diff --git a/sys/arch/amd64/amd64/fpu.c b/sys/arch/amd64/amd64/fpu.c
index 7cc706d45d02..e1d9507985f2 100644
--- a/sys/arch/amd64/amd64/fpu.c
+++ b/sys/arch/amd64/amd64/fpu.c
@@ -1,4 +1,4 @@
-/*	$NetBSD: fpu.c,v 1.28 2008/11/11 14:40:18 ad Exp $	*/
+/*	$NetBSD: fpu.c,v 1.29 2008/11/11 15:53:53 ad Exp $	*/
 
 /*-
  * Copyright (c) 2008 The NetBSD Foundation, Inc.  All
@@ -100,7 +100,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: fpu.c,v 1.28 2008/11/11 14:40:18 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: fpu.c,v 1.29 2008/11/11 15:53:53 ad Exp $");
 
 #include "opt_multiprocessor.h"
 
@@ -394,13 +394,13 @@ fpusave_lwp(struct lwp *l, bool save)
 		oci = l->l_addr->u_pcb.pcb_fpcpu;
 		if (oci == NULL) {
 			splx(s);
-			return;
+			break;
 		}
 		if (oci == curcpu()) {
 			KASSERT(oci->ci_fpcurlwp == l);
 			fpusave_cpu(save);
 			splx(s);
-			return;
+			break;
 		}
 		splx(s);
 		x86_send_ipi(oci, X86_IPI_SYNCH_FPU);
@@ -413,4 +413,9 @@ fpusave_lwp(struct lwp *l, bool save)
 			panic("fpusave_lwp: did not");
 		}
 	}
+
+	if (!save) {
+		/* Ensure we restart with a clean slate. */
+	 	l->l_md.md_flags &= ~MDP_USEDFPU;
+	}
 }
diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index 1d540141fcd1..9fdc16fd9d44 100644
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1,4 +1,4 @@
-/*	$NetBSD: machdep.c,v 1.104 2008/11/11 14:40:18 ad Exp $	*/
+/*	$NetBSD: machdep.c,v 1.105 2008/11/11 15:53:53 ad Exp $	*/
 
 /*-
  * Copyright (c) 1996, 1997, 1998, 2000, 2006, 2007, 2008
@@ -112,7 +112,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.104 2008/11/11 14:40:18 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.105 2008/11/11 15:53:53 ad Exp $");
 
 /* #define XENDEBUG_LOW  */
 
@@ -537,6 +537,9 @@ buildcontext(struct lwp *l, void *catcher, void *f)
 	tf->tf_rflags &= ~PSL_CLEARSIG;
 	tf->tf_rsp = (uint64_t)f;
 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
+
+	/* Ensure FP state is reset, if FP is used. */
+	l->l_md.md_flags &= ~MDP_USEDFPU;
 }
 
 void
@@ -1947,9 +1950,10 @@ cpu_setmcontext(struct lwp *l, const mcontext_t *mcp, unsigned int flags)
 		l->l_md.md_flags |= MDP_IRET;
 	}
 
+	if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
+		fpusave_lwp(l, false);
+
 	if ((flags & _UC_FPU) != 0) {
-		if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
-			fpusave_lwp(l, false);
 		memcpy(&l->l_addr->u_pcb.pcb_savefpu.fp_fxsave, mcp->__fpregs,
 		    sizeof (mcp->__fpregs));
 		l->l_md.md_flags |= MDP_USEDFPU;
diff --git a/sys/arch/i386/i386/machdep.c b/sys/arch/i386/i386/machdep.c
index 923c570e6ffa..49494fe26eb8 100644
--- a/sys/arch/i386/i386/machdep.c
+++ b/sys/arch/i386/i386/machdep.c
@@ -1,4 +1,4 @@
-/*	$NetBSD: machdep.c,v 1.646 2008/11/11 14:40:18 ad Exp $	*/
+/*	$NetBSD: machdep.c,v 1.647 2008/11/11 15:53:53 ad Exp $	*/
 
 /*-
  * Copyright (c) 1996, 1997, 1998, 2000, 2004, 2006, 2008 The NetBSD Foundation, Inc.
@@ -65,7 +65,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.646 2008/11/11 14:40:18 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.647 2008/11/11 15:53:53 ad Exp $");
 
 #include "opt_beep.h"
 #include "opt_compat_ibcs2.h"
@@ -770,6 +770,9 @@ buildcontext(struct lwp *l, int sel, void *catcher, void *fp)
 	tf->tf_eflags &= ~PSL_CLEARSIG;
 	tf->tf_esp = (int)fp;
 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
+
+	/* Ensure FP state is reset, if FP is used. */
+	l->l_md.md_flags &= ~MDL_USEDFPU;
 }
 
 static void
@@ -2211,8 +2214,6 @@ cpu_getmcontext(struct lwp *l, mcontext_t *mcp, unsigned int *flags)
 		/*
 		 * If this process is the current FP owner, dump its
 		 * context to the PCB first.
-		 * XXX npxsave() also clears the FPU state; depending on the
-		 * XXX application this might be a penalty.
 		 */
 		if (l->l_addr->u_pcb.pcb_fpcpu) {
 			npxsave_lwp(l, true);
@@ -2294,15 +2295,16 @@ cpu_setmcontext(struct lwp *l, const mcontext_t *mcp, unsigned int flags)
 		tf->tf_ss     = gr[_REG_SS];
 	}
 
+#if NNPX > 0
+	/*
+	 * If we were using the FPU, forget that we were.
+	 */
+	if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
+		npxsave_lwp(l, false);
+#endif
+
 	/* Restore floating point register context, if any. */
 	if ((flags & _UC_FPU) != 0) {
-#if NNPX > 0
-		/*
-		 * If we were using the FPU, forget that we were.
-		 */
-		if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
-			npxsave_lwp(l, false);
-#endif
 		if (flags & _UC_FXSAVE) {
 			if (i386_use_fxsave) {
 				memcpy(
@@ -2326,12 +2328,7 @@ cpu_setmcontext(struct lwp *l, const mcontext_t *mcp, unsigned int flags)
 				    sizeof (l->l_addr->u_pcb.pcb_savefpu.sv_87));
 			}
 		}
-		/* If not set already. */
 		l->l_md.md_flags |= MDL_USEDFPU;
-#if 0
-		/* Apparently unused. */
-		l->l_addr->u_pcb.pcb_saveemc = mcp->mc_fp.fp_emcsts;
-#endif
 	}
 	mutex_enter(p->p_lock);
 	if (flags & _UC_SETSTACK)
diff --git a/sys/arch/i386/isa/npx.c b/sys/arch/i386/isa/npx.c
index 027b384a3809..197f3503c53a 100644
--- a/sys/arch/i386/isa/npx.c
+++ b/sys/arch/i386/isa/npx.c
@@ -1,4 +1,4 @@
-/*	$NetBSD: npx.c,v 1.131 2008/11/11 14:40:19 ad Exp $	*/
+/*	$NetBSD: npx.c,v 1.132 2008/11/11 15:53:53 ad Exp $	*/
 
 /*-
  * Copyright (c) 2008 The NetBSD Foundation, Inc.
@@ -96,7 +96,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: npx.c,v 1.131 2008/11/11 14:40:19 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: npx.c,v 1.132 2008/11/11 15:53:53 ad Exp $");
 
 #if 0
 #define IPRINTF(x)	printf x
@@ -700,13 +700,13 @@ npxsave_lwp(struct lwp *l, bool save)
 		oci = l->l_addr->u_pcb.pcb_fpcpu;
 		if (oci == NULL) {
 			splx(s);
-			return;
+			break;
 		}
 		if (oci == curcpu()) {
 			KASSERT(oci->ci_fpcurlwp == l);
 			npxsave_cpu(save);
 			splx(s);
-			return;
+			break;
 		}
 		splx(s);
 		x86_send_ipi(oci, X86_IPI_SYNCH_FPU);
@@ -719,4 +719,9 @@ npxsave_lwp(struct lwp *l, bool save)
 			panic("npxsave_lwp: did not");
 		}
 	}
+
+	if (!save) {
+		/* Ensure we restart with a clean slate. */
+	 	l->l_md.md_flags &= ~MDL_USEDFPU;
+	}
 }