diff --git a/sys/amd64/acpica/acpi_wakeup.c b/sys/amd64/acpica/acpi_wakeup.c
index d53d8bb..937b914 100644
--- a/sys/amd64/acpica/acpi_wakeup.c
+++ b/sys/amd64/acpica/acpi_wakeup.c
@@ -245,7 +245,7 @@ acpi_sleep_machdep(struct acpi_softc *sc, int state)
 	cr3 = rcr3();
 	load_cr3(KPML4phys);
 
-	stopfpu = &stopxpcbs[0]->xpcb_pcb.pcb_save;
+	stopfpu = stopxpcbs[0]->xpcb_pcb.pcb_save;
 	if (acpi_savecpu(stopxpcbs[0])) {
 		fpugetregs(curthread, stopfpu);
 
diff --git a/sys/amd64/amd64/cpu_switch.S b/sys/amd64/amd64/cpu_switch.S
index 364875e..c71bcd0 100644
--- a/sys/amd64/amd64/cpu_switch.S
+++ b/sys/amd64/amd64/cpu_switch.S
@@ -116,7 +116,7 @@ done_store_dr:
 	/* have we used fp, and need a save? */
 	cmpq	%rdi,PCPU(FPCURTHREAD)
 	jne	1f
-	addq	$PCB_SAVEFPU,%r8
+	movq	PCB_SAVEFPU(%r8),%r8
 	clts
 	fxsave	(%r8)
 	smsw	%ax
@@ -341,7 +341,7 @@ ENTRY(savectx)
 	je	1f
 
 	movq	TD_PCB(%rax),%rdi
-	leaq	PCB_SAVEFPU(%rdi),%rdi
+	movq	PCB_SAVEFPU(%rdi),%rdi
 	clts
 	fxsave	(%rdi)
 	smsw	%ax
@@ -349,7 +349,7 @@ ENTRY(savectx)
 	lmsw	%ax
 
 	movq	$PCB_SAVEFPU_SIZE,%rdx	/* arg 3 */
-	leaq	PCB_SAVEFPU(%rcx),%rsi	/* arg 2 */
+	movq	PCB_SAVEFPU(%rcx),%rsi	/* arg 2 */
 	/* arg 1 (%rdi) already loaded */
 	call	bcopy
 1:
diff --git a/sys/amd64/amd64/fpu.c b/sys/amd64/amd64/fpu.c
index 10cb6c2..3e99b08 100644
--- a/sys/amd64/amd64/fpu.c
+++ b/sys/amd64/amd64/fpu.c
@@ -91,8 +91,8 @@ void	stop_emulating(void);
 
 #endif	/* __GNUCLIKE_ASM && !lint */
 
-#define GET_FPU_CW(thread) ((thread)->td_pcb->pcb_save.sv_env.en_cw)
-#define GET_FPU_SW(thread) ((thread)->td_pcb->pcb_save.sv_env.en_sw)
+#define GET_FPU_CW(thread) ((thread)->td_pcb->pcb_save->sv_env.en_cw)
+#define GET_FPU_SW(thread) ((thread)->td_pcb->pcb_save->sv_env.en_sw)
 
 typedef u_char bool_t;
 
@@ -146,7 +146,7 @@ fpuexit(struct thread *td)
 	savecrit = intr_disable();
 	if (curthread == PCPU_GET(fpcurthread)) {
 		stop_emulating();
-		fxsave(&PCPU_GET(curpcb)->pcb_save);
+		fxsave(PCPU_GET(curpcb)->pcb_save);
 		start_emulating();
 		PCPU_SET(fpcurthread, 0);
 	}
@@ -424,8 +424,10 @@ fpudna(void)
 		if (pcb->pcb_initial_fpucw != __INITIAL_FPUCW__)
 			fldcw(&pcb->pcb_initial_fpucw);
 		pcb->pcb_flags |= PCB_FPUINITDONE;
+		if (PCB_USER_FPU(pcb))
+			pcb->pcb_flags |= PCB_USERFPUINITDONE;
 	} else
-		fxrstor(&pcb->pcb_save);
+		fxrstor(pcb->pcb_save);
 	intr_restore(s);
 }
 
@@ -449,13 +451,39 @@ fpudrop()
  * It returns the FPU ownership status.
  */
 int
+fpugetuserregs(struct thread *td, struct savefpu *addr)
+{
+	register_t s;
+	struct pcb *pcb;
+
+	pcb = td->td_pcb;
+	if ((pcb->pcb_flags & PCB_USERFPUINITDONE) == 0) {
+		bcopy(&fpu_initialstate, addr, sizeof(fpu_initialstate));
+		addr->sv_env.en_cw = pcb->pcb_initial_fpucw;
+		return (_MC_FPOWNED_NONE);
+	}
+	s = intr_disable();
+	if (td == PCPU_GET(fpcurthread) && PCB_USER_FPU(pcb)) {
+		fxsave(addr);
+		intr_restore(s);
+		return (_MC_FPOWNED_FPU);
+	} else {
+		intr_restore(s);
+		bcopy(&pcb->pcb_user_save, addr, sizeof(*addr));
+		return (_MC_FPOWNED_PCB);
+	}
+}
+
+int
 fpugetregs(struct thread *td, struct savefpu *addr)
 {
 	register_t s;
+	struct pcb *pcb;
 
-	if ((td->td_pcb->pcb_flags & PCB_FPUINITDONE) == 0) {
+	pcb = td->td_pcb;
+	if ((pcb->pcb_flags & PCB_FPUINITDONE) == 0) {
 		bcopy(&fpu_initialstate, addr, sizeof(fpu_initialstate));
-		addr->sv_env.en_cw = td->td_pcb->pcb_initial_fpucw;
+		addr->sv_env.en_cw = pcb->pcb_initial_fpucw;
 		return (_MC_FPOWNED_NONE);
 	}
 	s = intr_disable();
@@ -465,7 +493,7 @@ fpugetregs(struct thread *td, struct savefpu *addr)
 		return (_MC_FPOWNED_FPU);
 	} else {
 		intr_restore(s);
-		bcopy(&td->td_pcb->pcb_save, addr, sizeof(*addr));
+		bcopy(pcb->pcb_save, addr, sizeof(*addr));
 		return (_MC_FPOWNED_PCB);
 	}
 }
@@ -474,19 +502,42 @@ fpugetregs(struct thread *td, struct savefpu *addr)
  * Set the state of the FPU.
  */
 void
+fpusetuserregs(struct thread *td, struct savefpu *addr)
+{
+	register_t s;
+	struct pcb *pcb;
+
+	pcb = td->td_pcb;
+	s = intr_disable();
+	if (td == PCPU_GET(fpcurthread) && PCB_USER_FPU(pcb)) {
+		fxrstor(addr);
+		intr_restore(s);
+		pcb->pcb_flags |= PCB_FPUINITDONE | PCB_USERFPUINITDONE;
+	} else {
+		intr_restore(s);
+		bcopy(addr, &td->td_pcb->pcb_user_save, sizeof(*addr));
+		pcb->pcb_flags |= PCB_USERFPUINITDONE;
+	}
+}
+
+void
 fpusetregs(struct thread *td, struct savefpu *addr)
 {
 	register_t s;
+	struct pcb *pcb;
 
+	pcb = td->td_pcb;
 	s = intr_disable();
 	if (td == PCPU_GET(fpcurthread)) {
 		fxrstor(addr);
 		intr_restore(s);
 	} else {
 		intr_restore(s);
-		bcopy(addr, &td->td_pcb->pcb_save, sizeof(*addr));
+		bcopy(addr, td->td_pcb->pcb_save, sizeof(*addr));
+		if (PCB_USER_FPU(pcb))
+			pcb->pcb_flags |= PCB_USERFPUINITDONE;
 	}
-	curthread->td_pcb->pcb_flags |= PCB_FPUINITDONE;
+	pcb->pcb_flags |= PCB_FPUINITDONE;
 }
 
 /*
@@ -575,3 +626,22 @@ static devclass_t fpupnp_devclass;
 
 DRIVER_MODULE(fpupnp, acpi, fpupnp_driver, fpupnp_devclass, 0, 0);
 #endif	/* DEV_ISA */
+
+int
+fpu_kern_enter(struct thread *td, struct fpu_kern_ctx *ctx)
+{
+
+	fpuexit(td);
+	ctx->prev = td->td_pcb->pcb_save;
+	td->td_pcb->pcb_save = &ctx->hwstate;
+	return (0);
+}
+
+int
+fpu_kern_leave(struct thread *td, struct fpu_kern_ctx *ctx)
+{
+
+	fpudrop();
+	td->td_pcb->pcb_save = ctx->prev;
+	return (0);
+}
diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c
index 2318975..62e0be5 100644
--- a/sys/amd64/amd64/machdep.c
+++ b/sys/amd64/amd64/machdep.c
@@ -1962,7 +1962,7 @@ int
 fill_fpregs(struct thread *td, struct fpreg *fpregs)
 {
 
-	fill_fpregs_xmm(&td->td_pcb->pcb_save, fpregs);
+	fill_fpregs_xmm(&td->td_pcb->pcb_user_save, fpregs);
 	return (0);
 }
 
@@ -1971,7 +1971,7 @@ int
 set_fpregs(struct thread *td, struct fpreg *fpregs)
 {
 
-	set_fpregs_xmm(fpregs, &td->td_pcb->pcb_save);
+	set_fpregs_xmm(fpregs, &td->td_pcb->pcb_user_save);
 	return (0);
 }
 
@@ -2086,7 +2086,8 @@ static void
 get_fpcontext(struct thread *td, mcontext_t *mcp)
 {
 
-	mcp->mc_ownedfp = fpugetregs(td, (struct savefpu *)&mcp->mc_fpstate);
+	mcp->mc_ownedfp = fpugetuserregs(td,
+	    (struct savefpu *)&mcp->mc_fpstate);
 	mcp->mc_fpformat = fpuformat();
 }
 
@@ -2111,7 +2112,7 @@ set_fpcontext(struct thread *td, const mcontext_t *mcp)
 		 */
 		fpstate = (struct savefpu *)&mcp->mc_fpstate;
 		fpstate->sv_env.en_mxcsr &= cpu_mxcsr_mask;
-		fpusetregs(td, fpstate);
+		fpusetuserregs(td, fpstate);
 	} else
 		return (EINVAL);
 	return (0);
diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 337c028..325ffe5 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -1250,7 +1250,7 @@ cpususpend_handler(void)
 
 	rf = intr_disable();
 	cr3 = rcr3();
-	stopfpu = &stopxpcbs[cpu]->xpcb_pcb.pcb_save;
+	stopfpu = stopxpcbs[cpu]->xpcb_pcb.pcb_save;
 	if (savectx2(stopxpcbs[cpu])) {
 		fpugetregs(curthread, stopfpu);
 		wbinvd();
diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c
index bd7ee63..1c42fe3 100644
--- a/sys/amd64/amd64/trap.c
+++ b/sys/amd64/amd64/trap.c
@@ -476,6 +476,8 @@ trap(struct trapframe *frame)
 
 		case T_DNA:
 			/* transparent fault (due to context switch "late") */
+			KASSERT(PCB_USER_FPU(td->td_pcb),
+			    ("kernel FPU ctx has leaked"));
 			fpudna();
 			goto userout;
 
@@ -500,16 +502,19 @@ trap(struct trapframe *frame)
 			goto out;
 
 		case T_DNA:
+			KASSERT(!PCB_USER_FPU(td->td_pcb),
+			    ("Unregistered use of FPU in kernel"));
+			fpudna();
+			goto out;
+
+		case T_ARITHTRAP:	/* arithmetic trap */
+		case T_XMMFLT:		/* SIMD floating-point exception */
+		case T_FPOPFLT:		/* FPU operand fetch fault */
 			/*
-			 * The kernel is apparently using fpu for copying.
-			 * XXX this should be fatal unless the kernel has
-			 * registered such use.
+			 * XXXKIB for now disable any FPU traps in kernel
+			 * handler registration seems to be overkill
 			 */
-			printf("fpudna in kernel mode!\n");
-#ifdef KDB
-			kdb_backtrace();
-#endif
-			fpudna();
+			trap_fatal(frame, 0);
 			goto out;
 
 		case T_STKFLT:		/* stack fault */
diff --git a/sys/amd64/amd64/vm_machdep.c b/sys/amd64/amd64/vm_machdep.c
index 3f7d76a..41d496d 100644
--- a/sys/amd64/amd64/vm_machdep.c
+++ b/sys/amd64/amd64/vm_machdep.c
@@ -122,7 +122,7 @@ cpu_fork(td1, p2, td2, flags)
 		return;
 	}
 
-	/* Ensure that p1's pcb is up to date. */
+	/* Ensure that td1's pcb is up to date. */
 	fpuexit(td1);
 
 	/* Point the pcb to the top of the stack */
@@ -133,6 +133,9 @@ cpu_fork(td1, p2, td2, flags)
 	/* Copy p1's pcb */
 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
 
+	/* Properly initialize pcb_save */
+	pcb2->pcb_save = &pcb2->pcb_user_save;
+
 	/* Point mdproc and then copy over td1's contents */
 	mdp2 = &p2->p_md;
 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
@@ -381,7 +384,7 @@ cpu_set_upcall(struct thread *td, struct thread *td0)
 	 * values here.
 	 */
 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
-	pcb2->pcb_flags &= ~PCB_FPUINITDONE;
+	pcb2->pcb_flags &= ~(PCB_FPUINITDONE | PCB_USERFPUINITDONE);
 	pcb2->pcb_full_iret = 1;
 
 	/*
diff --git a/sys/amd64/ia32/ia32_reg.c b/sys/amd64/ia32/ia32_reg.c
index da5190f..30fcffb 100644
--- a/sys/amd64/ia32/ia32_reg.c
+++ b/sys/amd64/ia32/ia32_reg.c
@@ -147,7 +147,7 @@ fill_fpregs32(struct thread *td, struct fpreg32 *regs)
 {
 	struct save87 *sv_87 = (struct save87 *)regs;
 	struct env87 *penv_87 = &sv_87->sv_env;
-	struct savefpu *sv_fpu = &td->td_pcb->pcb_save;
+	struct savefpu *sv_fpu = &td->td_pcb->pcb_user_save;
 	struct envxmm *penv_xmm = &sv_fpu->sv_env;
 	int i;
 
@@ -182,7 +182,7 @@ set_fpregs32(struct thread *td, struct fpreg32 *regs)
 {
 	struct save87 *sv_87 = (struct save87 *)regs;
 	struct env87 *penv_87 = &sv_87->sv_env;
-	struct savefpu *sv_fpu = &td->td_pcb->pcb_save;
+	struct savefpu *sv_fpu = &td->td_pcb->pcb_user_save;
 	struct envxmm *penv_xmm = &sv_fpu->sv_env;
 	int i;
 
diff --git a/sys/amd64/include/fpu.h b/sys/amd64/include/fpu.h
index 272f94a..3ce8e9a 100644
--- a/sys/amd64/include/fpu.h
+++ b/sys/amd64/include/fpu.h
@@ -73,6 +73,15 @@ struct  savefpu {
 	u_char sv_pad[96];
 } __aligned(16);
 
+#ifdef _KERNEL
+struct fpu_kern_ctx {
+	struct savefpu hwstate;
+	struct savefpu *prev;
+};
+
+#define	PCB_USER_FPU(pcb) ((pcb)->pcb_save == &(pcb)->pcb_user_save)
+#endif
+
 /*
  * The hardware default control word for i387's and later coprocessors is
  * 0x37F, giving:
@@ -102,9 +111,13 @@ void	fpudrop(void);
 void	fpuexit(struct thread *td);
 int	fpuformat(void);
 int	fpugetregs(struct thread *td, struct savefpu *addr);
+int	fpugetuserregs(struct thread *td, struct savefpu *addr);
 void	fpuinit(void);
 void	fpusetregs(struct thread *td, struct savefpu *addr);
+void	fpusetuserregs(struct thread *td, struct savefpu *addr);
 int	fputrap(void);
+int	fpu_kern_enter(struct thread *td, struct fpu_kern_ctx *ctx);
+int	fpu_kern_leave(struct thread *td, struct fpu_kern_ctx *ctx);
 #endif
 
 #endif /* !_MACHINE_FPU_H_ */
diff --git a/sys/amd64/include/pcb.h b/sys/amd64/include/pcb.h
index b26188a..344d56b 100644
--- a/sys/amd64/include/pcb.h
+++ b/sys/amd64/include/pcb.h
@@ -58,6 +58,7 @@ struct pcb {
 	u_long		pcb_flags;
 #define	PCB_DBREGS	0x02	/* process using debug registers */
 #define	PCB_FPUINITDONE	0x08	/* fpu state is initialized */
+#define	PCB_USERFPUINITDONE 0x10 /* fpu user state is initialized */
 #define	PCB_GS32BIT	0x20	/* linux gs switch */
 #define	PCB_32BIT	0x40	/* process has 32 bit context (segs etc) */
 #define	PCB_FULLCTX	0x80	/* full context restore on sysret */
@@ -69,7 +70,7 @@ struct pcb {
 	u_int64_t	pcb_dr6;
 	u_int64_t	pcb_dr7;
 
-	struct	savefpu	pcb_save;
+	struct	savefpu pcb_user_save;
 	uint16_t	pcb_initial_fpucw;
 
 	caddr_t		pcb_onfault; /* copyin/out fault recovery */
@@ -78,6 +79,7 @@ struct pcb {
 	struct user_segment_descriptor	pcb_gs32sd;
 	/* local tss, with i/o bitmap; NULL for common */
 	struct amd64tss *pcb_tssp;
+	struct	savefpu	*pcb_save;
 	char		pcb_full_iret;
 };