Note that this patches don't have x86/acpica/acpi_wakeup.c so you need to do
	cp amd64/acpica/acpi_wakeup.c x86/acpica/
after patching.

Index: amd64/acpica/acpi_wakecode.S
===================================================================
RCS file: /home/ncvs/src/sys/amd64/acpica/acpi_wakecode.S,v
retrieving revision 1.13
diff -u -r1.13 acpi_wakecode.S
--- amd64/acpica/acpi_wakecode.S	15 Feb 2012 22:10:33 -0000	1.13
+++ amd64/acpica/acpi_wakecode.S	6 Jun 2012 17:08:55 -0000
@@ -219,10 +219,15 @@
 	mov	$bootdata64 - bootgdt, %eax
 	mov	%ax, %ds
 
-	/* Restore arguments and return. */
-	movq	wakeup_kpml4 - wakeup_start(%rbx), %rdi
-	movq	wakeup_ctx - wakeup_start(%rbx), %rsi
-	movq	wakeup_retaddr - wakeup_start(%rbx), %rax
+	/* Restore arguments. */
+	movq	wakeup_cr3 - wakeup_start(%rbx), %rsi
+	movq	wakeup_pcb - wakeup_start(%rbx), %rdi
+	movq	wakeup_ret - wakeup_start(%rbx), %rax
+
+	/* Restore GDT. */
+	lgdt	wakeup_gdt - wakeup_start(%rbx)
+
+	/* Jump to return address. */
 	jmp	*%rax
 
 	.data
@@ -268,34 +273,15 @@
 	.long	bootgdt - wakeup_start	/* Offset plus %ds << 4 */
 
 	ALIGN_DATA
-wakeup_retaddr:
+wakeup_cr4:		/* not used */
 	.quad	0
-wakeup_kpml4:
-	.quad	0
-
-wakeup_ctx:
+wakeup_cr3:
 	.quad	0
 wakeup_pcb:
 	.quad	0
-wakeup_fpusave:
+wakeup_ret:
 	.quad	0
 wakeup_gdt:
 	.word	0
 	.quad	0
-
-	ALIGN_DATA
-wakeup_efer:
-	.quad	0
-wakeup_star:
-	.quad	0
-wakeup_lstar:
-	.quad	0
-wakeup_cstar:
-	.quad	0
-wakeup_sfmask:
-	.quad	0
-wakeup_xsmask:
-	.quad	0
-wakeup_cpu:
-	.long	0
 dummy:
Index: amd64/acpica/acpi_wakeup.c
===================================================================
RCS file: /home/ncvs/src/sys/amd64/acpica/acpi_wakeup.c,v
retrieving revision 1.58
diff -u -r1.58 acpi_wakeup.c
--- amd64/acpica/acpi_wakeup.c	1 Jun 2012 21:33:33 -0000	1.58
+++ amd64/acpica/acpi_wakeup.c	6 Jun 2012 17:08:55 -0000
@@ -1,6 +1,6 @@
 /*-
  * Copyright (c) 2001 Takanori Watanabe <takawata@jp.freebsd.org>
- * Copyright (c) 2001 Mitsuru IWASAKI <iwasaki@jp.freebsd.org>
+ * Copyright (c) 2001-2012 Mitsuru IWASAKI <iwasaki@jp.freebsd.org>
  * Copyright (c) 2003 Peter Wemm
  * Copyright (c) 2008-2012 Jung-uk Kim <jkim@FreeBSD.org>
  * All rights reserved.
@@ -28,7 +28,7 @@
  */
 
 #include <sys/cdefs.h>
-__FBSDID("$FreeBSD: src/sys/amd64/acpica/acpi_wakeup.c,v 1.58 2012/06/01 21:33:33 jkim Exp $");
+__FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
 #include <sys/bus.h>
@@ -70,15 +70,11 @@
 
 #ifdef SMP
 extern struct pcb	**susppcbs;
-extern void		**suspfpusave;
 static cpuset_t		suspcpus;
 #else
 static struct pcb	**susppcbs;
-static void		**suspfpusave;
 #endif
 
-int			acpi_restorecpu(uint64_t, vm_offset_t);
-
 static void		*acpi_alloc_wakeup_handler(void);
 static void		acpi_stop_beep(void *);
 
@@ -87,8 +83,14 @@
 static void		acpi_wakeup_cpus(struct acpi_softc *);
 #endif
 
-#define	WAKECODE_VADDR(sc)	((sc)->acpi_wakeaddr + (3 * PAGE_SIZE))
-#define	WAKECODE_PADDR(sc)	((sc)->acpi_wakephys + (3 * PAGE_SIZE))
+#ifdef __amd64__
+#define ACPI_PAGETABLES	3
+#else
+#define ACPI_PAGETABLES	0
+#endif
+
+#define	WAKECODE_VADDR(sc)	((sc)->acpi_wakeaddr + (ACPI_PAGETABLES * PAGE_SIZE))
+#define	WAKECODE_PADDR(sc)	((sc)->acpi_wakephys + (ACPI_PAGETABLES * PAGE_SIZE))
 #define	WAKECODE_FIXUP(offset, type, val) do	{	\
 	type	*addr;					\
 	addr = (type *)(WAKECODE_VADDR(sc) + offset);	\
@@ -112,11 +114,9 @@
 	int		ms;
 
 	WAKECODE_FIXUP(wakeup_pcb, struct pcb *, susppcbs[cpu]);
-	WAKECODE_FIXUP(wakeup_fpusave, void *, suspfpusave[cpu]);
 	WAKECODE_FIXUP(wakeup_gdt, uint16_t, susppcbs[cpu]->pcb_gdt.rd_limit);
 	WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
 	    susppcbs[cpu]->pcb_gdt.rd_base);
-	WAKECODE_FIXUP(wakeup_cpu, int, cpu);
 
 	/* do an INIT IPI: assert RESET */
 	lapic_ipi_raw(APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE |
@@ -162,9 +162,9 @@
 	lapic_ipi_wait(-1);
 	DELAY(200);		/* wait ~200uS */
 
-	/* Wait up to 5 seconds for it to start. */
+	/* Wait up to 5 seconds for it to resume. */
 	for (ms = 0; ms < 5000; ms++) {
-		if (*(int *)(WAKECODE_VADDR(sc) + wakeup_cpu) == 0)
+		if (!CPU_ISSET(cpu, &suspended_cpus))
 			return (1);	/* return SUCCESS */
 		DELAY(1000);
 	}
@@ -239,7 +239,9 @@
 	intr_suspend();
 
 	if (savectx(susppcbs[0])) {
-		ctx_fpusave(suspfpusave[0]);
+#ifdef __amd64__
+		ctx_fpusave(susppcbs[0]->pcb_fpususpend);
+#endif
 #ifdef SMP
 		if (!CPU_EMPTY(&suspcpus) && suspend_cpus(suspcpus) == 0) {
 			device_printf(sc->acpi_dev, "Failed to suspend APs\n");
@@ -250,13 +252,12 @@
 		WAKECODE_FIXUP(resume_beep, uint8_t, (acpi_resume_beep != 0));
 		WAKECODE_FIXUP(reset_video, uint8_t, (acpi_reset_video != 0));
 
+		WAKECODE_FIXUP(wakeup_cr4, register_t, susppcbs[0]->pcb_cr4);
 		WAKECODE_FIXUP(wakeup_pcb, struct pcb *, susppcbs[0]);
-		WAKECODE_FIXUP(wakeup_fpusave, void *, suspfpusave[0]);
 		WAKECODE_FIXUP(wakeup_gdt, uint16_t,
 		    susppcbs[0]->pcb_gdt.rd_limit);
 		WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
 		    susppcbs[0]->pcb_gdt.rd_base);
-		WAKECODE_FIXUP(wakeup_cpu, int, 0);
 
 		/* Call ACPICA to enter the desired sleep state */
 		if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
@@ -289,7 +290,9 @@
 		/* Wakeup MD procedures in interrupt disabled context */
 		if (sleep_result == 1) {
 			pmap_init_pat();
+#if 0
 			load_cr3(susppcbs[0]->pcb_cr3);
+#endif
 			initializecpu();
 			PCPU_SET(switchtime, 0);
 			PCPU_SET(switchticks, ticks);
@@ -330,8 +333,8 @@
 	 * and ROM area (0xa0000 and above).  The temporary page tables must be
 	 * page-aligned.
 	 */
-	wakeaddr = contigmalloc(4 * PAGE_SIZE, M_DEVBUF, M_WAITOK, 0x500,
-	    0xa0000, PAGE_SIZE, 0ul);
+	wakeaddr = contigmalloc((ACPI_PAGETABLES + 1) * PAGE_SIZE, M_DEVBUF,
+	    M_WAITOK, 0x500, 0xa0000, PAGE_SIZE, 0ul);
 	if (wakeaddr == NULL) {
 		printf("%s: can't alloc wake memory\n", __func__);
 		return (NULL);
@@ -339,14 +342,15 @@
 	if (EVENTHANDLER_REGISTER(power_resume, acpi_stop_beep, NULL,
 	    EVENTHANDLER_PRI_LAST) == NULL) {
 		printf("%s: can't register event handler\n", __func__);
-		contigfree(wakeaddr, 4 * PAGE_SIZE, M_DEVBUF);
+		contigfree(wakeaddr, (ACPI_PAGETABLES + 1) * PAGE_SIZE, M_DEVBUF);
 		return (NULL);
 	}
 	susppcbs = malloc(mp_ncpus * sizeof(*susppcbs), M_DEVBUF, M_WAITOK);
-	suspfpusave = malloc(mp_ncpus * sizeof(void *), M_DEVBUF, M_WAITOK);
 	for (i = 0; i < mp_ncpus; i++) {
 		susppcbs[i] = malloc(sizeof(**susppcbs), M_DEVBUF, M_WAITOK);
-		suspfpusave[i] = alloc_fpusave(M_WAITOK);
+#ifdef __amd64__
+		susppcbs[i]->pcb_fpususpend = alloc_fpusave(M_WAITOK);
+#endif
 	}
 
 	return (wakeaddr);
@@ -356,8 +360,10 @@
 acpi_install_wakeup_handler(struct acpi_softc *sc)
 {
 	static void	*wakeaddr = NULL;
+#ifdef __amd64__
 	uint64_t	*pt4, *pt3, *pt2;
 	int		i;
+#endif
 
 	if (wakeaddr != NULL)
 		return;
@@ -371,27 +377,30 @@
 
 	bcopy(wakecode, (void *)WAKECODE_VADDR(sc), sizeof(wakecode));
 
-	/* Patch GDT base address, ljmp targets and page table base address. */
+	/* Patch GDT base address, ljmp targets. */
 	WAKECODE_FIXUP((bootgdtdesc + 2), uint32_t,
 	    WAKECODE_PADDR(sc) + bootgdt);
 	WAKECODE_FIXUP((wakeup_sw32 + 2), uint32_t,
 	    WAKECODE_PADDR(sc) + wakeup_32);
+#ifdef __amd64__
 	WAKECODE_FIXUP((wakeup_sw64 + 1), uint32_t,
 	    WAKECODE_PADDR(sc) + wakeup_64);
 	WAKECODE_FIXUP(wakeup_pagetables, uint32_t, sc->acpi_wakephys);
+#endif
 
 	/* Save pointers to some global data. */
-	WAKECODE_FIXUP(wakeup_retaddr, void *, acpi_restorecpu);
-	WAKECODE_FIXUP(wakeup_kpml4, uint64_t, KPML4phys);
-	WAKECODE_FIXUP(wakeup_ctx, vm_offset_t,
-	    WAKECODE_VADDR(sc) + wakeup_ctx);
-	WAKECODE_FIXUP(wakeup_efer, uint64_t, rdmsr(MSR_EFER));
-	WAKECODE_FIXUP(wakeup_star, uint64_t, rdmsr(MSR_STAR));
-	WAKECODE_FIXUP(wakeup_lstar, uint64_t, rdmsr(MSR_LSTAR));
-	WAKECODE_FIXUP(wakeup_cstar, uint64_t, rdmsr(MSR_CSTAR));
-	WAKECODE_FIXUP(wakeup_sfmask, uint64_t, rdmsr(MSR_SF_MASK));
-	WAKECODE_FIXUP(wakeup_xsmask, uint64_t, xsave_mask);
+	WAKECODE_FIXUP(wakeup_ret, void *, resumectx);
+#ifdef __amd64__
+	WAKECODE_FIXUP(wakeup_cr3, uint64_t, KPML4phys);
+#else
+#ifdef PAE
+	WAKECODE_FIXUP(wakeup_cr3, register_t, vtophys(kernel_pmap->pm_pdpt));
+#else
+	WAKECODE_FIXUP(wakeup_cr3, register_t, vtophys(kernel_pmap->pm_pdir));
+#endif
+#endif
 
+#ifdef __amd64__
 	/* Build temporary page tables below realmode code. */
 	pt4 = wakeaddr;
 	pt3 = pt4 + (PAGE_SIZE) / sizeof(uint64_t);
@@ -417,8 +426,9 @@
 		pt2[i] = i * (2 * 1024 * 1024);
 		pt2[i] |= PG_V | PG_RW | PG_PS | PG_U;
 	}
+#endif
 
 	if (bootverbose)
-		device_printf(sc->acpi_dev, "wakeup code va %p pa %p\n",
-		    (void *)sc->acpi_wakeaddr, (void *)sc->acpi_wakephys);
+		device_printf(sc->acpi_dev, "wakeup code va %#jx pa %#jx\n",
+		    (uintmax_t)sc->acpi_wakeaddr, (uintmax_t)sc->acpi_wakephys);
 }
Index: amd64/amd64/cpu_switch.S
===================================================================
RCS file: /home/ncvs/src/sys/amd64/amd64/cpu_switch.S,v
retrieving revision 1.183
diff -u -r1.183 cpu_switch.S
--- amd64/amd64/cpu_switch.S	27 Feb 2012 17:28:22 -0000	1.183
+++ amd64/amd64/cpu_switch.S	6 Jun 2012 17:08:55 -0000
@@ -357,6 +357,30 @@
 	rdmsr
 	movl	%eax,PCB_KGSBASE(%rdi)
 	movl	%edx,PCB_KGSBASE+4(%rdi)
+	movl	$MSR_EFER,%ecx
+	rdmsr
+	movl	%eax,PCB_EFER(%rdi)
+	movl	%edx,PCB_EFER+4(%rdi)
+	movl	$MSR_STAR,%ecx
+	rdmsr
+	movl	%eax,PCB_STAR(%rdi)
+	movl	%edx,PCB_STAR+4(%rdi)
+	movl	$MSR_LSTAR,%ecx
+	rdmsr
+	movl	%eax,PCB_LSTAR(%rdi)
+	movl	%edx,PCB_LSTAR+4(%rdi)
+	movl	$MSR_CSTAR,%ecx
+	rdmsr
+	movl	%eax,PCB_CSTAR(%rdi)
+	movl	%edx,PCB_CSTAR+4(%rdi)
+	movl	$MSR_SF_MASK,%ecx
+	rdmsr
+	movl	%eax,PCB_SFMASK(%rdi)
+	movl	%edx,PCB_SFMASK+4(%rdi)
+	movl	xsave_mask,%eax
+	movl	%eax,PCB_XSMASK(%rdi)
+	movl	xsave_mask+4,%eax
+	movl	%eax,PCB_XSMASK+4(%rdi)
 
 	sgdt	PCB_GDT(%rdi)
 	sidt	PCB_IDT(%rdi)
@@ -370,6 +394,141 @@
 END(savectx)
 
 /*
+ * resumectx(pcb in %rdi, cr3 in %rsi)
+ * Resuming processor state from pcb.
+ */     
+ENTRY(resumectx)
+	/* Switch to KPML4phys. */
+	movq	%rsi,%cr3
+
+	/* Force kernel segment registers. */
+	movl	$KDSEL,%eax
+	movw	%ax,%ds
+	movw	%ax,%es
+	movw	%ax,%ss
+	movl	$KUF32SEL,%eax
+	movw	%ax,%fs
+	movl	$KUG32SEL,%eax
+	movw	%ax,%gs
+
+	movl	$MSR_FSBASE,%ecx
+	movl	PCB_FSBASE(%rdi),%eax
+	movl	4 + PCB_FSBASE(%rdi),%edx
+	wrmsr
+	movl	$MSR_GSBASE,%ecx
+	movl	PCB_GSBASE(%rdi),%eax
+	movl	4 + PCB_GSBASE(%rdi),%edx
+	wrmsr
+	movl	$MSR_KGSBASE,%ecx
+	movl	PCB_KGSBASE(%rdi),%eax
+	movl	4 + PCB_KGSBASE(%rdi),%edx
+	wrmsr
+
+	/* Restore EFER. */
+	movl	$MSR_EFER,%ecx
+	movl	PCB_EFER(%rdi),%eax
+	wrmsr
+
+	/* Restore fast syscall stuff. */
+	movl	$MSR_STAR,%ecx
+	movl	PCB_STAR(%rdi),%eax
+	movl	4 + PCB_STAR(%rdi),%edx
+	wrmsr
+	movl	$MSR_LSTAR,%ecx
+	movl	PCB_LSTAR(%rdi),%eax
+	movl	4 + PCB_LSTAR(%rdi),%edx
+	wrmsr
+	movl	$MSR_CSTAR,%ecx
+	movl	PCB_CSTAR(%rdi),%eax
+	movl	4 + PCB_CSTAR(%rdi),%edx
+	wrmsr
+	movl	$MSR_SF_MASK,%ecx
+	movl	PCB_SFMASK(%rdi),%eax
+	wrmsr
+
+	/* Restore CR0 except for FPU mode. */
+	movq	PCB_CR0(%rdi),%rax
+	andq	$~(CR0_EM | CR0_TS),%rax
+	movq	%rax,%cr0
+
+	/* Restore CR2, CR4 and CR3. */
+	movq	PCB_CR2(%rdi),%rax
+	movq	%rax,%cr2
+	movq	PCB_CR4(%rdi),%rax
+	movq	%rax,%cr4
+	movq	PCB_CR3(%rdi),%rax
+	movq	%rax,%cr3
+
+	/* Restore descriptor tables. */
+	lidt	PCB_IDT(%rdi)
+	lldt	PCB_LDT(%rdi)
+
+#define	SDT_SYSTSS	9
+#define	SDT_SYSBSY	11
+
+	/* Clear "task busy" bit and reload TR. */
+	movq	PCPU(TSS),%rax
+	andb	$(~SDT_SYSBSY | SDT_SYSTSS),5(%rax)
+	movw	PCB_TR(%rdi),%ax
+	ltr	%ax
+
+#undef	SDT_SYSTSS
+#undef	SDT_SYSBSY
+
+	/* Restore debug registers. */
+	movq	PCB_DR0(%rdi),%rax
+	movq	%rax,%dr0
+	movq	PCB_DR1(%rdi),%rax
+	movq	%rax,%dr1
+	movq	PCB_DR2(%rdi),%rax
+	movq	%rax,%dr2
+	movq	PCB_DR3(%rdi),%rax
+	movq	%rax,%dr3
+	movq	PCB_DR6(%rdi),%rax
+	movq	%rax,%dr6
+	movq	PCB_DR7(%rdi),%rax
+	movq	%rax,%dr7
+
+	/* Restore FPU state. */
+	fninit
+	movq	PCB_FPUSUSPEND(%rdi),%rbx
+	movq	PCB_XSMASK(%rdi),%rax
+	testq	%rax,%rax
+	jz	1f
+	movq	%rax,%rdx
+	shrq	$32,%rdx
+	movl	$XCR0,%ecx
+/*	xsetbv	*/
+	.byte	0x0f, 0x01, 0xd1
+/*	xrstor	(%rbx) */
+	.byte	0x0f, 0xae, 0x2b
+	jmp	2f
+1:
+	fxrstor	(%rbx)
+2:
+
+	/* Reload CR0. */
+	movq	PCB_CR0(%rdi),%rax
+	movq	%rax,%cr0
+
+	/* Restore other callee saved registers. */
+	movq	PCB_R15(%rdi),%r15
+	movq	PCB_R14(%rdi),%r14
+	movq	PCB_R13(%rdi),%r13
+	movq	PCB_R12(%rdi),%r12
+	movq	PCB_RBP(%rdi),%rbp
+	movq	PCB_RSP(%rdi),%rsp
+	movq	PCB_RBX(%rdi),%rbx
+
+	/* Restore return address. */
+	movq	PCB_RIP(%rdi),%rax
+	movq	%rax,(%rsp)
+
+	xorl	%eax,%eax
+	ret
+END(resumectx)
+
+/*
  * Wrapper around fpusave to care about TS0_CR.
  */
 ENTRY(ctx_fpusave)
Index: amd64/amd64/genassym.c
===================================================================
RCS file: /home/ncvs/src/sys/amd64/amd64/genassym.c,v
retrieving revision 1.188
diff -u -r1.188 genassym.c
--- amd64/amd64/genassym.c	21 Jan 2012 17:45:27 -0000	1.188
+++ amd64/amd64/genassym.c	6 Jun 2012 17:08:55 -0000
@@ -157,6 +157,13 @@
 ASSYM(PCB_SAVEFPU, offsetof(struct pcb, pcb_save));
 ASSYM(PCB_SAVEFPU_SIZE, sizeof(struct savefpu));
 ASSYM(PCB_USERFPU, sizeof(struct pcb));
+ASSYM(PCB_EFER, offsetof(struct pcb, pcb_efer));
+ASSYM(PCB_STAR, offsetof(struct pcb, pcb_star));
+ASSYM(PCB_LSTAR, offsetof(struct pcb, pcb_lstar));
+ASSYM(PCB_CSTAR, offsetof(struct pcb, pcb_cstar));
+ASSYM(PCB_SFMASK, offsetof(struct pcb, pcb_sfmask));
+ASSYM(PCB_XSMASK, offsetof(struct pcb, pcb_xsmask));
+ASSYM(PCB_FPUSUSPEND, offsetof(struct pcb, pcb_fpususpend));
 ASSYM(PCB_SIZE, sizeof(struct pcb));
 ASSYM(PCB_FULL_IRET, PCB_FULL_IRET);
 ASSYM(PCB_DBREGS, PCB_DBREGS);
Index: amd64/amd64/mp_machdep.c
===================================================================
RCS file: /home/ncvs/src/sys/amd64/amd64/mp_machdep.c,v
retrieving revision 1.351
diff -u -r1.351 mp_machdep.c
--- amd64/amd64/mp_machdep.c	13 Apr 2012 07:18:19 -0000	1.351
+++ amd64/amd64/mp_machdep.c	6 Jun 2012 17:08:55 -0000
@@ -100,7 +100,6 @@
 
 struct pcb stoppcbs[MAXCPU];
 struct pcb **susppcbs;
-void **suspfpusave;
 
 /* Variables needed for SMP tlb shootdown. */
 vm_offset_t smp_tlb_addr1;
@@ -1415,15 +1414,19 @@
 	cpu = PCPU_GET(cpuid);
 
 	if (savectx(susppcbs[cpu])) {
-		ctx_fpusave(suspfpusave[cpu]);
+		ctx_fpusave(susppcbs[cpu]->pcb_fpususpend);
 		wbinvd();
 		CPU_SET_ATOMIC(cpu, &stopped_cpus);
+		CPU_SET_ATOMIC(cpu, &suspended_cpus);
 	} else {
 		pmap_init_pat();
+#if 0
 		load_cr3(susppcbs[cpu]->pcb_cr3);
+#endif
 		initializecpu();
 		PCPU_SET(switchtime, 0);
 		PCPU_SET(switchticks, ticks);
+		CPU_CLR_ATOMIC(cpu, &suspended_cpus);
 	}
 
 	/* Wait for resume */
Index: amd64/include/pcb.h
===================================================================
RCS file: /home/ncvs/src/sys/amd64/include/pcb.h,v
retrieving revision 1.80
diff -u -r1.80 pcb.h
--- amd64/include/pcb.h	29 Apr 2012 11:04:31 -0000	1.80
+++ amd64/include/pcb.h	6 Jun 2012 17:08:55 -0000
@@ -91,9 +91,20 @@
 	/* local tss, with i/o bitmap; NULL for common */
 	struct amd64tss *pcb_tssp;
 
+	/* model specific registers */
+	register_t	pcb_efer;
+	register_t	pcb_star;
+	register_t	pcb_lstar;
+	register_t	pcb_cstar;
+	register_t	pcb_sfmask;
+	register_t	pcb_xsmask;
+
+	/* fpu context for suspend/resume */
+	void *		pcb_fpususpend;
+
 	struct savefpu	*pcb_save;
 
-	uint64_t	pcb_pad[2];
+	uint64_t	pcb_pad[3];
 };
 
 #ifdef _KERNEL
@@ -131,6 +142,7 @@
 
 void	makectx(struct trapframe *, struct pcb *);
 int	savectx(struct pcb *) __returns_twice;
+void	resumectx(struct pcb *);
 
 #endif
 
Index: conf/files.amd64
===================================================================
RCS file: /home/ncvs/src/sys/conf/files.amd64,v
retrieving revision 1.194
diff -u -r1.194 files.amd64
--- conf/files.amd64	11 Apr 2012 02:42:01 -0000	1.194
+++ conf/files.amd64	6 Jun 2012 17:08:55 -0000
@@ -72,7 +72,6 @@
 	no-implicit-rule
 #
 amd64/acpica/acpi_machdep.c	optional	acpi
-amd64/acpica/acpi_switch.S	optional	acpi
 acpi_wakecode.o			optional	acpi			\
 	dependency	"$S/amd64/acpica/acpi_wakecode.S assym.s"	\
 	compile-with	"${NORMAL_S}"					\
@@ -94,7 +93,7 @@
 	no-obj no-implicit-rule	before-depend				\
 	clean		"acpi_wakedata.h"
 #
-amd64/acpica/acpi_wakeup.c	optional	acpi
+x86/acpica/acpi_wakeup.c	optional	acpi
 amd64/amd64/amd64_mem.c		optional	mem
 #amd64/amd64/apic_vector.S	standard
 amd64/amd64/atomic.c		standard
Index: conf/files.i386
===================================================================
RCS file: /home/ncvs/src/sys/conf/files.i386,v
retrieving revision 1.677
diff -u -r1.677 files.i386
--- conf/files.i386	11 Apr 2012 02:42:01 -0000	1.677
+++ conf/files.i386	6 Jun 2012 17:08:55 -0000
@@ -381,7 +381,7 @@
 	no-obj no-implicit-rule	before-depend				\
 	clean		"acpi_wakedata.h"
 #
-i386/acpica/acpi_wakeup.c	optional acpi
+x86/acpica/acpi_wakeup.c	optional acpi
 i386/bios/apm.c			optional apm
 i386/bios/mca_machdep.c		optional mca
 i386/bios/smapi.c		optional smapi
Index: dev/acpica/acpivar.h
===================================================================
RCS file: /home/ncvs/src/sys/dev/acpica/acpivar.h,v
retrieving revision 1.132
diff -u -r1.132 acpivar.h
--- dev/acpica/acpivar.h	1 Jun 2012 21:33:33 -0000	1.132
+++ dev/acpica/acpivar.h	6 Jun 2012 17:08:55 -0000
@@ -71,8 +71,6 @@
     int			acpi_verbose;
     int			acpi_handle_reboot;
 
-    bus_dma_tag_t	acpi_waketag;
-    bus_dmamap_t	acpi_wakemap;
     vm_offset_t		acpi_wakeaddr;
     vm_paddr_t		acpi_wakephys;
 
Index: i386/acpica/acpi_wakecode.S
===================================================================
RCS file: /home/ncvs/src/sys/i386/acpica/acpi_wakecode.S,v
retrieving revision 1.15
diff -u -r1.15 acpi_wakecode.S
--- i386/acpica/acpi_wakecode.S	18 May 2012 18:55:58 -0000	1.15
+++ i386/acpica/acpi_wakecode.S	6 Jun 2012 17:08:55 -0000
@@ -202,4 +202,7 @@
 	.long	0
 wakeup_ret:
 	.long	0
+wakeup_gdt:		/* not used */
+	.word	0
+	.long	0
 dummy:
Index: i386/acpica/acpi_wakeup.c
===================================================================
RCS file: /home/ncvs/src/sys/i386/acpica/acpi_wakeup.c,v
retrieving revision 1.62
diff -u -r1.62 acpi_wakeup.c
--- i386/acpica/acpi_wakeup.c	1 Jun 2012 21:33:33 -0000	1.62
+++ i386/acpica/acpi_wakeup.c	6 Jun 2012 17:08:55 -0000
@@ -28,7 +28,7 @@
  */
 
 #include <sys/cdefs.h>
-__FBSDID("$FreeBSD: src/sys/i386/acpica/acpi_wakeup.c,v 1.62 2012/06/01 21:33:33 jkim Exp $");
+__FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
 #include <sys/bus.h>
@@ -83,7 +83,12 @@
 static void		acpi_wakeup_cpus(struct acpi_softc *);
 #endif
 
+#ifdef __amd64__
+#define ACPI_PAGETABLES	3
+#else
 #define ACPI_PAGETABLES	0
+#endif
+
 #define	WAKECODE_VADDR(sc)	((sc)->acpi_wakeaddr + (ACPI_PAGETABLES * PAGE_SIZE))
 #define	WAKECODE_PADDR(sc)	((sc)->acpi_wakephys + (ACPI_PAGETABLES * PAGE_SIZE))
 #define	WAKECODE_FIXUP(offset, type, val) do	{	\
@@ -109,6 +114,9 @@
 	int		ms;
 
 	WAKECODE_FIXUP(wakeup_pcb, struct pcb *, susppcbs[cpu]);
+	WAKECODE_FIXUP(wakeup_gdt, uint16_t, susppcbs[cpu]->pcb_gdt.rd_limit);
+	WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
+	    susppcbs[cpu]->pcb_gdt.rd_base);
 
 	/* do an INIT IPI: assert RESET */
 	lapic_ipi_raw(APIC_DEST_DESTFLD | APIC_TRIGMOD_EDGE |
@@ -154,9 +162,9 @@
 	lapic_ipi_wait(-1);
 	DELAY(200);		/* wait ~200uS */
 
-	/* Wait up to 5 seconds for it to start. */
+	/* Wait up to 5 seconds for it to resume. */
 	for (ms = 0; ms < 5000; ms++) {
-		if (susppcbs[cpu]->pcb_eip == 0)
+		if (!CPU_ISSET(cpu, &suspended_cpus))
 			return (1);	/* return SUCCESS */
 		DELAY(1000);
 	}
@@ -230,7 +238,10 @@
 
 	intr_suspend();
 
-	if (suspendctx(susppcbs[0])) {
+	if (savectx(susppcbs[0])) {
+#ifdef __amd64__
+		ctx_fpusave(susppcbs[0]->pcb_fpususpend);
+#endif
 #ifdef SMP
 		if (!CPU_EMPTY(&suspcpus) && suspend_cpus(suspcpus) == 0) {
 			device_printf(sc->acpi_dev, "Failed to suspend APs\n");
@@ -243,6 +254,10 @@
 
 		WAKECODE_FIXUP(wakeup_cr4, register_t, susppcbs[0]->pcb_cr4);
 		WAKECODE_FIXUP(wakeup_pcb, struct pcb *, susppcbs[0]);
+		WAKECODE_FIXUP(wakeup_gdt, uint16_t,
+		    susppcbs[0]->pcb_gdt.rd_limit);
+		WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
+		    susppcbs[0]->pcb_gdt.rd_base);
 
 		/* Call ACPICA to enter the desired sleep state */
 		if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
@@ -275,7 +290,9 @@
 		/* Wakeup MD procedures in interrupt disabled context */
 		if (sleep_result == 1) {
 			pmap_init_pat();
+#if 0
 			load_cr3(susppcbs[0]->pcb_cr3);
+#endif
 			initializecpu();
 			PCPU_SET(switchtime, 0);
 			PCPU_SET(switchticks, ticks);
@@ -317,7 +334,7 @@
 	 * page-aligned.
 	 */
 	wakeaddr = contigmalloc((ACPI_PAGETABLES + 1) * PAGE_SIZE, M_DEVBUF,
-	    M_NOWAIT, 0x500, 0xa0000, PAGE_SIZE, 0ul);
+	    M_WAITOK, 0x500, 0xa0000, PAGE_SIZE, 0ul);
 	if (wakeaddr == NULL) {
 		printf("%s: can't alloc wake memory\n", __func__);
 		return (NULL);
@@ -331,6 +348,9 @@
 	susppcbs = malloc(mp_ncpus * sizeof(*susppcbs), M_DEVBUF, M_WAITOK);
 	for (i = 0; i < mp_ncpus; i++) {
 		susppcbs[i] = malloc(sizeof(**susppcbs), M_DEVBUF, M_WAITOK);
+#ifdef __amd64__
+		susppcbs[i]->pcb_fpususpend = alloc_fpusave(M_WAITOK);
+#endif
 	}
 
 	return (wakeaddr);
@@ -340,6 +360,10 @@
 acpi_install_wakeup_handler(struct acpi_softc *sc)
 {
 	static void	*wakeaddr = NULL;
+#ifdef __amd64__
+	uint64_t	*pt4, *pt3, *pt2;
+	int		i;
+#endif
 
 	if (wakeaddr != NULL)
 		return;
@@ -353,26 +377,58 @@
 
 	bcopy(wakecode, (void *)WAKECODE_VADDR(sc), sizeof(wakecode));
 
-	/* Patch GDT base address, ljmp target. */
+	/* Patch GDT base address, ljmp targets. */
 	WAKECODE_FIXUP((bootgdtdesc + 2), uint32_t,
 	    WAKECODE_PADDR(sc) + bootgdt);
 	WAKECODE_FIXUP((wakeup_sw32 + 2), uint32_t,
 	    WAKECODE_PADDR(sc) + wakeup_32);
+#ifdef __amd64__
+	WAKECODE_FIXUP((wakeup_sw64 + 1), uint32_t,
+	    WAKECODE_PADDR(sc) + wakeup_64);
+	WAKECODE_FIXUP(wakeup_pagetables, uint32_t, sc->acpi_wakephys);
+#endif
 
 	/* Save pointers to some global data. */
 	WAKECODE_FIXUP(wakeup_ret, void *, resumectx);
-	/*
-	 * Temporarily switch to the kernel pmap because it provides
-	 * an identity mapping (setup at boot) for the low physical
-	 * memory region containing the wakeup code.
-	 */
+#ifdef __amd64__
+	WAKECODE_FIXUP(wakeup_cr3, uint64_t, KPML4phys);
+#else
 #ifdef PAE
 	WAKECODE_FIXUP(wakeup_cr3, register_t, vtophys(kernel_pmap->pm_pdpt));
 #else
 	WAKECODE_FIXUP(wakeup_cr3, register_t, vtophys(kernel_pmap->pm_pdir));
 #endif
+#endif
+
+#ifdef __amd64__
+	/* Build temporary page tables below realmode code. */
+	pt4 = wakeaddr;
+	pt3 = pt4 + (PAGE_SIZE) / sizeof(uint64_t);
+	pt2 = pt3 + (PAGE_SIZE) / sizeof(uint64_t);
+
+	/* Create the initial 1GB replicated page tables */
+	for (i = 0; i < 512; i++) {
+		/*
+		 * Each slot of the level 4 pages points
+		 * to the same level 3 page
+		 */
+		pt4[i] = (uint64_t)(sc->acpi_wakephys + PAGE_SIZE);
+		pt4[i] |= PG_V | PG_RW | PG_U;
+
+		/*
+		 * Each slot of the level 3 pages points
+		 * to the same level 2 page
+		 */
+		pt3[i] = (uint64_t)(sc->acpi_wakephys + (2 * PAGE_SIZE));
+		pt3[i] |= PG_V | PG_RW | PG_U;
+
+		/* The level 2 page slots are mapped with 2MB pages for 1GB. */
+		pt2[i] = i * (2 * 1024 * 1024);
+		pt2[i] |= PG_V | PG_RW | PG_PS | PG_U;
+	}
+#endif
 
 	if (bootverbose)
-		device_printf(sc->acpi_dev, "wakeup code va %#x pa %#jx\n",
-		    sc->acpi_wakeaddr, (uintmax_t)sc->acpi_wakephys);
+		device_printf(sc->acpi_dev, "wakeup code va %#jx pa %#jx\n",
+		    (uintmax_t)sc->acpi_wakeaddr, (uintmax_t)sc->acpi_wakephys);
 }
Index: i386/i386/mp_machdep.c
===================================================================
RCS file: /home/ncvs/src/sys/i386/i386/mp_machdep.c,v
retrieving revision 1.333
diff -u -r1.333 mp_machdep.c
--- i386/i386/mp_machdep.c	18 May 2012 18:55:58 -0000	1.333
+++ i386/i386/mp_machdep.c	6 Jun 2012 17:08:55 -0000
@@ -1512,14 +1512,15 @@
 
 	cpu = PCPU_GET(cpuid);
 
-	if (suspendctx(susppcbs[cpu])) {
+	if (savectx(susppcbs[cpu])) {
 		wbinvd();
 		CPU_SET_ATOMIC(cpu, &stopped_cpus);
+		CPU_SET_ATOMIC(cpu, &suspended_cpus);
 	} else {
 		pmap_init_pat();
 		PCPU_SET(switchtime, 0);
 		PCPU_SET(switchticks, ticks);
-		susppcbs[cpu]->pcb_eip = 0;
+		CPU_CLR_ATOMIC(cpu, &suspended_cpus);
 	}
 
 	/* Wait for resume */
Index: i386/i386/swtch.s
===================================================================
RCS file: /home/ncvs/src/sys/i386/i386/swtch.s,v
retrieving revision 1.161
diff -u -r1.161 swtch.s
--- i386/i386/swtch.s	18 May 2012 18:55:58 -0000	1.161
+++ i386/i386/swtch.s	6 Jun 2012 17:08:55 -0000
@@ -386,6 +386,36 @@
 	pushfl
 	popl	PCB_PSL(%ecx)
 
+	movl	%cr0,%eax
+	movl	%eax,PCB_CR0(%ecx)
+	movl	%cr2,%eax
+	movl	%eax,PCB_CR2(%ecx)
+	movl	%cr4,%eax
+	movl	%eax,PCB_CR4(%ecx)
+
+	movl	%dr0,%eax
+	movl	%eax,PCB_DR0(%ecx)
+	movl	%dr1,%eax
+	movl	%eax,PCB_DR1(%ecx)
+	movl	%dr2,%eax
+	movl	%eax,PCB_DR2(%ecx)
+	movl	%dr3,%eax
+	movl	%eax,PCB_DR3(%ecx)
+	movl	%dr6,%eax
+	movl	%eax,PCB_DR6(%ecx)
+	movl	%dr7,%eax
+	movl	%eax,PCB_DR7(%ecx)
+
+	mov	%ds,PCB_DS(%ecx)
+	mov	%es,PCB_ES(%ecx)
+	mov	%fs,PCB_FS(%ecx)
+	mov	%ss,PCB_SS(%ecx)
+	
+	sgdt	PCB_GDT(%ecx)
+	sidt	PCB_IDT(%ecx)
+	sldt	PCB_LDT(%ecx)
+	str	PCB_TR(%ecx)
+
 #ifdef DEV_NPX
 	/*
 	 * If fpcurthread == NULL, then the npx h/w state is irrelevant and the
@@ -425,64 +455,9 @@
 	popfl
 #endif	/* DEV_NPX */
 
-	ret
-END(savectx)
-
-/*
- * suspendctx(pcb)
- * Update pcb, suspending current processor state.
- */
-ENTRY(suspendctx)
-	/* Fetch PCB. */
-	movl	4(%esp),%ecx
-
-	/* Save context by calling savectx(). */
-	pushl	%ecx
-	call	savectx
-	addl	$4,%esp
-
-	/* Fetch PCB again. */
-	movl	4(%esp),%ecx
-
-	/* Update caller's return address and stack pointer. */
-	movl	(%esp),%eax
-	movl	%eax,PCB_EIP(%ecx)
-	movl	%esp,PCB_ESP(%ecx)
-
-	/* Save other registers and descriptor tables. */
-	movl	%cr0,%eax
-	movl	%eax,PCB_CR0(%ecx)
-	movl	%cr2,%eax
-	movl	%eax,PCB_CR2(%ecx)
-	movl	%cr4,%eax
-	movl	%eax,PCB_CR4(%ecx)
-
-	movl	%dr0,%eax
-	movl	%eax,PCB_DR0(%ecx)
-	movl	%dr1,%eax
-	movl	%eax,PCB_DR1(%ecx)
-	movl	%dr2,%eax
-	movl	%eax,PCB_DR2(%ecx)
-	movl	%dr3,%eax
-	movl	%eax,PCB_DR3(%ecx)
-	movl	%dr6,%eax
-	movl	%eax,PCB_DR6(%ecx)
-	movl	%dr7,%eax
-	movl	%eax,PCB_DR7(%ecx)
-
-	mov	%ds,PCB_DS(%ecx)
-	mov	%es,PCB_ES(%ecx)
-	mov	%fs,PCB_FS(%ecx)
-	mov	%ss,PCB_SS(%ecx)
-	
-	sgdt	PCB_GDT(%ecx)
-	sidt	PCB_IDT(%ecx)
-	sldt	PCB_LDT(%ecx)
-	str	PCB_TR(%ecx)
-
 	movl	$1,%eax
 	ret
-END(suspendctx)
+END(savectx)
 
 /*
  * resumectx(pcb in %esi)
Index: i386/include/pcb.h
===================================================================
RCS file: /home/ncvs/src/sys/i386/include/pcb.h,v
retrieving revision 1.61
diff -u -r1.61 pcb.h
--- i386/include/pcb.h	18 May 2012 18:55:58 -0000	1.61
+++ i386/include/pcb.h	6 Jun 2012 17:15:08 -0000
@@ -96,8 +96,7 @@
 struct trapframe;
 
 void	makectx(struct trapframe *, struct pcb *);
-void	savectx(struct pcb *) __returns_twice;
-int	suspendctx(struct pcb *) __returns_twice;
+int	savectx(struct pcb *) __returns_twice;
 void	resumectx(struct pcb *);
 #endif
 
Index: kern/subr_smp.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/subr_smp.c,v
retrieving revision 1.236
diff -u -r1.236 subr_smp.c
--- kern/subr_smp.c	18 May 2012 18:55:58 -0000	1.236
+++ kern/subr_smp.c	6 Jun 2012 17:08:55 -0000
@@ -55,6 +55,7 @@
 #ifdef SMP
 volatile cpuset_t stopped_cpus;
 volatile cpuset_t started_cpus;
+volatile cpuset_t suspended_cpus;
 cpuset_t hlt_cpus_mask;
 cpuset_t logical_cpus_mask;
 
Index: sys/smp.h
===================================================================
RCS file: /home/ncvs/src/sys/sys/smp.h,v
retrieving revision 1.98
diff -u -r1.98 smp.h
--- sys/smp.h	18 May 2012 18:55:58 -0000	1.98
+++ sys/smp.h	6 Jun 2012 17:08:55 -0000
@@ -75,6 +75,7 @@
 extern int smp_cpus;
 extern volatile cpuset_t started_cpus;
 extern volatile cpuset_t stopped_cpus;
+extern volatile cpuset_t suspended_cpus;
 extern cpuset_t hlt_cpus_mask;
 extern cpuset_t logical_cpus_mask;
 #endif /* SMP */