intermediate changes

ref:cde9a383711a11544ce7e107a78147fb96cc4029

5 files changed, 1646 insertions, 0 deletions

diff --git a/contrib/restricted/libffi/src/arm/ffi.c b/contrib/restricted/libffi/src/arm/ffi.c
new file mode 100644
index 0000000000..4e270718a3
--- /dev/null
+++ b/contrib/restricted/libffi/src/arm/ffi.c
@@ -0,0 +1,854 @@
+/* -----------------------------------------------------------------------
+   ffi.c - Copyright (c) 2011 Timothy Wall
+           Copyright (c) 2011 Plausible Labs Cooperative, Inc.
+           Copyright (c) 2011 Anthony Green
+	   Copyright (c) 2011 Free Software Foundation
+           Copyright (c) 1998, 2008, 2011  Red Hat, Inc.
+
+   ARM 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.
+   ----------------------------------------------------------------------- */
+
+#if defined(__arm__) || defined(_M_ARM)
+#include <fficonfig.h>
+#include <ffi.h>
+#include <ffi_common.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include "internal.h"
+
+#if defined(_MSC_VER) && defined(_M_ARM)
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#endif
+
+#if FFI_EXEC_TRAMPOLINE_TABLE
+
+#ifdef __MACH__
+#include <mach/machine/vm_param.h>
+#endif
+
+#else
+#ifndef _M_ARM
+extern unsigned int ffi_arm_trampoline[2] FFI_HIDDEN;
+#else
+extern unsigned int ffi_arm_trampoline[3] FFI_HIDDEN;
+#endif
+#endif
+
+/* Forward declares. */
+static int vfp_type_p (const ffi_type *);
+static void layout_vfp_args (ffi_cif *);
+
+static void *
+ffi_align (ffi_type *ty, void *p)
+{
+  /* Align if necessary */
+  size_t alignment;
+#ifdef _WIN32_WCE
+  alignment = 4;
+#else
+  alignment = ty->alignment;
+  if (alignment < 4)
+    alignment = 4;
+#endif
+  return (void *) FFI_ALIGN (p, alignment);
+}
+
+static size_t
+ffi_put_arg (ffi_type *ty, void *src, void *dst)
+{
+  size_t z = ty->size;
+
+  switch (ty->type)
+    {
+    case FFI_TYPE_SINT8:
+      *(UINT32 *)dst = *(SINT8 *)src;
+      break;
+    case FFI_TYPE_UINT8:
+      *(UINT32 *)dst = *(UINT8 *)src;
+      break;
+    case FFI_TYPE_SINT16:
+      *(UINT32 *)dst = *(SINT16 *)src;
+      break;
+    case FFI_TYPE_UINT16:
+      *(UINT32 *)dst = *(UINT16 *)src;
+      break;
+
+    case FFI_TYPE_INT:
+    case FFI_TYPE_SINT32:
+    case FFI_TYPE_UINT32:
+    case FFI_TYPE_POINTER:
+#ifndef _MSC_VER
+    case FFI_TYPE_FLOAT:
+#endif
+      *(UINT32 *)dst = *(UINT32 *)src;
+      break;
+
+#ifdef _MSC_VER
+    // casting a float* to a UINT32* doesn't work on Windows
+    case FFI_TYPE_FLOAT:
+        *(uintptr_t *)dst = 0;
+        *(float *)dst = *(float *)src;
+        break;
+#endif
+
+    case FFI_TYPE_SINT64:
+    case FFI_TYPE_UINT64:
+    case FFI_TYPE_DOUBLE:
+      *(UINT64 *)dst = *(UINT64 *)src;
+      break;
+
+    case FFI_TYPE_STRUCT:
+    case FFI_TYPE_COMPLEX:
+      memcpy (dst, src, z);
+      break;
+
+    default:
+      abort();
+    }
+
+  return FFI_ALIGN (z, 4);
+}
+
+/* ffi_prep_args is called once stack space has been allocated
+   for the function's arguments.
+
+   The vfp_space parameter is the load area for VFP regs, the return
+   value is cif->vfp_used (word bitset of VFP regs used for passing
+   arguments). These are only used for the VFP hard-float ABI.
+*/
+static void
+ffi_prep_args_SYSV (ffi_cif *cif, int flags, void *rvalue,
+		    void **avalue, char *argp)
+{
+  ffi_type **arg_types = cif->arg_types;
+  int i, n;
+
+  if (flags == ARM_TYPE_STRUCT)
+    {
+      *(void **) argp = rvalue;
+      argp += 4;
+    }
+
+  for (i = 0, n = cif->nargs; i < n; i++)
+    {
+      ffi_type *ty = arg_types[i];
+      argp = ffi_align (ty, argp);
+      argp += ffi_put_arg (ty, avalue[i], argp);
+    }
+}
+
+static void
+ffi_prep_args_VFP (ffi_cif *cif, int flags, void *rvalue,
+                   void **avalue, char *stack, char *vfp_space)
+{
+  ffi_type **arg_types = cif->arg_types;
+  int i, n, vi = 0;
+  char *argp, *regp, *eo_regp;
+  char stack_used = 0;
+  char done_with_regs = 0;
+
+  /* The first 4 words on the stack are used for values
+     passed in core registers.  */
+  regp = stack;
+  eo_regp = argp = regp + 16;
+
+  /* If the function returns an FFI_TYPE_STRUCT in memory,
+     that address is passed in r0 to the function.  */
+  if (flags == ARM_TYPE_STRUCT)
+    {
+      *(void **) regp = rvalue;
+      regp += 4;
+    }
+
+  for (i = 0, n = cif->nargs; i < n; i++)
+    {
+      ffi_type *ty = arg_types[i];
+      void *a = avalue[i];
+      int is_vfp_type = vfp_type_p (ty);
+
+      /* Allocated in VFP registers. */
+      if (vi < cif->vfp_nargs && is_vfp_type)
+	{
+	  char *vfp_slot = vfp_space + cif->vfp_args[vi++] * 4;
+	  ffi_put_arg (ty, a, vfp_slot);
+	  continue;
+	}
+      /* Try allocating in core registers. */
+      else if (!done_with_regs && !is_vfp_type)
+	{
+	  char *tregp = ffi_align (ty, regp);
+	  size_t size = ty->size;
+	  size = (size < 4) ? 4 : size;	// pad
+	  /* Check if there is space left in the aligned register
+	     area to place the argument.  */
+	  if (tregp + size <= eo_regp)
+	    {
+	      regp = tregp + ffi_put_arg (ty, a, tregp);
+	      done_with_regs = (regp == argp);
+	      // ensure we did not write into the stack area
+	      FFI_ASSERT (regp <= argp);
+	      continue;
+	    }
+	  /* In case there are no arguments in the stack area yet,
+	     the argument is passed in the remaining core registers
+	     and on the stack.  */
+	  else if (!stack_used)
+	    {
+	      stack_used = 1;
+	      done_with_regs = 1;
+	      argp = tregp + ffi_put_arg (ty, a, tregp);
+	      FFI_ASSERT (eo_regp < argp);
+	      continue;
+	    }
+	}
+      /* Base case, arguments are passed on the stack */
+      stack_used = 1;
+      argp = ffi_align (ty, argp);
+      argp += ffi_put_arg (ty, a, argp);
+    }
+}
+
+/* Perform machine dependent cif processing */
+ffi_status FFI_HIDDEN
+ffi_prep_cif_machdep (ffi_cif *cif)
+{
+  int flags = 0, cabi = cif->abi;
+  size_t bytes = cif->bytes;
+
+  /* Map out the register placements of VFP register args.  The VFP
+     hard-float calling conventions are slightly more sophisticated
+     than the base calling conventions, so we do it here instead of
+     in ffi_prep_args(). */
+  if (cabi == FFI_VFP)
+    layout_vfp_args (cif);
+
+  /* Set the return type flag */
+  switch (cif->rtype->type)
+    {
+    case FFI_TYPE_VOID:
+      flags = ARM_TYPE_VOID;
+      break;
+
+    case FFI_TYPE_INT:
+    case FFI_TYPE_UINT8:
+    case FFI_TYPE_SINT8:
+    case FFI_TYPE_UINT16:
+    case FFI_TYPE_SINT16:
+    case FFI_TYPE_UINT32:
+    case FFI_TYPE_SINT32:
+    case FFI_TYPE_POINTER:
+      flags = ARM_TYPE_INT;
+      break;
+
+    case FFI_TYPE_SINT64:
+    case FFI_TYPE_UINT64:
+      flags = ARM_TYPE_INT64;
+      break;
+
+    case FFI_TYPE_FLOAT:
+      flags = (cabi == FFI_VFP ? ARM_TYPE_VFP_S : ARM_TYPE_INT);
+      break;
+    case FFI_TYPE_DOUBLE:
+      flags = (cabi == FFI_VFP ? ARM_TYPE_VFP_D : ARM_TYPE_INT64);
+      break;
+
+    case FFI_TYPE_STRUCT:
+    case FFI_TYPE_COMPLEX:
+      if (cabi == FFI_VFP)
+	{
+	  int h = vfp_type_p (cif->rtype);
+
+	  flags = ARM_TYPE_VFP_N;
+	  if (h == 0x100 + FFI_TYPE_FLOAT)
+	    flags = ARM_TYPE_VFP_S;
+	  if (h == 0x100 + FFI_TYPE_DOUBLE)
+	    flags = ARM_TYPE_VFP_D;
+	  if (h != 0)
+	      break;
+	}
+
+      /* A Composite Type not larger than 4 bytes is returned in r0.
+	 A Composite Type larger than 4 bytes, or whose size cannot
+	 be determined statically ... is stored in memory at an
+	 address passed [in r0].  */
+      if (cif->rtype->size <= 4)
+	flags = ARM_TYPE_INT;
+      else
+	{
+	  flags = ARM_TYPE_STRUCT;
+	  bytes += 4;
+	}
+      break;
+
+    default:
+      abort();
+    }
+
+  /* Round the stack up to a multiple of 8 bytes.  This isn't needed
+     everywhere, but it is on some platforms, and it doesn't harm anything
+     when it isn't needed.  */
+  bytes = FFI_ALIGN (bytes, 8);
+
+  /* Minimum stack space is the 4 register arguments that we pop.  */
+  if (bytes < 4*4)
+    bytes = 4*4;
+
+  cif->bytes = bytes;
+  cif->flags = flags;
+
+  return FFI_OK;
+}
+
+/* Perform machine dependent cif processing for variadic calls */
+ffi_status FFI_HIDDEN
+ffi_prep_cif_machdep_var (ffi_cif * cif,
+			  unsigned int nfixedargs, unsigned int ntotalargs)
+{
+  /* VFP variadic calls actually use the SYSV ABI */
+  if (cif->abi == FFI_VFP)
+    cif->abi = FFI_SYSV;
+
+  return ffi_prep_cif_machdep (cif);
+}
+
+/* Prototypes for assembly functions, in sysv.S.  */
+
+struct call_frame
+{
+  void *fp;
+  void *lr;
+  void *rvalue;
+  int flags;
+  void *closure;
+};
+
+extern void ffi_call_SYSV (void *stack, struct call_frame *,
+			   void (*fn) (void)) FFI_HIDDEN;
+extern void ffi_call_VFP (void *vfp_space, struct call_frame *,
+			   void (*fn) (void), unsigned vfp_used) FFI_HIDDEN;
+
+static void
+ffi_call_int (ffi_cif * cif, void (*fn) (void), void *rvalue,
+	      void **avalue, void *closure)
+{
+  int flags = cif->flags;
+  ffi_type *rtype = cif->rtype;
+  size_t bytes, rsize, vfp_size;
+  char *stack, *vfp_space, *new_rvalue;
+  struct call_frame *frame;
+
+  rsize = 0;
+  if (rvalue == NULL)
+    {
+      /* If the return value is a struct and we don't have a return
+	 value address then we need to make one.  Otherwise the return
+	 value is in registers and we can ignore them.  */
+      if (flags == ARM_TYPE_STRUCT)
+	rsize = rtype->size;
+      else
+	flags = ARM_TYPE_VOID;
+    }
+  else if (flags == ARM_TYPE_VFP_N)
+    {
+      /* Largest case is double x 4. */
+      rsize = 32;
+    }
+  else if (flags == ARM_TYPE_INT && rtype->type == FFI_TYPE_STRUCT)
+    rsize = 4;
+
+  /* Largest case.  */
+  vfp_size = (cif->abi == FFI_VFP && cif->vfp_used ? 8*8: 0);
+
+  bytes = cif->bytes;
+  stack = alloca (vfp_size + bytes + sizeof(struct call_frame) + rsize);
+
+  vfp_space = NULL;
+  if (vfp_size)
+    {
+      vfp_space = stack;
+      stack += vfp_size;
+    }
+
+  frame = (struct call_frame *)(stack + bytes);
+
+  new_rvalue = rvalue;
+  if (rsize)
+    new_rvalue = (void *)(frame + 1);
+
+  frame->rvalue = new_rvalue;
+  frame->flags = flags;
+  frame->closure = closure;
+
+  if (vfp_space)
+    {
+      ffi_prep_args_VFP (cif, flags, new_rvalue, avalue, stack, vfp_space);
+      ffi_call_VFP (vfp_space, frame, fn, cif->vfp_used);
+    }
+  else
+    {
+      ffi_prep_args_SYSV (cif, flags, new_rvalue, avalue, stack);
+      ffi_call_SYSV (stack, frame, fn);
+    }
+
+  if (rvalue && rvalue != new_rvalue)
+    memcpy (rvalue, new_rvalue, rtype->size);
+}
+
+void
+ffi_call (ffi_cif *cif, void (*fn) (void), void *rvalue, void **avalue)
+{
+  ffi_call_int (cif, fn, rvalue, avalue, NULL);
+}
+
+void
+ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue,
+	     void **avalue, void *closure)
+{
+  ffi_call_int (cif, fn, rvalue, avalue, closure);
+}
+
+static void *
+ffi_prep_incoming_args_SYSV (ffi_cif *cif, void *rvalue,
+			     char *argp, void **avalue)
+{
+  ffi_type **arg_types = cif->arg_types;
+  int i, n;
+
+  if (cif->flags == ARM_TYPE_STRUCT)
+    {
+      rvalue = *(void **) argp;
+      argp += 4;
+    }
+  else
+    {
+      if (cif->rtype->size && cif->rtype->size < 4)
+        *(uint32_t *) rvalue = 0;
+    }
+
+  for (i = 0, n = cif->nargs; i < n; i++)
+    {
+      ffi_type *ty = arg_types[i];
+      size_t z = ty->size;
+
+      argp = ffi_align (ty, argp);
+      avalue[i] = (void *) argp;
+      argp += z;
+    }
+
+  return rvalue;
+}
+
+static void *
+ffi_prep_incoming_args_VFP (ffi_cif *cif, void *rvalue, char *stack,
+			    char *vfp_space, void **avalue)
+{
+  ffi_type **arg_types = cif->arg_types;
+  int i, n, vi = 0;
+  char *argp, *regp, *eo_regp;
+  char done_with_regs = 0;
+  char stack_used = 0;
+
+  regp = stack;
+  eo_regp = argp = regp + 16;
+
+  if (cif->flags == ARM_TYPE_STRUCT)
+    {
+      rvalue = *(void **) regp;
+      regp += 4;
+    }
+
+  for (i = 0, n = cif->nargs; i < n; i++)
+    {
+      ffi_type *ty = arg_types[i];
+      int is_vfp_type = vfp_type_p (ty);
+      size_t z = ty->size;
+
+      if (vi < cif->vfp_nargs && is_vfp_type)
+	{
+	  avalue[i] = vfp_space + cif->vfp_args[vi++] * 4;
+	  continue;
+	}
+      else if (!done_with_regs && !is_vfp_type)
+	{
+	  char *tregp = ffi_align (ty, regp);
+
+	  z = (z < 4) ? 4 : z;	// pad
+
+	  /* If the arguments either fits into the registers or uses registers
+	     and stack, while we haven't read other things from the stack */
+	  if (tregp + z <= eo_regp || !stack_used)
+	    {
+	      /* Because we're little endian, this is what it turns into.  */
+	      avalue[i] = (void *) tregp;
+	      regp = tregp + z;
+
+	      /* If we read past the last core register, make sure we
+		 have not read from the stack before and continue
+		 reading after regp.  */
+	      if (regp > eo_regp)
+		{
+		  FFI_ASSERT (!stack_used);
+		  argp = regp;
+		}
+	      if (regp >= eo_regp)
+		{
+		  done_with_regs = 1;
+		  stack_used = 1;
+		}
+	      continue;
+	    }
+	}
+
+      stack_used = 1;
+      argp = ffi_align (ty, argp);
+      avalue[i] = (void *) argp;
+      argp += z;
+    }
+
+  return rvalue;
+}
+
+struct closure_frame
+{
+  char vfp_space[8*8] __attribute__((aligned(8)));
+  char result[8*4];
+  char argp[];
+};
+
+int FFI_HIDDEN
+ffi_closure_inner_SYSV (ffi_cif *cif,
+		        void (*fun) (ffi_cif *, void *, void **, void *),
+		        void *user_data,
+		        struct closure_frame *frame)
+{
+  void **avalue = (void **) alloca (cif->nargs * sizeof (void *));
+  void *rvalue = ffi_prep_incoming_args_SYSV (cif, frame->result,
+					      frame->argp, avalue);
+  fun (cif, rvalue, avalue, user_data);
+  return cif->flags;
+}
+
+int FFI_HIDDEN
+ffi_closure_inner_VFP (ffi_cif *cif,
+		       void (*fun) (ffi_cif *, void *, void **, void *),
+		       void *user_data,
+		       struct closure_frame *frame)
+{
+  void **avalue = (void **) alloca (cif->nargs * sizeof (void *));
+  void *rvalue = ffi_prep_incoming_args_VFP (cif, frame->result, frame->argp,
+					     frame->vfp_space, avalue);
+  fun (cif, rvalue, avalue, user_data);
+  return cif->flags;
+}
+
+void ffi_closure_SYSV (void) FFI_HIDDEN;
+void ffi_closure_VFP (void) FFI_HIDDEN;
+void ffi_go_closure_SYSV (void) FFI_HIDDEN;
+void ffi_go_closure_VFP (void) FFI_HIDDEN;
+
+/* the cif must already be prep'ed */
+
+ffi_status
+ffi_prep_closure_loc (ffi_closure * closure,
+		      ffi_cif * cif,
+		      void (*fun) (ffi_cif *, void *, void **, void *),
+		      void *user_data, void *codeloc)
+{
+  void (*closure_func) (void) = ffi_closure_SYSV;
+
+  if (cif->abi == FFI_VFP)
+    {
+      /* We only need take the vfp path if there are vfp arguments.  */
+      if (cif->vfp_used)
+	closure_func = ffi_closure_VFP;
+    }
+  else if (cif->abi != FFI_SYSV)
+    return FFI_BAD_ABI;
+
+#if FFI_EXEC_TRAMPOLINE_TABLE
+  void **config = (void **)((uint8_t *)codeloc - PAGE_MAX_SIZE);
+  config[0] = closure;
+  config[1] = closure_func;
+#else
+
+#ifndef _M_ARM
+  memcpy(closure->tramp, ffi_arm_trampoline, 8);
+#else
+  // cast away function type so MSVC doesn't set the lower bit of the function pointer
+  memcpy(closure->tramp, (void*)((uintptr_t)ffi_arm_trampoline & 0xFFFFFFFE), FFI_TRAMPOLINE_CLOSURE_OFFSET);
+#endif
+
+#if defined (__QNX__)
+  msync(closure->tramp, 8, 0x1000000);	/* clear data map */
+  msync(codeloc, 8, 0x1000000);	/* clear insn map */
+#elif defined(_MSC_VER)
+  FlushInstructionCache(GetCurrentProcess(), closure->tramp, FFI_TRAMPOLINE_SIZE);
+#else
+  __clear_cache(closure->tramp, closure->tramp + 8);	/* clear data map */
+  __clear_cache(codeloc, codeloc + 8);			/* clear insn map */
+#endif
+#ifdef _M_ARM
+  *(void(**)(void))(closure->tramp + FFI_TRAMPOLINE_CLOSURE_FUNCTION) = closure_func;
+#else
+  *(void (**)(void))(closure->tramp + 8) = closure_func;
+#endif
+#endif
+
+  closure->cif = cif;
+  closure->fun = fun;
+  closure->user_data = user_data;
+
+  return FFI_OK;
+}
+
+ffi_status
+ffi_prep_go_closure (ffi_go_closure *closure, ffi_cif *cif,
+		     void (*fun) (ffi_cif *, void *, void **, void *))
+{
+  void (*closure_func) (void) = ffi_go_closure_SYSV;
+
+  if (cif->abi == FFI_VFP)
+    {
+      /* We only need take the vfp path if there are vfp arguments.  */
+      if (cif->vfp_used)
+	closure_func = ffi_go_closure_VFP;
+    }
+  else if (cif->abi != FFI_SYSV)
+    return FFI_BAD_ABI;
+
+  closure->tramp = closure_func;
+  closure->cif = cif;
+  closure->fun = fun;
+
+  return FFI_OK;
+}
+
+/* Below are routines for VFP hard-float support. */
+
+/* A subroutine of vfp_type_p.  Given a structure type, return the type code
+   of the first non-structure element.  Recurse for structure elements.
+   Return -1 if the structure is in fact empty, i.e. no nested elements.  */
+
+static int
+is_hfa0 (const ffi_type *ty)
+{
+  ffi_type **elements = ty->elements;
+  int i, ret = -1;
+
+  if (elements != NULL)
+    for (i = 0; elements[i]; ++i)
+      {
+        ret = elements[i]->type;
+        if (ret == FFI_TYPE_STRUCT || ret == FFI_TYPE_COMPLEX)
+          {
+            ret = is_hfa0 (elements[i]);
+            if (ret < 0)
+              continue;
+          }
+        break;
+      }
+
+  return ret;
+}
+
+/* A subroutine of vfp_type_p.  Given a structure type, return true if all
+   of the non-structure elements are the same as CANDIDATE.  */
+
+static int
+is_hfa1 (const ffi_type *ty, int candidate)
+{
+  ffi_type **elements = ty->elements;
+  int i;
+
+  if (elements != NULL)
+    for (i = 0; elements[i]; ++i)
+      {
+        int t = elements[i]->type;
+        if (t == FFI_TYPE_STRUCT || t == FFI_TYPE_COMPLEX)
+          {
+            if (!is_hfa1 (elements[i], candidate))
+              return 0;
+          }
+        else if (t != candidate)
+          return 0;
+      }
+
+  return 1;
+}
+
+/* Determine if TY is an homogenous floating point aggregate (HFA).
+   That is, a structure consisting of 1 to 4 members of all the same type,
+   where that type is a floating point scalar.
+
+   Returns non-zero iff TY is an HFA.  The result is an encoded value where
+   bits 0-7 contain the type code, and bits 8-10 contain the element count.  */
+
+static int
+vfp_type_p (const ffi_type *ty)
+{
+  ffi_type **elements;
+  int candidate, i;
+  size_t size, ele_count;
+
+  /* Quickest tests first.  */
+  candidate = ty->type;
+  switch (ty->type)
+    {
+    default:
+      return 0;
+    case FFI_TYPE_FLOAT:
+    case FFI_TYPE_DOUBLE:
+      ele_count = 1;
+      goto done;
+    case FFI_TYPE_COMPLEX:
+      candidate = ty->elements[0]->type;
+      if (candidate != FFI_TYPE_FLOAT && candidate != FFI_TYPE_DOUBLE)
+	return 0;
+      ele_count = 2;
+      goto done;
+    case FFI_TYPE_STRUCT:
+      break;
+    }
+
+  /* No HFA types are smaller than 4 bytes, or larger than 32 bytes.  */
+  size = ty->size;
+  if (size < 4 || size > 32)
+    return 0;
+
+  /* Find the type of the first non-structure member.  */
+  elements = ty->elements;
+  candidate = elements[0]->type;
+  if (candidate == FFI_TYPE_STRUCT || candidate == FFI_TYPE_COMPLEX)
+    {
+      for (i = 0; ; ++i)
+        {
+          candidate = is_hfa0 (elements[i]);
+          if (candidate >= 0)
+            break;
+        }
+    }
+
+  /* If the first member is not a floating point type, it's not an HFA.
+     Also quickly re-check the size of the structure.  */
+  switch (candidate)
+    {
+    case FFI_TYPE_FLOAT:
+      ele_count = size / sizeof(float);
+      if (size != ele_count * sizeof(float))
+        return 0;
+      break;
+    case FFI_TYPE_DOUBLE:
+      ele_count = size / sizeof(double);
+      if (size != ele_count * sizeof(double))
+        return 0;
+      break;
+    default:
+      return 0;
+    }
+  if (ele_count > 4)
+    return 0;
+
+  /* Finally, make sure that all scalar elements are the same type.  */
+  for (i = 0; elements[i]; ++i)
+    {
+      int t = elements[i]->type;
+      if (t == FFI_TYPE_STRUCT || t == FFI_TYPE_COMPLEX)
+        {
+          if (!is_hfa1 (elements[i], candidate))
+            return 0;
+        }
+      else if (t != candidate)
+        return 0;
+    }
+
+  /* All tests succeeded.  Encode the result.  */
+ done:
+  return (ele_count << 8) | candidate;
+}
+
+static int
+place_vfp_arg (ffi_cif *cif, int h)
+{
+  unsigned short reg = cif->vfp_reg_free;
+  int align = 1, nregs = h >> 8;
+
+  if ((h & 0xff) == FFI_TYPE_DOUBLE)
+    align = 2, nregs *= 2;
+
+  /* Align register number. */
+  if ((reg & 1) && align == 2)
+    reg++;
+
+  while (reg + nregs <= 16)
+    {
+      int s, new_used = 0;
+      for (s = reg; s < reg + nregs; s++)
+	{
+	  new_used |= (1 << s);
+	  if (cif->vfp_used & (1 << s))
+	    {
+	      reg += align;
+	      goto next_reg;
+	    }
+	}
+      /* Found regs to allocate. */
+      cif->vfp_used |= new_used;
+      cif->vfp_args[cif->vfp_nargs++] = (signed char)reg;
+
+      /* Update vfp_reg_free. */
+      if (cif->vfp_used & (1 << cif->vfp_reg_free))
+	{
+	  reg += nregs;
+	  while (cif->vfp_used & (1 << reg))
+	    reg += 1;
+	  cif->vfp_reg_free = reg;
+	}
+      return 0;
+    next_reg:;
+    }
+  // done, mark all regs as used
+  cif->vfp_reg_free = 16;
+  cif->vfp_used = 0xFFFF;
+  return 1;
+}
+
+static void
+layout_vfp_args (ffi_cif * cif)
+{
+  unsigned int i;
+  /* Init VFP fields */
+  cif->vfp_used = 0;
+  cif->vfp_nargs = 0;
+  cif->vfp_reg_free = 0;
+  memset (cif->vfp_args, -1, 16);	/* Init to -1. */
+
+  for (i = 0; i < cif->nargs; i++)
+    {
+      int h = vfp_type_p (cif->arg_types[i]);
+      if (h && place_vfp_arg (cif, h) == 1)
+	break;
+    }
+}
+
+#endif /* __arm__ or _M_ARM */
diff --git a/contrib/restricted/libffi/src/arm/ffitarget.h b/contrib/restricted/libffi/src/arm/ffitarget.h
new file mode 100644
index 0000000000..cb57b84880
--- /dev/null
+++ b/contrib/restricted/libffi/src/arm/ffitarget.h
@@ -0,0 +1,89 @@
+/* -----------------------------------------------------------------*-C-*-
+   ffitarget.h - Copyright (c) 2012  Anthony Green
+                 Copyright (c) 2010  CodeSourcery
+                 Copyright (c) 1996-2003  Red Hat, Inc.
+
+   Target configuration macros for ARM.
+
+   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 LIBFFI_ASM
+typedef unsigned long          ffi_arg;
+typedef signed long            ffi_sarg;
+
+typedef enum ffi_abi {
+  FFI_FIRST_ABI = 0,
+  FFI_SYSV,
+  FFI_VFP,
+  FFI_LAST_ABI,
+#if defined(__ARM_PCS_VFP) || defined(_M_ARM)
+  FFI_DEFAULT_ABI = FFI_VFP,
+#else
+  FFI_DEFAULT_ABI = FFI_SYSV,
+#endif
+} ffi_abi;
+#endif
+
+#define FFI_EXTRA_CIF_FIELDS			\
+  int vfp_used;					\
+  unsigned short vfp_reg_free, vfp_nargs;	\
+  signed char vfp_args[16]			\
+
+#define FFI_TARGET_SPECIFIC_VARIADIC
+#ifndef _M_ARM
+#define FFI_TARGET_HAS_COMPLEX_TYPE
+#endif
+
+/* ---- Definitions for closures ----------------------------------------- */
+
+#define FFI_CLOSURES 1
+#define FFI_GO_CLOSURES 1
+#define FFI_NATIVE_RAW_API 0
+
+#if defined (FFI_EXEC_TRAMPOLINE_TABLE) && FFI_EXEC_TRAMPOLINE_TABLE
+
+#ifdef __MACH__
+#define FFI_TRAMPOLINE_SIZE 12
+#define FFI_TRAMPOLINE_CLOSURE_OFFSET 8
+#else
+#error "No trampoline table implementation"
+#endif
+
+#else
+#ifdef _MSC_VER
+#define FFI_TRAMPOLINE_SIZE 16
+#define FFI_TRAMPOLINE_CLOSURE_FUNCTION 12
+#else
+#define FFI_TRAMPOLINE_SIZE 12
+#endif
+#define FFI_TRAMPOLINE_CLOSURE_OFFSET FFI_TRAMPOLINE_SIZE
+#endif
+
+#endif
diff --git a/contrib/restricted/libffi/src/arm/internal.h b/contrib/restricted/libffi/src/arm/internal.h
new file mode 100644
index 0000000000..6cf0b2ae5d
--- /dev/null
+++ b/contrib/restricted/libffi/src/arm/internal.h
@@ -0,0 +1,7 @@
+#define ARM_TYPE_VFP_S	0
+#define ARM_TYPE_VFP_D	1
+#define ARM_TYPE_VFP_N	2
+#define ARM_TYPE_INT64	3
+#define ARM_TYPE_INT	4
+#define ARM_TYPE_VOID	5
+#define ARM_TYPE_STRUCT	6
diff --git a/contrib/restricted/libffi/src/arm/sysv.S b/contrib/restricted/libffi/src/arm/sysv.S
new file mode 100644
index 0000000000..63180a4639
--- /dev/null
+++ b/contrib/restricted/libffi/src/arm/sysv.S
@@ -0,0 +1,385 @@
+/* -----------------------------------------------------------------------
+   sysv.S - Copyright (c) 1998, 2008, 2011 Red Hat, Inc.
+	    Copyright (c) 2011 Plausible Labs Cooperative, Inc.
+
+   ARM 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.
+   ----------------------------------------------------------------------- */
+
+#ifdef __arm__
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+#include <ffi_cfi.h>
+#include "internal.h"
+
+/* GCC 4.8 provides __ARM_ARCH; construct it otherwise.  */
+#ifndef __ARM_ARCH
+# if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) \
+     || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) \
+     || defined(__ARM_ARCH_7EM__)
+#  define __ARM_ARCH 7
+# elif defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) \
+        || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) \
+        || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) \
+	|| defined(__ARM_ARCH_6M__)
+#  define __ARM_ARCH 6
+# elif defined(__ARM_ARCH_5__) || defined(__ARM_ARCH_5T__) \
+	|| defined(__ARM_ARCH_5E__) || defined(__ARM_ARCH_5TE__) \
+	|| defined(__ARM_ARCH_5TEJ__)
+#  define __ARM_ARCH 5
+# else
+#  define __ARM_ARCH 4
+# endif
+#endif
+
+/* Conditionally compile unwinder directives.  */
+#ifdef __ARM_EABI__
+# define UNWIND(...)	__VA_ARGS__
+#else
+# define UNWIND(...)
+#endif
+
+#if defined(HAVE_AS_CFI_PSEUDO_OP) && defined(__ARM_EABI__)
+	.cfi_sections	.debug_frame
+#endif
+
+#define CONCAT(a, b)	CONCAT2(a, b)
+#define CONCAT2(a, b)	a ## b
+
+#ifdef __USER_LABEL_PREFIX__
+# define CNAME(X)	CONCAT (__USER_LABEL_PREFIX__, X)
+#else
+# define CNAME(X)	X
+#endif
+#ifdef __ELF__
+# define SIZE(X)	.size CNAME(X), . - CNAME(X)
+# define TYPE(X, Y)	.type CNAME(X), Y
+#else
+# define SIZE(X)
+# define TYPE(X, Y)
+#endif
+
+#define ARM_FUNC_START_LOCAL(name)	\
+	.align	3;			\
+	TYPE(CNAME(name), %function);	\
+	CNAME(name):
+
+#define ARM_FUNC_START(name)		\
+	.globl CNAME(name);		\
+	FFI_HIDDEN(CNAME(name));	\
+	ARM_FUNC_START_LOCAL(name)
+
+#define ARM_FUNC_END(name) \
+	SIZE(name)
+
+/* Aid in defining a jump table with 8 bytes between entries.  */
+/* ??? The clang assembler doesn't handle .if with symbolic expressions.  */
+#ifdef __clang__
+# define E(index)
+#else
+# define E(index)				\
+	.if . - 0b - 8*index;			\
+	.error "type table out of sync";	\
+	.endif
+#endif
+
+	.text
+	.syntax unified
+	.arm
+
+#ifndef __clang__
+	/* We require interworking on LDM, which implies ARMv5T,
+	   which implies the existance of BLX.  */
+ 	.arch	armv5t
+#endif
+
+	/* Note that we use STC and LDC to encode VFP instructions,
+	   so that we do not need ".fpu vfp", nor get that added to
+	   the object file attributes.  These will not be executed
+	   unless the FFI_VFP abi is used.  */
+
+	@ r0:   stack
+	@ r1:   frame
+	@ r2:   fn
+	@ r3:	vfp_used
+
+ARM_FUNC_START(ffi_call_VFP)
+	UNWIND(.fnstart)
+	cfi_startproc
+
+	cmp	r3, #3			@ load only d0 if possible
+#ifdef __clang__
+	vldrle d0, [sp]
+	vldmgt sp, {d0-d7}
+#else
+	ldcle	p11, cr0, [r0]		@ vldrle d0, [sp]
+	ldcgt	p11, cr0, [r0], {16}	@ vldmgt sp, {d0-d7}
+#endif
+	add	r0, r0, #64		@ discard the vfp register args
+	/* FALLTHRU */
+ARM_FUNC_END(ffi_call_VFP)
+
+ARM_FUNC_START(ffi_call_SYSV)
+	stm	r1, {fp, lr}
+	mov	fp, r1
+
+	@ This is a bit of a lie wrt the origin of the unwind info, but
+	@ now we've got the usual frame pointer and two saved registers.
+	UNWIND(.save {fp,lr})
+	UNWIND(.setfp fp, sp)
+	cfi_def_cfa(fp, 8)
+	cfi_rel_offset(fp, 0)
+	cfi_rel_offset(lr, 4)
+
+	mov	sp, r0		@ install the stack pointer
+	mov	lr, r2		@ move the fn pointer out of the way
+	ldr	ip, [fp, #16]	@ install the static chain
+	ldmia	sp!, {r0-r3}	@ move first 4 parameters in registers.
+	blx	lr		@ call fn
+
+	@ Load r2 with the pointer to storage for the return value
+	@ Load r3 with the return type code
+	ldr	r2, [fp, #8]
+	ldr	r3, [fp, #12]
+
+	@ Deallocate the stack with the arguments.
+	mov	sp, fp
+	cfi_def_cfa_register(sp)
+
+	@ Store values stored in registers.
+	.align	3
+	add	pc, pc, r3, lsl #3
+	nop
+0:
+E(ARM_TYPE_VFP_S)
+#ifdef __clang__
+	vstr s0, [r2]
+#else
+	stc	p10, cr0, [r2]		@ vstr s0, [r2]
+#endif
+	pop	{fp,pc}
+E(ARM_TYPE_VFP_D)
+#ifdef __clang__
+	vstr d0, [r2]
+#else
+	stc	p11, cr0, [r2]		@ vstr d0, [r2]
+#endif
+	pop	{fp,pc}
+E(ARM_TYPE_VFP_N)
+#ifdef __clang__
+	vstm r2, {d0-d3}
+#else
+	stc	p11, cr0, [r2], {8}	@ vstm r2, {d0-d3}
+#endif
+	pop	{fp,pc}
+E(ARM_TYPE_INT64)
+	str	r1, [r2, #4]
+	nop
+E(ARM_TYPE_INT)
+	str	r0, [r2]
+	pop	{fp,pc}
+E(ARM_TYPE_VOID)
+	pop	{fp,pc}
+	nop
+E(ARM_TYPE_STRUCT)
+	pop	{fp,pc}
+
+	cfi_endproc
+	UNWIND(.fnend)
+ARM_FUNC_END(ffi_call_SYSV)
+
+
+/*
+	int ffi_closure_inner_* (cif, fun, user_data, frame)
+*/
+
+ARM_FUNC_START(ffi_go_closure_SYSV)
+	cfi_startproc
+	stmdb	sp!, {r0-r3}			@ save argument regs
+	cfi_adjust_cfa_offset(16)
+	ldr	r0, [ip, #4]			@ load cif
+	ldr	r1, [ip, #8]			@ load fun
+	mov	r2, ip				@ load user_data
+	b	0f
+	cfi_endproc
+ARM_FUNC_END(ffi_go_closure_SYSV)
+
+ARM_FUNC_START(ffi_closure_SYSV)
+	UNWIND(.fnstart)
+	cfi_startproc
+	stmdb	sp!, {r0-r3}			@ save argument regs
+	cfi_adjust_cfa_offset(16)
+
+#if FFI_EXEC_TRAMPOLINE_TABLE
+	ldr ip, [ip]				@ ip points to the config page, dereference to get the ffi_closure*
+#endif
+	ldr	r0, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET]	@ load cif
+	ldr	r1, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+4]  @ load fun
+	ldr	r2, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+8]  @ load user_data
+0:
+	add	ip, sp, #16			@ compute entry sp
+	sub	sp, sp, #64+32			@ allocate frame
+	cfi_adjust_cfa_offset(64+32)
+	stmdb	sp!, {ip,lr}
+
+	/* Remember that EABI unwind info only applies at call sites.
+	   We need do nothing except note the save of the stack pointer
+	   and the link registers.  */
+	UNWIND(.save {sp,lr})
+	cfi_adjust_cfa_offset(8)
+	cfi_rel_offset(lr, 4)
+
+	add	r3, sp, #8			@ load frame
+	bl	CNAME(ffi_closure_inner_SYSV)
+
+	@ Load values returned in registers.
+	add	r2, sp, #8+64			@ load result
+	adr	r3, CNAME(ffi_closure_ret)
+	add	pc, r3, r0, lsl #3
+	cfi_endproc
+	UNWIND(.fnend)
+ARM_FUNC_END(ffi_closure_SYSV)
+
+ARM_FUNC_START(ffi_go_closure_VFP)
+	cfi_startproc
+	stmdb	sp!, {r0-r3}			@ save argument regs
+	cfi_adjust_cfa_offset(16)
+	ldr	r0, [ip, #4]			@ load cif
+	ldr	r1, [ip, #8]			@ load fun
+	mov	r2, ip				@ load user_data
+	b	0f
+	cfi_endproc
+ARM_FUNC_END(ffi_go_closure_VFP)
+
+ARM_FUNC_START(ffi_closure_VFP)
+	UNWIND(.fnstart)
+	cfi_startproc
+	stmdb	sp!, {r0-r3}			@ save argument regs
+	cfi_adjust_cfa_offset(16)
+
+#if FFI_EXEC_TRAMPOLINE_TABLE
+	ldr ip, [ip]				@ ip points to the config page, dereference to get the ffi_closure*
+#endif
+	ldr	r0, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET]	@ load cif
+	ldr	r1, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+4]  @ load fun
+	ldr	r2, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+8]  @ load user_data
+0:
+	add	ip, sp, #16
+	sub	sp, sp, #64+32			@ allocate frame
+	cfi_adjust_cfa_offset(64+32)
+#ifdef __clang__
+	vstm sp, {d0-d7}
+#else
+	stc	p11, cr0, [sp], {16}		@ vstm sp, {d0-d7}
+#endif
+	stmdb	sp!, {ip,lr}
+
+	/* See above.  */
+	UNWIND(.save {sp,lr})
+	cfi_adjust_cfa_offset(8)
+	cfi_rel_offset(lr, 4)
+
+	add	r3, sp, #8			@ load frame
+	bl	CNAME(ffi_closure_inner_VFP)
+
+	@ Load values returned in registers.
+	add	r2, sp, #8+64			@ load result
+	adr	r3, CNAME(ffi_closure_ret)
+	add	pc, r3, r0, lsl #3
+	cfi_endproc
+	UNWIND(.fnend)
+ARM_FUNC_END(ffi_closure_VFP)
+
+/* Load values returned in registers for both closure entry points.
+   Note that we use LDM with SP in the register set.  This is deprecated
+   by ARM, but not yet unpredictable.  */
+
+ARM_FUNC_START_LOCAL(ffi_closure_ret)
+	cfi_startproc
+	cfi_rel_offset(sp, 0)
+	cfi_rel_offset(lr, 4)
+0:
+E(ARM_TYPE_VFP_S)
+#ifdef __clang__
+	vldr s0, [r2]
+#else
+	ldc	p10, cr0, [r2]			@ vldr s0, [r2]
+#endif
+	ldm	sp, {sp,pc}
+E(ARM_TYPE_VFP_D)
+#ifdef __clang__
+	vldr d0, [r2]
+#else
+	ldc	p11, cr0, [r2]			@ vldr d0, [r2]
+#endif
+	ldm	sp, {sp,pc}
+E(ARM_TYPE_VFP_N)
+#ifdef __clang__
+	vldm r2, {d0-d3}
+#else
+	ldc	p11, cr0, [r2], {8}		@ vldm r2, {d0-d3}
+#endif
+	ldm	sp, {sp,pc}
+E(ARM_TYPE_INT64)
+	ldr	r1, [r2, #4]
+	nop
+E(ARM_TYPE_INT)
+	ldr	r0, [r2]
+	ldm	sp, {sp,pc}
+E(ARM_TYPE_VOID)
+	ldm	sp, {sp,pc}
+	nop
+E(ARM_TYPE_STRUCT)
+	ldm	sp, {sp,pc}
+	cfi_endproc
+ARM_FUNC_END(ffi_closure_ret)
+
+#if FFI_EXEC_TRAMPOLINE_TABLE
+
+#ifdef __MACH__
+#include <mach/machine/vm_param.h>
+
+.align	PAGE_MAX_SHIFT
+ARM_FUNC_START(ffi_closure_trampoline_table_page)
+.rept	PAGE_MAX_SIZE / FFI_TRAMPOLINE_SIZE
+	adr ip, #-PAGE_MAX_SIZE   @ the config page is PAGE_MAX_SIZE behind the trampoline page
+	sub ip, #8				  @ account for pc bias
+	ldr	pc, [ip, #4]		  @ jump to ffi_closure_SYSV or ffi_closure_VFP
+.endr
+ARM_FUNC_END(ffi_closure_trampoline_table_page)
+#endif
+
+#else
+
+ARM_FUNC_START(ffi_arm_trampoline)
+0:	adr	ip, 0b
+	ldr	pc, 1f
+1:	.long	0
+ARM_FUNC_END(ffi_arm_trampoline)
+
+#endif /* FFI_EXEC_TRAMPOLINE_TABLE */
+#endif /* __arm__ */
+
+#if defined __ELF__ && defined __linux__
+	.section	.note.GNU-stack,"",%progbits
+#endif
diff --git a/contrib/restricted/libffi/src/arm/sysv_msvc_arm32.S b/contrib/restricted/libffi/src/arm/sysv_msvc_arm32.S
new file mode 100644
index 0000000000..5c99d0207a
--- /dev/null
+++ b/contrib/restricted/libffi/src/arm/sysv_msvc_arm32.S
@@ -0,0 +1,311 @@
+/* -----------------------------------------------------------------------
+   sysv.S - Copyright (c) 1998, 2008, 2011 Red Hat, Inc.
+        Copyright (c) 2011 Plausible Labs Cooperative, Inc.
+        Copyright (c) 2019 Microsoft Corporation.
+
+   ARM 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>
+#include <ffi_cfi.h>
+#include "internal.h"
+#include "ksarm.h"
+
+
+        ; 8 byte aligned AREA to support 8 byte aligned jump tables
+        MACRO
+        NESTED_ENTRY_FFI $FuncName, $AreaName, $ExceptHandler
+
+        ; compute the function's labels
+        __DeriveFunctionLabels $FuncName
+
+        ; determine the area we will put the function into
+__FuncArea   SETS    "|.text|"
+        IF "$AreaName" != ""
+__FuncArea   SETS    "$AreaName"
+        ENDIF
+
+        ; set up the exception handler itself
+__FuncExceptionHandler SETS ""
+        IF "$ExceptHandler" != ""
+__FuncExceptionHandler SETS    "|$ExceptHandler|"
+        ENDIF
+
+        ; switch to the specified area, jump tables require 8 byte alignment
+        AREA    $__FuncArea,CODE,CODEALIGN,ALIGN=3,READONLY
+
+        ; export the function name
+        __ExportProc $FuncName
+
+        ; flush any pending literal pool stuff
+        ROUT
+
+        ; reset the state of the unwind code tracking
+        __ResetUnwindState
+
+        MEND
+
+;        MACRO
+;        TABLE_ENTRY $Type, $Table
+;$Type_$Table
+;        MEND
+
+#define E(index,table) return_##index##_##table
+
+    ; r0:   stack
+    ; r1:   frame
+    ; r2:   fn
+    ; r3:   vfp_used
+
+    ; fake entry point exists only to generate exists only to 
+    ; generate .pdata for exception unwinding
+    NESTED_ENTRY_FFI ffi_call_VFP_fake
+    PROLOG_PUSH  {r11, lr}          ; save fp and lr for unwind
+
+    ALTERNATE_ENTRY ffi_call_VFP
+    cmp    r3, #3                   ; load only d0 if possible
+    vldrle d0, [r0]
+    vldmgt r0, {d0-d7}
+    add    r0, r0, #64              ; discard the vfp register args
+    b ffi_call_SYSV
+    NESTED_END ffi_call_VFP_fake
+
+    ; fake entry point exists only to generate exists only to 
+    ; generate .pdata for exception unwinding
+    NESTED_ENTRY_FFI ffi_call_SYSV_fake
+    PROLOG_PUSH  {r11, lr}          ; save fp and lr for unwind
+
+    ALTERNATE_ENTRY ffi_call_SYSV
+    stm    r1, {fp, lr}
+    mov    fp, r1
+
+    mov    sp, r0                   ; install the stack pointer
+    mov    lr, r2                   ; move the fn pointer out of the way
+    ldr    ip, [fp, #16]            ; install the static chain
+    ldmia  sp!, {r0-r3}             ; move first 4 parameters in registers.
+    blx    lr                       ; call fn
+
+    ; Load r2 with the pointer to storage for the return value
+    ; Load r3 with the return type code
+    ldr    r2, [fp, #8]
+    ldr    r3, [fp, #12]
+
+    ; Deallocate the stack with the arguments.
+    mov    sp, fp
+
+    ; Store values stored in registers.
+    ALIGN 8
+    lsl     r3, #3
+    add     r3, r3, pc
+    add     r3, #8
+    mov     pc, r3
+
+
+E(ARM_TYPE_VFP_S, ffi_call)
+    ALIGN 8
+    vstr s0, [r2]
+    pop    {fp,pc}
+E(ARM_TYPE_VFP_D, ffi_call)
+    ALIGN 8
+    vstr d0, [r2]
+    pop    {fp,pc}
+E(ARM_TYPE_VFP_N, ffi_call)
+    ALIGN 8
+    vstm r2, {d0-d3}
+    pop    {fp,pc}
+E(ARM_TYPE_INT64, ffi_call)
+    ALIGN 8
+    str    r1, [r2, #4]
+    nop
+E(ARM_TYPE_INT, ffi_call)
+    ALIGN 8
+    str    r0, [r2]
+    pop    {fp,pc}
+E(ARM_TYPE_VOID, ffi_call)
+    ALIGN 8
+    pop    {fp,pc}
+    nop
+E(ARM_TYPE_STRUCT, ffi_call)
+    ALIGN 8
+    cmp r3, #ARM_TYPE_STRUCT
+    pop    {fp,pc}
+    NESTED_END ffi_call_SYSV_fake
+
+    IMPORT |ffi_closure_inner_SYSV|
+    /*
+    int ffi_closure_inner_SYSV
+    (
+        cif,        ; r0
+        fun,        ; r1
+        user_data,  ; r2
+        frame       ; r3
+    )
+    */
+
+    NESTED_ENTRY_FFI ffi_go_closure_SYSV
+    stmdb   sp!, {r0-r3}            ; save argument regs
+    ldr     r0, [ip, #4]            ; load cif
+    ldr     r1, [ip, #8]            ; load fun
+    mov     r2, ip                  ; load user_data
+    b       ffi_go_closure_SYSV_0
+    NESTED_END ffi_go_closure_SYSV
+
+    ; r3:    ffi_closure
+
+    ; fake entry point exists only to generate exists only to 
+    ; generate .pdata for exception unwinding
+    NESTED_ENTRY_FFI ffi_closure_SYSV_fake  
+    PROLOG_PUSH  {r11, lr}          ; save fp and lr for unwind
+    ALTERNATE_ENTRY ffi_closure_SYSV
+    ldmfd   sp!, {ip,r0}            ; restore fp (r0 is used for stack alignment)
+    stmdb   sp!, {r0-r3}            ; save argument regs
+
+    ldr     r0, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET]    ; ffi_closure->cif
+    ldr     r1, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+4]  ; ffi_closure->fun
+    ldr     r2, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+8]  ; ffi_closure->user_data
+
+    ALTERNATE_ENTRY ffi_go_closure_SYSV_0
+    add     ip, sp, #16             ; compute entry sp
+
+    sub     sp, sp, #64+32          ; allocate frame parameter (sizeof(vfp_space) = 64, sizeof(result) = 32)
+    mov     r3, sp                  ; set frame parameter
+    stmdb   sp!, {ip,lr}
+
+    bl      ffi_closure_inner_SYSV  ; call the Python closure
+
+                                    ; Load values returned in registers.
+    add     r2, sp, #64+8           ; address of closure_frame->result
+    bl      ffi_closure_ret         ; move result to correct register or memory for type
+
+    ldmfd   sp!, {ip,lr}
+    mov     sp, ip                  ; restore stack pointer
+    mov     pc, lr
+    NESTED_END ffi_closure_SYSV_fake
+
+    IMPORT |ffi_closure_inner_VFP|
+    /*
+    int ffi_closure_inner_VFP
+    (
+        cif,        ; r0
+        fun,        ; r1
+        user_data,  ; r2
+        frame       ; r3
+    )
+    */
+
+    NESTED_ENTRY_FFI ffi_go_closure_VFP
+    stmdb   sp!, {r0-r3}			; save argument regs
+    ldr	r0, [ip, #4]			; load cif
+    ldr	r1, [ip, #8]			; load fun
+    mov	r2, ip				; load user_data
+    b	ffi_go_closure_VFP_0
+    NESTED_END ffi_go_closure_VFP
+
+    ; fake entry point exists only to generate exists only to 
+    ; generate .pdata for exception unwinding
+    ; r3:    closure
+    NESTED_ENTRY_FFI ffi_closure_VFP_fake
+    PROLOG_PUSH  {r11, lr}          ; save fp and lr for unwind
+
+    ALTERNATE_ENTRY ffi_closure_VFP
+    ldmfd   sp!, {ip,r0}            ; restore fp (r0 is used for stack alignment)
+    stmdb   sp!, {r0-r3}            ; save argument regs
+
+    ldr     r0, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET]    ; load cif
+    ldr     r1, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+4]  ; load fun
+    ldr     r2, [ip, #FFI_TRAMPOLINE_CLOSURE_OFFSET+8]  ; load user_data
+
+    ALTERNATE_ENTRY ffi_go_closure_VFP_0
+    add     ip, sp, #16             ; compute entry sp
+    sub     sp, sp, #32             ; save space for closure_frame->result
+    vstmdb  sp!, {d0-d7}            ; push closure_frame->vfp_space
+
+    mov     r3, sp                  ; save closure_frame
+    stmdb   sp!, {ip,lr}
+
+    bl      ffi_closure_inner_VFP
+
+    ; Load values returned in registers.
+    add     r2, sp, #64+8           ; load result
+    bl      ffi_closure_ret
+    ldmfd   sp!, {ip,lr}
+    mov     sp, ip                  ; restore stack pointer
+    mov     pc, lr
+    NESTED_END ffi_closure_VFP_fake
+
+/* Load values returned in registers for both closure entry points.
+   Note that we use LDM with SP in the register set.  This is deprecated
+   by ARM, but not yet unpredictable.  */
+
+    NESTED_ENTRY_FFI ffi_closure_ret
+    stmdb sp!, {fp,lr}
+
+    ALIGN 8
+    lsl     r0, #3
+    add     r0, r0, pc
+    add     r0, #8
+    mov     pc, r0
+
+E(ARM_TYPE_VFP_S, ffi_closure)
+    ALIGN 8
+    vldr s0, [r2]
+    b call_epilogue
+E(ARM_TYPE_VFP_D, ffi_closure)
+    ALIGN 8
+    vldr d0, [r2]
+    b call_epilogue
+E(ARM_TYPE_VFP_N, ffi_closure)
+    ALIGN 8
+    vldm r2, {d0-d3}
+    b call_epilogue
+E(ARM_TYPE_INT64, ffi_closure)
+    ALIGN 8
+    ldr    r1, [r2, #4]
+    nop
+E(ARM_TYPE_INT, ffi_closure)
+    ALIGN 8
+    ldr    r0, [r2]
+    b call_epilogue
+E(ARM_TYPE_VOID, ffi_closure)
+    ALIGN 8
+    b call_epilogue
+    nop
+E(ARM_TYPE_STRUCT, ffi_closure)
+    ALIGN 8
+    b call_epilogue
+call_epilogue
+    ldmfd sp!, {fp,pc}
+    NESTED_END ffi_closure_ret
+
+    AREA |.trampoline|, DATA, THUMB, READONLY
+    EXPORT |ffi_arm_trampoline|
+|ffi_arm_trampoline| DATA
+thisproc    adr    ip, thisproc
+            stmdb  sp!, {ip, r0}
+            ldr    pc, [pc, #0]
+            DCD    0
+            ;ENDP
+
+    END
\ No newline at end of file