ffi 1.9.25 → 1.15.0

data/ext/ffi_c/libffi/msvcc.sh

@@ -60,7 +60,7 @@ do
   case $1
   in
     --verbose)
-      $verbose=1
+      verbose=1
       shift 1
     ;;
     --version)
@@ -80,6 +80,16 @@ do
       safeseh=
       shift 1
     ;;
+    -marm)
+      ml='armasm'
+      safeseh=
+      shift 1
+    ;;
+    -marm64)
+      ml='armasm64'
+      safeseh=
+      shift 1
+    ;;
     -clang-cl)
       cl="clang-cl"
       shift 1
@@ -155,24 +165,24 @@ do
       shift 1
     ;;
     -I)
-      p=$(cygpath -m $2)
-      args="$args -I$p"
-      includes="$includes -I$p"
+      p=$(cygpath -ma "$2")
+      args="$args -I\"$p\""
+      includes="$includes -I\"$p\""
       shift 2
     ;;
     -I*)
-      p=$(cygpath -m ${1#-I})
-      args="$args -I$p"
-      includes="$includes -I$p"
+      p=$(cygpath -ma "${1#-I}")
+      args="$args -I\"$p\""
+      includes="$includes -I\"$p\""
       shift 1
     ;;
     -L)
-      p=$(cygpath -m $2)
+      p=$(cygpath -ma $2)
       linkargs="$linkargs -LIBPATH:$p"
       shift 2
     ;;
     -L*)
-      p=$(cygpath -m ${1#-L})
+      p=$(cygpath -ma ${1#-L})
       linkargs="$linkargs -LIBPATH:$p"
       shift 1
     ;;
@@ -237,6 +247,7 @@ do
       else
         output="-Fe$2"
       fi
+      armasm_output="-o $2"
       if [ -n "$assembly" ]; then
         args="$args $output"
       else
@@ -245,12 +256,12 @@ do
       shift 2
     ;;
     *.S)
-      src=$1
+      src="$(cygpath -ma $1)"
       assembly="true"
       shift 1
     ;;
     *.c)
-      args="$args $1"
+      args="$args $(cygpath -ma $1)"
       shift 1
     ;;
     *)
@@ -289,13 +300,27 @@ if [ -n "$assembly" ]; then
     fi
     ppsrc="$outdir/$(basename $src|sed 's/.S$/.asm/g')"
 
+    if [ $ml = "armasm" ]; then
+      defines="$defines -D_M_ARM"
+    fi
+
+    if [ $ml = "armasm64" ]; then
+      defines="$defines -D_M_ARM64"
+    fi
+
     if test -n "$verbose"; then
       echo "$cl -nologo -EP $includes $defines $src > $ppsrc"
     fi
 
-    "$cl" -nologo -EP $includes $defines $src > $ppsrc || exit $?
+    eval "\"$cl\" -nologo -EP $includes $defines $src" > $ppsrc || exit $?
     output="$(echo $output | sed 's%/F[dpa][^ ]*%%g')"
-    args="-nologo $safeseh $single $output $ppsrc"
+    if [ $ml = "armasm" ]; then
+      args="-nologo -g -oldit $armasm_output $ppsrc -errorReport:prompt"
+    elif [ $ml = "armasm64" ]; then
+      args="-nologo -g $armasm_output $ppsrc -errorReport:prompt"
+    else
+      args="-nologo $safeseh $single $output $ppsrc"
+    fi
 
     if test -n "$verbose"; then
       echo "$ml $args"

data/ext/ffi_c/libffi/src/aarch64/ffi.c

@@ -19,6 +19,7 @@ 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(__aarch64__) || defined(__arm64__)|| defined (_M_ARM64)
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdint.h>
@@ -26,6 +27,9 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.  */
 #include <ffi.h>
 #include <ffi_common.h>
 #include "internal.h"
+#ifdef _WIN32
+#include <windows.h> /* FlushInstructionCache */
+#endif
 
 /* Force FFI_TYPE_LONGDOUBLE to be different than FFI_TYPE_DOUBLE;
    all further uses in this file will refer to the 128-bit type.  */
@@ -58,6 +62,9 @@ struct call_context
 #if FFI_EXEC_TRAMPOLINE_TABLE
 
 #ifdef __MACH__
+#ifdef HAVE_PTRAUTH
+#include <ptrauth.h>
+#endif
 #include <mach/vm_param.h>
 #endif
 
@@ -74,6 +81,8 @@ ffi_clear_cache (void *start, void *end)
   sys_icache_invalidate (start, (char *)end - (char *)start);
 #elif defined (__GNUC__)
   __builtin___clear_cache (start, end);
+#elif defined (_WIN32)
+  FlushInstructionCache(GetCurrentProcess(), start, (char*)end - (char*)start);
 #else
 #error "Missing builtin to flush instruction cache"
 #endif
@@ -315,6 +324,9 @@ extend_integer_type (void *source, int type)
     }
 }
 
+#if defined(_MSC_VER)
+void extend_hfa_type (void *dest, void *src, int h);
+#else
 static void
 extend_hfa_type (void *dest, void *src, int h)
 {
@@ -350,10 +362,10 @@ extend_hfa_type (void *dest, void *src, int h)
 "	b	1f\n"
 "	nop\n"
 "	ldp	q16, q17, [%3]\n"	/* Q4 */
-"	ldp	q18, q19, [%3, #16]\n"
+"	ldp	q18, q19, [%3, #32]\n"
 "	b	4f\n"
 "	ldp	q16, q17, [%3]\n"	/* Q3 */
-"	ldr	q18, [%3, #16]\n"
+"	ldr	q18, [%3, #32]\n"
 "	b	3f\n"
 "	ldp	q16, q17, [%3]\n"	/* Q2 */
 "	b	2f\n"
@@ -368,7 +380,11 @@ extend_hfa_type (void *dest, void *src, int h)
     : "r"(f * 12), "r"(dest), "r"(src)
     : "memory", "v16", "v17", "v18", "v19");
 }
+#endif
 
+#if defined(_MSC_VER)
+void* compress_hfa_type (void *dest, void *src, int h);
+#else
 static void *
 compress_hfa_type (void *dest, void *reg, int h)
 {
@@ -437,6 +453,7 @@ compress_hfa_type (void *dest, void *reg, int h)
     }
   return dest;
 }
+#endif
 
 /* Either allocate an appropriate register for the argument type, or if
    none are available, allocate a stack slot and return a pointer
@@ -454,7 +471,7 @@ allocate_int_to_reg_or_stack (struct call_context *context,
   return allocate_to_stack (state, stack, size, size);
 }
 
-ffi_status
+ffi_status FFI_HIDDEN
 ffi_prep_cif_machdep (ffi_cif *cif)
 {
   ffi_type *rtype = cif->rtype;
@@ -539,14 +556,22 @@ ffi_prep_cif_machdep (ffi_cif *cif)
 
 #if defined (__APPLE__)
 /* Perform Apple-specific cif processing for variadic calls */
-ffi_status ffi_prep_cif_machdep_var(ffi_cif *cif,
-				    unsigned int nfixedargs,
-				    unsigned int ntotalargs)
+ffi_status FFI_HIDDEN
+ffi_prep_cif_machdep_var(ffi_cif *cif, unsigned int nfixedargs,
+			 unsigned int ntotalargs)
 {
   ffi_status status = ffi_prep_cif_machdep (cif);
   cif->aarch64_nfixedargs = nfixedargs;
   return status;
 }
+#else
+ffi_status FFI_HIDDEN
+ffi_prep_cif_machdep_var(ffi_cif *cif, unsigned int nfixedargs, unsigned int ntotalargs)
+{
+  ffi_status status = ffi_prep_cif_machdep (cif);
+  cif->flags |= AARCH64_FLAG_VARARG;
+  return status;
+}
 #endif /* __APPLE__ */
 
 extern void ffi_call_SYSV (struct call_context *context, void *frame,
@@ -563,7 +588,7 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *orig_rvalue,
   void *stack, *frame, *rvalue;
   struct arg_state state;
   size_t stack_bytes, rtype_size, rsize;
-  int i, nargs, flags;
+  int i, nargs, flags, isvariadic = 0;
   ffi_type *rtype;
 
   flags = cif->flags;
@@ -571,6 +596,12 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *orig_rvalue,
   rtype_size = rtype->size;
   stack_bytes = cif->bytes;
 
+  if (flags & AARCH64_FLAG_VARARG)
+  {
+    isvariadic = 1;
+    flags &= ~AARCH64_FLAG_VARARG;
+  }
+
   /* If the target function returns a structure via hidden pointer,
      then we cannot allow a null rvalue.  Otherwise, mash a null
      rvalue to void return type.  */
@@ -588,8 +619,8 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *orig_rvalue,
   /* Allocate consectutive stack for everything we'll need.  */
   context = alloca (sizeof(struct call_context) + stack_bytes + 32 + rsize);
   stack = context + 1;
-  frame = stack + stack_bytes;
-  rvalue = (rsize ? frame + 32 : orig_rvalue);
+  frame = (void*)((uintptr_t)stack + (uintptr_t)stack_bytes);
+  rvalue = (rsize ? (void*)((uintptr_t)frame + 32) : orig_rvalue);
 
   arg_init (&state);
   for (i = 0, nargs = cif->nargs; i < nargs; i++)
@@ -650,16 +681,31 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *orig_rvalue,
 	    h = is_vfp_type (ty);
 	    if (h)
 	      {
-		int elems = 4 - (h & 3);
-	        if (state.nsrn + elems <= N_V_ARG_REG)
-		  {
-		    dest = &context->v[state.nsrn];
-		    state.nsrn += elems;
-		    extend_hfa_type (dest, a, h);
-		    break;
-		  }
-		state.nsrn = N_V_ARG_REG;
-		dest = allocate_to_stack (&state, stack, ty->alignment, s);
+              int elems = 4 - (h & 3);
+              if (cif->abi == FFI_WIN64 && isvariadic)
+              {
+                if (state.ngrn + elems <= N_X_ARG_REG)
+                {
+                  dest = &context->x[state.ngrn];
+                  state.ngrn += elems;
+                  extend_hfa_type(dest, a, h);
+                  break;
+                }
+                state.nsrn = N_X_ARG_REG;
+                dest = allocate_to_stack(&state, stack, ty->alignment, s);
+              }
+              else
+              {
+                if (state.nsrn + elems <= N_V_ARG_REG)
+                {
+                  dest = &context->v[state.nsrn];
+                  state.nsrn += elems;
+                  extend_hfa_type (dest, a, h);
+                  break;
+                }
+                state.nsrn = N_V_ARG_REG;
+                dest = allocate_to_stack (&state, stack, ty->alignment, s);
+              }
 	      }
 	    else if (s > 16)
 	      {
@@ -681,7 +727,7 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *orig_rvalue,
 		       X registers, then the argument is copied into
 		       consecutive X registers.  */
 		    dest = &context->x[state.ngrn];
-		    state.ngrn += n;
+                    state.ngrn += (unsigned int)n;
 		  }
 		else
 		  {
@@ -756,6 +802,9 @@ ffi_prep_closure_loc (ffi_closure *closure,
 
 #if FFI_EXEC_TRAMPOLINE_TABLE
 #ifdef __MACH__
+#ifdef HAVE_PTRAUTH
+  codeloc = ptrauth_strip (codeloc, ptrauth_key_asia);
+#endif
   void **config = (void **)((uint8_t *)codeloc - PAGE_MAX_SIZE);
   config[0] = closure;
   config[1] = start;
@@ -773,6 +822,16 @@ ffi_prep_closure_loc (ffi_closure *closure,
   *(UINT64 *)(tramp + 16) = (uintptr_t)start;
 
   ffi_clear_cache(tramp, tramp + FFI_TRAMPOLINE_SIZE);
+
+  /* Also flush the cache for code mapping.  */
+#ifdef _WIN32
+  // Not using dlmalloc.c for Windows ARM64 builds
+  // so calling ffi_data_to_code_pointer() isn't necessary
+  unsigned char *tramp_code = tramp;
+  #else
+  unsigned char *tramp_code = ffi_data_to_code_pointer (tramp);
+  #endif
+  ffi_clear_cache (tramp_code, tramp_code + FFI_TRAMPOLINE_SIZE);
 #endif
 
   closure->cif = cif;
@@ -871,55 +930,78 @@ ffi_closure_SYSV_inner (ffi_cif *cif,
 	  if (h)
 	    {
 	      n = 4 - (h & 3);
-	      if (state.nsrn + n <= N_V_ARG_REG)
-		{
-		  void *reg = &context->v[state.nsrn];
-		  state.nsrn += n;
-
-		  /* Eeek! We need a pointer to the structure, however the
-		     homogeneous float elements are being passed in individual
-		     registers, therefore for float and double the structure
-		     is not represented as a contiguous sequence of bytes in
-		     our saved register context.  We don't need the original
-		     contents of the register storage, so we reformat the
-		     structure into the same memory.  */
-		  avalue[i] = compress_hfa_type (reg, reg, h);
-		}
-	      else
-		{
-		  state.nsrn = N_V_ARG_REG;
-		  avalue[i] = allocate_to_stack (&state, stack,
-						 ty->alignment, s);
-		}
-	    }
-	  else if (s > 16)
-	    {
-	      /* Replace Composite type of size greater than 16 with a
-		 pointer.  */
-	      avalue[i] = *(void **)
-		allocate_int_to_reg_or_stack (context, &state, stack,
-					      sizeof (void *));
-	    }
-	  else
-	    {
-	      n = (s + 7) / 8;
-	      if (state.ngrn + n <= N_X_ARG_REG)
-		{
-		  avalue[i] = &context->x[state.ngrn];
-		  state.ngrn += n;
-		}
-	      else
-		{
-		  state.ngrn = N_X_ARG_REG;
-		  avalue[i] = allocate_to_stack (&state, stack,
-						 ty->alignment, s);
-		}
-	    }
-	  break;
-
-	default:
-	  abort();
-	}
+#ifdef _WIN32  /* for handling armasm calling convention */
+              if (cif->is_variadic)
+                {
+                  if (state.ngrn + n <= N_X_ARG_REG)
+                    {
+                      void *reg = &context->x[state.ngrn];
+                      state.ngrn += (unsigned int)n;
+    
+                      /* Eeek! We need a pointer to the structure, however the
+                       homogeneous float elements are being passed in individual
+                       registers, therefore for float and double the structure
+                       is not represented as a contiguous sequence of bytes in
+                       our saved register context.  We don't need the original
+                       contents of the register storage, so we reformat the
+                       structure into the same memory.  */
+                      avalue[i] = compress_hfa_type(reg, reg, h);
+                    }
+                  else
+                    {
+                      state.ngrn = N_X_ARG_REG;
+                      state.nsrn = N_V_ARG_REG;
+                      avalue[i] = allocate_to_stack(&state, stack,
+                             ty->alignment, s);
+                    }
+                }
+              else
+                {
+#endif  /* for handling armasm calling convention */
+                  if (state.nsrn + n <= N_V_ARG_REG)
+                    {
+                      void *reg = &context->v[state.nsrn];
+                      state.nsrn += (unsigned int)n;
+                      avalue[i] = compress_hfa_type(reg, reg, h);
+                    }
+                  else
+                    {
+                      state.nsrn = N_V_ARG_REG;
+                      avalue[i] = allocate_to_stack(&state, stack,
+                                                   ty->alignment, s);
+                    }
+#ifdef _WIN32  /* for handling armasm calling convention */
+                }
+#endif  /* for handling armasm calling convention */
+            }
+          else if (s > 16)
+            {
+              /* Replace Composite type of size greater than 16 with a
+                  pointer.  */
+              avalue[i] = *(void **)
+              allocate_int_to_reg_or_stack (context, &state, stack,
+                                         sizeof (void *));
+            }
+          else
+            {
+              n = (s + 7) / 8;
+              if (state.ngrn + n <= N_X_ARG_REG)
+                {
+                  avalue[i] = &context->x[state.ngrn];
+                  state.ngrn += (unsigned int)n;
+                }
+              else
+                {
+                  state.ngrn = N_X_ARG_REG;
+                  avalue[i] = allocate_to_stack(&state, stack,
+                                           ty->alignment, s);
+                }
+            }
+          break;
+
+        default:
+          abort();
+      }
 
 #if defined (__APPLE__)
       if (i + 1 == cif->aarch64_nfixedargs)
@@ -939,3 +1021,5 @@ ffi_closure_SYSV_inner (ffi_cif *cif,
 
   return flags;
 }
+
+#endif /* (__aarch64__) || defined(__arm64__)|| defined (_M_ARM64)*/