diff --git a/sys/amd64/amd64/exception.S b/sys/amd64/amd64/exception.S
index 68a87d36d540..a005c3b70591 100644
--- a/sys/amd64/amd64/exception.S
+++ b/sys/amd64/amd64/exception.S
@@ -345,10 +345,11 @@ IDTVEC(dblfault)
 	pushfq
 	andq	$~(PSL_D | PSL_AC),(%rsp)
 	popfq
-	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
-	jz	1f			/* already running with kernel GS.base */
-	swapgs
-1:	lfence
+	movq	TF_SIZE(%rsp),%rdx
+	movl	%edx,%eax
+	shrq	$32,%rdx
+	movl	$MSR_GSBASE,%ecx
+	wrmsr
 	movq	%cr3,%rax
 	movq	%rax,PCPU(SAVED_UCR3)
 	movq	PCPU(KCR3),%rax
diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c
index fbc5342b4c47..0575fe5e9611 100644
--- a/sys/amd64/amd64/machdep.c
+++ b/sys/amd64/amd64/machdep.c
@@ -1575,7 +1575,9 @@ amd64_bsp_ist_init(struct pcpu *pc)
 	tssp = &pc->pc_common_tss;
 
 	/* doublefault stack space, runs on ist1 */
-	tssp->tss_ist1 = (long)&dblfault_stack[sizeof(dblfault_stack)];
+	np = ((struct nmi_pcpu *)&dblfault_stack[sizeof(dblfault_stack)]) - 1;
+	np->np_pcpu = (register_t)pc;
+	tssp->tss_ist1 = (long)np;
 
 	/*
 	 * NMI stack, runs on ist2.  The pcpu pointer is stored just
diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 7ac40f507caa..9511a9c018c8 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -314,18 +314,24 @@ init_secondary(void)
 	    IOPERM_BITMAP_SIZE;
 	pc->pc_common_tss.tss_rsp0 = 0;
 
-	pc->pc_common_tss.tss_ist1 = (long)&doublefault_stack[PAGE_SIZE];
+	/* The doublefault stack runs on IST1. */
+	np = ((struct nmi_pcpu *)&doublefault_stack[PAGE_SIZE]) - 1;
+	np->np_pcpu = (register_t)pc;
+	pc->pc_common_tss.tss_ist1 = (long)np;
 
 	/* The NMI stack runs on IST2. */
 	np = ((struct nmi_pcpu *) &nmi_stack[PAGE_SIZE]) - 1;
+	np->np_pcpu = (register_t)pc;
 	pc->pc_common_tss.tss_ist2 = (long)np;
 
 	/* The MC# stack runs on IST3. */
 	np = ((struct nmi_pcpu *) &mce_stack[PAGE_SIZE]) - 1;
+	np->np_pcpu = (register_t)pc;
 	pc->pc_common_tss.tss_ist3 = (long)np;
 
 	/* The DB# stack runs on IST4. */
 	np = ((struct nmi_pcpu *) &dbg_stack[PAGE_SIZE]) - 1;
+	np->np_pcpu = (register_t)pc;
 	pc->pc_common_tss.tss_ist4 = (long)np;
 
 	/* Prepare private GDT */
@@ -341,18 +347,6 @@ init_secondary(void)
 	ap_gdt.rd_base = (u_long)gdt;
 	lgdt(&ap_gdt);			/* does magic intra-segment return */
 
-	/* Save the per-cpu pointer for use by the NMI handler. */
-	np = ((struct nmi_pcpu *) &nmi_stack[PAGE_SIZE]) - 1;
-	np->np_pcpu = (register_t)pc;
-
-	/* Save the per-cpu pointer for use by the MC# handler. */
-	np = ((struct nmi_pcpu *) &mce_stack[PAGE_SIZE]) - 1;
-	np->np_pcpu = (register_t)pc;
-
-	/* Save the per-cpu pointer for use by the DB# handler. */
-	np = ((struct nmi_pcpu *) &dbg_stack[PAGE_SIZE]) - 1;
-	np->np_pcpu = (register_t)pc;
-
 	wrmsr(MSR_FSBASE, 0);		/* User value */
 	wrmsr(MSR_GSBASE, (u_int64_t)pc);
 	wrmsr(MSR_KGSBASE, (u_int64_t)pc);	/* XXX User value while we're in the kernel */