Merge with upstream libffi db1b34b7e1f5e473d17557e454a29933dfecd1af

Includes build fixes for Solaris and FreeBSD.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@220222 138bc75d-0d04-0410-961f-82ee72b054a4

9 files changed, 467 insertions, 174 deletions

diff --git a/libffi/ChangeLog b/libffi/ChangeLog
index 9d1ae071c50a..3bc3036ca231 100644
--- a/libffi/ChangeLog
+++ b/libffi/ChangeLog
@@ -1,3 +1,7 @@
+2015-01-28  Richard Henderson  <rth@redhat.com>
+
+	* Merge to upstream commit db1b34b7e1f5e473d17557e454a29933dfecd1af.
+
 2015-01-27  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
 
 	* testsuite/lib/libffi.exp: Load target-supports.exp.
diff --git a/libffi/configure b/libffi/configure
index f580af0f69bc..751ed015f3a9 100755
--- a/libffi/configure
+++ b/libffi/configure
@@ -15643,7 +15643,8 @@ $as_echo "#define HAVE_AS_CFI_PSEUDO_OP 1" >>confdefs.h
  fi
 
 
-if test x$TARGET = xSPARC; then
+case "$TARGET" in
+  SPARC)
     { $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler and linker support unaligned pc related relocs" >&5
 $as_echo_n "checking assembler and linker support unaligned pc related relocs... " >&6; }
 if test "${libffi_cv_as_sparc_ua_pcrel+set}" = set; then :
@@ -15717,9 +15718,9 @@ $as_echo "$libffi_cv_as_register_pseudo_op" >&6; }
 $as_echo "#define HAVE_AS_REGISTER_PSEUDO_OP 1" >>confdefs.h
 
     fi
-fi
+    ;;
 
-if test x$TARGET = xX86 || test x$TARGET = xX86_WIN32 || test x$TARGET = xX86_64; then
+  X86*)
     { $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler supports pc related relocs" >&5
 $as_echo_n "checking assembler supports pc related relocs... " >&6; }
 if test "${libffi_cv_as_x86_pcrel+set}" = set; then :
@@ -15740,79 +15741,9 @@ $as_echo "$libffi_cv_as_x86_pcrel" >&6; }
 $as_echo "#define HAVE_AS_X86_PCREL 1" >>confdefs.h
 
     fi
+    ;;
 
-    { $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler .ascii pseudo-op support" >&5
-$as_echo_n "checking assembler .ascii pseudo-op support... " >&6; }
-if test "${libffi_cv_as_ascii_pseudo_op+set}" = set; then :
-  $as_echo_n "(cached) " >&6
-else
-
-       libffi_cv_as_ascii_pseudo_op=unknown
-       # Check if we have .ascii
-       cat confdefs.h - <<_ACEOF >conftest.$ac_ext
-/* end confdefs.h.  */
-
-int
-main ()
-{
-asm (".ascii \\"string\\"");
-  ;
-  return 0;
-}
-_ACEOF
-if ac_fn_c_try_compile "$LINENO"; then :
-  libffi_cv_as_ascii_pseudo_op=yes
-else
-  libffi_cv_as_ascii_pseudo_op=no
-fi
-rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-
-fi
-{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libffi_cv_as_ascii_pseudo_op" >&5
-$as_echo "$libffi_cv_as_ascii_pseudo_op" >&6; }
-    if test "x$libffi_cv_as_ascii_pseudo_op" = xyes; then
-
-$as_echo "#define HAVE_AS_ASCII_PSEUDO_OP 1" >>confdefs.h
-
-    fi
-
-    { $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler .string pseudo-op support" >&5
-$as_echo_n "checking assembler .string pseudo-op support... " >&6; }
-if test "${libffi_cv_as_string_pseudo_op+set}" = set; then :
-  $as_echo_n "(cached) " >&6
-else
-
-       libffi_cv_as_string_pseudo_op=unknown
-       # Check if we have .string
-       cat confdefs.h - <<_ACEOF >conftest.$ac_ext
-/* end confdefs.h.  */
-
-int
-main ()
-{
-asm (".string \\"string\\"");
-  ;
-  return 0;
-}
-_ACEOF
-if ac_fn_c_try_compile "$LINENO"; then :
-  libffi_cv_as_string_pseudo_op=yes
-else
-  libffi_cv_as_string_pseudo_op=no
-fi
-rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-
-fi
-{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libffi_cv_as_string_pseudo_op" >&5
-$as_echo "$libffi_cv_as_string_pseudo_op" >&6; }
-    if test "x$libffi_cv_as_string_pseudo_op" = xyes; then
-
-$as_echo "#define HAVE_AS_STRING_PSEUDO_OP 1" >>confdefs.h
-
-    fi
-fi
-
-if test x$TARGET = xS390; then
+  S390)
     { $as_echo "$as_me:${as_lineno-$LINENO}: checking compiler uses zarch features" >&5
 $as_echo_n "checking compiler uses zarch features... " >&6; }
 if test "${libffi_cv_as_s390_zarch+set}" = set; then :
@@ -15835,7 +15766,8 @@ $as_echo "$libffi_cv_as_s390_zarch" >&6; }
 $as_echo "#define HAVE_AS_S390_ZARCH 1" >>confdefs.h
 
     fi
-fi
+    ;;
+esac
 
 # On PaX enable kernels that have MPROTECT enable we can't use PROT_EXEC.
 # Check whether --enable-pax_emutramp was given.
@@ -15850,7 +15782,7 @@ fi
 
 FFI_EXEC_TRAMPOLINE_TABLE=0
 case "$target" in
-     *arm*-apple-darwin*)
+     *arm*-apple-darwin* | aarch64-apple-darwin*)
        FFI_EXEC_TRAMPOLINE_TABLE=1
 
 $as_echo "#define FFI_EXEC_TRAMPOLINE_TABLE 1" >>confdefs.h
diff --git a/libffi/configure.ac b/libffi/configure.ac
index 07ec10d42e72..68501d100dcb 100644
--- a/libffi/configure.ac
+++ b/libffi/configure.ac
@@ -112,7 +112,8 @@ AC_C_BIGENDIAN
 
 GCC_AS_CFI_PSEUDO_OP
 
-if test x$TARGET = xSPARC; then
+case "$TARGET" in
+  SPARC)
     AC_CACHE_CHECK([assembler and linker support unaligned pc related relocs],
 	libffi_cv_as_sparc_ua_pcrel, [
 	save_CFLAGS="$CFLAGS"
@@ -141,9 +142,9 @@ if test x$TARGET = xSPARC; then
        AC_DEFINE(HAVE_AS_REGISTER_PSEUDO_OP, 1,
 	       [Define if your assembler supports .register.])
     fi
-fi
+    ;;
 
-if test x$TARGET = xX86 || test x$TARGET = xX86_WIN32 || test x$TARGET = xX86_64; then
+  X86*)
     AC_CACHE_CHECK([assembler supports pc related relocs],
 	libffi_cv_as_x86_pcrel, [
 	libffi_cv_as_x86_pcrel=no
@@ -156,35 +157,9 @@ if test x$TARGET = xX86 || test x$TARGET = xX86_WIN32 || test x$TARGET = xX86_64
 	AC_DEFINE(HAVE_AS_X86_PCREL, 1,
 		  [Define if your assembler supports PC relative relocs.])
     fi
+    ;;
 
-    AC_CACHE_CHECK([assembler .ascii pseudo-op support],
-       libffi_cv_as_ascii_pseudo_op, [
-       libffi_cv_as_ascii_pseudo_op=unknown
-       # Check if we have .ascii
-       AC_TRY_COMPILE(,[asm (".ascii \\"string\\"");],
-		       [libffi_cv_as_ascii_pseudo_op=yes],
-		       [libffi_cv_as_ascii_pseudo_op=no])
-    ])
-    if test "x$libffi_cv_as_ascii_pseudo_op" = xyes; then
-       AC_DEFINE(HAVE_AS_ASCII_PSEUDO_OP, 1,
-	       [Define if your assembler supports .ascii.])
-    fi
-
-    AC_CACHE_CHECK([assembler .string pseudo-op support],
-       libffi_cv_as_string_pseudo_op, [
-       libffi_cv_as_string_pseudo_op=unknown
-       # Check if we have .string
-       AC_TRY_COMPILE(,[asm (".string \\"string\\"");],
-		       [libffi_cv_as_string_pseudo_op=yes],
-		       [libffi_cv_as_string_pseudo_op=no])
-    ])
-    if test "x$libffi_cv_as_string_pseudo_op" = xyes; then
-       AC_DEFINE(HAVE_AS_STRING_PSEUDO_OP, 1,
-	       [Define if your assembler supports .string.])
-    fi
-fi
-
-if test x$TARGET = xS390; then
+  S390)
     AC_CACHE_CHECK([compiler uses zarch features],
 	libffi_cv_as_s390_zarch, [
 	libffi_cv_as_s390_zarch=no
@@ -199,7 +174,8 @@ if test x$TARGET = xS390; then
 	AC_DEFINE(HAVE_AS_S390_ZARCH, 1,
 		  [Define if the compiler uses zarch features.])
     fi
-fi
+    ;;
+esac
 
 # On PaX enable kernels that have MPROTECT enable we can't use PROT_EXEC.
 AC_ARG_ENABLE(pax_emutramp,
@@ -211,7 +187,7 @@ AC_ARG_ENABLE(pax_emutramp,
 
 FFI_EXEC_TRAMPOLINE_TABLE=0
 case "$target" in
-     *arm*-apple-darwin*)
+     *arm*-apple-darwin* | aarch64-apple-darwin*)
        FFI_EXEC_TRAMPOLINE_TABLE=1
        AC_DEFINE(FFI_EXEC_TRAMPOLINE_TABLE, 1,
                  [Cannot use PROT_EXEC on this target, so, we revert to
diff --git a/libffi/doc/libffi.texi b/libffi/doc/libffi.texi
index b1c9bc367bb9..b9887a8c65b7 100644
--- a/libffi/doc/libffi.texi
+++ b/libffi/doc/libffi.texi
@@ -748,11 +748,6 @@ Variadic closures.
 There is no support for bit fields in structures.
 
 @item
-The closure API is
-
-@c FIXME: ...
-
-@item
 The ``raw'' API is undocumented.
 @c argument promotion?
 @c unions?
diff --git a/libffi/src/aarch64/ffi.c b/libffi/src/aarch64/ffi.c
index 0cace9d8e7c3..f79602bb7b8b 100644
--- a/libffi/src/aarch64/ffi.c
+++ b/libffi/src/aarch64/ffi.c
@@ -725,6 +725,240 @@ ffi_call_go (ffi_cif *cif, void (*fn) (void), void *rvalue,
 extern void ffi_closure_SYSV (void) FFI_HIDDEN;
 extern void ffi_closure_SYSV_V (void) FFI_HIDDEN;
 
+#if FFI_EXEC_TRAMPOLINE_TABLE
+
+#include <mach/mach.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+extern void *ffi_closure_trampoline_table_page;
+
+typedef struct ffi_trampoline_table ffi_trampoline_table;
+typedef struct ffi_trampoline_table_entry ffi_trampoline_table_entry;
+
+struct ffi_trampoline_table
+{
+  /* contiguous writable and executable pages */
+  vm_address_t config_page;
+  vm_address_t trampoline_page;
+
+  /* free list tracking */
+  uint16_t free_count;
+  ffi_trampoline_table_entry *free_list;
+  ffi_trampoline_table_entry *free_list_pool;
+
+  ffi_trampoline_table *prev;
+  ffi_trampoline_table *next;
+};
+
+struct ffi_trampoline_table_entry
+{
+  void *(*trampoline) ();
+  ffi_trampoline_table_entry *next;
+};
+
+/* The trampoline configuration is placed a page prior to the trampoline's entry point */
+#define FFI_TRAMPOLINE_CODELOC_CONFIG(codeloc) ((void **) (((uint8_t *) codeloc) - PAGE_SIZE));
+
+/* Total number of trampolines that fit in one trampoline table */
+#define FFI_TRAMPOLINE_COUNT (PAGE_SIZE / FFI_TRAMPOLINE_SIZE)
+
+static pthread_mutex_t ffi_trampoline_lock = PTHREAD_MUTEX_INITIALIZER;
+static ffi_trampoline_table *ffi_trampoline_tables = NULL;
+
+static ffi_trampoline_table *
+ffi_trampoline_table_alloc ()
+{
+  ffi_trampoline_table *table = NULL;
+
+  /* Loop until we can allocate two contiguous pages */
+  while (table == NULL)
+    {
+      vm_address_t config_page = 0x0;
+      kern_return_t kt;
+
+      /* Try to allocate two pages */
+      kt =
+	vm_allocate (mach_task_self (), &config_page, PAGE_SIZE * 2,
+		     VM_FLAGS_ANYWHERE);
+      if (kt != KERN_SUCCESS)
+	{
+	  fprintf (stderr, "vm_allocate() failure: %d at %s:%d\n", kt,
+		   __FILE__, __LINE__);
+	  break;
+	}
+
+      /* Now drop the second half of the allocation to make room for the trampoline table */
+      vm_address_t trampoline_page = config_page + PAGE_SIZE;
+      kt = vm_deallocate (mach_task_self (), trampoline_page, PAGE_SIZE);
+      if (kt != KERN_SUCCESS)
+	{
+	  fprintf (stderr, "vm_deallocate() failure: %d at %s:%d\n", kt,
+		   __FILE__, __LINE__);
+	  break;
+	}
+
+      /* Remap the trampoline table to directly follow the config page */
+      vm_prot_t cur_prot;
+      vm_prot_t max_prot;
+
+      kt =
+	vm_remap (mach_task_self (), &trampoline_page, PAGE_SIZE, 0x0, FALSE,
+		  mach_task_self (),
+		  (vm_address_t) & ffi_closure_trampoline_table_page, FALSE,
+		  &cur_prot, &max_prot, VM_INHERIT_SHARE);
+
+      /* If we lost access to the destination trampoline page, drop our config allocation mapping and retry */
+      if (kt != KERN_SUCCESS)
+	{
+	  /* Log unexpected failures */
+	  if (kt != KERN_NO_SPACE)
+	    {
+	      fprintf (stderr, "vm_remap() failure: %d at %s:%d\n", kt,
+		       __FILE__, __LINE__);
+	    }
+
+	  vm_deallocate (mach_task_self (), config_page, PAGE_SIZE);
+	  continue;
+	}
+
+      /* We have valid trampoline and config pages */
+      table = calloc (1, sizeof (ffi_trampoline_table));
+      table->free_count = FFI_TRAMPOLINE_COUNT;
+      table->config_page = config_page;
+      table->trampoline_page = trampoline_page;
+
+      /* Create and initialize the free list */
+      table->free_list_pool =
+	calloc (FFI_TRAMPOLINE_COUNT, sizeof (ffi_trampoline_table_entry));
+
+      uint16_t i;
+      for (i = 0; i < table->free_count; i++)
+	{
+	  ffi_trampoline_table_entry *entry = &table->free_list_pool[i];
+	  entry->trampoline =
+	    (void *) (table->trampoline_page + (i * FFI_TRAMPOLINE_SIZE));
+
+	  if (i < table->free_count - 1)
+	    entry->next = &table->free_list_pool[i + 1];
+	}
+
+      table->free_list = table->free_list_pool;
+    }
+
+  return table;
+}
+
+void *
+ffi_closure_alloc (size_t size, void **code)
+{
+  /* Create the closure */
+  ffi_closure *closure = malloc (size);
+  if (closure == NULL)
+    return NULL;
+
+  pthread_mutex_lock (&ffi_trampoline_lock);
+
+  /* Check for an active trampoline table with available entries. */
+  ffi_trampoline_table *table = ffi_trampoline_tables;
+  if (table == NULL || table->free_list == NULL)
+    {
+      table = ffi_trampoline_table_alloc ();
+      if (table == NULL)
+	{
+	  free (closure);
+	  return NULL;
+	}
+
+      /* Insert the new table at the top of the list */
+      table->next = ffi_trampoline_tables;
+      if (table->next != NULL)
+	table->next->prev = table;
+
+      ffi_trampoline_tables = table;
+    }
+
+  /* Claim the free entry */
+  ffi_trampoline_table_entry *entry = ffi_trampoline_tables->free_list;
+  ffi_trampoline_tables->free_list = entry->next;
+  ffi_trampoline_tables->free_count--;
+  entry->next = NULL;
+
+  pthread_mutex_unlock (&ffi_trampoline_lock);
+
+  /* Initialize the return values */
+  *code = entry->trampoline;
+  closure->trampoline_table = table;
+  closure->trampoline_table_entry = entry;
+
+  return closure;
+}
+
+void
+ffi_closure_free (void *ptr)
+{
+  ffi_closure *closure = ptr;
+
+  pthread_mutex_lock (&ffi_trampoline_lock);
+
+  /* Fetch the table and entry references */
+  ffi_trampoline_table *table = closure->trampoline_table;
+  ffi_trampoline_table_entry *entry = closure->trampoline_table_entry;
+
+  /* Return the entry to the free list */
+  entry->next = table->free_list;
+  table->free_list = entry;
+  table->free_count++;
+
+  /* If all trampolines within this table are free, and at least one other table exists, deallocate
+   * the table */
+  if (table->free_count == FFI_TRAMPOLINE_COUNT
+      && ffi_trampoline_tables != table)
+    {
+      /* Remove from the list */
+      if (table->prev != NULL)
+	table->prev->next = table->next;
+
+      if (table->next != NULL)
+	table->next->prev = table->prev;
+
+      /* Deallocate pages */
+      kern_return_t kt;
+      kt = vm_deallocate (mach_task_self (), table->config_page, PAGE_SIZE);
+      if (kt != KERN_SUCCESS)
+	fprintf (stderr, "vm_deallocate() failure: %d at %s:%d\n", kt,
+		 __FILE__, __LINE__);
+
+      kt =
+	vm_deallocate (mach_task_self (), table->trampoline_page, PAGE_SIZE);
+      if (kt != KERN_SUCCESS)
+	fprintf (stderr, "vm_deallocate() failure: %d at %s:%d\n", kt,
+		 __FILE__, __LINE__);
+
+      /* Deallocate free list */
+      free (table->free_list_pool);
+      free (table);
+    }
+  else if (ffi_trampoline_tables != table)
+    {
+      /* Otherwise, bump this table to the top of the list */
+      table->prev = NULL;
+      table->next = ffi_trampoline_tables;
+      if (ffi_trampoline_tables != NULL)
+	ffi_trampoline_tables->prev = table;
+
+      ffi_trampoline_tables = table;
+    }
+
+  pthread_mutex_unlock (&ffi_trampoline_lock);
+
+  /* Free the closure */
+  free (closure);
+}
+
+#endif
+
 ffi_status
 ffi_prep_closure_loc (ffi_closure *closure,
                       ffi_cif* cif,
@@ -732,31 +966,39 @@ ffi_prep_closure_loc (ffi_closure *closure,
                       void *user_data,
                       void *codeloc)
 {
+  if (cif->abi != FFI_SYSV)
+    return FFI_BAD_ABI;
+
+  void (*start)(void);
+  
+  if (cif->flags & AARCH64_FLAG_ARG_V)
+    start = ffi_closure_SYSV_V;
+  else
+    start = ffi_closure_SYSV;
+
+#if FFI_EXEC_TRAMPOLINE_TABLE
+  void **config = FFI_TRAMPOLINE_CODELOC_CONFIG (codeloc);
+  config[0] = closure;
+  config[1] = start;
+#else
   static const unsigned char trampoline[16] = {
     0x90, 0x00, 0x00, 0x58,	/* ldr	x16, tramp+16	*/
     0xf1, 0xff, 0xff, 0x10,	/* adr	x17, tramp+0	*/
     0x00, 0x02, 0x1f, 0xd6	/* br	x16		*/
   };
   char *tramp = closure->tramp;
-  void (*start)(void);
+  
+  memcpy (tramp, trampoline, sizeof(trampoline));
+  
+  *(UINT64 *)(tramp + 16) = (uintptr_t)start;
 
-  if (cif->abi != FFI_SYSV)
-    return FFI_BAD_ABI;
+  ffi_clear_cache(tramp, tramp + FFI_TRAMPOLINE_SIZE);
+#endif
 
   closure->cif = cif;
   closure->fun = fun;
   closure->user_data = user_data;
 
-  memcpy (tramp, trampoline, sizeof(trampoline));
-
-  if (cif->flags & AARCH64_FLAG_ARG_V)
-    start = ffi_closure_SYSV_V;
-  else
-    start = ffi_closure_SYSV;
-  *(UINT64 *)(tramp + 16) = (uintptr_t)start;
-
-  ffi_clear_cache(tramp, tramp + FFI_TRAMPOLINE_SIZE);
-
   return FFI_OK;
 }
 
diff --git a/libffi/src/aarch64/ffitarget.h b/libffi/src/aarch64/ffitarget.h
index 80d09af16812..fca2811ae214 100644
--- a/libffi/src/aarch64/ffitarget.h
+++ b/libffi/src/aarch64/ffitarget.h
@@ -42,7 +42,13 @@ typedef enum ffi_abi
 /* ---- Definitions for closures ----------------------------------------- */
 
 #define FFI_CLOSURES 1
+#if defined (__APPLE__)
+#define FFI_TRAMPOLINE_SIZE 20
+#define FFI_TRAMPOLINE_CLOSURE_OFFSET 16
+#else
 #define FFI_TRAMPOLINE_SIZE 24
+#define FFI_TRAMPOLINE_CLOSURE_OFFSET FFI_TRAMPOLINE_SIZE
+#endif
 #define FFI_NATIVE_RAW_API 0
 
 /* ---- Internal ---- */
diff --git a/libffi/src/aarch64/sysv.S b/libffi/src/aarch64/sysv.S
index 5c9cdda18c5e..46f50b9ef218 100644
--- a/libffi/src/aarch64/sysv.S
+++ b/libffi/src/aarch64/sysv.S
@@ -134,17 +134,17 @@ CNAME(ffi_call_SYSV):
 	ret
 7:	brk	#1000			/* UNUSED */
 	ret
-8:	st4	{ v0.s-v3.s }[0], [x3]	/* S4 */
+8:	st4	{ v0.s, v1.s, v2.s, v3.s }[0], [x3]	/* S4 */
 	ret
-9:	st3	{ v0.s-v2.s }[0], [x3]	/* S3 */
+9:	st3	{ v0.s, v1.s, v2.s }[0], [x3]	/* S3 */
 	ret
 10:	stp	s0, s1, [x3]		/* S2 */
 	ret
 11:	str	s0, [x3]		/* S1 */
 	ret
-12:	st4	{ v0.d-v3.d }[0], [x3]	/* D4 */
+12:	st4	{ v0.d, v1.d, v2.d, v3.d }[0], [x3]	/* D4 */
 	ret
-13:	st3	{ v0.d-v2.d }[0], [x3]	/* D3 */
+13:	st3	{ v0.d, v1.d, v2.d }[0], [x3]	/* D3 */
 	ret
 14:	stp	d0, d1, [x3]		/* D2 */
 	ret
@@ -248,8 +248,8 @@ CNAME(ffi_closure_SYSV):
 	stp     x6, x7, [sp, #16 + 16*N_V_ARG_REG + 48]
 
 	/* Load ffi_closure_inner arguments.  */
-	ldp	x0, x1, [x17, #FFI_TRAMPOLINE_SIZE]	/* load cif, fn */
-	ldr	x2, [x17, #FFI_TRAMPOLINE_SIZE+16]	/* load user_data */
+	ldp	x0, x1, [x17, #FFI_TRAMPOLINE_CLOSURE_OFFSET]	/* load cif, fn */
+	ldr	x2, [x17, #FFI_TRAMPOLINE_CLOSURE_OFFSET+16]	/* load user_data */
 .Ldo_closure:
 	add	x3, sp, #16				/* load context */
 	add	x4, sp, #ffi_closure_SYSV_FS		/* load stack */
@@ -343,6 +343,25 @@ CNAME(ffi_closure_SYSV):
 	.size	CNAME(ffi_closure_SYSV), . - CNAME(ffi_closure_SYSV)
 #endif
 
+#if FFI_EXEC_TRAMPOLINE_TABLE
+    .align 12
+CNAME(ffi_closure_trampoline_table_page):
+    .rept 16384 / FFI_TRAMPOLINE_SIZE
+    adr	x17, -16384
+    adr	x16, -16380
+    ldr x16, [x16]
+    ldr x17, [x17]
+    br	x16
+    .endr
+    
+    .globl CNAME(ffi_closure_trampoline_table_page)
+    #ifdef __ELF__
+    	.type	CNAME(ffi_closure_trampoline_table_page), #function
+    	.hidden	CNAME(ffi_closure_trampoline_table_page)
+    	.size	CNAME(ffi_closure_trampoline_table_page), . - CNAME(ffi_closure_trampoline_table_page)
+    #endif
+#endif
+
 #ifdef FFI_GO_CLOSURES
 	.align 4
 CNAME(ffi_go_closure_SYSV_V):
diff --git a/libffi/src/sparc/v8.S b/libffi/src/sparc/v8.S
index 3f483826c1ab..a2e4908fd417 100644
--- a/libffi/src/sparc/v8.S
+++ b/libffi/src/sparc/v8.S
@@ -28,7 +28,6 @@
 #define LIBFFI_ASM
 #include <fficonfig.h>
 #include <ffi.h>
-#include <ffi_cfi.h>
 #include "internal.h"
 
 #ifndef SPARC64
@@ -52,7 +51,6 @@
 	FFI_HIDDEN(C(ffi_flush_icache))
 
 C(ffi_flush_icache):
-	cfi_startproc
 1:	iflush %o0
 	iflush %o+8
 	nop
@@ -62,7 +60,6 @@ C(ffi_flush_icache):
 	nop
 	retl
 	 nop
-	cfi_endproc
 	.size	C(ffi_flush_icache), . - C(ffi_flush_icache)
 #endif
 
@@ -78,13 +75,10 @@ C(ffi_flush_icache):
 	FFI_HIDDEN(C(ffi_call_v8))
 
 C(ffi_call_v8):
-	cfi_startproc
+.LUW0:
 	! Allocate a stack frame sized by ffi_call.
 	save	%sp, %o4, %sp
-	cfi_def_cfa_register(%fp)
-	cfi_window_save
-	cfi_register(%o7, %i7)
-
+.LUW1:
 	mov	%i0, %o0		! copy cif
 	add	%sp, 64+32, %o1		! load args area
 	mov	%i2, %o2		! copy rvalue
@@ -233,7 +227,7 @@ E(SPARC_RET_F_1)
 	rept256; rept256; rept256; rept256
 	rept256; rept256; rept256; rept256
 
-	cfi_endproc
+.LUW2:
 	.size	C(ffi_call_v8),. - C(ffi_call_v8)
 
 
@@ -255,17 +249,14 @@ E(SPARC_RET_F_1)
 	FFI_HIDDEN(C(ffi_go_closure_v8))
 
 C(ffi_go_closure_v8):
-	cfi_startproc
+.LUW3:
 	save	%sp, -STACKFRAME, %sp
-	cfi_def_cfa_register(%fp)
-	cfi_window_save
-	cfi_register(%o7, %i7)
-
+.LUW4:
 	ld	[%g2+4], %o0			! load cif
 	ld	[%g2+8], %o1			! load fun
 	b	0f
 	 mov	%g2, %o2			! load user_data
-	cfi_endproc
+.LUW5:
 	.size	C(ffi_go_closure_v8), . - C(ffi_go_closure_v8)
 
 	.align 8
@@ -274,12 +265,9 @@ C(ffi_go_closure_v8):
 	FFI_HIDDEN(C(ffi_closure_v8))
 
 C(ffi_closure_v8):
-	cfi_startproc
+.LUW6:
 	save	%sp, -STACKFRAME, %sp
-	cfi_def_cfa_register(%fp)
-	cfi_window_save
-	cfi_register(%o7, %i7)
-
+.LUW7:
 	ld	[%g2+FFI_TRAMPOLINE_SIZE], %o0		! load cif
 	ld	[%g2+FFI_TRAMPOLINE_SIZE+4], %o1	! load fun
 	ld	[%g2+FFI_TRAMPOLINE_SIZE+8], %o2	! load user_data
@@ -375,8 +363,80 @@ E(SPARC_RET_F_1)
 	ret
 	 restore
 
-	cfi_endproc
+.LUW8:
 	.size	C(ffi_closure_v8), . - C(ffi_closure_v8)
+
+#ifdef HAVE_RO_EH_FRAME
+        .section        ".eh_frame",#alloc
+#else
+        .section        ".eh_frame",#alloc,#write
+#endif
+
+#ifdef HAVE_AS_SPARC_UA_PCREL
+# define FDE_ADDR(X)	%r_disp32(X)
+#else
+# define FDE_ADDR(X)	X
+#endif
+
+	.align 4
+.LCIE:
+	.long	.LECIE - .LSCIE		! CIE Length
+.LSCIE:
+	.long	0			! CIE Identifier Tag
+	.byte	1			! CIE Version
+	.ascii	"zR\0"			! CIE Augmentation
+	.byte	4			! CIE Code Alignment Factor
+	.byte	0x7c			! CIE Data Alignment Factor
+	.byte	15			! CIE RA Column
+	.byte	1			! Augmentation size
+#ifdef HAVE_AS_SPARC_UA_PCREL
+	.byte	0x1b			! FDE Encoding (pcrel sdata4)
+#else
+	.byte	0x50			! FDE Encoding (aligned absolute)
+#endif
+	.byte	0xc, 14, 0		! DW_CFA_def_cfa, %o6, offset 0
+	.align	4
+.LECIE:
+
+	.long	.LEFDE1 - .LSFDE1	! FDE Length
+.LSFDE1:
+	.long	.LSFDE1 - .LCIE		! FDE CIE offset
+	.long	FDE_ADDR(.LUW0)		! Initial location
+	.long	.LUW2 - .LUW0		! Address range
+	.byte	0			! Augmentation size
+	.byte	0x40+1			! DW_CFA_advance_loc 4
+	.byte	0xd, 30			! DW_CFA_def_cfa_register, %i6
+	.byte	0x2d			! DW_CFA_GNU_window_save
+	.byte	0x9, 15, 31		! DW_CFA_register, %o7, %i7
+	.align	4
+.LEFDE1:
+
+	.long	.LEFDE2 - .LSFDE2	! FDE Length
+.LSFDE2:
+	.long	.LSFDE2 - .LCIE		! FDE CIE offset
+	.long	FDE_ADDR(.LUW3)		! Initial location
+	.long	.LUW5 - .LUW3		! Address range
+	.byte	0			! Augmentation size
+	.byte	0x40+1			! DW_CFA_advance_loc 4
+	.byte	0xd, 30			! DW_CFA_def_cfa_register, %i6
+	.byte	0x2d			! DW_CFA_GNU_window_save
+	.byte	0x9, 15, 31		! DW_CFA_register, %o7, %i7
+	.align	4
+.LEFDE2:
+
+	.long	.LEFDE3 - .LSFDE3	! FDE Length
+.LSFDE3:
+	.long	.LSFDE3 - .LCIE		! FDE CIE offset
+	.long	FDE_ADDR(.LUW6)		! Initial location
+	.long	.LUW8 - .LUW6		! Address range
+	.byte	0			! Augmentation size
+	.byte	0x40+1			! DW_CFA_advance_loc 4
+	.byte	0xd, 30			! DW_CFA_def_cfa_register, %i6
+	.byte	0x2d			! DW_CFA_GNU_window_save
+	.byte	0x9, 15, 31		! DW_CFA_register, %o7, %i7
+	.align	4
+.LEFDE3:
+
 #endif /* !SPARC64 */
 #if defined __ELF__ && defined __linux__
 	.section	.note.GNU-stack,"",@progbits
diff --git a/libffi/src/sparc/v9.S b/libffi/src/sparc/v9.S
index 05ef54c3e158..55f8f4324cfe 100644
--- a/libffi/src/sparc/v9.S
+++ b/libffi/src/sparc/v9.S
@@ -27,7 +27,6 @@
 #define LIBFFI_ASM	
 #include <fficonfig.h>
 #include <ffi.h>
-#include <ffi_cfi.h>
 #include "internal.h"
 
 #ifdef SPARC64
@@ -57,12 +56,9 @@
 	FFI_HIDDEN(C(ffi_call_v9))
 
 C(ffi_call_v9):
-	cfi_startproc
+.LUW0:
 	save	%sp, %o4, %sp
-	cfi_def_cfa_register(%fp)
-	cfi_window_save
-	cfi_register(%o7, %i7)
-	
+.LUW1:
 	mov	%i0, %o0			! copy cif
 	add	%sp, STACK_BIAS+128+48, %o1	! load args area
 	mov	%i2, %o2			! copy rvalue
@@ -199,7 +195,7 @@ E(SPARC_RET_F_1)
 	return	%i7+8
 	 nop
 
-	cfi_endproc
+.LUW2:
 	.size	C(ffi_call_v9), . - C(ffi_call_v9)
 
 
@@ -219,18 +215,15 @@ E(SPARC_RET_F_1)
 	FFI_HIDDEN(C(ffi_go_closure_v9))
 
 C(ffi_go_closure_v9):
-	cfi_startproc
+.LUW3:
 	save	%sp, -STACKFRAME, %sp
-	cfi_def_cfa_register(%fp)
-	cfi_window_save
-	cfi_register(%o7, %i7)
-
+.LUW4:
 	ldx	[%g5+8], %o0
 	ldx	[%g5+16], %o1
 	b	0f
 	 mov	%g5, %o2
 
-	cfi_endproc
+.LUW5:
 	.size	C(ffi_go_closure_v9), . - C(ffi_go_closure_v9)
 
 	.align 8
@@ -239,12 +232,9 @@ C(ffi_go_closure_v9):
 	FFI_HIDDEN(C(ffi_closure_v9))
 
 C(ffi_closure_v9):
-	cfi_startproc
+.LUW6:
 	save	%sp, -STACKFRAME, %sp
-	cfi_def_cfa_register(%fp)
-	cfi_window_save
-	cfi_register(%o7, %i7)
-
+.LUW7:
 	ldx	[%g1+FFI_TRAMPOLINE_SIZE], %o0
 	ldx	[%g1+FFI_TRAMPOLINE_SIZE+8], %o1
 	ldx	[%g1+FFI_TRAMPOLINE_SIZE+16], %o2
@@ -373,8 +363,77 @@ E(SPARC_RET_F_1)
 	return	%i7+8
 	 nop
 
-	cfi_endproc
+.LUW8:
 	.size	C(ffi_closure_v9), . - C(ffi_closure_v9)
+
+#ifdef HAVE_RO_EH_FRAME
+        .section        ".eh_frame",#alloc
+#else
+        .section        ".eh_frame",#alloc,#write
+#endif
+
+#ifdef HAVE_AS_SPARC_UA_PCREL
+# define FDE_RANGE(B, E)  .long %r_disp32(B), E - B
+#else
+# define FDE_RANGE(B, E)  .align 8; .xword B, E - B
+#endif
+
+	.align 8
+.LCIE:
+	.long	.LECIE - .LSCIE		! CIE Length
+.LSCIE:
+	.long	0			! CIE Identifier Tag
+	.byte	1			! CIE Version
+	.ascii	"zR\0"			! CIE Augmentation
+	.byte	4			! CIE Code Alignment Factor
+	.byte	0x78			! CIE Data Alignment Factor
+	.byte	15			! CIE RA Column
+	.byte	1			! Augmentation size
+#ifdef HAVE_AS_SPARC_UA_PCREL
+	.byte	0x1b			! FDE Encoding (pcrel sdata4)
+#else
+	.byte	0x50			! FDE Encoding (aligned absolute)
+#endif
+	.byte	0xc, 14, 0xff, 0xf	! DW_CFA_def_cfa, %o6, offset 0x7ff
+	.align	8
+.LECIE:
+
+	.long	.LEFDE1 - .LSFDE1	! FDE Length
+.LSFDE1:
+	.long	.LSFDE1 - .LCIE		! FDE CIE offset
+	FDE_RANGE(.LUW0, .LUW2)
+	.byte	0			! Augmentation size
+	.byte	0x40+1			! DW_CFA_advance_loc 4
+	.byte	0xd, 30			! DW_CFA_def_cfa_register, %i6
+	.byte	0x2d			! DW_CFA_GNU_window_save
+	.byte	0x9, 15, 31		! DW_CFA_register, %o7, %i7
+	.align	8
+.LEFDE1:
+
+	.long	.LEFDE2 - .LSFDE2	! FDE Length
+.LSFDE2:
+	.long	.LSFDE2 - .LCIE		! FDE CIE offset
+	FDE_RANGE(.LUW3, .LUW5)
+	.byte	0			! Augmentation size
+	.byte	0x40+1			! DW_CFA_advance_loc 4
+	.byte	0xd, 30			! DW_CFA_def_cfa_register, %i6
+	.byte	0x2d			! DW_CFA_GNU_window_save
+	.byte	0x9, 15, 31		! DW_CFA_register, %o7, %i7
+	.align	8
+.LEFDE2:
+
+	.long	.LEFDE3 - .LSFDE3	! FDE Length
+.LSFDE3:
+	.long	.LSFDE3 - .LCIE		! FDE CIE offset
+	FDE_RANGE(.LUW6, .LUW8)
+	.byte	0			! Augmentation size
+	.byte	0x40+1			! DW_CFA_advance_loc 4
+	.byte	0xd, 30			! DW_CFA_def_cfa_register, %i6
+	.byte	0x2d			! DW_CFA_GNU_window_save
+	.byte	0x9, 15, 31		! DW_CFA_register, %o7, %i7
+	.align	8
+.LEFDE3:
+
 #endif /* SPARC64 */
 #ifdef __linux__
 	.section	.note.GNU-stack,"",@progbits