diff --git a/Makefile b/Makefile
index 5904a3f967e3..7e76ac915fc7 100644
--- a/Makefile
+++ b/Makefile
@@ -239,7 +239,7 @@ _MAKE+=	MK_META_MODE=no
 _TARGET_ARCH=	${TARGET:S/pc98/i386/:S/arm64/aarch64/}
 .elif !defined(TARGET) && defined(TARGET_ARCH) && \
     ${TARGET_ARCH} != ${MACHINE_ARCH}
-_TARGET=		${TARGET_ARCH:C/mips(n32|64)?(el)?(hf)?/mips/:C/arm(v6)?(eb)?/arm/:C/aarch64/arm64/:C/powerpc64/powerpc/:C/powerpcspe/powerpc/:C/riscv64/riscv/}
+_TARGET=		${TARGET_ARCH:C/mips(n32|64)?(el)?(hf)?/mips/:C/arm(v6)?(eb)?/arm/:C/aarch64/arm64/:C/powerpc64/powerpc/:C/powerpcspe/powerpc/:C/riscv64(sf)?/riscv/}
 .endif
 .if defined(TARGET) && !defined(_TARGET)
 _TARGET=${TARGET}
diff --git a/Makefile.inc1 b/Makefile.inc1
index d78cfd26a162..a5b44cb6c8b9 100644
--- a/Makefile.inc1
+++ b/Makefile.inc1
@@ -364,6 +364,7 @@ KNOWN_ARCHES?=	aarch64/arm64 \
 		powerpc64/powerpc \
 		powerpcspe/powerpc \
 		riscv64/riscv \
+		riscv64sf/riscv \
 		sparc64
 
 .if ${TARGET} == ${TARGET_ARCH}
diff --git a/gnu/usr.bin/cc/Makefile.tgt b/gnu/usr.bin/cc/Makefile.tgt
index 3acd4d7c9295..eb8ee7b4db8e 100644
--- a/gnu/usr.bin/cc/Makefile.tgt
+++ b/gnu/usr.bin/cc/Makefile.tgt
@@ -4,7 +4,7 @@
 # MACHINE_CPUARCH, but there's no easy way to export make functions...
 
 .if defined(TARGET_ARCH)
-TARGET_CPUARCH=${TARGET_ARCH:C/mips(n32|64)?(el)?(hf)?/mips/:C/arm(v6)?(eb)?/arm/:C/powerpc(64|spe)/powerpc/}
+TARGET_CPUARCH=${TARGET_ARCH:C/mips(n32|64)?(el)?(hf)?/mips/:C/arm(v6)?(eb)?/arm/:C/powerpc(64|spe)/powerpc/:C/riscv64(sf)?/riscv64/}
 .else
 TARGET_CPUARCH=${MACHINE_CPUARCH}
 .endif
diff --git a/lib/libc/Makefile b/lib/libc/Makefile
index 99ddfc2e8f7f..377a6ee6cd11 100644
--- a/lib/libc/Makefile
+++ b/lib/libc/Makefile
@@ -111,7 +111,8 @@ NOASM=
 .include "${LIBC_SRCTOP}/xdr/Makefile.inc"
 .if (${LIBC_ARCH} == "arm" && \
 	(${MACHINE_ARCH:Marmv6*} == "" || (defined(CPUTYPE) && ${CPUTYPE:M*soft*}))) || \
-     (${LIBC_ARCH} == "mips" && ${MACHINE_ARCH:Mmips*hf} == "")
+    (${LIBC_ARCH} == "mips" && ${MACHINE_ARCH:Mmips*hf} == "") || \
+    (${LIBC_ARCH} == "riscv" && ${MACHINE_ARCH:Mriscv*sf} != "")
 .include "${LIBC_SRCTOP}/softfloat/Makefile.inc"
 .endif
 .if ${LIBC_ARCH} == "i386" || ${LIBC_ARCH} == "amd64"
diff --git a/lib/libc/riscv/Makefile.inc b/lib/libc/riscv/Makefile.inc
index b22190353e32..105839021b96 100644
--- a/lib/libc/riscv/Makefile.inc
+++ b/lib/libc/riscv/Makefile.inc
@@ -3,6 +3,10 @@
 # Machine dependent definitions for the RISC-V architecture.
 #
 
+.if ${MACHINE_ARCH:Mriscv*sf} != ""
+CFLAGS+=-DSOFTFLOAT
+.endif
+
 # Long double is quad precision
 GDTOASRCS+=strtorQ.c
 MDSRCS+=machdep_ldisQ.c
diff --git a/lib/libc/riscv/Symbol.map b/lib/libc/riscv/Symbol.map
index 669dad37983f..b959072e7e0c 100644
--- a/lib/libc/riscv/Symbol.map
+++ b/lib/libc/riscv/Symbol.map
@@ -35,4 +35,22 @@ FBSDprivate_1.0 {
 	_set_tp;
 	_end;
 	__makecontext;
+
+	/* softfloat */
+	__addsf3;
+	__adddf3;
+	__subsf3;
+	__subdf3;
+	__mulsf3;
+	__muldf3;
+	__divsf3;
+	__divdf3;
+	__floatsisf;
+	__floatsidf;
+	__fixsfsi;
+	__fixdfsi;
+	__fixunssfsi;
+	__fixunsdfsi;
+	__extendsfdf2;
+	__truncdfsf2;
 };
diff --git a/lib/libc/riscv/gen/_setjmp.S b/lib/libc/riscv/gen/_setjmp.S
index 1fa064d2ecf2..4e82ae420850 100644
--- a/lib/libc/riscv/gen/_setjmp.S
+++ b/lib/libc/riscv/gen/_setjmp.S
@@ -1,5 +1,5 @@
 /*-
- * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2015-2016 Ruslan Bukin <br@bsdpad.com>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -61,24 +61,21 @@ ENTRY(_setjmp)
 	sd	ra, (12 * 8)(a0)
 	addi	a0, a0, (13 * 8)
 
-#ifndef _STANDALONE
-#if 0
-	/* RISCVTODO */
-	/* Store the vfp registers */
-	fsq	fs0, (0 * 16)(a0)
-	fsq	fs1, (1 * 16)(a0)
-	fsq	fs2, (2 * 16)(a0)
-	fsq	fs3, (3 * 16)(a0)
-	fsq	fs4, (4 * 16)(a0)
-	fsq	fs5, (5 * 16)(a0)
-	fsq	fs6, (6 * 16)(a0)
-	fsq	fs7, (7 * 16)(a0)
-	fsq	fs8, (8 * 16)(a0)
-	fsq	fs9, (9 * 16)(a0)
-	fsq	fs10, (10 * 16)(a0)
-	fsq	fs11, (11 * 16)(a0)
+#if !defined(_STANDALONE) && !defined(SOFTFLOAT)
+	/* Store the fpe registers */
+	fsd	fs0, (0 * 16)(a0)
+	fsd	fs1, (1 * 16)(a0)
+	fsd	fs2, (2 * 16)(a0)
+	fsd	fs3, (3 * 16)(a0)
+	fsd	fs4, (4 * 16)(a0)
+	fsd	fs5, (5 * 16)(a0)
+	fsd	fs6, (6 * 16)(a0)
+	fsd	fs7, (7 * 16)(a0)
+	fsd	fs8, (8 * 16)(a0)
+	fsd	fs9, (9 * 16)(a0)
+	fsd	fs10, (10 * 16)(a0)
+	fsd	fs11, (11 * 16)(a0)
 	addi	a0, a0, (12 * 16)
-#endif
 #endif
 
 	/* Return value */
@@ -117,24 +114,21 @@ ENTRY(_longjmp)
 	ld	ra, (12 * 8)(a0)
 	addi	a0, a0, (13 * 8)
 
-#ifndef _STANDALONE
-#if 0
-	/* RISCVTODO */
-	/* Restore the vfp registers */
-	flq	fs0, (0 * 16)(a0)
-	flq	fs1, (1 * 16)(a0)
-	flq	fs2, (2 * 16)(a0)
-	flq	fs3, (3 * 16)(a0)
-	flq	fs4, (4 * 16)(a0)
-	flq	fs5, (5 * 16)(a0)
-	flq	fs6, (6 * 16)(a0)
-	flq	fs7, (7 * 16)(a0)
-	flq	fs8, (8 * 16)(a0)
-	flq	fs9, (9 * 16)(a0)
-	flq	fs10, (10 * 16)(a0)
-	flq	fs11, (11 * 16)(a0)
+#if !defined(_STANDALONE) && !defined(SOFTFLOAT)
+	/* Restore the fpe registers */
+	fld	fs0, (0 * 16)(a0)
+	fld	fs1, (1 * 16)(a0)
+	fld	fs2, (2 * 16)(a0)
+	fld	fs3, (3 * 16)(a0)
+	fld	fs4, (4 * 16)(a0)
+	fld	fs5, (5 * 16)(a0)
+	fld	fs6, (6 * 16)(a0)
+	fld	fs7, (7 * 16)(a0)
+	fld	fs8, (8 * 16)(a0)
+	fld	fs9, (9 * 16)(a0)
+	fld	fs10, (10 * 16)(a0)
+	fld	fs11, (11 * 16)(a0)
 	addi	a0, a0, (12 * 16)
-#endif
 #endif
 
 	/* Load the return value */
diff --git a/lib/libc/riscv/gen/flt_rounds.c b/lib/libc/riscv/gen/flt_rounds.c
index 179963cfa29e..13bb8b2a1796 100644
--- a/lib/libc/riscv/gen/flt_rounds.c
+++ b/lib/libc/riscv/gen/flt_rounds.c
@@ -1,5 +1,5 @@
 /*-
- * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2015-2016 Ruslan Bukin <br@bsdpad.com>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -40,23 +40,24 @@ __FBSDID("$FreeBSD$");
 #include <fenv.h>
 #include <float.h>
 
+#ifdef	SOFTFLOAT
+#include "softfloat-for-gcc.h"
+#include "milieu.h"
+#include "softfloat.h"
+#endif
+
 int
 __flt_rounds(void)
 {
-#if 0
-	uint64_t fcsr;
-#endif
-	int mode;
+	uint64_t mode;
 
-#if 0
-	__asm __volatile("csrr    %0, fcsr" : "=r" (fcsr));
-	mode = (fcsr & _ROUND_MASK);
+#ifdef SOFTFLOAT
+	mode = __softfloat_float_rounding_mode;
+#else
+	__asm __volatile("csrr %0, fcsr" : "=r" (mode));
 #endif
 
-	/* RISCVTODO */
-	mode = FE_TOWARDZERO; /* softfloat rounding mode */
-
-	switch (mode) {
+	switch (mode & _ROUND_MASK) {
 	case FE_TOWARDZERO:
 		return (0);
 	case FE_TONEAREST:
diff --git a/lib/libc/riscv/gen/setjmp.S b/lib/libc/riscv/gen/setjmp.S
index de8d7b5d5190..9d97b51de6b6 100644
--- a/lib/libc/riscv/gen/setjmp.S
+++ b/lib/libc/riscv/gen/setjmp.S
@@ -1,5 +1,5 @@
 /*-
- * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2015-2016 Ruslan Bukin <br@bsdpad.com>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -75,21 +75,20 @@ ENTRY(setjmp)
 	sd	ra, (12 * 8)(a0)
 	addi	a0, a0, (13 * 8)
 
-#if 0
-	/* RISCVTODO */
-	/* Store the vfp registers */
-	fsq	fs0, (0 * 16)(a0)
-	fsq	fs1, (1 * 16)(a0)
-	fsq	fs2, (2 * 16)(a0)
-	fsq	fs3, (3 * 16)(a0)
-	fsq	fs4, (4 * 16)(a0)
-	fsq	fs5, (5 * 16)(a0)
-	fsq	fs6, (6 * 16)(a0)
-	fsq	fs7, (7 * 16)(a0)
-	fsq	fs8, (8 * 16)(a0)
-	fsq	fs9, (9 * 16)(a0)
-	fsq	fs10, (10 * 16)(a0)
-	fsq	fs11, (11 * 16)(a0)
+#ifndef SOFTFLOAT
+	/* Store the fpe registers */
+	fsd	fs0, (0 * 16)(a0)
+	fsd	fs1, (1 * 16)(a0)
+	fsd	fs2, (2 * 16)(a0)
+	fsd	fs3, (3 * 16)(a0)
+	fsd	fs4, (4 * 16)(a0)
+	fsd	fs5, (5 * 16)(a0)
+	fsd	fs6, (6 * 16)(a0)
+	fsd	fs7, (7 * 16)(a0)
+	fsd	fs8, (8 * 16)(a0)
+	fsd	fs9, (9 * 16)(a0)
+	fsd	fs10, (10 * 16)(a0)
+	fsd	fs11, (11 * 16)(a0)
 	addi	a0, a0, (12 * 16)
 #endif
 
@@ -145,21 +144,20 @@ ENTRY(longjmp)
 	ld	ra, (12 * 8)(a0)
 	addi	a0, a0, (13 * 8)
 
-#if 0
-	/* RISCVTODO */
-	/* Restore the vfp registers */
-	flq	fs0, (0 * 16)(a0)
-	flq	fs1, (1 * 16)(a0)
-	flq	fs2, (2 * 16)(a0)
-	flq	fs3, (3 * 16)(a0)
-	flq	fs4, (4 * 16)(a0)
-	flq	fs5, (5 * 16)(a0)
-	flq	fs6, (6 * 16)(a0)
-	flq	fs7, (7 * 16)(a0)
-	flq	fs8, (8 * 16)(a0)
-	flq	fs9, (9 * 16)(a0)
-	flq	fs10, (10 * 16)(a0)
-	flq	fs11, (11 * 16)(a0)
+#ifndef SOFTFLOAT
+	/* Restore the fpe registers */
+	fld	fs0, (0 * 16)(a0)
+	fld	fs1, (1 * 16)(a0)
+	fld	fs2, (2 * 16)(a0)
+	fld	fs3, (3 * 16)(a0)
+	fld	fs4, (4 * 16)(a0)
+	fld	fs5, (5 * 16)(a0)
+	fld	fs6, (6 * 16)(a0)
+	fld	fs7, (7 * 16)(a0)
+	fld	fs8, (8 * 16)(a0)
+	fld	fs9, (9 * 16)(a0)
+	fld	fs10, (10 * 16)(a0)
+	fld	fs11, (11 * 16)(a0)
 	addi	a0, a0, (12 * 16)
 #endif
 
diff --git a/lib/libc/riscv/softfloat/milieu.h b/lib/libc/riscv/softfloat/milieu.h
new file mode 100644
index 000000000000..e97437bdc2d1
--- /dev/null
+++ b/lib/libc/riscv/softfloat/milieu.h
@@ -0,0 +1,48 @@
+/* $FreeBSD$ */
+
+/*
+===============================================================================
+
+This C header file is part of the SoftFloat IEC/IEEE Floating-point
+Arithmetic Package, Release 2a.
+
+Written by John R. Hauser.  This work was made possible in part by the
+International Computer Science Institute, located at Suite 600, 1947 Center
+Street, Berkeley, California 94704.  Funding was partially provided by the
+National Science Foundation under grant MIP-9311980.  The original version
+of this code was written as part of a project to build a fixed-point vector
+processor in collaboration with the University of California at Berkeley,
+overseen by Profs. Nelson Morgan and John Wawrzynek.  More information
+is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
+arithmetic/SoftFloat.html'.
+
+THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE.  Although reasonable effort
+has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
+TIMES RESULT IN INCORRECT BEHAVIOR.  USE OF THIS SOFTWARE IS RESTRICTED TO
+PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
+AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
+
+Derivative works are acceptable, even for commercial purposes, so long as
+(1) they include prominent notice that the work is derivative, and (2) they
+include prominent notice akin to these four paragraphs for those parts of
+this code that are retained.
+
+===============================================================================
+*/
+
+/*
+-------------------------------------------------------------------------------
+Include common integer types and flags.
+-------------------------------------------------------------------------------
+*/
+#include "riscv-gcc.h"
+
+/*
+-------------------------------------------------------------------------------
+Symbolic Boolean literals.
+-------------------------------------------------------------------------------
+*/
+enum {
+    FALSE = 0,
+    TRUE  = 1
+};
diff --git a/lib/libc/riscv/softfloat/riscv-gcc.h b/lib/libc/riscv/softfloat/riscv-gcc.h
new file mode 100644
index 000000000000..3fe1767e63cc
--- /dev/null
+++ b/lib/libc/riscv/softfloat/riscv-gcc.h
@@ -0,0 +1,86 @@
+/* $NetBSD: arm-gcc.h,v 1.2 2001/02/21 18:09:25 bjh21 Exp $ */
+/* $FreeBSD$ */
+
+/*
+-------------------------------------------------------------------------------
+One of the macros `BIGENDIAN' or `LITTLEENDIAN' must be defined.
+-------------------------------------------------------------------------------
+*/
+#define LITTLEENDIAN
+
+/*
+-------------------------------------------------------------------------------
+The macro `BITS64' can be defined to indicate that 64-bit integer types are
+supported by the compiler.
+-------------------------------------------------------------------------------
+*/
+#define BITS64
+
+/*
+-------------------------------------------------------------------------------
+Each of the following `typedef's defines the most convenient type that holds
+integers of at least as many bits as specified.  For example, `uint8' should
+be the most convenient type that can hold unsigned integers of as many as
+8 bits.  The `flag' type must be able to hold either a 0 or 1.  For most
+implementations of C, `flag', `uint8', and `int8' should all be `typedef'ed
+to the same as `int'.
+-------------------------------------------------------------------------------
+*/
+typedef int flag;
+typedef int uint8;
+typedef int int8;
+typedef int uint16;
+typedef int int16;
+typedef unsigned int uint32;
+typedef signed int int32;
+#ifdef BITS64
+typedef unsigned long long int uint64;
+typedef signed long long int int64;
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Each of the following `typedef's defines a type that holds integers
+of _exactly_ the number of bits specified.  For instance, for most
+implementation of C, `bits16' and `sbits16' should be `typedef'ed to
+`unsigned short int' and `signed short int' (or `short int'), respectively.
+-------------------------------------------------------------------------------
+*/
+typedef unsigned char bits8;
+typedef signed char sbits8;
+typedef unsigned short int bits16;
+typedef signed short int sbits16;
+typedef unsigned int bits32;
+typedef signed int sbits32;
+#ifdef BITS64
+typedef unsigned long long int bits64;
+typedef signed long long int sbits64;
+#endif
+
+#ifdef BITS64
+/*
+-------------------------------------------------------------------------------
+The `LIT64' macro takes as its argument a textual integer literal and
+if necessary ``marks'' the literal as having a 64-bit integer type.
+For example, the GNU C Compiler (`gcc') requires that 64-bit literals be
+appended with the letters `LL' standing for `long long', which is `gcc's
+name for the 64-bit integer type.  Some compilers may allow `LIT64' to be
+defined as the identity macro:  `#define LIT64( a ) a'.
+-------------------------------------------------------------------------------
+*/
+#define LIT64( a ) a##LL
+#endif
+
+/*
+-------------------------------------------------------------------------------
+The macro `INLINE' can be used before functions that should be inlined.  If
+a compiler does not support explicit inlining, this macro should be defined
+to be `static'.
+-------------------------------------------------------------------------------
+*/
+#define INLINE static __inline
+
+#if defined(SOFTFLOAT_FOR_GCC)
+#define FLOAT64_DEMANGLE(a)	(a)
+#define FLOAT64_MANGLE(a)	(a)
+#endif
diff --git a/lib/libc/riscv/softfloat/softfloat.h b/lib/libc/riscv/softfloat/softfloat.h
new file mode 100644
index 000000000000..6aef49975a38
--- /dev/null
+++ b/lib/libc/riscv/softfloat/softfloat.h
@@ -0,0 +1,315 @@
+/*	$NetBSD: softfloat.h,v 1.6 2002/05/12 13:12:46 bjh21 Exp $	*/
+/* $FreeBSD$ */
+
+/* This is a derivative work. */
+
+/*
+===============================================================================
+
+This C header file is part of the SoftFloat IEC/IEEE Floating-point
+Arithmetic Package, Release 2a.
+
+Written by John R. Hauser.  This work was made possible in part by the
+International Computer Science Institute, located at Suite 600, 1947 Center
+Street, Berkeley, California 94704.  Funding was partially provided by the
+National Science Foundation under grant MIP-9311980.  The original version
+of this code was written as part of a project to build a fixed-point vector
+processor in collaboration with the University of California at Berkeley,
+overseen by Profs. Nelson Morgan and John Wawrzynek.  More information
+is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
+arithmetic/SoftFloat.html'.
+
+THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE.  Although reasonable effort
+has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
+TIMES RESULT IN INCORRECT BEHAVIOR.  USE OF THIS SOFTWARE IS RESTRICTED TO
+PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
+AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
+
+Derivative works are acceptable, even for commercial purposes, so long as
+(1) they include prominent notice that the work is derivative, and (2) they
+include prominent notice akin to these four paragraphs for those parts of
+this code that are retained.
+
+===============================================================================
+*/
+
+/*
+-------------------------------------------------------------------------------
+The macro `FLOATX80' must be defined to enable the extended double-precision
+floating-point format `floatx80'.  If this macro is not defined, the
+`floatx80' type will not be defined, and none of the functions that either
+input or output the `floatx80' type will be defined.  The same applies to
+the `FLOAT128' macro and the quadruple-precision format `float128'.
+-------------------------------------------------------------------------------
+*/
+/* #define FLOATX80 */
+/* #define FLOAT128 */
+
+#include <fenv.h>
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE floating-point types.
+-------------------------------------------------------------------------------
+*/
+typedef unsigned int float32;
+typedef unsigned long long float64;
+#ifdef FLOATX80
+typedef struct {
+    unsigned short high;
+    unsigned long long low;
+} floatx80;
+#endif
+#ifdef FLOAT128
+typedef struct {
+    unsigned long long high, low;
+} float128;
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE floating-point underflow tininess-detection mode.
+-------------------------------------------------------------------------------
+*/
+#ifndef SOFTFLOAT_FOR_GCC
+extern int float_detect_tininess;
+#endif
+enum {
+    float_tininess_after_rounding  = 0,
+    float_tininess_before_rounding = 1
+};
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE floating-point rounding mode.
+-------------------------------------------------------------------------------
+*/
+extern int float_rounding_mode;
+enum {
+    float_round_nearest_even = FE_TONEAREST,
+    float_round_to_zero      = FE_TOWARDZERO,
+    float_round_down         = FE_DOWNWARD,
+    float_round_up           = FE_UPWARD
+};
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE floating-point exception flags.
+-------------------------------------------------------------------------------
+*/
+extern int float_exception_flags;
+extern int float_exception_mask;
+enum {
+    float_flag_inexact   = FE_INEXACT,
+    float_flag_underflow = FE_UNDERFLOW,
+    float_flag_overflow  = FE_OVERFLOW,
+    float_flag_divbyzero = FE_DIVBYZERO,
+    float_flag_invalid   = FE_INVALID
+};
+
+/*
+-------------------------------------------------------------------------------
+Routine to raise any or all of the software IEC/IEEE floating-point
+exception flags.
+-------------------------------------------------------------------------------
+*/
+void float_raise( int );
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE integer-to-floating-point conversion routines.
+-------------------------------------------------------------------------------
+*/
+float32 int32_to_float32( int );
+float64 int32_to_float64( int );
+#ifdef FLOATX80
+floatx80 int32_to_floatx80( int );
+#endif
+#ifdef FLOAT128
+float128 int32_to_float128( int );
+#endif
+#ifndef SOFTFLOAT_FOR_GCC /* __floatdi?f is in libgcc2.c */
+float32 int64_to_float32( long long );
+float64 int64_to_float64( long long );
+#ifdef FLOATX80
+floatx80 int64_to_floatx80( long long );
+#endif
+#ifdef FLOAT128
+float128 int64_to_float128( long long );
+#endif
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE single-precision conversion routines.
+-------------------------------------------------------------------------------
+*/
+int float32_to_int32( float32 );
+int float32_to_int32_round_to_zero( float32 );
+#if defined(SOFTFLOAT_FOR_GCC) && defined(SOFTFLOAT_NEED_FIXUNS)
+unsigned int float32_to_uint32_round_to_zero( float32 );
+#endif
+#ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */
+long long float32_to_int64( float32 );
+long long float32_to_int64_round_to_zero( float32 );
+#endif
+float64 float32_to_float64( float32 );
+#ifdef FLOATX80
+floatx80 float32_to_floatx80( float32 );
+#endif
+#ifdef FLOAT128
+float128 float32_to_float128( float32 );
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE single-precision operations.
+-------------------------------------------------------------------------------
+*/
+float32 float32_round_to_int( float32 );
+float32 float32_add( float32, float32 );
+float32 float32_sub( float32, float32 );
+float32 float32_mul( float32, float32 );
+float32 float32_div( float32, float32 );
+float32 float32_rem( float32, float32 );
+float32 float32_sqrt( float32 );
+int float32_eq( float32, float32 );
+int float32_le( float32, float32 );
+int float32_lt( float32, float32 );
+int float32_eq_signaling( float32, float32 );
+int float32_le_quiet( float32, float32 );
+int float32_lt_quiet( float32, float32 );
+#ifndef SOFTFLOAT_FOR_GCC
+int float32_is_signaling_nan( float32 );
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE double-precision conversion routines.
+-------------------------------------------------------------------------------
+*/
+int float64_to_int32( float64 );
+int float64_to_int32_round_to_zero( float64 );
+#if defined(SOFTFLOAT_FOR_GCC) && defined(SOFTFLOAT_NEED_FIXUNS)
+unsigned int float64_to_uint32_round_to_zero( float64 );
+#endif
+#ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */
+long long float64_to_int64( float64 );
+long long float64_to_int64_round_to_zero( float64 );
+#endif
+float32 float64_to_float32( float64 );
+#ifdef FLOATX80
+floatx80 float64_to_floatx80( float64 );
+#endif
+#ifdef FLOAT128
+float128 float64_to_float128( float64 );
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE double-precision operations.
+-------------------------------------------------------------------------------
+*/
+float64 float64_round_to_int( float64 );
+float64 float64_add( float64, float64 );
+float64 float64_sub( float64, float64 );
+float64 float64_mul( float64, float64 );
+float64 float64_div( float64, float64 );
+float64 float64_rem( float64, float64 );
+float64 float64_sqrt( float64 );
+int float64_eq( float64, float64 );
+int float64_le( float64, float64 );
+int float64_lt( float64, float64 );
+int float64_eq_signaling( float64, float64 );
+int float64_le_quiet( float64, float64 );
+int float64_lt_quiet( float64, float64 );
+#ifndef SOFTFLOAT_FOR_GCC
+int float64_is_signaling_nan( float64 );
+#endif
+
+#ifdef FLOATX80
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE extended double-precision conversion routines.
+-------------------------------------------------------------------------------
+*/
+int floatx80_to_int32( floatx80 );
+int floatx80_to_int32_round_to_zero( floatx80 );
+long long floatx80_to_int64( floatx80 );
+long long floatx80_to_int64_round_to_zero( floatx80 );
+float32 floatx80_to_float32( floatx80 );
+float64 floatx80_to_float64( floatx80 );
+#ifdef FLOAT128
+float128 floatx80_to_float128( floatx80 );
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE extended double-precision rounding precision.  Valid
+values are 32, 64, and 80.
+-------------------------------------------------------------------------------
+*/
+extern int floatx80_rounding_precision;
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE extended double-precision operations.
+-------------------------------------------------------------------------------
+*/
+floatx80 floatx80_round_to_int( floatx80 );
+floatx80 floatx80_add( floatx80, floatx80 );
+floatx80 floatx80_sub( floatx80, floatx80 );
+floatx80 floatx80_mul( floatx80, floatx80 );
+floatx80 floatx80_div( floatx80, floatx80 );
+floatx80 floatx80_rem( floatx80, floatx80 );
+floatx80 floatx80_sqrt( floatx80 );
+int floatx80_eq( floatx80, floatx80 );
+int floatx80_le( floatx80, floatx80 );
+int floatx80_lt( floatx80, floatx80 );
+int floatx80_eq_signaling( floatx80, floatx80 );
+int floatx80_le_quiet( floatx80, floatx80 );
+int floatx80_lt_quiet( floatx80, floatx80 );
+int floatx80_is_signaling_nan( floatx80 );
+
+#endif
+
+#ifdef FLOAT128
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE quadruple-precision conversion routines.
+-------------------------------------------------------------------------------
+*/
+int float128_to_int32( float128 );
+int float128_to_int32_round_to_zero( float128 );
+long long float128_to_int64( float128 );
+long long float128_to_int64_round_to_zero( float128 );
+float32 float128_to_float32( float128 );
+float64 float128_to_float64( float128 );
+#ifdef FLOATX80
+floatx80 float128_to_floatx80( float128 );
+#endif
+
+/*
+-------------------------------------------------------------------------------
+Software IEC/IEEE quadruple-precision operations.
+-------------------------------------------------------------------------------
+*/
+float128 float128_round_to_int( float128 );
+float128 float128_add( float128, float128 );
+float128 float128_sub( float128, float128 );
+float128 float128_mul( float128, float128 );
+float128 float128_div( float128, float128 );
+float128 float128_rem( float128, float128 );
+float128 float128_sqrt( float128 );
+int float128_eq( float128, float128 );
+int float128_le( float128, float128 );
+int float128_lt( float128, float128 );
+int float128_eq_signaling( float128, float128 );
+int float128_le_quiet( float128, float128 );
+int float128_lt_quiet( float128, float128 );
+int float128_is_signaling_nan( float128 );
+
+#endif
+
diff --git a/lib/libc/softfloat/Makefile.inc b/lib/libc/softfloat/Makefile.inc
index f15e4d2a5a41..fa0ee9d2977f 100644
--- a/lib/libc/softfloat/Makefile.inc
+++ b/lib/libc/softfloat/Makefile.inc
@@ -12,8 +12,11 @@ CFLAGS+=	-DSOFTFLOAT_FOR_GCC
 
 SRCS+=		softfloat.c
 
+# Deprecated FPU control interface
+.if ${LIBC_ARCH} != "riscv"
 SRCS+=		fpgetround.c fpsetround.c fpgetmask.c fpsetmask.c \
 		fpgetsticky.c
+.endif
 
 SRCS+=		eqsf2.c nesf2.c gtsf2.c gesf2.c ltsf2.c lesf2.c negsf2.c \
 		eqdf2.c nedf2.c gtdf2.c gedf2.c ltdf2.c ledf2.c negdf2.c \
diff --git a/lib/libcompiler_rt/Makefile.inc b/lib/libcompiler_rt/Makefile.inc
index 27a5879dd24a..5913faab838b 100644
--- a/lib/libcompiler_rt/Makefile.inc
+++ b/lib/libcompiler_rt/Makefile.inc
@@ -125,7 +125,7 @@ SRCF+=		umodti3
 
 #
 # 128-bit quad precision long double support,
-# only used on arm64 and riscv.
+# only used on some architectures.
 #
 .if ${MACHINE_CPUARCH} == "aarch64" || ${MACHINE_CPUARCH} == "riscv"
 SRCF+=		addtf3
@@ -146,8 +146,9 @@ SRCF+=		trunctfdf2
 SRCF+=		trunctfsf2
 .endif
 
-# These are already shipped by libc.a on arm and mips
-.if ${MACHINE_CPUARCH} != "arm" && ${MACHINE_CPUARCH} != "mips"
+# These are already shipped by libc.a on some architectures.
+.if ${MACHINE_CPUARCH} != "arm" && ${MACHINE_CPUARCH} != "mips" && \
+    ${MACHINE_CPUARCH} != "riscv"
 SRCF+=		adddf3
 SRCF+=		addsf3
 SRCF+=		divdf3
diff --git a/lib/msun/riscv/Makefile.inc b/lib/msun/riscv/Makefile.inc
index 97d3302b96da..cd624918bed1 100644
--- a/lib/msun/riscv/Makefile.inc
+++ b/lib/msun/riscv/Makefile.inc
@@ -1,3 +1,8 @@
 # $FreeBSD$
 
+.if ${MACHINE_ARCH:Mriscv*sf} != ""
+CFLAGS+=-DSOFTFLOAT
+.endif
+
 LDBL_PREC = 113
+SYM_MAPS += ${.CURDIR}/riscv/Symbol.map
diff --git a/lib/msun/riscv/Symbol.map b/lib/msun/riscv/Symbol.map
new file mode 100644
index 000000000000..081294cf5562
--- /dev/null
+++ b/lib/msun/riscv/Symbol.map
@@ -0,0 +1,21 @@
+/*
+ * $FreeBSD$
+ */
+FBSD_1.0 {
+};
+
+FBSD_1.3 {
+	feclearexcept;
+	fegetexceptflag;
+	fesetexceptflag;
+	feraiseexcept;
+	fetestexcept;
+	fegetround;
+	fesetround;
+	fegetenv;
+	feholdexcept;
+	feupdateenv;
+	feenableexcept;
+	fedisableexcept;
+	fegetexcept;
+};
diff --git a/lib/msun/riscv/fenv.c b/lib/msun/riscv/fenv.c
index a5a5c03d3a74..9bc0d7e35740 100644
--- a/lib/msun/riscv/fenv.c
+++ b/lib/msun/riscv/fenv.c
@@ -39,6 +39,14 @@
  */
 const fenv_t __fe_dfl_env = 0;
 
+#ifdef SOFTFLOAT
+#define __set_env(env, flags, mask, rnd) env = ((flags) | (rnd) << 5)
+#define __env_flags(env)                ((env) & FE_ALL_EXCEPT)
+#define __env_mask(env)                 (0) /* No exception traps. */
+#define __env_round(env)                (((env) >> 5) & _ROUND_MASK)
+#include "fenv-softfloat.h"
+#endif
+
 extern inline int feclearexcept(int __excepts);
 extern inline int fegetexceptflag(fexcept_t *__flagp, int __excepts);
 extern inline int fesetexceptflag(const fexcept_t *__flagp, int __excepts);
@@ -50,3 +58,6 @@ extern inline int fegetenv(fenv_t *__envp);
 extern inline int feholdexcept(fenv_t *__envp);
 extern inline int fesetenv(const fenv_t *__envp);
 extern inline int feupdateenv(const fenv_t *__envp);
+extern inline int feenableexcept(int __mask);
+extern inline int fedisableexcept(int __mask);
+extern inline int fegetexcept(void);
diff --git a/lib/msun/riscv/fenv.h b/lib/msun/riscv/fenv.h
index 3eae6c2ed164..e66aa1c57753 100644
--- a/lib/msun/riscv/fenv.h
+++ b/lib/msun/riscv/fenv.h
@@ -1,6 +1,6 @@
 /*-
  * Copyright (c) 2004-2005 David Schultz <das@FreeBSD.ORG>
- * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2015-2016 Ruslan Bukin <br@bsdpad.com>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -59,11 +59,11 @@ typedef	__uint64_t	fexcept_t;
 /*
  * RISC-V Rounding modes
  */
-#define	FE_TONEAREST	(0x00 << 5)
-#define	FE_TOWARDZERO	(0x01 << 5)
-#define	FE_DOWNWARD	(0x02 << 5)
-#define	FE_UPWARD	(0x03 << 5)
 #define	_ROUND_SHIFT	5
+#define	FE_TONEAREST	(0x00 << _ROUND_SHIFT)
+#define	FE_TOWARDZERO	(0x01 << _ROUND_SHIFT)
+#define	FE_DOWNWARD	(0x02 << _ROUND_SHIFT)
+#define	FE_UPWARD	(0x03 << _ROUND_SHIFT)
 #define	_ROUND_MASK	(FE_TONEAREST | FE_DOWNWARD | \
 			 FE_UPWARD | FE_TOWARDZERO)
 
@@ -73,96 +73,117 @@ __BEGIN_DECLS
 extern const fenv_t	__fe_dfl_env;
 #define	FE_DFL_ENV	(&__fe_dfl_env)
 
-/* We need to be able to map status flag positions to mask flag positions */
-#define	_FPUSW_SHIFT	0
-#define	_ENABLE_MASK	(FE_ALL_EXCEPT << _FPUSW_SHIFT)
-
-#define	__rfs(__fpsr)	__asm __volatile("csrr %0, fcsr" : "=r" (*(__fpsr)))
-#define	__wfs(__fpsr)	__asm __volatile("csrw fcsr, %0" :: "r" (__fpsr))
+#ifndef SOFTFLOAT
+#define	__rfs(__fcsr)	__asm __volatile("csrr %0, fcsr" : "=r" (__fcsr))
+#define	__wfs(__fcsr)	__asm __volatile("csrw fcsr, %0" :: "r" (__fcsr))
+#endif
 
+#ifdef SOFTFLOAT
+int feclearexcept(int __excepts);
+int fegetexceptflag(fexcept_t *__flagp, int __excepts);
+int fesetexceptflag(const fexcept_t *__flagp, int __excepts);
+int feraiseexcept(int __excepts);
+int fetestexcept(int __excepts);
+int fegetround(void);
+int fesetround(int __round);
+int fegetenv(fenv_t *__envp);
+int feholdexcept(fenv_t *__envp);
+int fesetenv(const fenv_t *__envp);
+int feupdateenv(const fenv_t *__envp);
+#else
 __fenv_static inline int
 feclearexcept(int __excepts)
 {
-	fexcept_t __fpsr;
 
-	__rfs(&__fpsr);
-	__fpsr &= ~__excepts;
-	__wfs(__fpsr);
+	__asm __volatile("csrc fflags, %0" :: "r"(__excepts));
+
 	return (0);
 }
 
 __fenv_static inline int
 fegetexceptflag(fexcept_t *__flagp, int __excepts)
 {
-	fexcept_t __fpsr;
+	fexcept_t __fcsr;
+
+	__rfs(__fcsr);
+	*__flagp = __fcsr & __excepts;
 
-	__rfs(&__fpsr);
-	*__flagp = __fpsr & __excepts;
 	return (0);
 }
 
 __fenv_static inline int
 fesetexceptflag(const fexcept_t *__flagp, int __excepts)
 {
-	fexcept_t __fpsr;
+	fexcept_t __fcsr;
+
+	__fcsr = *__flagp;
+	__asm __volatile("csrc fflags, %0" :: "r"(__excepts));
+	__asm __volatile("csrs fflags, %0" :: "r"(__fcsr & __excepts));
 
-	__rfs(&__fpsr);
-	__fpsr &= ~__excepts;
-	__fpsr |= *__flagp & __excepts;
-	__wfs(__fpsr);
 	return (0);
 }
 
 __fenv_static inline int
 feraiseexcept(int __excepts)
 {
-	fexcept_t __ex = __excepts;
 
-	fesetexceptflag(&__ex, __excepts);	/* XXX */
+	__asm __volatile("csrs fflags, %0" :: "r"(__excepts));
+
 	return (0);
 }
 
 __fenv_static inline int
 fetestexcept(int __excepts)
 {
-	fexcept_t __fpsr;
+	fexcept_t __fcsr;
 
-	__rfs(&__fpsr);
-	return (__fpsr & __excepts);
+	__rfs(__fcsr);
+
+	return (__fcsr & __excepts);
 }
 
 __fenv_static inline int
 fegetround(void)
 {
+	fexcept_t __fcsr;
 
-	return (-1);
+	__rfs(__fcsr);
+
+	return (__fcsr & _ROUND_MASK);
 }
 
 __fenv_static inline int
 fesetround(int __round)
 {
+	fexcept_t __fcsr;
 
-	return (-1);
+	if (__round & ~_ROUND_MASK)
+		return (-1);
+
+	__rfs(__fcsr);
+	__fcsr &= ~_ROUND_MASK;
+	__fcsr |= __round;
+	__wfs(__fcsr);
+
+	return (0);
 }
 
 __fenv_static inline int
 fegetenv(fenv_t *__envp)
 {
 
-	__rfs(__envp);
+	__rfs(*__envp);
+
 	return (0);
 }
 
 __fenv_static inline int
 feholdexcept(fenv_t *__envp)
 {
-	fenv_t __env;
 
-	__rfs(&__env);
-	*__envp = __env;
-	__env &= ~(FE_ALL_EXCEPT | _ENABLE_MASK);
-	__wfs(__env);
-	return (0);
+	/* No exception traps. */
+
+	return (-1);
 }
 
 __fenv_static inline int
@@ -170,56 +191,59 @@ fesetenv(const fenv_t *__envp)
 {
 
 	__wfs(*__envp);
+
 	return (0);
 }
 
 __fenv_static inline int
 feupdateenv(const fenv_t *__envp)
 {
-	fexcept_t __fpsr;
+	fexcept_t __fcsr;
 
-	__rfs(&__fpsr);
+	__rfs(__fcsr);
 	__wfs(*__envp);
-	feraiseexcept(__fpsr & FE_ALL_EXCEPT);
+	feraiseexcept(__fcsr & FE_ALL_EXCEPT);
+
 	return (0);
 }
+#endif /* !SOFTFLOAT */
 
 #if __BSD_VISIBLE
 
 /* We currently provide no external definitions of the functions below. */
 
+#ifdef SOFTFLOAT
+int feenableexcept(int __mask);
+int fedisableexcept(int __mask);
+int fegetexcept(void);
+#else
 static inline int
 feenableexcept(int __mask)
 {
-	fenv_t __old_fpsr;
-	fenv_t __new_fpsr;
 
-	__rfs(&__old_fpsr);
-	__new_fpsr = __old_fpsr | (__mask & FE_ALL_EXCEPT) << _FPUSW_SHIFT;
-	__wfs(__new_fpsr);
-	return ((__old_fpsr >> _FPUSW_SHIFT) & FE_ALL_EXCEPT);
+	/* No exception traps. */
+
+	return (-1);
 }
 
 static inline int
 fedisableexcept(int __mask)
 {
-	fenv_t __old_fpsr;
-	fenv_t __new_fpsr;
 
-	__rfs(&__old_fpsr);
-	__new_fpsr = __old_fpsr & ~((__mask & FE_ALL_EXCEPT) << _FPUSW_SHIFT);
-	__wfs(__new_fpsr);
-	return ((__old_fpsr >> _FPUSW_SHIFT) & FE_ALL_EXCEPT);
+	/* No exception traps. */
+
+	return (0);
 }
 
 static inline int
 fegetexcept(void)
 {
-	fenv_t __fpsr;
 
-	__rfs(&__fpsr);
-	return ((__fpsr & _ENABLE_MASK) >> _FPUSW_SHIFT);
+	/* No exception traps. */
+
+	return (0);
 }
+#endif /* !SOFTFLOAT */
 
 #endif /* __BSD_VISIBLE */
 
diff --git a/share/mk/bsd.cpu.mk b/share/mk/bsd.cpu.mk
index 874beed525ad..79c37664849f 100644
--- a/share/mk/bsd.cpu.mk
+++ b/share/mk/bsd.cpu.mk
@@ -155,8 +155,6 @@ _CPUCFLAGS = -march=${CPUTYPE}
 #	sb1, xlp, xlr
 _CPUCFLAGS = -march=${CPUTYPE:S/^mips//}
 . endif
-. elif ${MACHINE_CPUARCH} == "riscv"
-_CPUCFLAGS = -mno-float -march="IMAFD"
 . elif ${MACHINE_ARCH} == "sparc64"
 .  if ${CPUTYPE} == "v9"
 _CPUCFLAGS = -mcpu=v9
@@ -350,9 +348,11 @@ CFLAGS += -mcpu=8540 -Wa,-me500 -mspe=yes -mabi=spe -mfloat-gprs=double
 .endif
 
 .if ${MACHINE_CPUARCH} == "riscv"
+.if ${TARGET_ARCH:Mriscv*sf}
 CFLAGS += -mno-float
 ACFLAGS += -mno-float
 .endif
+.endif
 
 # NB: COPTFLAGS is handled in /usr/src/sys/conf/kern.pre.mk
 
diff --git a/share/mk/local.meta.sys.mk b/share/mk/local.meta.sys.mk
index 49f625ccefa9..3a4e8f1352bd 100644
--- a/share/mk/local.meta.sys.mk
+++ b/share/mk/local.meta.sys.mk
@@ -48,7 +48,7 @@ TARGET_ARCHES_arm64?=   aarch64
 TARGET_ARCHES_mips?=    mipsel mips mips64el mips64 mipsn32 mipsn32el
 TARGET_ARCHES_powerpc?= powerpc powerpc64 powerpcspe
 TARGET_ARCHES_pc98?=    i386
-TARGET_ARCHES_riscv?=   riscv64
+TARGET_ARCHES_riscv?=   riscv64 riscv64sf
 
 # some corner cases
 BOOT_MACHINE_DIR.amd64 = boot/i386
diff --git a/share/mk/src.opts.mk b/share/mk/src.opts.mk
index 89125d695130..64ffbe50ded8 100644
--- a/share/mk/src.opts.mk
+++ b/share/mk/src.opts.mk
@@ -236,16 +236,16 @@ __DEFAULT_YES_OPTIONS+=GCC GCC_BOOTSTRAP GNUCXX
 __DEFAULT_NO_OPTIONS+=CLANG CLANG_BOOTSTRAP CLANG_FULL CLANG_IS_CC
 .endif
 # In-tree binutils/gcc are older versions without modern architecture support.
-.if ${__T} == "aarch64" || ${__T} == "riscv64"
+.if ${__T} == "aarch64" || ${__T:Mriscv*} != ""
 BROKEN_OPTIONS+=BINUTILS BINUTILS_BOOTSTRAP GCC GCC_BOOTSTRAP GDB
 .endif
-.if ${__T} == "riscv64"
+.if ${__T:Mriscv*} != ""
 BROKEN_OPTIONS+=PROFILE # "sorry, unimplemented: profiler support for RISC-V"
 BROKEN_OPTIONS+=TESTS   # "undefined reference to `_Unwind_Resume'"
 BROKEN_OPTIONS+=CXX     # "libcxxrt.so: undefined reference to `_Unwind_Resume_or_Rethrow'"
 .endif
 .if ${__T} == "aarch64" || ${__T} == "amd64" || ${__T} == "i386" || \
-    ${__T} == "riscv64"
+    ${__T:Mriscv*} != ""
 __DEFAULT_YES_OPTIONS+=LLVM_LIBUNWIND
 .else
 __DEFAULT_NO_OPTIONS+=LLVM_LIBUNWIND
diff --git a/share/mk/sys.mk b/share/mk/sys.mk
index 2c3743342018..04c342c134f0 100644
--- a/share/mk/sys.mk
+++ b/share/mk/sys.mk
@@ -13,7 +13,7 @@ unix		?=	We run FreeBSD, not UNIX.
 # and/or endian.  This is called MACHINE_CPU in NetBSD, but that's used
 # for something different in FreeBSD.
 #
-MACHINE_CPUARCH=${MACHINE_ARCH:C/mips(n32|64)?(el)?(hf)?/mips/:C/arm(v6)?(eb|hf)?/arm/:C/powerpc(64|spe)/powerpc/:C/riscv64/riscv/}
+MACHINE_CPUARCH=${MACHINE_ARCH:C/mips(n32|64)?(el)?(hf)?/mips/:C/arm(v6)?(eb|hf)?/arm/:C/powerpc(64|spe)/powerpc/:C/riscv64(sf)?/riscv/}
 .endif
 
 
diff --git a/sys/conf/options.riscv b/sys/conf/options.riscv
index c263bd860875..af6a89e80f19 100644
--- a/sys/conf/options.riscv
+++ b/sys/conf/options.riscv
@@ -1,4 +1,4 @@
 # $FreeBSD$
 
 RISCV				opt_global.h
-VFP				opt_global.h
+FPE				opt_global.h
diff --git a/sys/modules/dtrace/dtrace/Makefile b/sys/modules/dtrace/dtrace/Makefile
index 543be8c20191..0889646f88c0 100644
--- a/sys/modules/dtrace/dtrace/Makefile
+++ b/sys/modules/dtrace/dtrace/Makefile
@@ -60,7 +60,11 @@ assym.o: assym.s
 
 .if ${MACHINE_CPUARCH} == "riscv"
 assym.o: assym.s
+.if ${TARGET_ARCH:Mriscv*sf}
 	${AS} -mfloat-abi=soft -o assym.o assym.s
+.else
+	${AS} -mfloat-abi=double -o assym.o assym.s
+.endif
 .endif
 
 .include <bsd.kmod.mk>
diff --git a/sys/riscv/conf/GENERIC b/sys/riscv/conf/GENERIC
index ad23c6647e73..a9c3c6fbdad0 100644
--- a/sys/riscv/conf/GENERIC
+++ b/sys/riscv/conf/GENERIC
@@ -69,7 +69,7 @@ options 	CAPABILITIES		# Capsicum capabilities
 options 	MAC			# TrustedBSD MAC Framework
 options 	KDTRACE_FRAME		# Ensure frames are compiled in
 options 	KDTRACE_HOOKS		# Kernel DTrace hooks
-# options 	VFP			# Floating-point support
+options 	FPE			# Floating-point extension support
 options 	RACCT			# Resource accounting framework
 options 	RACCT_DEFAULT_TO_DISABLED # Set kern.racct.enable=0 by default
 options 	RCTL			# Resource limits
diff --git a/sys/riscv/include/fpe.h b/sys/riscv/include/fpe.h
new file mode 100644
index 000000000000..1294ab0de8fa
--- /dev/null
+++ b/sys/riscv/include/fpe.h
@@ -0,0 +1,38 @@
+/*-
+ * Copyright (c) 2016 Ruslan Bukin <br@bsdpad.com>
+ * All rights reserved.
+ *
+ * This software was developed by SRI International and the University of
+ * Cambridge Computer Laboratory under DARPA/AFRL contract FA8750-10-C-0237
+ * ("CTSRD"), as part of the DARPA CRASH research programme.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+
+#ifndef	_MACHINE_FPE_H_
+#define	_MACHINE_FPE_H_
+
+void fpe_state_save(struct thread *td);
+
+#endif /* !_MACHINE_FPE_H_ */
diff --git a/sys/riscv/include/pcb.h b/sys/riscv/include/pcb.h
index a6cecb73434c..27737a4bdf24 100644
--- a/sys/riscv/include/pcb.h
+++ b/sys/riscv/include/pcb.h
@@ -1,5 +1,5 @@
 /*-
- * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2015-2016 Ruslan Bukin <br@bsdpad.com>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -42,16 +42,21 @@
 struct trapframe;
 
 struct pcb {
-	uint64_t	pcb_ra;
-	uint64_t	pcb_sp;
-	uint64_t	pcb_gp;
-	uint64_t	pcb_tp;
-	uint64_t	pcb_t[7];
-	uint64_t	pcb_s[12];
-	uint64_t	pcb_a[8];
-	uint64_t	pcb_sepc;
+	uint64_t	pcb_ra;		/* Return address */
+	uint64_t	pcb_sp;		/* Stack pointer */
+	uint64_t	pcb_gp;		/* Global pointer */
+	uint64_t	pcb_tp;		/* Thread pointer */
+	uint64_t	pcb_t[7];	/* Temporary registers */
+	uint64_t	pcb_s[12];	/* Saved registers */
+	uint64_t	pcb_a[8];	/* Argument registers */
+	uint64_t	pcb_x[32][2];	/* Floating point registers */
+	uint64_t	pcb_fcsr;	/* Floating point control reg */
+	uint64_t	pcb_fpflags;	/* Floating point flags */
+#define	PCB_FP_STARTED	0x1
+#define	PCB_FP_USERMASK	0x1
+	uint64_t	pcb_sepc;	/* Supervisor exception pc */
 	vm_offset_t	pcb_l1addr;	/* L1 page tables base address */
-	vm_offset_t	pcb_onfault;
+	vm_offset_t	pcb_onfault;	/* Copyinout fault handler */
 };
 
 #ifdef _KERNEL
diff --git a/sys/riscv/include/reg.h b/sys/riscv/include/reg.h
index 35d79f10a8f5..9e94b4a5768d 100644
--- a/sys/riscv/include/reg.h
+++ b/sys/riscv/include/reg.h
@@ -1,5 +1,5 @@
 /*-
- * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2015-2016 Ruslan Bukin <br@bsdpad.com>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -50,7 +50,8 @@ struct reg {
 };
 
 struct fpreg {
-	int dummy;
+	uint64_t	fp_x[32][2];	/* Floating point registers */
+	uint64_t	fp_fcsr;	/* Floating point control reg */
 };
 
 struct dbreg {
diff --git a/sys/riscv/include/riscvreg.h b/sys/riscv/include/riscvreg.h
index 2e3a4a2e9f5d..77f35bfa64b0 100644
--- a/sys/riscv/include/riscvreg.h
+++ b/sys/riscv/include/riscvreg.h
@@ -60,10 +60,14 @@
 #define	SSTATUS_SPIE_SHIFT		5
 #define	SSTATUS_SPP			(1 << 8)
 #define	SSTATUS_SPP_SHIFT		8
-#define	SSTATUS_FS_MASK			0x3
 #define	SSTATUS_FS_SHIFT		13
-#define	SSTATUS_XS_MASK			0x3
+#define	SSTATUS_FS_OFF			(0x0 << SSTATUS_FS_SHIFT)
+#define	SSTATUS_FS_INITIAL		(0x1 << SSTATUS_FS_SHIFT)
+#define	SSTATUS_FS_CLEAN		(0x2 << SSTATUS_FS_SHIFT)
+#define	SSTATUS_FS_DIRTY		(0x3 << SSTATUS_FS_SHIFT)
+#define	SSTATUS_FS_MASK			(0x3 << SSTATUS_FS_SHIFT)
 #define	SSTATUS_XS_SHIFT		15
+#define	SSTATUS_XS_MASK			(0x3 << SSTATUS_XS_SHIFT)
 #define	SSTATUS_PUM			(1 << 18)
 #define	SSTATUS32_SD			(1 << 63)
 #define	SSTATUS64_SD			(1 << 31)
diff --git a/sys/riscv/riscv/genassym.c b/sys/riscv/riscv/genassym.c
index c9e9bc1c991a..33f1d41750a7 100644
--- a/sys/riscv/riscv/genassym.c
+++ b/sys/riscv/riscv/genassym.c
@@ -72,6 +72,8 @@ ASSYM(PCB_TP, offsetof(struct pcb, pcb_tp));
 ASSYM(PCB_T, offsetof(struct pcb, pcb_t));
 ASSYM(PCB_S, offsetof(struct pcb, pcb_s));
 ASSYM(PCB_A, offsetof(struct pcb, pcb_a));
+ASSYM(PCB_X, offsetof(struct pcb, pcb_x));
+ASSYM(PCB_FCSR, offsetof(struct pcb, pcb_fcsr));
 
 ASSYM(SF_UC, offsetof(struct sigframe, sf_uc));
 
diff --git a/sys/riscv/riscv/machdep.c b/sys/riscv/riscv/machdep.c
index 7bc4402597de..f08527a7f2f3 100644
--- a/sys/riscv/riscv/machdep.c
+++ b/sys/riscv/riscv/machdep.c
@@ -84,8 +84,8 @@ __FBSDID("$FreeBSD$");
 
 #include <machine/asm.h>
 
-#ifdef VFP
-#include <machine/vfp.h>
+#ifdef FPE
+#include <machine/fpe.h>
 #endif
 
 #ifdef FDT
@@ -203,17 +203,39 @@ set_regs(struct thread *td, struct reg *regs)
 int
 fill_fpregs(struct thread *td, struct fpreg *regs)
 {
+#ifdef FPE
+	struct pcb *pcb;
+
+	pcb = td->td_pcb;
+
+	if ((pcb->pcb_fpflags & PCB_FP_STARTED) != 0) {
+		/*
+		 * If we have just been running FPE instructions we will
+		 * need to save the state to memcpy it below.
+		 */
+		fpe_state_save(td);
+
+		memcpy(regs->fp_x, pcb->pcb_x, sizeof(regs->fp_x));
+		regs->fp_fcsr = pcb->pcb_fcsr;
+	} else
+#endif
+		memset(regs->fp_x, 0, sizeof(regs->fp_x));
 
-	/* TODO */
-	bzero(regs, sizeof(*regs));
 	return (0);
 }
 
 int
 set_fpregs(struct thread *td, struct fpreg *regs)
 {
+#ifdef FPE
+	struct pcb *pcb;
+
+	pcb = td->td_pcb;
+
+	memcpy(pcb->pcb_x, regs->fp_x, sizeof(regs->fp_x));
+	pcb->pcb_fcsr = regs->fp_fcsr;
+#endif
 
-	/* TODO */
 	return (0);
 }
 
@@ -259,8 +281,10 @@ void
 exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
 {
 	struct trapframe *tf;
+	struct pcb *pcb;
 
 	tf = td->td_frame;
+	pcb = td->td_pcb;
 
 	memset(tf, 0, sizeof(struct trapframe));
 
@@ -273,6 +297,8 @@ exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
 	tf->tf_sp = STACKALIGN(stack);
 	tf->tf_ra = imgp->entry_addr;
 	tf->tf_sepc = imgp->entry_addr;
+
+	pcb->pcb_fpflags &= ~PCB_FP_STARTED;
 }
 
 /* Sanity check these are the same size, they will be memcpy'd to and fro */
@@ -337,13 +363,54 @@ set_mcontext(struct thread *td, mcontext_t *mcp)
 static void
 get_fpcontext(struct thread *td, mcontext_t *mcp)
 {
-	/* TODO */
+#ifdef FPE
+	struct pcb *curpcb;
+
+	critical_enter();
+
+	curpcb = curthread->td_pcb;
+
+	KASSERT(td->td_pcb == curpcb, ("Invalid fpe pcb"));
+
+	if ((curpcb->pcb_fpflags & PCB_FP_STARTED) != 0) {
+		/*
+		 * If we have just been running FPE instructions we will
+		 * need to save the state to memcpy it below.
+		 */
+		fpe_state_save(td);
+
+		KASSERT((curpcb->pcb_fpflags & ~PCB_FP_USERMASK) == 0,
+		    ("Non-userspace FPE flags set in get_fpcontext"));
+		memcpy(mcp->mc_fpregs.fp_x, curpcb->pcb_x,
+		    sizeof(mcp->mc_fpregs));
+		mcp->mc_fpregs.fp_fcsr = curpcb->pcb_fcsr;
+		mcp->mc_fpregs.fp_flags = curpcb->pcb_fpflags;
+		mcp->mc_flags |= _MC_FP_VALID;
+	}
+
+	critical_exit();
+#endif
 }
 
 static void
 set_fpcontext(struct thread *td, mcontext_t *mcp)
 {
-	/* TODO */
+#ifdef FPE
+	struct pcb *curpcb;
+
+	critical_enter();
+
+	if ((mcp->mc_flags & _MC_FP_VALID) != 0) {
+		curpcb = curthread->td_pcb;
+		/* FPE usage is enabled, override registers. */
+		memcpy(curpcb->pcb_x, mcp->mc_fpregs.fp_x,
+		    sizeof(mcp->mc_fpregs));
+		curpcb->pcb_fcsr = mcp->mc_fpregs.fp_fcsr;
+		curpcb->pcb_fpflags = mcp->mc_fpregs.fp_flags & PCB_FP_USERMASK;
+	}
+
+	critical_exit();
+#endif
 }
 
 void
@@ -572,6 +639,7 @@ init_proc0(vm_offset_t kstack)
 	proc_linkup0(&proc0, &thread0);
 	thread0.td_kstack = kstack;
 	thread0.td_pcb = (struct pcb *)(thread0.td_kstack) - 1;
+	thread0.td_pcb->pcb_fpflags = 0;
 	thread0.td_frame = &proc0_tf;
 	pcpup->pc_curpcb = thread0.td_pcb;
 }
diff --git a/sys/riscv/riscv/mp_machdep.c b/sys/riscv/riscv/mp_machdep.c
index 33353ac2a3c8..adad7b4b533b 100644
--- a/sys/riscv/riscv/mp_machdep.c
+++ b/sys/riscv/riscv/mp_machdep.c
@@ -62,9 +62,6 @@ __FBSDID("$FreeBSD$");
 #include <machine/intr.h>
 #include <machine/smp.h>
 #include <machine/sbi.h>
-#ifdef VFP
-#include <machine/vfp.h>
-#endif
 
 #ifdef FDT
 #include <dev/ofw/openfirm.h>
@@ -242,10 +239,6 @@ init_secondary(uint64_t cpu)
 	/* Start per-CPU event timers. */
 	cpu_initclocks_ap();
 
-#ifdef VFP
-	/* TODO: init FPU */
-#endif
-
 	/* Enable interrupts */
 	intr_enable();
 
diff --git a/sys/riscv/riscv/swtch.S b/sys/riscv/riscv/swtch.S
index b5cced30d8f7..5cc1edd69a02 100644
--- a/sys/riscv/riscv/swtch.S
+++ b/sys/riscv/riscv/swtch.S
@@ -42,6 +42,117 @@
 
 __FBSDID("$FreeBSD$");
 
+#ifdef FPE
+.macro __fpe_state_save p
+	/*
+	 * Enable FPE usage in supervisor mode,
+	 * so we can access registers.
+	 */
+	li	t0, SSTATUS_FS_INITIAL
+	csrs	sstatus, t0
+
+	/* Store registers */
+	frcsr	t0
+	sd	t0, (PCB_FCSR)(\p)
+	fsd	f0, (PCB_X + 0 * 16)(\p)
+	fsd	f1, (PCB_X + 1 * 16)(\p)
+	fsd	f2, (PCB_X + 2 * 16)(\p)
+	fsd	f3, (PCB_X + 3 * 16)(\p)
+	fsd	f4, (PCB_X + 4 * 16)(\p)
+	fsd	f5, (PCB_X + 5 * 16)(\p)
+	fsd	f6, (PCB_X + 6 * 16)(\p)
+	fsd	f7, (PCB_X + 7 * 16)(\p)
+	fsd	f8, (PCB_X + 8 * 16)(\p)
+	fsd	f9, (PCB_X + 9 * 16)(\p)
+	fsd	f10, (PCB_X + 10 * 16)(\p)
+	fsd	f11, (PCB_X + 11 * 16)(\p)
+	fsd	f12, (PCB_X + 12 * 16)(\p)
+	fsd	f13, (PCB_X + 13 * 16)(\p)
+	fsd	f14, (PCB_X + 14 * 16)(\p)
+	fsd	f15, (PCB_X + 15 * 16)(\p)
+	fsd	f16, (PCB_X + 16 * 16)(\p)
+	fsd	f17, (PCB_X + 17 * 16)(\p)
+	fsd	f18, (PCB_X + 18 * 16)(\p)
+	fsd	f19, (PCB_X + 19 * 16)(\p)
+	fsd	f20, (PCB_X + 20 * 16)(\p)
+	fsd	f21, (PCB_X + 21 * 16)(\p)
+	fsd	f22, (PCB_X + 22 * 16)(\p)
+	fsd	f23, (PCB_X + 23 * 16)(\p)
+	fsd	f24, (PCB_X + 24 * 16)(\p)
+	fsd	f25, (PCB_X + 25 * 16)(\p)
+	fsd	f26, (PCB_X + 26 * 16)(\p)
+	fsd	f27, (PCB_X + 27 * 16)(\p)
+	fsd	f28, (PCB_X + 28 * 16)(\p)
+	fsd	f29, (PCB_X + 29 * 16)(\p)
+	fsd	f30, (PCB_X + 30 * 16)(\p)
+	fsd	f31, (PCB_X + 31 * 16)(\p)
+
+	/* Disable FPE usage in supervisor mode. */
+	li	t0, SSTATUS_FS_MASK
+	csrc	sstatus, t0
+.endm
+
+.macro __fpe_state_load p
+	/*
+	 * Enable FPE usage in supervisor mode,
+	 * so we can access registers.
+	 */
+	li	t0, SSTATUS_FS_INITIAL
+	csrs	sstatus, t0
+
+	/* Restore registers */
+	ld	t0, (PCB_FCSR)(\p)
+	fscsr	t0
+	fld	f0, (PCB_X + 0 * 16)(\p)
+	fld	f1, (PCB_X + 1 * 16)(\p)
+	fld	f2, (PCB_X + 2 * 16)(\p)
+	fld	f3, (PCB_X + 3 * 16)(\p)
+	fld	f4, (PCB_X + 4 * 16)(\p)
+	fld	f5, (PCB_X + 5 * 16)(\p)
+	fld	f6, (PCB_X + 6 * 16)(\p)
+	fld	f7, (PCB_X + 7 * 16)(\p)
+	fld	f8, (PCB_X + 8 * 16)(\p)
+	fld	f9, (PCB_X + 9 * 16)(\p)
+	fld	f10, (PCB_X + 10 * 16)(\p)
+	fld	f11, (PCB_X + 11 * 16)(\p)
+	fld	f12, (PCB_X + 12 * 16)(\p)
+	fld	f13, (PCB_X + 13 * 16)(\p)
+	fld	f14, (PCB_X + 14 * 16)(\p)
+	fld	f15, (PCB_X + 15 * 16)(\p)
+	fld	f16, (PCB_X + 16 * 16)(\p)
+	fld	f17, (PCB_X + 17 * 16)(\p)
+	fld	f18, (PCB_X + 18 * 16)(\p)
+	fld	f19, (PCB_X + 19 * 16)(\p)
+	fld	f20, (PCB_X + 20 * 16)(\p)
+	fld	f21, (PCB_X + 21 * 16)(\p)
+	fld	f22, (PCB_X + 22 * 16)(\p)
+	fld	f23, (PCB_X + 23 * 16)(\p)
+	fld	f24, (PCB_X + 24 * 16)(\p)
+	fld	f25, (PCB_X + 25 * 16)(\p)
+	fld	f26, (PCB_X + 26 * 16)(\p)
+	fld	f27, (PCB_X + 27 * 16)(\p)
+	fld	f28, (PCB_X + 28 * 16)(\p)
+	fld	f29, (PCB_X + 29 * 16)(\p)
+	fld	f30, (PCB_X + 30 * 16)(\p)
+	fld	f31, (PCB_X + 31 * 16)(\p)
+
+	/* Disable FPE usage in supervisor mode. */
+	li	t0, SSTATUS_FS_MASK
+	csrc	sstatus, t0
+.endm
+
+/*
+ * void
+ * fpe_state_save(struct thread *td)
+ */
+ENTRY(fpe_state_save)
+	/* Get pointer to PCB */
+	ld	a0, TD_PCB(a0)
+	__fpe_state_save a0
+	ret
+END(fpe_state_save)
+#endif /* FPE */
+
 /*
  * void cpu_throw(struct thread *old, struct thread *new)
  */
@@ -81,8 +192,20 @@ ENTRY(cpu_throw)
 	ld	s10, (PCB_S + 10 * 8)(x13)
 	ld	s11, (PCB_S + 11 * 8)(x13)
 
-	ret
+#ifdef FPE
+	/* Is FPE enabled for new thread? */
+	ld	t0, TD_FRAME(a1)
+	ld	t1, (TF_SSTATUS)(t0)
+	li	t2, SSTATUS_FS_MASK
+	and	t3, t1, t2
+	beqz	t3, 1f		/* No, skip. */
 
+	/* Restore registers. */
+	__fpe_state_load x13
+1:
+#endif
+
+	ret
 .Lcpu_throw_panic_str:
 	.asciz "cpu_throw: %p\0"
 END(cpu_throw)
@@ -123,6 +246,29 @@ ENTRY(cpu_switch)
 	sd	s10, (PCB_S + 10 * 8)(x13)
 	sd	s11, (PCB_S + 11 * 8)(x13)
 
+#ifdef FPE
+	/*
+	 * Is FPE enabled and is it in dirty state
+	 * for the old thread?
+	 */
+	ld	t0, TD_FRAME(a0)
+	ld	t1, (TF_SSTATUS)(t0)
+	li	t2, SSTATUS_FS_MASK
+	and	t3, t1, t2
+	li	t2, SSTATUS_FS_DIRTY
+	bne	t3, t2, 1f		/* No, skip. */
+
+	/* Yes, mark FPE state clean and save registers. */
+	li	t2, ~SSTATUS_FS_MASK
+	and	t3, t1, t2
+	li	t2, SSTATUS_FS_CLEAN
+	or	t3, t3, t2
+	sd	t3, (TF_SSTATUS)(t0)
+
+	__fpe_state_save x13
+1:
+#endif
+
 	/*
 	 * Restore the saved context.
 	 */
@@ -171,6 +317,20 @@ ENTRY(cpu_switch)
 	ld	s9, (PCB_S + 9 * 8)(x13)
 	ld	s10, (PCB_S + 10 * 8)(x13)
 	ld	s11, (PCB_S + 11 * 8)(x13)
+
+#ifdef FPE
+	/* Is FPE enabled for new thread? */
+	ld	t0, TD_FRAME(a1)
+	ld	t1, (TF_SSTATUS)(t0)
+	li	t2, SSTATUS_FS_MASK
+	and	t3, t1, t2
+	beqz	t3, 1f		/* No, skip. */
+
+	/* Restore registers. */
+	__fpe_state_load x13
+1:
+#endif
+
 	ret
 .Lcpu_switch_panic_str:
 	.asciz "cpu_switch: %p\0"
@@ -269,9 +429,8 @@ ENTRY(savectx)
 	sd	s10, (PCB_S + 10 * 8)(a0)
 	sd	s11, (PCB_S + 11 * 8)(a0)
 
-	/* Store the VFP registers */
-#ifdef VFP
-	/* TODO */
+#ifdef FPE
+	__fpe_state_save a0
 #endif
 	ret
 END(savectx)
diff --git a/sys/riscv/riscv/trap.c b/sys/riscv/riscv/trap.c
index 97d9a2c1efdd..cd2b7c4bfb74 100644
--- a/sys/riscv/riscv/trap.c
+++ b/sys/riscv/riscv/trap.c
@@ -328,9 +328,11 @@ do_trap_user(struct trapframe *frame)
 	uint64_t exception;
 	struct thread *td;
 	uint64_t sstatus;
+	struct pcb *pcb;
 
 	td = curthread;
 	td->td_frame = frame;
+	pcb = td->td_pcb;
 
 	/* Ensure we came from usermode, interrupts disabled */
 	__asm __volatile("csrr %0, sstatus" : "=&r" (sstatus));
@@ -358,6 +360,17 @@ do_trap_user(struct trapframe *frame)
 		svc_handler(frame);
 		break;
 	case EXCP_ILLEGAL_INSTRUCTION:
+#ifdef FPE
+		if ((pcb->pcb_fpflags & PCB_FP_STARTED) == 0) {
+			/*
+			 * May be a FPE trap. Enable FPE usage
+			 * for this thread and try again.
+			 */
+			frame->tf_sstatus |= SSTATUS_FS_INITIAL;
+			pcb->pcb_fpflags |= PCB_FP_STARTED;
+			break;
+		}
+#endif
 		call_trapsignal(td, SIGILL, ILL_ILLTRP, (void *)frame->tf_sepc);
 		userret(td, frame);
 		break;
diff --git a/sys/riscv/riscv/vm_machdep.c b/sys/riscv/riscv/vm_machdep.c
index d2d4c6e3d3ec..c5c62fe24db2 100644
--- a/sys/riscv/riscv/vm_machdep.c
+++ b/sys/riscv/riscv/vm_machdep.c
@@ -87,8 +87,8 @@ cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
 	/* Arguments for child */
 	tf->tf_a[0] = 0;
 	tf->tf_a[1] = 0;
-	tf->tf_sstatus = (SSTATUS_SPIE);
-	tf->tf_sstatus |= (MSTATUS_PRV_U << MSTATUS_SPP_SHIFT);
+	tf->tf_sstatus |= (SSTATUS_SPIE); /* Enable interrupts. */
+	tf->tf_sstatus &= ~(SSTATUS_SPP); /* User mode. */
 
 	td2->td_frame = tf;