From 8ac73103bf12ce4f776940cb17f3ced15a362f23 Mon Sep 17 00:00:00 2001
From: Stef O'Rear <sorear2@gmail.com>
Date: Sun, 11 Mar 2018 05:55:15 -0700
Subject: [PATCH] New RISC-V port (#281)

* Add RISC-V support

This patch adds support for the RISC-V architecture (https://riscv.org).

This patch has been tested using QEMU user-mode emulation and GCC 7.2.0
in the following configurations:

* -march=rv32imac -mabi=ilp32
* -march=rv32g -mabi=ilp32d
* -march=rv64imac -mabi=lp64
* -march=rv64g -mabi=lp64d

The ABI currently can be found at
https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md .

* Add RISC-V to README

* RISC-V: fix configure.host

Upstream-Status: Backport [https://github.com/libffi/libffi/commit/3840d49aaa831d649b1597518a2903dfed0d57f3]
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
---
 Makefile.am           |   4 +
 configure.ac          |   5 +
 src/riscv/ffi.c       | 445 ++++++++++++++++++++++++++++++++++++++++++
 src/riscv/ffitarget.h |  68 +++++++
 src/riscv/sysv.S      | 214 ++++++++++++++++++++
 5 files changed, 736 insertions(+)
 create mode 100644 src/riscv/ffi.c
 create mode 100644 src/riscv/ffitarget.h
 create mode 100644 src/riscv/sysv.S

diff --git a/Makefile.am b/Makefile.am
index 0e40451..3837650 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -32,6 +32,7 @@ EXTRA_DIST = LICENSE ChangeLog.v1 ChangeLog.libgcj			\
 	 src/powerpc/asm.h src/powerpc/aix.S src/powerpc/darwin.S	\
 	 src/powerpc/aix_closure.S src/powerpc/darwin_closure.S		\
 	 src/powerpc/ffi_darwin.c src/powerpc/ffitarget.h		\
+	 src/riscv/ffi.c src/riscv/ffitarget.h src/riscv/sysv.S		\
 	 src/s390/ffi.c src/s390/sysv.S src/s390/ffitarget.h		\
 	 src/sh/ffi.c src/sh/sysv.S src/sh/ffitarget.h src/sh64/ffi.c	\
 	 src/sh64/sysv.S src/sh64/ffitarget.h src/sparc/v8.S		\
@@ -122,6 +123,9 @@ endif
 if MIPS
 nodist_libffi_la_SOURCES += src/mips/ffi.c src/mips/o32.S src/mips/n32.S
 endif
+if RISCV
+nodist_libffi_la_SOURCES += src/riscv/ffi.c src/riscv/sysv.S
+endif
 if BFIN
 nodist_libffi_la_SOURCES += src/bfin/ffi.c src/bfin/sysv.S
 endif
diff --git a/configure.ac b/configure.ac
index ce30853..33375aa 100644
--- a/configure.ac
+++ b/configure.ac
@@ -226,6 +226,10 @@ case "$host" in
 	TARGET=MIPS; TARGETDIR=mips
 	;;
 
+  riscv*-*-*)
+	TARGET=RISCV; TARGETDIR=riscv
+	;;
+
   nios2*-linux*)
 	TARGET=NIOS2; TARGETDIR=nios2
 	;;
@@ -298,6 +302,7 @@ if test $TARGETDIR = unknown; then
 fi
 
 AM_CONDITIONAL(MIPS, test x$TARGET = xMIPS)
+AM_CONDITIONAL(RISCV, test x$TARGET = xRISCV)
 AM_CONDITIONAL(BFIN, test x$TARGET = xBFIN)
 AM_CONDITIONAL(SPARC, test x$TARGET = xSPARC)
 AM_CONDITIONAL(X86, test x$TARGET = xX86)
diff --git a/src/riscv/ffi.c b/src/riscv/ffi.c
new file mode 100644
index 0000000..b744fdd
--- /dev/null
+++ b/src/riscv/ffi.c
@@ -0,0 +1,445 @@
+/* -----------------------------------------------------------------------
+   ffi.c - Copyright (c) 2015 Michael Knyszek <mknyszek@berkeley.edu>
+                         2015 Andrew Waterman <waterman@cs.berkeley.edu>
+                         2018 Stef O'Rear <sorear2@gmail.com>
+   Based on MIPS N32/64 port
+
+   RISC-V Foreign Function Interface
+
+   Permission is hereby granted, free of charge, to any person obtaining
+   a copy of this software and associated documentation files (the
+   ``Software''), to deal in the Software without restriction, including
+   without limitation the rights to use, copy, modify, merge, publish,
+   distribute, sublicense, and/or sell copies of the Software, and to
+   permit persons to whom the Software is furnished to do so, subject to
+   the following conditions:
+
+   The above copyright notice and this permission notice shall be included
+   in all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND,
+   EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+   MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+   NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+   HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+   WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+   OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+   ----------------------------------------------------------------------- */
+
+#include <ffi.h>
+#include <ffi_common.h>
+
+#include <stdlib.h>
+#include <stdint.h>
+
+#if __riscv_float_abi_double
+#define ABI_FLEN 64
+#define ABI_FLOAT double
+#elif __riscv_float_abi_single
+#define ABI_FLEN 32
+#define ABI_FLOAT float
+#endif
+
+#define NARGREG 8
+#define STKALIGN 16
+#define MAXCOPYARG (2 * sizeof(double))
+
+typedef struct call_context
+{
+#if ABI_FLEN
+    ABI_FLOAT fa[8];
+#endif
+    size_t a[8];
+    /* used by the assembly code to in-place construct its own stack frame */
+    char frame[16];
+} call_context;
+
+typedef struct call_builder
+{
+    call_context *aregs;
+    int used_integer;
+    int used_float;
+    size_t *used_stack;
+} call_builder;
+
+/* integer (not pointer) less than ABI XLEN */
+/* FFI_TYPE_INT does not appear to be used */
+#if __SIZEOF_POINTER__ == 8
+#define IS_INT(type) ((type) >= FFI_TYPE_UINT8 && (type) <= FFI_TYPE_SINT64)
+#else
+#define IS_INT(type) ((type) >= FFI_TYPE_UINT8 && (type) <= FFI_TYPE_SINT32)
+#endif
+
+#if ABI_FLEN
+typedef struct {
+    char as_elements, type1, offset2, type2;
+} float_struct_info;
+
+#if ABI_FLEN >= 64
+#define IS_FLOAT(type) ((type) >= FFI_TYPE_FLOAT && (type) <= FFI_TYPE_DOUBLE)
+#else
+#define IS_FLOAT(type) ((type) == FFI_TYPE_FLOAT)
+#endif
+
+static ffi_type **flatten_struct(ffi_type *in, ffi_type **out, ffi_type **out_end) {
+    int i;
+    if (out == out_end) return out;
+    if (in->type != FFI_TYPE_STRUCT) {
+        *(out++) = in;
+    } else {
+        for (i = 0; in->elements[i]; i++)
+            out = flatten_struct(in->elements[i], out, out_end);
+    }
+    return out;
+}
+
+/* Structs with at most two fields after flattening, one of which is of
+   floating point type, are passed in multiple registers if sufficient
+   registers are available. */
+static float_struct_info struct_passed_as_elements(call_builder *cb, ffi_type *top) {
+    float_struct_info ret = {0, 0, 0, 0};
+    ffi_type *fields[3];
+    int num_floats, num_ints;
+    int num_fields = flatten_struct(top, fields, fields + 3) - fields;
+
+    if (num_fields == 1) {
+        if (IS_FLOAT(fields[0]->type)) {
+            ret.as_elements = 1;
+            ret.type1 = fields[0]->type;
+        }
+    } else if (num_fields == 2) {
+        num_floats = IS_FLOAT(fields[0]->type) + IS_FLOAT(fields[1]->type);
+        num_ints = IS_INT(fields[0]->type) + IS_INT(fields[1]->type);
+        if (num_floats == 0 || num_floats + num_ints != 2)
+            return ret;
+        if (cb->used_float + num_floats > NARGREG || cb->used_integer + (2 - num_floats) > NARGREG)
+            return ret;
+        if (!IS_FLOAT(fields[0]->type) && !IS_FLOAT(fields[1]->type))
+            return ret;
+
+        ret.type1 = fields[0]->type;
+        ret.type2 = fields[1]->type;
+        ret.offset2 = ALIGN(fields[0]->size, fields[1]->alignment);
+        ret.as_elements = 1;
+    }
+
+    return ret;
+}
+#endif
+
+/* allocates a single register, float register, or XLEN-sized stack slot to a datum */
+static void marshal_atom(call_builder *cb, int type, void *data) {
+    size_t value = 0;
+    switch (type) {
+        case FFI_TYPE_UINT8: value = *(uint8_t *)data; break;
+        case FFI_TYPE_SINT8: value = *(int8_t *)data; break;
+        case FFI_TYPE_UINT16: value = *(uint16_t *)data; break;
+        case FFI_TYPE_SINT16: value = *(int16_t *)data; break;
+        /* 32-bit quantities are always sign-extended in the ABI */
+        case FFI_TYPE_UINT32: value = *(int32_t *)data; break;
+        case FFI_TYPE_SINT32: value = *(int32_t *)data; break;
+#if __SIZEOF_POINTER__ == 8
+        case FFI_TYPE_UINT64: value = *(uint64_t *)data; break;
+        case FFI_TYPE_SINT64: value = *(int64_t *)data; break;
+#endif
+        case FFI_TYPE_POINTER: value = *(size_t *)data; break;
+
+        /* float values may be recoded in an implementation-defined way
+           by hardware conforming to 2.1 or earlier, so use asm to
+           reinterpret floats as doubles */
+#if ABI_FLEN >= 32
+        case FFI_TYPE_FLOAT:
+            asm("" : "=f"(cb->aregs->fa[cb->used_float++]) : "0"(*(float *)data));
+            return;
+#endif
+#if ABI_FLEN >= 64
+        case FFI_TYPE_DOUBLE:
+            asm("" : "=f"(cb->aregs->fa[cb->used_float++]) : "0"(*(double *)data));
+            return;
+#endif
+        default: FFI_ASSERT(0); break;
+    }
+
+    if (cb->used_integer == NARGREG) {
+        *cb->used_stack++ = value;
+    } else {
+        cb->aregs->a[cb->used_integer++] = value;
+    }
+}
+
+static void unmarshal_atom(call_builder *cb, int type, void *data) {
+    size_t value;
+    switch (type) {
+#if ABI_FLEN >= 32
+        case FFI_TYPE_FLOAT:
+            asm("" : "=f"(*(float *)data) : "0"(cb->aregs->fa[cb->used_float++]));
+            return;
+#endif
+#if ABI_FLEN >= 64
+        case FFI_TYPE_DOUBLE:
+            asm("" : "=f"(*(double *)data) : "0"(cb->aregs->fa[cb->used_float++]));
+            return;
+#endif
+    }
+
+    if (cb->used_integer == NARGREG) {
+        value = *cb->used_stack++;
+    } else {
+        value = cb->aregs->a[cb->used_integer++];
+    }
+
+    switch (type) {
+        case FFI_TYPE_UINT8: *(uint8_t *)data = value; break;
+        case FFI_TYPE_SINT8: *(uint8_t *)data = value; break;
+        case FFI_TYPE_UINT16: *(uint16_t *)data = value; break;
+        case FFI_TYPE_SINT16: *(uint16_t *)data = value; break;
+        case FFI_TYPE_UINT32: *(uint32_t *)data = value; break;
+        case FFI_TYPE_SINT32: *(uint32_t *)data = value; break;
+#if __SIZEOF_POINTER__ == 8
+        case FFI_TYPE_UINT64: *(uint64_t *)data = value; break;
+        case FFI_TYPE_SINT64: *(uint64_t *)data = value; break;
+#endif
+        case FFI_TYPE_POINTER: *(size_t *)data = value; break;
+        default: FFI_ASSERT(0); break;
+    }
+}
+
+/* adds an argument to a call, or a not by reference return value */
+static void marshal(call_builder *cb, ffi_type *type, int var, void *data) {
+    size_t realign[2];
+
+#if ABI_FLEN
+    if (!var && type->type == FFI_TYPE_STRUCT) {
+        float_struct_info fsi = struct_passed_as_elements(cb, type);
+        if (fsi.as_elements) {
+            marshal_atom(cb, fsi.type1, data);
+            if (fsi.offset2)
+                marshal_atom(cb, fsi.type2, ((char*)data) + fsi.offset2);
+            return;
+        }
+    }
+
+    if (!var && cb->used_float < NARGREG && IS_FLOAT(type->type)) {
+        marshal_atom(cb, type->type, data);
+        return;
+    }
+#endif
+
+    if (type->size > 2 * __SIZEOF_POINTER__) {
+        /* pass by reference */
+        marshal_atom(cb, FFI_TYPE_POINTER, &data);
+    } else if (IS_INT(type->type) || type->type == FFI_TYPE_POINTER) {
+        marshal_atom(cb, type->type, data);
+    } else {
+        /* overlong integers, soft-float floats, and structs without special
+           float handling are treated identically from this point on */
+
+        /* variadics are aligned even in registers */
+        if (type->alignment > __SIZEOF_POINTER__) {
+            if (var)
+                cb->used_integer = ALIGN(cb->used_integer, 2);
+            cb->used_stack = (size_t *)ALIGN(cb->used_stack, 2*__SIZEOF_POINTER__);
+        }
+
+        memcpy(realign, data, type->size);
+        if (type->size > 0)
+            marshal_atom(cb, FFI_TYPE_POINTER, realign);
+        if (type->size > __SIZEOF_POINTER__)
+            marshal_atom(cb, FFI_TYPE_POINTER, realign + 1);
+    }
+}
+
+/* for arguments passed by reference returns the pointer, otherwise the arg is copied (up to MAXCOPYARG bytes) */
+static void *unmarshal(call_builder *cb, ffi_type *type, int var, void *data) {
+    size_t realign[2];
+    void *pointer;
+
+#if ABI_FLEN
+    if (!var && type->type == FFI_TYPE_STRUCT) {
+        float_struct_info fsi = struct_passed_as_elements(cb, type);
+        if (fsi.as_elements) {
+            unmarshal_atom(cb, fsi.type1, data);
+            if (fsi.offset2)
+                unmarshal_atom(cb, fsi.type2, ((char*)data) + fsi.offset2);
+            return data;
+        }
+    }
+
+    if (!var && cb->used_float < NARGREG && IS_FLOAT(type->type)) {
+        unmarshal_atom(cb, type->type, data);
+        return data;
+    }
+#endif
+
+    if (type->size > 2 * __SIZEOF_POINTER__) {
+        /* pass by reference */
+        unmarshal_atom(cb, FFI_TYPE_POINTER, (char*)&pointer);
+        return pointer;
+    } else if (IS_INT(type->type) || type->type == FFI_TYPE_POINTER) {
+        unmarshal_atom(cb, type->type, data);
+        return data;
+    } else {
+        /* overlong integers, soft-float floats, and structs without special
+           float handling are treated identically from this point on */
+
+        /* variadics are aligned even in registers */
+        if (type->alignment > __SIZEOF_POINTER__) {
+            if (var)
+                cb->used_integer = ALIGN(cb->used_integer, 2);
+            cb->used_stack = (size_t *)ALIGN(cb->used_stack, 2*__SIZEOF_POINTER__);
+        }
+
+        if (type->size > 0)
+            unmarshal_atom(cb, FFI_TYPE_POINTER, realign);
+        if (type->size > __SIZEOF_POINTER__)
+            unmarshal_atom(cb, FFI_TYPE_POINTER, realign + 1);
+        memcpy(data, realign, type->size);
+        return data;
+    }
+}
+
+static int passed_by_ref(call_builder *cb, ffi_type *type, int var) {
+#if ABI_FLEN
+    if (!var && type->type == FFI_TYPE_STRUCT) {
+        float_struct_info fsi = struct_passed_as_elements(cb, type);
+        if (fsi.as_elements) return 0;
+    }
+#endif
+
+    return type->size > 2 * __SIZEOF_POINTER__;
+}
+
+/* Perform machine dependent cif processing */
+ffi_status ffi_prep_cif_machdep(ffi_cif *cif) {
+    cif->riscv_nfixedargs = cif->nargs;
+    return FFI_OK;
+}
+
+/* Perform machine dependent cif processing when we have a variadic function */
+
+ffi_status ffi_prep_cif_machdep_var(ffi_cif *cif, unsigned int nfixedargs, unsigned int ntotalargs) {
+    cif->riscv_nfixedargs = nfixedargs;
+    return FFI_OK;
+}
+
+/* Low level routine for calling functions */
+extern void ffi_call_asm(void *stack, struct call_context *regs, void (*fn)(void)) FFI_HIDDEN;
+
+void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
+{
+    /* this is a conservative estimate, assuming a complex return value and
+       that all remaining arguments are long long / __int128 */
+    size_t arg_bytes = cif->nargs <= 3 ? 0 :
+        ALIGN(2 * sizeof(size_t) * (cif->nargs - 3), STKALIGN);
+    size_t rval_bytes = 0;
+    if (rvalue == NULL && cif->rtype->size > 2*__SIZEOF_POINTER__)
+        rval_bytes = ALIGN(cif->rtype->size, STKALIGN);
+    size_t alloc_size = arg_bytes + rval_bytes + sizeof(call_context);
+
+    /* the assembly code will deallocate all stack data at lower addresses
+       than the argument region, so we need to allocate the frame and the
+       return value after the arguments in a single allocation */
+    size_t alloc_base;
+    /* Argument region must be 16-byte aligned */
+    if (_Alignof(max_align_t) >= STKALIGN) {
+        /* since sizeof long double is normally 16, the compiler will
+           guarantee alloca alignment to at least that much */
+        alloc_base = (size_t)alloca(alloc_size);
+    } else {
+        alloc_base = ALIGN(alloca(alloc_size + STKALIGN - 1), STKALIGN);
+    }
+
+    if (rval_bytes)
+        rvalue = (void*)(alloc_base + arg_bytes);
+
+    call_builder cb;
+    cb.used_float = cb.used_integer = 0;
+    cb.aregs = (call_context*)(alloc_base + arg_bytes + rval_bytes);
+    cb.used_stack = (void*)alloc_base;
+
+    int return_by_ref = passed_by_ref(&cb, cif->rtype, 0);
+    if (return_by_ref)
+        marshal(&cb, &ffi_type_pointer, 0, &rvalue);
+
+    int i;
+    for (i = 0; i < cif->nargs; i++)
+        marshal(&cb, cif->arg_types[i], i >= cif->riscv_nfixedargs, avalue[i]);
+
+    ffi_call_asm((void*)alloc_base, cb.aregs, fn);
+
+    cb.used_float = cb.used_integer = 0;
+    if (!return_by_ref && rvalue)
+        unmarshal(&cb, cif->rtype, 0, rvalue);
+}
+
+extern void ffi_closure_asm(void) FFI_HIDDEN;
+
+ffi_status ffi_prep_closure_loc(ffi_closure *closure, ffi_cif *cif, void (*fun)(ffi_cif*,void*,void**,void*), void *user_data, void *codeloc)
+{
+    uint32_t *tramp = (uint32_t *) &closure->tramp[0];
+    uint64_t fn = (uint64_t) (uintptr_t) ffi_closure_asm;
+
+    if (cif->abi <= FFI_FIRST_ABI || cif->abi >= FFI_LAST_ABI)
+        return FFI_BAD_ABI;
+
+    /* we will call ffi_closure_inner with codeloc, not closure, but as long
+       as the memory is readable it should work */
+
+    tramp[0] = 0x00000317; /* auipc t1, 0 (i.e. t0 <- codeloc) */
+#if __SIZEOF_POINTER__ == 8
+    tramp[1] = 0x01033383; /* ld t2, 16(t1) */
+#else
+    tramp[1] = 0x01032383; /* lw t2, 16(t1) */
+#endif
+    tramp[2] = 0x00038067; /* jr t2 */
+    tramp[3] = 0x00000013; /* nop */
+    tramp[4] = fn;
+    tramp[5] = fn >> 32;
+
+    closure->cif = cif;
+    closure->fun = fun;
+    closure->user_data = user_data;
+
+    __builtin___clear_cache(codeloc, codeloc + FFI_TRAMPOLINE_SIZE);
+
+    return FFI_OK;
+}
+
+/* Called by the assembly code with aregs pointing to saved argument registers
+   and stack pointing to the stacked arguments.  Return values passed in
+   registers will be reloaded from aregs. */
+void FFI_HIDDEN ffi_closure_inner(size_t *stack, call_context *aregs, ffi_closure *closure) {
+    ffi_cif *cif = closure->cif;
+    void **avalue = alloca(cif->nargs * sizeof(void*));
+    /* storage for arguments which will be copied by unmarshal().  We could
+       theoretically avoid the copies in many cases and use at most 128 bytes
+       of memory, but allocating disjoint storage for each argument is
+       simpler. */
+    char *astorage = alloca(cif->nargs * MAXCOPYARG);
+    void *rvalue;
+    call_builder cb;
+    int return_by_ref;
+    int i;
+
+    cb.aregs = aregs;
+    cb.used_integer = cb.used_float = 0;
+    cb.used_stack = stack;
+
+    return_by_ref = passed_by_ref(&cb, cif->rtype, 0);
+    if (return_by_ref)
+        unmarshal(&cb, &ffi_type_pointer, 0, &rvalue);
+    else
+        rvalue = alloca(cif->rtype->size);
+
+    for (i = 0; i < cif->nargs; i++)
+        avalue[i] = unmarshal(&cb, cif->arg_types[i],
+            i >= cif->riscv_nfixedargs, astorage + i*MAXCOPYARG);
+
+    (closure->fun)(cif, rvalue, avalue, closure->user_data);
+
+    if (!return_by_ref && cif->rtype->type != FFI_TYPE_VOID) {
+        cb.used_integer = cb.used_float = 0;
+        marshal(&cb, cif->rtype, 0, rvalue);
+    }
+}
diff --git a/src/riscv/ffitarget.h b/src/riscv/ffitarget.h
new file mode 100644
index 0000000..fcaa899
--- /dev/null
+++ b/src/riscv/ffitarget.h
@@ -0,0 +1,68 @@
+/* -----------------------------------------------------------------*-C-*-
+   ffitarget.h - 2014 Michael Knyszek
+
+   Target configuration macros for RISC-V.
+
+   Permission is hereby granted, free of charge, to any person obtaining
+   a copy of this software and associated documentation files (the
+   ``Software''), to deal in the Software without restriction, including
+   without limitation the rights to use, copy, modify, merge, publish,
+   distribute, sublicense, and/or sell copies of the Software, and to
+   permit persons to whom the Software is furnished to do so, subject to
+   the following conditions:
+
+   The above copyright notice and this permission notice shall be included
+   in all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND,
+   EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+   MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+   NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+   HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+   WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+   OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+
+   ----------------------------------------------------------------------- */
+
+#ifndef LIBFFI_TARGET_H
+#define LIBFFI_TARGET_H
+
+#ifndef LIBFFI_H
+#error "Please do not include ffitarget.h directly into your source.  Use ffi.h instead."
+#endif
+
+#ifndef __riscv
+#error "libffi was configured for a RISC-V target but this does not appear to be a RISC-V compiler."
+#endif
+
+#ifndef LIBFFI_ASM
+
+typedef unsigned long ffi_arg;
+typedef   signed long ffi_sarg;
+
+/* FFI_UNUSED_NN and riscv_unused are to maintain ABI compatibility with a
+   distributed Berkeley patch from 2014, and can be removed at SONAME bump */
+typedef enum ffi_abi {
+    FFI_FIRST_ABI = 0,
+    FFI_SYSV,
+    FFI_UNUSED_1,
+    FFI_UNUSED_2,
+    FFI_UNUSED_3,
+    FFI_LAST_ABI,
+
+    FFI_DEFAULT_ABI = FFI_SYSV
+} ffi_abi;
+
+#endif /* LIBFFI_ASM */
+
+/* ---- Definitions for closures ----------------------------------------- */
+
+#define FFI_CLOSURES 1
+#define FFI_TRAMPOLINE_SIZE 24
+#define FFI_NATIVE_RAW_API 0
+#define FFI_EXTRA_CIF_FIELDS unsigned riscv_nfixedargs; unsigned riscv_unused;
+#define FFI_TARGET_SPECIFIC_VARIADIC
+
+#endif
+
diff --git a/src/riscv/sysv.S b/src/riscv/sysv.S
new file mode 100644
index 0000000..2d09865
--- /dev/null
+++ b/src/riscv/sysv.S
@@ -0,0 +1,214 @@
+/* -----------------------------------------------------------------------
+   ffi.c - Copyright (c) 2015 Michael Knyszek <mknyszek@berkeley.edu>
+                         2015 Andrew Waterman <waterman@cs.berkeley.edu>
+                         2018 Stef O'Rear <sorear2@gmail.com>
+
+   RISC-V Foreign Function Interface
+
+   Permission is hereby granted, free of charge, to any person obtaining
+   a copy of this software and associated documentation files (the
+   ``Software''), to deal in the Software without restriction, including
+   without limitation the rights to use, copy, modify, merge, publish,
+   distribute, sublicense, and/or sell copies of the Software, and to
+   permit persons to whom the Software is furnished to do so, subject to
+   the following conditions:
+
+   The above copyright notice and this permission notice shall be included
+   in all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND,
+   EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+   MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+   NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+   HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+   WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+   OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+   ----------------------------------------------------------------------- */
+
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+
+/* Define aliases so that we can handle all ABIs uniformly */
+
+#if __SIZEOF_POINTER__ == 8
+#define PTRS 8
+#define LARG ld
+#define SARG sd
+#else
+#define PTRS 4
+#define LARG lw
+#define SARG sw
+#endif
+
+#if __riscv_float_abi_double
+#define FLTS 8
+#define FLARG fld
+#define FSARG fsd
+#elif __riscv_float_abi_single
+#define FLTS 4
+#define FLARG flw
+#define FSARG fsw
+#else
+#define FLTS 0
+#endif
+
+#define fp s0
+
+    .text
+    .globl  ffi_call_asm
+    .type   ffi_call_asm, @function
+    .hidden ffi_call_asm
+/*
+  struct call_context {
+      floatreg fa[8];
+      intreg a[8];
+      intreg pad[rv32 ? 2 : 0];
+      intreg save_fp, save_ra;
+  }
+  void ffi_call_asm(size_t *stackargs, struct call_context *regargs,
+      void (*fn)(void));
+*/
+
+#define FRAME_LEN (8 * FLTS + 8 * PTRS + 16)
+
+ffi_call_asm:
+    .cfi_startproc
+
+    /*
+      We are NOT going to set up an ordinary stack frame.  In order to pass
+      the stacked args to the called function, we adjust our stack pointer to
+      a0, which is in the _caller's_ alloca area.  We establish our own stack
+      frame at the end of the call_context.
+
+      Anything below the arguments will be freed at this point, although we
+      preserve the call_context so that it can be read back in the caller.
+    */
+
+    .cfi_def_cfa 11, FRAME_LEN # interim CFA based on a1
+    SARG    fp, FRAME_LEN - 2*PTRS(a1)
+    .cfi_offset 8, -2*PTRS
+    SARG    ra, FRAME_LEN - 1*PTRS(a1)
+    .cfi_offset 1, -1*PTRS
+
+    addi    fp, a1, FRAME_LEN
+    mv      sp, a0
+    .cfi_def_cfa 8, 0 # our frame is fully set up
+
+    # Load arguments
+    mv      t1, a2
+
+#if FLTS
+    FLARG   fa0, -FRAME_LEN+0*FLTS(fp)
+    FLARG   fa1, -FRAME_LEN+1*FLTS(fp)
+    FLARG   fa2, -FRAME_LEN+2*FLTS(fp)
+    FLARG   fa3, -FRAME_LEN+3*FLTS(fp)
+    FLARG   fa4, -FRAME_LEN+4*FLTS(fp)
+    FLARG   fa5, -FRAME_LEN+5*FLTS(fp)
+    FLARG   fa6, -FRAME_LEN+6*FLTS(fp)
+    FLARG   fa7, -FRAME_LEN+7*FLTS(fp)
+#endif
+
+    LARG    a0, -FRAME_LEN+8*FLTS+0*PTRS(fp)
+    LARG    a1, -FRAME_LEN+8*FLTS+1*PTRS(fp)
+    LARG    a2, -FRAME_LEN+8*FLTS+2*PTRS(fp)
+    LARG    a3, -FRAME_LEN+8*FLTS+3*PTRS(fp)
+    LARG    a4, -FRAME_LEN+8*FLTS+4*PTRS(fp)
+    LARG    a5, -FRAME_LEN+8*FLTS+5*PTRS(fp)
+    LARG    a6, -FRAME_LEN+8*FLTS+6*PTRS(fp)
+    LARG    a7, -FRAME_LEN+8*FLTS+7*PTRS(fp)
+
+    /* Call */
+    jalr    t1
+
+    /* Save return values - only a0/a1 (fa0/fa1) are used */
+#if FLTS
+    FSARG   fa0, -FRAME_LEN+0*FLTS(fp)
+    FSARG   fa1, -FRAME_LEN+1*FLTS(fp)
+#endif
+
+    SARG    a0, -FRAME_LEN+8*FLTS+0*PTRS(fp)
+    SARG    a1, -FRAME_LEN+8*FLTS+1*PTRS(fp)
+
+    /* Restore and return */
+    addi    sp, fp, -FRAME_LEN
+    .cfi_def_cfa 2, FRAME_LEN
+    LARG    ra, -1*PTRS(fp)
+    .cfi_restore 1
+    LARG    fp, -2*PTRS(fp)
+    .cfi_restore 8
+    ret
+    .cfi_endproc
+    .size   ffi_call_asm, .-ffi_call_asm
+
+
+/*
+  ffi_closure_asm. Expects address of the passed-in ffi_closure in t1.
+  void ffi_closure_inner(size_t *stackargs, struct call_context *regargs,
+      ffi_closure *closure);
+*/
+
+    .globl ffi_closure_asm
+    .hidden ffi_closure_asm
+    .type ffi_closure_asm, @function
+ffi_closure_asm:
+    .cfi_startproc
+
+    addi    sp,  sp, -FRAME_LEN
+    .cfi_def_cfa_offset FRAME_LEN
+
+    /* make a frame */
+    SARG    fp, FRAME_LEN - 2*PTRS(sp)
+    .cfi_offset 8, -2*PTRS
+    SARG    ra, FRAME_LEN - 1*PTRS(sp)
+    .cfi_offset 1, -1*PTRS
+    addi    fp, sp, FRAME_LEN
+
+    /* save arguments */
+#if FLTS
+    FSARG   fa0, 0*FLTS(sp)
+    FSARG   fa1, 1*FLTS(sp)
+    FSARG   fa2, 2*FLTS(sp)
+    FSARG   fa3, 3*FLTS(sp)
+    FSARG   fa4, 4*FLTS(sp)
+    FSARG   fa5, 5*FLTS(sp)
+    FSARG   fa6, 6*FLTS(sp)
+    FSARG   fa7, 7*FLTS(sp)
+#endif
+
+    SARG    a0, 8*FLTS+0*PTRS(sp)
+    SARG    a1, 8*FLTS+1*PTRS(sp)
+    SARG    a2, 8*FLTS+2*PTRS(sp)
+    SARG    a3, 8*FLTS+3*PTRS(sp)
+    SARG    a4, 8*FLTS+4*PTRS(sp)
+    SARG    a5, 8*FLTS+5*PTRS(sp)
+    SARG    a6, 8*FLTS+6*PTRS(sp)
+    SARG    a7, 8*FLTS+7*PTRS(sp)
+
+    /* enter C */
+    addi    a0, sp, FRAME_LEN
+    mv      a1, sp
+    mv      a2, t1
+
+    call    ffi_closure_inner
+
+    /* return values */
+#if FLTS
+    FLARG   fa0, 0*FLTS(sp)
+    FLARG   fa1, 1*FLTS(sp)
+#endif
+
+    LARG    a0, 8*FLTS+0*PTRS(sp)
+    LARG    a1, 8*FLTS+1*PTRS(sp)
+
+    /* restore and return */
+    LARG    ra, FRAME_LEN-1*PTRS(sp)
+    .cfi_restore 1
+    LARG    fp, FRAME_LEN-2*PTRS(sp)
+    .cfi_restore 8
+    addi    sp, sp, FRAME_LEN
+    .cfi_def_cfa_offset 0
+    ret
+    .cfi_endproc
+    .size ffi_closure_asm, .-ffi_closure_asm