1 = (long)&doublefault_stack[PAGE_SIZE];
 
+	/* Prepare private GDT */
 	gdt_segs[GPROC0_SEL].ssd_base = (long) &common_tss[cpu];
 	ssdtosyssd(&gdt_segs[GPROC0_SEL],
-	   (struct system_segment_descriptor *)&gdt[GPROC0_SEL]);
-
-	lgdt(&r_gdt);			/* does magic intra-segment return */
+	   (struct system_segment_descriptor *)&gdt[NGDT * cpu + GPROC0_SEL]);
+	for (x = 0; x < NGDT; x++) {
+		if (x != GPROC0_SEL && x != (GPROC0_SEL + 1))
+			ssdtosd(&gdt_segs[x], &gdt[NGDT * cpu + x]);
+	}
+	ap_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
+	ap_gdt.rd_base =  (long) &gdt[NGDT * cpu];
+	lgdt(&ap_gdt);			/* does magic intra-segment return */
 
 	/* Get per-cpu data */
 	pc = &__pcpu[cpu];
@@ -463,6 +470,7 @@ init_secondary(void)
 	pc->pc_curthread = 0;
 	pc->pc_tssp = &common_tss[cpu];
 	pc->pc_rsp0 = 0;
+	pc->pc_gs32p = &gdt[NGDT * cpu + GUGS32_SEL];
 
 	wrmsr(MSR_FSBASE, 0);		/* User value */
 	wrmsr(MSR_GSBASE, (u_int64_t)pc);
diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c
index 00e596f..8d710cd 100644
--- a/sys/amd64/amd64/trap.c
+++ b/sys/amd64/amd64/trap.c
@@ -692,7 +692,8 @@ trap_fatal(frame, eva)
 
 	code = frame->tf_err;
 	type = frame->tf_trapno;
-	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
+	sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
+	    &softseg);
 
 	if (type <= MAX_TRAP_MSG)
 		msg = trap_msg[type];
diff --git a/sys/amd64/include/pcb.h b/sys/amd64/include/pcb.h
index 647d5c9..43b59e5 100644
--- a/sys/amd64/include/pcb.h
+++ b/sys/amd64/include/pcb.h
@@ -77,7 +77,6 @@ struct pcb {
 	caddr_t	pcb_onfault;	/* copyin/out fault recovery */
 
 	/* 32-bit segment descriptor */
-	struct user_segment_descriptor	*pcb_gs32p;
 	struct user_segment_descriptor	pcb_gs32sd;
 };
 
diff --git a/sys/amd64/include/pcpu.h b/sys/amd64/include/pcpu.h
index fe811c5..e9faf28 100644
--- a/sys/amd64/include/pcpu.h
+++ b/sys/amd64/include/pcpu.h
@@ -48,7 +48,8 @@
 	register_t pc_rsp0;						\
 	register_t pc_scratch_rsp;	/* User %rsp in syscall */	\
 	u_int	pc_apic_id;						\
-	u_int   pc_acpi_id		/* ACPI CPU id */
+	u_int   pc_acpi_id;		/* ACPI CPU id */		\
+	struct user_segment_descriptor	*pc_gs32p
 
 #ifdef _KERNEL
 
diff --git a/sys/amd64/include/segments.h b/sys/amd64/include/segments.h
index 891404c..1c83d1c 100644
--- a/sys/amd64/include/segments.h
+++ b/sys/amd64/include/segments.h
@@ -201,7 +201,7 @@ struct region_descriptor {
 #define	GUDATA_SEL	4	/* User 32/64 bit Data Descriptor */
 #define	GUCODE_SEL	5	/* User 64 bit Code Descriptor */
 #define	GPROC0_SEL	6	/* TSS for entering kernel etc */
-/* slot 6 is second half of GPROC0_SEL */
+/* slot 7 is second half of GPROC0_SEL */
 #define	GUGS32_SEL	8	/* User 32 bit GS Descriptor */
 #define	NGDT 		9
 
diff --git a/sys/amd64/linux32/linux32_machdep.c b/sys/amd64/linux32/linux32_machdep.c
index 6f1401c..aecf869 100644
--- a/sys/amd64/linux32/linux32_machdep.c
+++ b/sys/amd64/linux32/linux32_machdep.c
@@ -694,7 +694,6 @@ linux_clone(struct thread *td, struct linux_clone_args *args)
 #endif
 			td2->td_pcb->pcb_gsbase = (register_t)info.base_addr;
 			td2->td_pcb->pcb_gs32sd = sd;
-			td2->td_pcb->pcb_gs32p = &gdt[GUGS32_SEL];
 			td2->td_pcb->pcb_gs = GSEL(GUGS32_SEL, SEL_UPL);
 			td2->td_pcb->pcb_flags |= PCB_GS32BIT | PCB_32BIT;
 		}
@@ -1352,9 +1351,8 @@ linux_set_thread_area(struct thread *td,
 
 	critical_enter();
 	td->td_pcb->pcb_gsbase = (register_t)info.base_addr;
-	td->td_pcb->pcb_gs32sd = gdt[GUGS32_SEL] = sd;
-	td->td_pcb->pcb_gs32p = &gdt[GUGS32_SEL];
-	td->td_pcb->pcb_flags |= PCB_32BIT;
+	td->td_pcb->pcb_gs32sd = *PCPU_GET(gs32p) = sd;
+	td->td_pcb->pcb_flags |= PCB_32BIT | PCB_GS32BIT;
 	wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase);
 	critical_exit();