Merge changes between r11842 and r11980 from /fsf/trunk.

git-svn-id: svn://svn.eglibc.org/trunk@11981 7b3dc134-2b1b-0410-93df-9e9f96275f8d

29 files changed, 2259 insertions, 134 deletions

diff --git a/libc/ChangeLog b/libc/ChangeLog
index e162666af..ffa4f6836 100644
--- a/libc/ChangeLog
+++ b/libc/ChangeLog
@@ -1,3 +1,110 @@
+2010-11-04  Luis Machado  <luisgpm@br.ibm.com>
+
+	* sysdeps/powerpc/powerpc32/power7/mempcpy.S: New file.
+	* sysdeps/powerpc/powerpc64/power7/mempcpy.S: New file.
+
+2010-11-03  H.J. Lu  <hongjiu.lu@intel.com>
+
+	[BZ #12191]
+	* sysdeps/i386/i686/cacheinfo.c (__x86_64_raw_data_cache_size): New.
+	(__x86_64_raw_data_cache_size_half): Likewise.
+	(__x86_64_raw_shared_cache_size): Likewise.
+	(__x86_64_raw_shared_cache_size_half): Likewise.
+
+	* sysdeps/x86_64/cacheinfo.c (__x86_64_raw_data_cache_size): New.
+	(__x86_64_raw_data_cache_size_half): Likewise.
+	(__x86_64_raw_shared_cache_size): Likewise.
+	(__x86_64_raw_shared_cache_size_half): Likewise.
+	(init_cacheinfo): Set __x86_64_raw_data_cache_size,
+	__x86_64_raw_data_cache_size_half, __x86_64_raw_shared_cache_size
+	and __x86_64_raw_shared_cache_size_half.  Round
+	__x86_64_data_cache_size_half, __x86_64_data_cache_size
+	__x86_64_shared_cache_size_half and __x86_64_shared_cache_size,
+	to multiple of 256 bytes.
+
+2010-11-03  Ulrich Drepper  <drepper@gmail.com>
+
+	[BZ #12167]
+	* sysdeps/unix/sysv/linux/ttyname.c (ttyname): Recognize new mangling
+	of inacessible symlinks.  Verify result of symlink before returning it.
+	* sysdeps/unix/sysv/linux/ttyname_r.c (__ttyname_r): Likewise.
+	Patch mostly by Miklos Szeredi <miklos@szeredi.hu>.
+
+2010-10-28  Erich Ritz  <erichritz@gmail.com>
+
+	* math/math.h (isinf): Fix typo in comment.
+
+2010-11-01  Ulrich Drepper  <drepper@gmail.com>
+
+	* po/da.po: Update from translation team.
+
+2010-10-26  Ulrich Drepper  <drepper@gmail.com>
+
+	* elf/rtld.c (dl_main): Move assertion after the point where rtld map
+	is added to the list.
+
+2010-10-20  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+	    Ulrich Drepper  <drepper@gmail.com>
+
+	* elf/dl-object.c (_dl_new_object): Don't append the new object to
+	the global list here.  Move code to...
+	(_dl_add_to_namespace_list): ...here.  New function.
+	* elf/rtld.c (dl_main): Invoke _dl_add_to_namespace_list.
+	* sysdeps/generic/ldsodefs.h (_dl_add_to_namespace_list): Declare.
+	* elf/dl-load.c (lose): Don't remove the element from the list.
+	(_dl_map_object_from_fd): Invoke _dl_add_to_namespace_list.
+	(_dl_map_object): Likewise.
+
+2010-10-25  Ulrich Drepper  <drepper@gmail.com>
+
+	[BZ #12159]
+	* sysdeps/x86_64/multiarch/strchr.S: Fix propagation of search byte
+	into all bytes of SSE register.
+	Patch by Richard Li <richardpku@gmail.com>.
+
+2010-10-24  Ulrich Drepper  <drepper@gmail.com>
+
+	[BZ #12140]
+	* malloc/malloc.c (_int_free): Fill correct number of bytes when
+	perturbing.
+
+2010-10-20  Michael B. Brutman  <brutman@us.ibm.com>
+
+	* sysdeps/powerpc/dl-procinfo.c: Add support for ppca2 platform
+	* sysdeps/powerpc/dl-procinfo.h: Add support for ppca2 platform
+	* sysdeps/powerpc/powerpc32/a2/memcpy.S: New file.
+	* sysdeps/powerpc/powerpc64/a2/memcpy.S: Likewise.
+	* sysdeps/unix/sysv/linux/powerpc/powerpc32/a2/Implies: New
+	submachine.
+	* sysdeps/unix/sysv/linux/powerpc/powerpc64/a2/Implies: Likewise.
+
+2010-10-22  Andreas Schwab  <schwab@redhat.com>
+
+	* include/dlfcn.h (__RTLD_SECURE): Define.
+	* elf/dl-load.c (_dl_map_object): Remove preloaded parameter.  Use
+	mode & __RTLD_SECURE instead.
+	(open_path): Rename preloaded parameter to secure.
+	* sysdeps/generic/ldsodefs.h (_dl_map_object): Adjust declaration.
+	* elf/dl-open.c (dl_open_worker): Adjust call to _dl_map_object.
+	* elf/dl-deps.c (openaux): Likewise.
+	* elf/rtld.c (struct map_args): Remove is_preloaded.
+	(map_doit): Don't use it.
+	(dl_main): Likewise.
+	(do_preload): Use __RTLD_SECURE instead of is_preloaded.
+	(dlmopen_doit): Add __RTLD_SECURE to mode bits.
+
+2010-09-09  Andreas Schwab  <schwab@redhat.com>
+
+	* Makeconfig (sysd-rules-patterns): Add rtld-%:rtld-%.
+	(sysd-rules-targets): Remove duplicates.
+	* elf/rtld-Rules ($(objpfx)rtld-%.os): Add pattern rules with
+	rtld-%.$o dependency.
+
+2010-10-18  Andreas Schwab  <schwab@redhat.com>
+
+	* elf/dl-open.c (dl_open_worker): Don't expand DST here, let
+	_dl_map_object do it.
+
 2010-10-19  Ulrich Drepper  <drepper@gmail.com>
 
 	* sysdeps/i386/bits/mathdef.h (FP_FAST_FMA): If the GCC 4.6 port has
diff --git a/libc/Makeconfig b/libc/Makeconfig
index 9db1ee466..c9fe1991a 100644
--- a/libc/Makeconfig
+++ b/libc/Makeconfig
@@ -703,7 +703,7 @@ CPPFLAGS = $($(subdir)-CPPFLAGS) $(+includes) $(defines) \
 	   -include $(..)include/libc-symbols.h $(sysdep-CPPFLAGS) \
 	   $(CPPFLAGS-$(suffix $@)) \
 	   $(foreach lib,$(libof-$(basename $(@F))) \
-		         $(libof-$(<F)) $(libof-$(@F)),$(CPPFLAGS-$(lib))) \
+			 $(libof-$(<F)) $(libof-$(@F)),$(CPPFLAGS-$(lib))) \
 	   $(CPPFLAGS-$(<F)) $(CPPFLAGS-$(@F)) $(CPPFLAGS-$(basename $(@F)))
 override CFLAGS	= -std=gnu99 $(gnu89-inline-CFLAGS) \
 		  $(filter-out %frame-pointer,$(+cflags)) $(+gccwarn-c) \
@@ -992,7 +992,7 @@ endif
 # emitted into sysd-rules.  A sysdeps Makeconfig fragment can
 # add its own special object file prefix to this list with e.g. foo-%:%
 # to have foo-*.? compiled from *.? using $(foo-CPPFLAGS).
-sysd-rules-patterns := %:% rtld-%:% m_%:s_%
+sysd-rules-patterns := %:% rtld-%:rtld-% rtld-%:% m_%:s_%
 
 # Let sysdeps/ subdirs contain a Makeconfig fragment for us to include here.
 sysdep-makeconfigs := $(wildcard $(+sysdep_dirs:=/Makeconfig))
@@ -1001,8 +1001,8 @@ include $(sysdep-makeconfigs)
 endif
 
 # Compute just the target patterns.  Makeconfig has set sysd-rules-patterns.
-sysd-rules-targets := $(foreach p,$(sysd-rules-patterns),\
-		      		$(firstword $(subst :, ,$p)))
+sysd-rules-targets := $(sort $(foreach p,$(sysd-rules-patterns),\
+					 $(firstword $(subst :, ,$p))))
 
 endif # Makeconfig not yet included
 
diff --git a/libc/NEWS b/libc/NEWS
index 48efcbd52..cbe6e511b 100644
--- a/libc/NEWS
+++ b/libc/NEWS
@@ -1,4 +1,4 @@
-GNU C Library NEWS -- history of user-visible changes.  2010-10-13
+GNU C Library NEWS -- history of user-visible changes.  2010-11-5
 Copyright (C) 1992-2009, 2010 Free Software Foundation, Inc.
 See the end for copying conditions.
 
@@ -11,7 +11,7 @@ Version 2.13
 
   3268, 7066, 10851, 11611, 11640, 11701, 11840, 11856, 11883, 11903, 11904,
   11968, 11979, 12005, 12037, 12067, 12077, 12078, 12092, 12093, 12107, 12108,
-  12113
+  12113, 12140, 12159, 12167, 12191
 
 * New Linux interfaces: prlimit, prlimit64, fanotify_init, fanotify_mark
 
diff --git a/libc/elf/dl-deps.c b/libc/elf/dl-deps.c
index 92840e43f..e4dea1aa1 100644
--- a/libc/elf/dl-deps.c
+++ b/libc/elf/dl-deps.c
@@ -62,7 +62,7 @@ openaux (void *a)
 {
   struct openaux_args *args = (struct openaux_args *) a;
 
-  args->aux = _dl_map_object (args->map, args->name, 0,
+  args->aux = _dl_map_object (args->map, args->name,
 			      (args->map->l_type == lt_executable
 			       ? lt_library : args->map->l_type),
 			      args->trace_mode, args->open_mode,
diff --git a/libc/elf/dl-load.c b/libc/elf/dl-load.c
index 0345efb93..697c1f171 100644
--- a/libc/elf/dl-load.c
+++ b/libc/elf/dl-load.c
@@ -798,22 +798,7 @@ lose (int code, int fd, const char *name, char *realname, struct link_map *l,
   /* The file might already be closed.  */
   if (fd != -1)
     (void) __close (fd);
-  if (l != NULL)
-    {
-      /* We modify the list of loaded objects.  */
-      __rtld_lock_lock_recursive (GL(dl_load_write_lock));
-      /* Remove the stillborn object from the list and free it.  */
-      assert (l->l_next == NULL);
-      if (l->l_prev == NULL)
-	/* No other module loaded. This happens only in the static library,
-	   or in rtld under --verify.  */
-	GL(dl_ns)[l->l_ns]._ns_loaded = NULL;
-      else
-	l->l_prev->l_next = NULL;
-      --GL(dl_ns)[l->l_ns]._ns_nloaded;
-      free (l);
-      __rtld_lock_unlock_recursive (GL(dl_load_write_lock));
-    }
+  free (l);
   free (realname);
 
   if (r != NULL)
@@ -898,6 +883,9 @@ _dl_map_object_from_fd (const char *name, int fd, struct filebuf *fbp,
 	 never be unloaded.  */
       __close (fd);
 
+      /* Add the map for the mirrored object to the object list.  */
+      _dl_add_to_namespace_list (l, nsid);
+
       return l;
     }
 #endif
@@ -1492,6 +1480,9 @@ cannot enable executable stack as shared object requires");
     add_name_to_object (l, ((const char *) D_PTR (l, l_info[DT_STRTAB])
 			    + l->l_info[DT_SONAME]->d_un.d_val));
 
+  /* Now that the object is fully initialized add it to the object list.  */
+  _dl_add_to_namespace_list (l, nsid);
+
 #ifdef SHARED
   /* Auditing checkpoint: we have a new object.  */
   if (__builtin_expect (GLRO(dl_naudit) > 0, 0)
@@ -1812,7 +1803,7 @@ open_verify (const char *name, struct filebuf *fbp, struct link_map *loader,
    if MAY_FREE_DIRS is true.  */
 
 static int
-open_path (const char *name, size_t namelen, int preloaded,
+open_path (const char *name, size_t namelen, int secure,
 	   struct r_search_path_struct *sps, char **realname,
 	   struct filebuf *fbp, struct link_map *loader, int whatcode,
 	   bool *found_other_class)
@@ -1894,7 +1885,7 @@ open_path (const char *name, size_t namelen, int preloaded,
 	  /* Remember whether we found any existing directory.  */
 	  here_any |= this_dir->status[cnt] != nonexisting;
 
-	  if (fd != -1 && __builtin_expect (preloaded, 0)
+	  if (fd != -1 && __builtin_expect (secure, 0)
 	      && INTUSE(__libc_enable_secure))
 	    {
 	      /* This is an extra security effort to make sure nobody can
@@ -1963,7 +1954,7 @@ open_path (const char *name, size_t namelen, int preloaded,
 
 struct link_map *
 internal_function
-_dl_map_object (struct link_map *loader, const char *name, int preloaded,
+_dl_map_object (struct link_map *loader, const char *name,
 		int type, int trace_mode, int mode, Lmid_t nsid)
 {
   int fd;
@@ -2067,7 +2058,8 @@ _dl_map_object (struct link_map *loader, const char *name, int preloaded,
 	  for (l = loader; l; l = l->l_loader)
 	    if (cache_rpath (l, &l->l_rpath_dirs, DT_RPATH, "RPATH"))
 	      {
-		fd = open_path (name, namelen, preloaded, &l->l_rpath_dirs,
+		fd = open_path (name, namelen, mode & __RTLD_SECURE,
+				&l->l_rpath_dirs,
 				&realname, &fb, loader, LA_SER_RUNPATH,
 				&found_other_class);
 		if (fd != -1)
@@ -2082,14 +2074,15 @@ _dl_map_object (struct link_map *loader, const char *name, int preloaded,
 	      && main_map != NULL && main_map->l_type != lt_loaded
 	      && cache_rpath (main_map, &main_map->l_rpath_dirs, DT_RPATH,
 			      "RPATH"))
-	    fd = open_path (name, namelen, preloaded, &main_map->l_rpath_dirs,
+	    fd = open_path (name, namelen, mode & __RTLD_SECURE,
+			    &main_map->l_rpath_dirs,
 			    &realname, &fb, loader ?: main_map, LA_SER_RUNPATH,
 			    &found_other_class);
 	}
 
       /* Try the LD_LIBRARY_PATH environment variable.  */
       if (fd == -1 && env_path_list.dirs != (void *) -1)
-	fd = open_path (name, namelen, preloaded, &env_path_list,
+	fd = open_path (name, namelen, mode & __RTLD_SECURE, &env_path_list,
 			&realname, &fb,
 			loader ?: GL(dl_ns)[LM_ID_BASE]._ns_loaded,
 			LA_SER_LIBPATH, &found_other_class);
@@ -2098,12 +2091,12 @@ _dl_map_object (struct link_map *loader, const char *name, int preloaded,
       if (fd == -1 && loader != NULL
 	  && cache_rpath (loader, &loader->l_runpath_dirs,
 			  DT_RUNPATH, "RUNPATH"))
-	fd = open_path (name, namelen, preloaded,
+	fd = open_path (name, namelen, mode & __RTLD_SECURE,
 			&loader->l_runpath_dirs, &realname, &fb, loader,
 			LA_SER_RUNPATH, &found_other_class);
 
       if (fd == -1
-	  && (__builtin_expect (! preloaded, 1)
+	  && (__builtin_expect (! (mode & __RTLD_SECURE), 1)
 	      || ! INTUSE(__libc_enable_secure)))
 	{
 	  /* Check the list of libraries in the file /etc/ld.so.cache,
@@ -2169,7 +2162,7 @@ _dl_map_object (struct link_map *loader, const char *name, int preloaded,
 	  && ((l = loader ?: GL(dl_ns)[nsid]._ns_loaded) == NULL
 	      || __builtin_expect (!(l->l_flags_1 & DF_1_NODEFLIB), 1))
 	  && rtld_search_dirs.dirs != (void *) -1)
-	fd = open_path (name, namelen, preloaded, &rtld_search_dirs,
+	fd = open_path (name, namelen, mode & __RTLD_SECURE, &rtld_search_dirs,
 			&realname, &fb, l, LA_SER_DEFAULT, &found_other_class);
 
       /* Add another newline when we are tracing the library loading.  */
@@ -2214,7 +2207,7 @@ _dl_map_object (struct link_map *loader, const char *name, int preloaded,
 	     have.  */
 	  static const Elf_Symndx dummy_bucket = STN_UNDEF;
 
-	  /* Enter the new object in the list of loaded objects.  */
+	  /* Allocate a new object map.  */
 	  if ((name_copy = local_strdup (name)) == NULL
 	      || (l = _dl_new_object (name_copy, name, type, loader,
 				      mode, nsid)) == NULL)
@@ -2232,6 +2225,9 @@ _dl_map_object (struct link_map *loader, const char *name, int preloaded,
 	  l->l_nbuckets = 1;
 	  l->l_relocated = 1;
 
+	  /* Enter the object in the object list.  */
+	  _dl_add_to_namespace_list (l, nsid);
+
 	  return l;
 	}
       else if (found_other_class)
diff --git a/libc/elf/dl-object.c b/libc/elf/dl-object.c
index fb20b308f..7a3602931 100644
--- a/libc/elf/dl-object.c
+++ b/libc/elf/dl-object.c
@@ -1,5 +1,5 @@
 /* Storage management for the chain of loaded shared objects.
-   Copyright (C) 1995-2002,2004,2006-2008,2009 Free Software Foundation, Inc.
+   Copyright (C) 1995-2002,2004,2006-2009,2010 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -26,16 +26,41 @@
 #include <assert.h>
 
 
+/* Add the new link_map NEW to the end of the namespace list.  */
+void
+internal_function
+_dl_add_to_namespace_list (struct link_map *new, Lmid_t nsid)
+{
+  /* We modify the list of loaded objects.  */
+  __rtld_lock_lock_recursive (GL(dl_load_write_lock));
+
+  if (GL(dl_ns)[nsid]._ns_loaded != NULL)
+    {
+      struct link_map *l = GL(dl_ns)[nsid]._ns_loaded;
+      while (l->l_next != NULL)
+	l = l->l_next;
+      new->l_prev = l;
+      /* new->l_next = NULL;   Would be necessary but we use calloc.  */
+      l->l_next = new;
+    }
+  else
+    GL(dl_ns)[nsid]._ns_loaded = new;
+  ++GL(dl_ns)[nsid]._ns_nloaded;
+  new->l_serial = GL(dl_load_adds);
+  ++GL(dl_load_adds);
+
+  __rtld_lock_unlock_recursive (GL(dl_load_write_lock));
+}
+
+
 /* Allocate a `struct link_map' for a new object being loaded,
    and enter it into the _dl_loaded list.  */
-
 struct link_map *
 internal_function
 _dl_new_object (char *realname, const char *libname, int type,
 		struct link_map *loader, int mode, Lmid_t nsid)
 {
   struct link_map *l;
-  int idx;
   size_t libname_len = strlen (libname) + 1;
   struct link_map *new;
   struct libname_list *newname;
@@ -93,31 +118,12 @@ _dl_new_object (char *realname, const char *libname, int type,
   new->l_scope = new->l_scope_mem;
   new->l_scope_max = sizeof (new->l_scope_mem) / sizeof (new->l_scope_mem[0]);
 
-  /* We modify the list of loaded objects.  */
-  __rtld_lock_lock_recursive (GL(dl_load_write_lock));
-
   /* Counter for the scopes we have to handle.  */
-  idx = 0;
+  int idx = 0;
 
   if (GL(dl_ns)[nsid]._ns_loaded != NULL)
-    {
-      l = GL(dl_ns)[nsid]._ns_loaded;
-      while (l->l_next != NULL)
-	l = l->l_next;
-      new->l_prev = l;
-      /* new->l_next = NULL;	Would be necessary but we use calloc.  */
-      l->l_next = new;
-
-      /* Add the global scope.  */
-      new->l_scope[idx++] = &GL(dl_ns)[nsid]._ns_loaded->l_searchlist;
-    }
-  else
-    GL(dl_ns)[nsid]._ns_loaded = new;
-  ++GL(dl_ns)[nsid]._ns_nloaded;
-  new->l_serial = GL(dl_load_adds);
-  ++GL(dl_load_adds);
-
-  __rtld_lock_unlock_recursive (GL(dl_load_write_lock));
+    /* Add the global scope.  */
+    new->l_scope[idx++] = &GL(dl_ns)[nsid]._ns_loaded->l_searchlist;
 
   /* If we have no loader the new object acts as it.  */
   if (loader == NULL)
diff --git a/libc/elf/dl-open.c b/libc/elf/dl-open.c
index 610ceaedb..ee09f0ec5 100644
--- a/libc/elf/dl-open.c
+++ b/libc/elf/dl-open.c
@@ -221,38 +221,9 @@ dl_open_worker (void *a)
 
   assert (_dl_debug_initialize (0, args->nsid)->r_state == RT_CONSISTENT);
 
-  /* Maybe we have to expand a DST.  */
-  if (__builtin_expect (dst != NULL, 0))
-    {
-      size_t len = strlen (file);
-
-      /* Determine how much space we need.  We have to allocate the
-	 memory locally.  */
-      size_t required = DL_DST_REQUIRED (call_map, file, len,
-					 _dl_dst_count (dst, 0));
-
-      /* Get space for the new file name.  */
-      char *new_file = (char *) alloca (required + 1);
-
-      /* Generate the new file name.  */
-      _dl_dst_substitute (call_map, file, new_file, 0);
-
-      /* If the substitution failed don't try to load.  */
-      if (*new_file == '\0')
-	_dl_signal_error (0, "dlopen", NULL,
-			  N_("empty dynamic string token substitution"));
-
-      /* Now we have a new file name.  */
-      file = new_file;
-
-      /* It does not matter whether call_map is set even if we
-	 computed it only because of the DST.  Since the path contains
-	 a slash the value is not used.  See dl-load.c.  */
-    }
-
   /* Load the named object.  */
   struct link_map *new;
-  args->map = new = _dl_map_object (call_map, file, 0, lt_loaded, 0,
+  args->map = new = _dl_map_object (call_map, file, lt_loaded, 0,
 				    mode | __RTLD_CALLMAP, args->nsid);
 
   /* If the pointer returned is NULL this means the RTLD_NOLOAD flag is
diff --git a/libc/elf/rtld-Rules b/libc/elf/rtld-Rules
index 9f31a560e..10c9452ce 100644
--- a/libc/elf/rtld-Rules
+++ b/libc/elf/rtld-Rules
@@ -1,6 +1,6 @@
 # Subroutine makefile for compiling libc modules linked into dynamic linker.
 
-# Copyright (C) 2002, 2003, 2005, 2006, 2008 Free Software Foundation, Inc.
+# Copyright (C) 2002,2003,2005,2006,2008,2010 Free Software Foundation, Inc.
 # This file is part of the GNU C Library.
 
 # The GNU C Library is free software; you can redistribute it and/or
@@ -56,7 +56,7 @@ $(objpfx)rtld-libc.a: $(foreach dir,$(rtld-subdirs),\
 # Use the verbose option of ar and tar when not running silently.
 ifeq	"$(findstring s,$(MAKEFLAGS))" ""	# if not -s
 verbose := v
-else	   					# -s
+else						# -s
 verbose	:=
 endif						# not -s
 
@@ -93,6 +93,12 @@ else
 # These are the basic compilation rules corresponding to the Makerules ones.
 # The sysd-rules generated makefile already defines pattern rules for rtld-%
 # targets built from sysdeps source files.
+$(objpfx)rtld-%.os: rtld-%.S $(before-compile)
+	$(compile-command.S) $(rtld-CPPFLAGS)
+$(objpfx)rtld-%.os: rtld-%.s $(before-compile)
+	$(compile-command.s) $(rtld-CPPFLAGS)
+$(objpfx)rtld-%.os: rtld-%.c $(before-compile)
+	$(compile-command.c) $(rtld-CPPFLAGS)
 $(objpfx)rtld-%.os: %.S $(before-compile)
 	$(compile-command.S) $(rtld-CPPFLAGS)
 $(objpfx)rtld-%.os: %.s $(before-compile)
@@ -101,6 +107,9 @@ $(objpfx)rtld-%.os: %.c $(before-compile)
 	$(compile-command.c) $(rtld-CPPFLAGS)
 
 # The rules for generated source files.
+$(objpfx)rtld-%.os: $(objpfx)rtld-%.S $(before-compile); $(compile-command.S)
+$(objpfx)rtld-%.os: $(objpfx)rtld-%.s $(before-compile); $(compile-command.s)
+$(objpfx)rtld-%.os: $(objpfx)rtld-%.c $(before-compile); $(compile-command.c)
 $(objpfx)rtld-%.os: $(objpfx)%.S $(before-compile); $(compile-command.S)
 $(objpfx)rtld-%.os: $(objpfx)%.s $(before-compile); $(compile-command.s)
 $(objpfx)rtld-%.os: $(objpfx)%.c $(before-compile); $(compile-command.c)
diff --git a/libc/elf/rtld.c b/libc/elf/rtld.c
index 4ca07f8ae..c714be101 100644
--- a/libc/elf/rtld.c
+++ b/libc/elf/rtld.c
@@ -589,7 +589,6 @@ struct map_args
   /* Argument to map_doit.  */
   char *str;
   struct link_map *loader;
-  int is_preloaded;
   int mode;
   /* Return value of map_doit.  */
   struct link_map *map;
@@ -627,16 +626,17 @@ static void
 map_doit (void *a)
 {
   struct map_args *args = (struct map_args *) a;
-  args->map = _dl_map_object (args->loader, args->str,
-			      args->is_preloaded, lt_library, 0, args->mode,
-			      LM_ID_BASE);
+  args->map = _dl_map_object (args->loader, args->str, lt_library, 0,
+			      args->mode, LM_ID_BASE);
 }
 
 static void
 dlmopen_doit (void *a)
 {
   struct dlmopen_args *args = (struct dlmopen_args *) a;
-  args->map = _dl_open (args->fname, RTLD_LAZY | __RTLD_DLOPEN | __RTLD_AUDIT,
+  args->map = _dl_open (args->fname,
+			(RTLD_LAZY | __RTLD_DLOPEN | __RTLD_AUDIT
+			 | __RTLD_SECURE),
 			dl_main, LM_ID_NEWLM, _dl_argc, INTUSE(_dl_argv),
 			__environ);
 }
@@ -806,8 +806,7 @@ do_preload (char *fname, struct link_map *main_map, const char *where)
 
   args.str = fname;
   args.loader = main_map;
-  args.is_preloaded = 1;
-  args.mode = 0;
+  args.mode = __RTLD_SECURE;
 
   unsigned int old_nloaded = GL(dl_ns)[LM_ID_BASE]._ns_nloaded;
 
@@ -1054,7 +1053,6 @@ of this helper program; chances are you did not intend to run this program.\n\
 
 	  args.str = rtld_progname;
 	  args.loader = NULL;
-	  args.is_preloaded = 0;
 	  args.mode = __RTLD_OPENEXEC;
 	  (void) _dl_catch_error (&objname, &err_str, &malloced, map_doit,
 				  &args);
@@ -1066,7 +1064,7 @@ of this helper program; chances are you did not intend to run this program.\n\
       else
 	{
 	  HP_TIMING_NOW (start);
-	  _dl_map_object (NULL, rtld_progname, 0, lt_library, 0,
+	  _dl_map_object (NULL, rtld_progname, lt_library, 0,
 			  __RTLD_OPENEXEC, LM_ID_BASE);
 	  HP_TIMING_NOW (stop);
 
@@ -1110,11 +1108,15 @@ of this helper program; chances are you did not intend to run this program.\n\
       main_map = _dl_new_object ((char *) "", "", lt_executable, NULL,
 				 __RTLD_OPENEXEC, LM_ID_BASE);
       assert (main_map != NULL);
-      assert (main_map == GL(dl_ns)[LM_ID_BASE]._ns_loaded);
       main_map->l_phdr = phdr;
       main_map->l_phnum = phnum;
       main_map->l_entry = *user_entry;
 
+      /* Even though the link map is not yet fully initialized we can add
+	 it to the map list since there are no possible users running yet.  */
+      _dl_add_to_namespace_list (main_map, LM_ID_BASE);
+      assert (main_map == GL(dl_ns)[LM_ID_BASE]._ns_loaded);
+
       /* At this point we are in a bit of trouble.  We would have to
 	 fill in the values for l_dev and l_ino.  But in general we
 	 do not know where the file is.  We also do not handle AT_EXECFD
@@ -1257,7 +1259,7 @@ of this helper program; chances are you did not intend to run this program.\n\
       /* We were invoked directly, so the program might not have a
 	 PT_INTERP.  */
       _dl_rtld_libname.name = GL(dl_rtld_map).l_name;
-      /* _dl_rtld_libname.next = NULL; 	Already zero.  */
+      /* _dl_rtld_libname.next = NULL;	Already zero.  */
       GL(dl_rtld_map).l_libname =  &_dl_rtld_libname;
     }
   else
@@ -1382,6 +1384,9 @@ of this helper program; chances are you did not intend to run this program.\n\
 	      l->l_libname->name = memcpy (copy, dsoname, len);
 	    }
 
+	  /* Add the vDSO to the object list.  */
+	  _dl_add_to_namespace_list (l, LM_ID_BASE);
+
 	  /* Rearrange the list so this DSO appears after rtld_map.  */
 	  assert (l->l_next == NULL);
 	  assert (l->l_prev == main_map);
diff --git a/libc/include/dlfcn.h b/libc/include/dlfcn.h
index a67426df3..af92483f5 100644
--- a/libc/include/dlfcn.h
+++ b/libc/include/dlfcn.h
@@ -9,6 +9,7 @@
 #define __RTLD_OPENEXEC	0x20000000
 #define __RTLD_CALLMAP	0x10000000
 #define __RTLD_AUDIT	0x08000000
+#define __RTLD_SECURE	0x04000000 /* Apply additional security checks.  */
 
 #define __LM_ID_CALLER	-2
 
diff --git a/libc/malloc/malloc.c b/libc/malloc/malloc.c
index 53ee1ccb8..f7770a394 100644
--- a/libc/malloc/malloc.c
+++ b/libc/malloc/malloc.c
@@ -4850,7 +4850,7 @@ _int_free(mstate av, mchunkptr p)
       }
 
     if (__builtin_expect (perturb_byte, 0))
-      free_perturb (chunk2mem(p), size - SIZE_SZ);
+      free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
 
     set_fastchunks(av);
     unsigned int idx = fastbin_index(size);
@@ -4954,7 +4954,7 @@ _int_free(mstate av, mchunkptr p)
       }
 
     if (__builtin_expect (perturb_byte, 0))
-      free_perturb (chunk2mem(p), size - SIZE_SZ);
+      free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
 
     /* consolidate backward */
     if (!prev_inuse(p)) {
diff --git a/libc/math/math.h b/libc/math/math.h
index 3abf7ae9a..272cea719 100644
--- a/libc/math/math.h
+++ b/libc/math/math.h
@@ -265,7 +265,7 @@ enum
       ? __isnan (x) : __isnanl (x))
 # endif
 
-/* Return nonzero value is X is positive or negative infinity.  */
+/* Return nonzero value if X is positive or negative infinity.  */
 # ifdef __NO_LONG_DOUBLE_MATH
 #  define isinf(x) \
      (sizeof (x) == sizeof (float) ? __isinff (x) : __isinf (x))
diff --git a/libc/po/da.po b/libc/po/da.po
index e823bfcc6..2923cf117 100644
--- a/libc/po/da.po
+++ b/libc/po/da.po
@@ -1,15 +1,14 @@
-# translation of libc-2.9.90.po to Danish
 # Danish messages for GNU libc.
 # Copyright (C) 1996, 2009 Free Software Foundation, Inc.
 # This file is distributed under the same license as the glibc package.
 #
-# Keld Simonsen <keld@dkuug.dk>, 2000-2002, 2009.
+# Keld Simonsen <keld@keldix.com>, 2000-2002, 2009-2010.
 msgid ""
 msgstr ""
-"Project-Id-Version: libc-2.9.90\n"
+"Project-Id-Version: libc-2.11.1\n"
 "POT-Creation-Date: 2009-02-06 12:40-0800\n"
-"PO-Revision-Date: 2009-05-20 11:37+0200\n"
-"Last-Translator: Keld Simonsen <keld@dkuug.dk>\n"
+"PO-Revision-Date: 2010-11-01 10:37+0100\n"
+"Last-Translator: Keld Simonsen <keld@keldix.com>\n"
 "Language-Team: Danish <dansk@dansk-gruppen.dk>\n"
 "MIME-Version: 1.0\n"
 "Content-Type: text/plain; charset=ISO-8859-1\n"
@@ -122,7 +121,7 @@ msgid ""
 "Generate message catalog.\vIf INPUT-FILE is -, input is read from standard input.  If OUTPUT-FILE\n"
 "is -, output is written to standard output.\n"
 msgstr ""
-"Generér meddelelseskatalog.\\vHvis INDFIL er '-' lćses inddata fra standard ind.\n"
+"Generér meddelelseskatalog.\vHvis INDFIL er '-' lćses inddata fra standard ind.\n"
 "Hvis UDFIL er '-' skrives uddata til standard ud.\n"
 
 #: catgets/gencat.c:124
diff --git a/libc/string/Makefile b/libc/string/Makefile
index f511eb378..0d5366559 100644
--- a/libc/string/Makefile
+++ b/libc/string/Makefile
@@ -59,7 +59,7 @@ tests		:= tester inl-tester noinl-tester testcopy test-ffs	\
 		   bug-strncat1 bug-strspn1 bug-strpbrk1 tst-bswap	\
 		   tst-strtok tst-strfry	\
 		   bug-strtok1 $(addprefix test-,$(strop-tests))	\
-		   tst-strxfrm2 tst-endian tst-svc2 bug-strstr1
+		   tst-strxfrm2 tst-endian tst-svc2 bug-strstr1 bug-strchr1
 tests-$(OPTION_EGLIBC_ENVZ) += bug-envz1 
 tests-$(OPTION_EGLIBC_LOCALE_CODE) \
 		+= tst-strxfrm bug-strcoll1
diff --git a/libc/string/bug-strchr1.c b/libc/string/bug-strchr1.c
new file mode 100644
index 000000000..21155d8a7
--- /dev/null
+++ b/libc/string/bug-strchr1.c
@@ -0,0 +1,14 @@
+#include <stdio.h>
+#include <string.h>
+
+static int
+do_test (void)
+{
+  char s[] __attribute__((aligned(16))) = "\xff";
+  char *p = strchr (s, '\xfe');
+  printf ("%p\n", p);
+  return p != NULL;
+}
+
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"
diff --git a/libc/sysdeps/generic/ldsodefs.h b/libc/sysdeps/generic/ldsodefs.h
index ec1e57f88..e79252fe0 100644
--- a/libc/sysdeps/generic/ldsodefs.h
+++ b/libc/sysdeps/generic/ldsodefs.h
@@ -832,11 +832,9 @@ extern void _dl_receive_error (receiver_fct fct, void (*operate) (void *),
 
 /* Open the shared object NAME and map in its segments.
    LOADER's DT_RPATH is used in searching for NAME.
-   If the object is already opened, returns its existing map.
-   For preloaded shared objects PRELOADED is set to a non-zero
-   value to allow additional security checks.  */
+   If the object is already opened, returns its existing map.  */
 extern struct link_map *_dl_map_object (struct link_map *loader,
-					const char *name, int preloaded,
+					const char *name,
 					int type, int trace_mode, int mode,
 					Lmid_t nsid)
      internal_function attribute_hidden;
@@ -901,8 +899,11 @@ extern lookup_t _dl_lookup_symbol_x (const char *undef,
 extern ElfW(Addr) _dl_symbol_value (struct link_map *map, const char *name)
      internal_function;
 
-/* Allocate a `struct link_map' for a new object being loaded,
-   and enter it into the _dl_main_map list.  */
+/* Add the new link_map NEW to the end of the namespace list.  */
+extern void _dl_add_to_namespace_list (struct link_map *new, Lmid_t nsid)
+     internal_function attribute_hidden;
+
+/* Allocate a `struct link_map' for a new object being loaded.  */
 extern struct link_map *_dl_new_object (char *realname, const char *libname,
 					int type, struct link_map *loader,
 					int mode, Lmid_t nsid)
diff --git a/libc/sysdeps/i386/i686/cacheinfo.c b/libc/sysdeps/i386/i686/cacheinfo.c
index f8b7f521c..363596172 100644
--- a/libc/sysdeps/i386/i686/cacheinfo.c
+++ b/libc/sysdeps/i386/i686/cacheinfo.c
@@ -1,7 +1,11 @@
 #define __x86_64_data_cache_size __x86_data_cache_size
+#define __x86_64_raw_data_cache_size __x86_raw_data_cache_size
 #define __x86_64_data_cache_size_half __x86_data_cache_size_half
+#define __x86_64_raw_data_cache_size_half __x86_raw_data_cache_size_half
 #define __x86_64_shared_cache_size __x86_shared_cache_size
+#define __x86_64_raw_shared_cache_size __x86_raw_shared_cache_size
 #define __x86_64_shared_cache_size_half __x86_shared_cache_size_half
+#define __x86_64_raw_shared_cache_size_half __x86_raw_shared_cache_size_half
 
 #define DISABLE_PREFETCHW
 #define DISABLE_PREFERRED_MEMORY_INSTRUCTION
diff --git a/libc/sysdeps/powerpc/dl-procinfo.c b/libc/sysdeps/powerpc/dl-procinfo.c
index 1c74c2a90..2ca76d8b7 100644
--- a/libc/sysdeps/powerpc/dl-procinfo.c
+++ b/libc/sysdeps/powerpc/dl-procinfo.c
@@ -68,7 +68,7 @@ PROCINFO_CLASS const char _dl_powerpc_cap_flags[25][10]
 #if !defined PROCINFO_DECL && defined SHARED
   ._dl_powerpc_platforms
 #else
-PROCINFO_CLASS const char _dl_powerpc_platforms[8][12]
+PROCINFO_CLASS const char _dl_powerpc_platforms[9][12]
 #endif
 #ifndef PROCINFO_DECL
 = {
@@ -79,7 +79,8 @@ PROCINFO_CLASS const char _dl_powerpc_platforms[8][12]
     [PPC_PLATFORM_POWER6] = "power6",
     [PPC_PLATFORM_CELL_BE] = "ppc-cell-be",
     [PPC_PLATFORM_POWER6X] = "power6x",
-    [PPC_PLATFORM_POWER7] = "power7"
+    [PPC_PLATFORM_POWER7] = "power7",
+    [PPC_PLATFORM_PPCA2] = "ppca2"
   }
 #endif
 #if !defined SHARED || defined PROCINFO_DECL
diff --git a/libc/sysdeps/powerpc/dl-procinfo.h b/libc/sysdeps/powerpc/dl-procinfo.h
index 254195a94..2ae35644a 100644
--- a/libc/sysdeps/powerpc/dl-procinfo.h
+++ b/libc/sysdeps/powerpc/dl-procinfo.h
@@ -31,7 +31,7 @@
 #define HWCAP_IMPORTANT		(PPC_FEATURE_HAS_ALTIVEC \
 				+ PPC_FEATURE_HAS_DFP)
 
-#define _DL_PLATFORMS_COUNT	8
+#define _DL_PLATFORMS_COUNT	9
 
 #define _DL_FIRST_PLATFORM      32
 /* Mask to filter out platforms.  */
@@ -47,6 +47,7 @@
 #define PPC_PLATFORM_CELL_BE		5
 #define PPC_PLATFORM_POWER6X		6
 #define PPC_PLATFORM_POWER7		7
+#define PPC_PLATFORM_PPCA2		8
 
 static inline const char *
 __attribute__ ((unused))
@@ -123,6 +124,10 @@ _dl_string_platform (const char *str)
 		       GLRO(dl_powerpc_platforms)[PPC_PLATFORM_CELL_BE] + 3)
 	       == 0)
 	return _DL_FIRST_PLATFORM + PPC_PLATFORM_CELL_BE;
+      else if (strcmp (str + 3,
+		       GLRO(dl_powerpc_platforms)[PPC_PLATFORM_PPCA2] + 3)
+	       == 0)
+	return _DL_FIRST_PLATFORM + PPC_PLATFORM_PPCA2;
     }
 
   return -1;
diff --git a/libc/sysdeps/powerpc/powerpc32/a2/memcpy.S b/libc/sysdeps/powerpc/powerpc32/a2/memcpy.S
new file mode 100644
index 000000000..472f7a393
--- /dev/null
+++ b/libc/sysdeps/powerpc/powerpc32/a2/memcpy.S
@@ -0,0 +1,511 @@
+/* Optimized memcpy implementation for PowerPC A2.
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   Contributed by Michael Brutman <brutman@us.ibm.com>.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+#define PREFETCH_AHEAD 4        /* no cache lines SRC prefetching ahead  */
+#define ZERO_AHEAD 2            /* no cache lines DST zeroing ahead  */
+
+	.machine  a2
+EALIGN (BP_SYM (memcpy), 5, 0)
+	CALL_MCOUNT
+
+	dcbt    0,r4            /* Prefetch ONE SRC cacheline  */
+	cmplwi  cr1,r5,16       /* is size < 16 ?  */
+	mr      r6,r3           /* Copy dest reg to r6; */
+	blt+    cr1,L(shortcopy)
+
+
+	/* Big copy (16 bytes or more)
+
+	   Figure out how far to the nearest quadword boundary, or if we are
+	   on one already.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	*/
+
+	neg     r8,r3           /* LS 4 bits = # bytes to 8-byte dest bdry  */
+	clrlwi  r8,r8,32-4      /* align to 16byte boundary  */
+	sub     r7,r4,r3        /* compute offset to src from dest */
+	cmplwi  cr0,r8,0        /* Were we aligned on a 16 byte bdy? */
+	beq+    L(dst_aligned)
+
+
+
+	/* Destination is not aligned on quadword boundary.  Get us to one.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	   r7 - offset to src from dest
+	   r8 - number of bytes to quadword boundary
+	*/
+
+	mtcrf   0x01,r8         /* put #bytes to boundary into cr7  */
+	subf    r5,r8,r5        /* adjust remaining len */
+
+	bf      cr7*4+3,1f
+	lbzx    r0,r7,r6        /* copy 1 byte addr */
+	stb     r0,0(r6)
+	addi    r6,r6,1
+1:
+	bf      cr7*4+2,2f
+	lhzx    r0,r7,r6        /* copy 2 byte addr */
+	sth     r0,0(r6)
+	addi    r6,r6,2
+2:
+	bf      cr7*4+1,4f
+	lwzx    r0,r7,r6        /* copy 4 byte addr */
+	stw     r0,0(r6)
+	addi    r6,r6,4
+4:
+	bf      cr7*4+0,8f
+	lfdx    r0,r7,r6        /* copy 8 byte addr */
+	stfd    r0,0(r6)
+	addi    r6,r6,8
+8:
+	add     r4,r7,r6        /* update src addr */
+
+
+
+	/* Dest is quadword aligned now.
+
+	   Lots of decisions to make.  If we are copying less than a cache
+	   line we won't be here long.  If we are not on a cache line
+	   boundary we need to get there.  And then we need to figure out
+	   how many cache lines ahead to pre-touch.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	*/
+
+
+	.align  4
+L(dst_aligned):
+
+
+#ifdef SHARED
+	mflr    r0
+/* Establishes GOT addressability so we can load __cache_line_size
+   from static. This value was set from the aux vector during startup.  */
+	bcl     20,31,1f
+1:
+	mflr    r9
+	addis   r9,r9,__cache_line_size-1b@ha
+	lwz     r9,__cache_line_size-1b@l(r9)
+	mtlr    r0
+#else
+/* Load __cache_line_size from static. This value was set from the
+   aux vector during startup.  */
+	lis     r9,__cache_line_size@ha
+	lwz     r9,__cache_line_size@l(r9)
+#endif
+
+	cmplwi  cr5, r9, 0
+	bne+    cr5,L(cachelineset)
+	li      r9,64
+
+
+
+L(cachelineset):
+
+	addi   r10,r9,-1
+
+	cmpw   cr5,r5,r10       /* Less than a cacheline to go? */
+
+	neg     r7,r6           /* How far to next cacheline bdy? */
+
+	addi    r6,r6,-8        /* prepare for stdu  */
+	cmpwi   cr0,r9,128
+	addi    r4,r4,-8        /* prepare for ldu  */
+
+
+	ble+    cr5,L(lessthancacheline)
+
+	beq-    cr0,L(big_lines) /* 128 byte line code */
+
+
+
+
+	/* More than a cacheline left to go, and using 64 byte cachelines */
+
+	clrlwi  r7,r7,32-6      /* How far to next cacheline bdy? */
+
+	cmplwi  cr6,r7,0        /* Are we on a cacheline bdy already? */
+
+	/* Reduce total len by what it takes to get to the next cache line */
+	subf    r5,r7,r5
+	srwi    r7,r7,4         /* How many qws to get to the line bdy? */
+
+	/* How many full cache lines to copy after getting to a line bdy? */
+	srwi    r10,r5,6
+
+	cmplwi  r10,0           /* If no full cache lines to copy ... */
+	li      r11,0           /* number cachelines to copy with prefetch  */
+	beq     L(nocacheprefetch)
+
+
+	/* We are here because we have at least one full cache line to copy,
+	   and therefore some pre-touching to do. */
+
+	cmplwi  r10,PREFETCH_AHEAD
+	li      r12,64+8        /* prefetch distance  */
+	ble     L(lessthanmaxprefetch)
+
+	/* We can only do so much pre-fetching.  R11 will have the count of
+	   lines left to prefetch after the initial batch of prefetches
+	   are executed. */
+
+	subi    r11,r10,PREFETCH_AHEAD
+	li      r10,PREFETCH_AHEAD
+
+L(lessthanmaxprefetch):
+	mtctr   r10
+
+	/* At this point r10/ctr hold the number of lines to prefetch in this
+	   initial batch, and r11 holds any remainder. */
+
+L(prefetchSRC):
+	dcbt    r12,r4
+	addi    r12,r12,64
+	bdnz    L(prefetchSRC)
+
+
+	/* Prefetching is done, or was not needed.
+
+	   cr6 - are we on a cacheline boundary already?
+	   r7  - number of quadwords to the next cacheline boundary
+	*/
+
+L(nocacheprefetch):
+	mtctr   r7
+
+	cmplwi  cr1,r5,64   /* Less than a cache line to copy? */
+
+	/* How many bytes are left after we copy whatever full
+	   cache lines we can get? */
+	clrlwi  r5,r5,32-6
+
+	beq     cr6,L(cachelinealigned)
+
+
+	/* Copy quadwords up to the next cacheline boundary */
+
+L(aligntocacheline):
+	lfd     fp9,0x08(r4)
+	lfdu    fp10,0x10(r4)
+	stfd    fp9,0x08(r6)
+	stfdu   fp10,0x10(r6)
+	bdnz    L(aligntocacheline)
+
+
+	.align 4
+L(cachelinealigned):            /* copy while cache lines  */
+
+	blt-    cr1,L(lessthancacheline) /* size <64  */
+
+L(outerloop):
+	cmpwi   r11,0
+	mtctr   r11
+	beq-    L(endloop)
+
+	li      r11,64*ZERO_AHEAD +8    /* DCBZ dist  */
+
+	.align  4
+	/* Copy whole cachelines, optimized by prefetching SRC cacheline  */
+L(loop):                        /* Copy aligned body  */
+	dcbt    r12,r4          /* PREFETCH SOURCE some cache lines ahead  */
+	lfd     fp9,  0x08(r4)
+	dcbz    r11,r6
+	lfd     fp10, 0x10(r4)
+	lfd     fp11, 0x18(r4)
+	lfd     fp12, 0x20(r4)
+	stfd    fp9,  0x08(r6)
+	stfd    fp10, 0x10(r6)
+	stfd    fp11, 0x18(r6)
+	stfd    fp12, 0x20(r6)
+	lfd     fp9,  0x28(r4)
+	lfd     fp10, 0x30(r4)
+	lfd     fp11, 0x38(r4)
+	lfdu    fp12, 0x40(r4)
+	stfd    fp9,  0x28(r6)
+	stfd    fp10, 0x30(r6)
+	stfd    fp11, 0x38(r6)
+	stfdu   fp12, 0x40(r6)
+
+	bdnz    L(loop)
+
+
+L(endloop):
+	cmpwi   r10,0
+	beq-    L(endloop2)
+	mtctr   r10
+
+L(loop2):                       /* Copy aligned body  */
+	lfd     fp9,  0x08(r4)
+	lfd     fp10, 0x10(r4)
+	lfd     fp11, 0x18(r4)
+	lfd     fp12, 0x20(r4)
+	stfd    fp9,  0x08(r6)
+	stfd    fp10, 0x10(r6)
+	stfd    fp11, 0x18(r6)
+	stfd    fp12, 0x20(r6)
+	lfd     fp9,  0x28(r4)
+	lfd     fp10, 0x30(r4)
+	lfd     fp11, 0x38(r4)
+	lfdu    fp12, 0x40(r4)
+	stfd    fp9,  0x28(r6)
+	stfd    fp10, 0x30(r6)
+	stfd    fp11, 0x38(r6)
+	stfdu   fp12, 0x40(r6)
+
+	bdnz    L(loop2)
+L(endloop2):
+
+
+	.align  4
+L(lessthancacheline):           /* Was there less than cache to do ?  */
+	cmplwi  cr0,r5,16
+	srwi    r7,r5,4         /* divide size by 16  */
+	blt-    L(do_lt16)
+	mtctr   r7
+
+L(copy_remaining):
+	lfd     fp9,  0x08(r4)
+	lfdu    fp10, 0x10(r4)
+	stfd    fp9,  0x08(r6)
+	stfdu   fp10, 0x10(r6)
+	bdnz    L(copy_remaining)
+
+L(do_lt16):                     /* less than 16 ?  */
+	cmplwi  cr0,r5,0        /* copy remaining bytes (0-15)  */
+	beqlr+                  /* no rest to copy  */
+	addi    r4,r4,8
+	addi    r6,r6,8
+
+L(shortcopy):                   /* SIMPLE COPY to handle size =< 15 bytes  */
+	mtcrf   0x01,r5
+	sub     r7,r4,r6
+	bf-     cr7*4+0,8f
+	lfdx    fp9,r7,r6       /* copy 8 byte  */
+	stfd    fp9,0(r6)
+	addi    r6,r6,8
+8:
+	bf      cr7*4+1,4f
+	lwzx    r0,r7,r6        /* copy 4 byte  */
+	stw     r0,0(r6)
+	addi    r6,r6,4
+4:
+	bf      cr7*4+2,2f
+	lhzx    r0,r7,r6        /* copy 2 byte  */
+	sth     r0,0(r6)
+	addi    r6,r6,2
+2:
+	bf      cr7*4+3,1f
+	lbzx    r0,r7,r6        /* copy 1 byte  */
+	stb     r0,0(r6)
+1:
+	blr
+
+
+
+
+
+	/* Similar to above, but for use with 128 byte lines. */
+
+
+L(big_lines):
+
+	clrlwi  r7,r7,32-7      /* How far to next cacheline bdy? */
+
+	cmplwi  cr6,r7,0        /* Are we on a cacheline bdy already? */
+
+	/* Reduce total len by what it takes to get to the next cache line */
+	subf    r5,r7,r5
+	srwi    r7,r7,4         /* How many qw to get to the line bdy? */
+
+	/* How many full cache lines to copy after getting to a line bdy? */
+	srwi    r10,r5,7
+
+	cmplwi  r10,0           /* If no full cache lines to copy ... */
+	li      r11,0           /* number cachelines to copy with prefetch  */
+	beq     L(nocacheprefetch_128)
+
+
+	/* We are here because we have at least one full cache line to copy,
+	   and therefore some pre-touching to do. */
+
+	cmplwi  r10,PREFETCH_AHEAD
+	li      r12,128+8       /* prefetch distance  */
+	ble     L(lessthanmaxprefetch_128)
+
+	/* We can only do so much pre-fetching.  R11 will have the count of
+	   lines left to prefetch after the initial batch of prefetches
+	   are executed. */
+
+	subi    r11,r10,PREFETCH_AHEAD
+	li      r10,PREFETCH_AHEAD
+
+L(lessthanmaxprefetch_128):
+	mtctr   r10
+
+	/* At this point r10/ctr hold the number of lines to prefetch in this
+	   initial batch, and r11 holds any remainder. */
+
+L(prefetchSRC_128):
+	dcbt    r12,r4
+	addi    r12,r12,128
+	bdnz    L(prefetchSRC_128)
+
+
+	/* Prefetching is done, or was not needed.
+
+	   cr6 - are we on a cacheline boundary already?
+	   r7  - number of quadwords to the next cacheline boundary
+	*/
+
+L(nocacheprefetch_128):
+	mtctr   r7
+
+	cmplwi  cr1,r5,128  /* Less than a cache line to copy? */
+
+	/* How many bytes are left after we copy whatever full
+	   cache lines we can get? */
+	clrlwi  r5,r5,32-7
+
+	beq     cr6,L(cachelinealigned_128)
+
+
+	/* Copy quadwords up to the next cacheline boundary */
+
+L(aligntocacheline_128):
+	lfd     fp9,0x08(r4)
+	lfdu    fp10,0x10(r4)
+	stfd    fp9,0x08(r6)
+	stfdu   fp10,0x10(r6)
+	bdnz    L(aligntocacheline_128)
+
+
+L(cachelinealigned_128):        /* copy while cache lines  */
+
+	blt-    cr1,L(lessthancacheline) /* size <128  */
+
+L(outerloop_128):
+	cmpwi   r11,0
+	mtctr   r11
+	beq-    L(endloop_128)
+
+	li      r11,128*ZERO_AHEAD +8    /* DCBZ dist  */
+
+	.align  4
+	/* Copy whole cachelines, optimized by prefetching SRC cacheline  */
+L(loop_128):                    /* Copy aligned body  */
+	dcbt    r12,r4          /* PREFETCH SOURCE some cache lines ahead  */
+	lfd     fp9,  0x08(r4)
+	dcbz    r11,r6
+	lfd     fp10, 0x10(r4)
+	lfd     fp11, 0x18(r4)
+	lfd     fp12, 0x20(r4)
+	stfd    fp9,  0x08(r6)
+	stfd    fp10, 0x10(r6)
+	stfd    fp11, 0x18(r6)
+	stfd    fp12, 0x20(r6)
+	lfd     fp9,  0x28(r4)
+	lfd     fp10, 0x30(r4)
+	lfd     fp11, 0x38(r4)
+	lfd     fp12, 0x40(r4)
+	stfd    fp9,  0x28(r6)
+	stfd    fp10, 0x30(r6)
+	stfd    fp11, 0x38(r6)
+	stfd    fp12, 0x40(r6)
+	lfd     fp9,  0x48(r4)
+	lfd     fp10, 0x50(r4)
+	lfd     fp11, 0x58(r4)
+	lfd     fp12, 0x60(r4)
+	stfd    fp9,  0x48(r6)
+	stfd    fp10, 0x50(r6)
+	stfd    fp11, 0x58(r6)
+	stfd    fp12, 0x60(r6)
+	lfd     fp9,  0x68(r4)
+	lfd     fp10, 0x70(r4)
+	lfd     fp11, 0x78(r4)
+	lfdu    fp12, 0x80(r4)
+	stfd    fp9,  0x68(r6)
+	stfd    fp10, 0x70(r6)
+	stfd    fp11, 0x78(r6)
+	stfdu   fp12, 0x80(r6)
+
+	bdnz    L(loop_128)
+
+
+L(endloop_128):
+	cmpwi   r10,0
+	beq-    L(endloop2_128)
+	mtctr   r10
+
+L(loop2_128):                   /* Copy aligned body  */
+	lfd     fp9,  0x08(r4)
+	lfd     fp10, 0x10(r4)
+	lfd     fp11, 0x18(r4)
+	lfd     fp12, 0x20(r4)
+	stfd    fp9,  0x08(r6)
+	stfd    fp10, 0x10(r6)
+	stfd    fp11, 0x18(r6)
+	stfd    fp12, 0x20(r6)
+	lfd     fp9,  0x28(r4)
+	lfd     fp10, 0x30(r4)
+	lfd     fp11, 0x38(r4)
+	lfd     fp12, 0x40(r4)
+	stfd    fp9,  0x28(r6)
+	stfd    fp10, 0x30(r6)
+	stfd    fp11, 0x38(r6)
+	stfd    fp12, 0x40(r6)
+	lfd     fp9,  0x48(r4)
+	lfd     fp10, 0x50(r4)
+	lfd     fp11, 0x58(r4)
+	lfd     fp12, 0x60(r4)
+	stfd    fp9,  0x48(r6)
+	stfd    fp10, 0x50(r6)
+	stfd    fp11, 0x58(r6)
+	stfd    fp12, 0x60(r6)
+	lfd     fp9,  0x68(r4)
+	lfd     fp10, 0x70(r4)
+	lfd     fp11, 0x78(r4)
+	lfdu    fp12, 0x80(r4)
+	stfd    fp9,  0x68(r6)
+	stfd    fp10, 0x70(r6)
+	stfd    fp11, 0x78(r6)
+	stfdu   fp12, 0x80(r6)
+	bdnz    L(loop2_128)
+L(endloop2_128):
+
+	b       L(lessthancacheline)
+
+
+END (BP_SYM (memcpy))
+libc_hidden_builtin_def (memcpy)
diff --git a/libc/sysdeps/powerpc/powerpc32/power7/mempcpy.S b/libc/sysdeps/powerpc/powerpc32/power7/mempcpy.S
new file mode 100644
index 000000000..5e0525645
--- /dev/null
+++ b/libc/sysdeps/powerpc/powerpc32/power7/mempcpy.S
@@ -0,0 +1,471 @@
+/* Optimized mempcpy implementation for POWER7.
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   Contributed by Luis Machado <luisgpm@br.ibm.com>.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA
+   02110-1301 USA.  */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+/* __ptr_t [r3] __mempcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
+	Returns 'dst' + 'len'.  */
+
+	.machine  power7
+EALIGN (BP_SYM (__mempcpy), 5, 0)
+	CALL_MCOUNT
+
+	stwu	1,-32(1)
+	cfi_adjust_cfa_offset(32)
+	stw	30,20(1)
+	cfi_offset(30,(20-32))
+	stw	31,24(1)
+	mr	30,3
+	cmplwi	cr1,5,31
+	neg	0,3
+	cfi_offset(31,-8)
+	ble	cr1,L(copy_LT_32)  /* If move < 32 bytes use short move
+					code.  */
+
+	andi.	11,3,7	      /* Check alignment of DST.  */
+	clrlwi	10,4,29	      /* Check alignment of SRC.  */
+	cmplw	cr6,10,11     /* SRC and DST alignments match?  */
+	mr	12,4
+	mr	31,5
+	bne	cr6,L(copy_GE_32_unaligned)
+
+	srwi	9,5,3	      /* Number of full quadwords remaining.  */
+
+	beq	L(copy_GE_32_aligned_cont)
+
+	clrlwi	0,0,29
+	mtcrf	0x01,0
+	subf	31,0,5
+
+	/* Get the SRC aligned to 8 bytes.  */
+
+1:	bf	31,2f
+	lbz	6,0(12)
+	addi	12,12,1
+	stb	6,0(3)
+	addi	3,3,1
+2:	bf	30,4f
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+4:	bf	29,0f
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+0:
+	clrlwi	10,12,29      /* Check alignment of SRC again.  */
+	srwi	9,31,3	      /* Number of full doublewords remaining.  */
+
+L(copy_GE_32_aligned_cont):
+
+	clrlwi	11,31,29
+	mtcrf	0x01,9
+
+	srwi	8,31,5
+	cmplwi	cr1,9,4
+	cmplwi	cr6,11,0
+	mr	11,12
+
+	/* Copy 1~3 doublewords so the main loop starts
+	at a multiple of 32 bytes.  */
+
+	bf	30,1f
+	lfd	6,0(12)
+	lfd	7,8(12)
+	addi	11,12,16
+	mtctr	8
+	stfd	6,0(3)
+	stfd	7,8(3)
+	addi	10,3,16
+	bf	31,4f
+	lfd	0,16(12)
+	stfd	0,16(3)
+	blt	cr1,3f
+	addi	11,12,24
+	addi	10,3,24
+	b	4f
+
+	.align	4
+1:	/* Copy 1 doubleword and set the counter.  */
+	mr	10,3
+	mtctr	8
+	bf	31,4f
+	lfd	6,0(12)
+	addi	11,12,8
+	stfd	6,0(3)
+	addi	10,3,8
+
+	.align	4
+4:	/* Main aligned copy loop. Copies 32-bytes at a time.  */
+	lfd	6,0(11)
+	lfd	7,8(11)
+	lfd	8,16(11)
+	lfd	0,24(11)
+	addi	11,11,32
+
+	stfd	6,0(10)
+	stfd	7,8(10)
+	stfd	8,16(10)
+	stfd	0,24(10)
+	addi	10,10,32
+	bdnz	4b
+3:
+
+	/* Check for tail bytes.  */
+
+	clrrwi	0,31,3
+	mtcrf	0x01,31
+	beq	cr6,0f
+
+.L9:
+	add	3,3,0
+	add	12,12,0
+
+	/*  At this point we have a tail of 0-7 bytes and we know that the
+	destination is doubleword-aligned.  */
+4:	/* Copy 4 bytes.  */
+	bf	29,2f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+2:	/* Copy 2 bytes.  */
+	bf	30,1f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+1:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,0(12)
+	stb	6,0(3)
+0:	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	lwz	31,24(1)
+	addi	1,1,32
+	blr
+
+	/* Handle copies of 0~31 bytes.  */
+	.align	4
+L(copy_LT_32):
+	cmplwi	cr6,5,8
+	mr	12,4
+	mtcrf	0x01,5
+	ble	cr6,L(copy_LE_8)
+
+	/* At least 9 bytes to go.  */
+	neg	8,4
+	clrrwi	11,4,2
+	andi.	0,8,3
+	cmplwi	cr1,5,16
+	mr	10,5
+	beq	L(copy_LT_32_aligned)
+
+	/* Force 4-bytes alignment for SRC.  */
+	mtocrf  0x01,0
+	subf	10,0,5
+2:	bf	30,1f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+1:	bf	31,L(end_4bytes_alignment)
+
+	lbz	6,0(12)
+	addi	12,12,1
+	stb	6,0(3)
+	addi	3,3,1
+
+	.align	4
+L(end_4bytes_alignment):
+	cmplwi	cr1,10,16
+	mtcrf	0x01,10
+
+L(copy_LT_32_aligned):
+	/* At least 6 bytes to go, and SRC is word-aligned.  */
+	blt	cr1,8f
+
+	/* Copy 16 bytes.  */
+	lwz	6,0(12)
+	lwz	7,4(12)
+	stw	6,0(3)
+	lwz	8,8(12)
+	stw	7,4(3)
+	lwz	6,12(12)
+	addi	12,12,16
+	stw	8,8(3)
+	stw	6,12(3)
+	addi	3,3,16
+8:	/* Copy 8 bytes.  */
+	bf	28,4f
+
+	lwz	6,0(12)
+	lwz	7,4(12)
+	addi	12,12,8
+	stw	6,0(3)
+	stw	7,4(3)
+	addi	3,3,8
+4:	/* Copy 4 bytes.  */
+	bf	29,2f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+2:	/* Copy 2-3 bytes.  */
+	bf	30,1f
+
+	lhz	6,0(12)
+	sth	6,0(3)
+	bf	31,0f
+	lbz	7,2(12)
+	stb	7,2(3)
+
+	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	addi	1,1,32
+	blr
+
+	.align	4
+1:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,0(12)
+	stb	6,0(3)
+0:	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	addi	1,1,32
+	blr
+
+	/* Handles copies of 0~8 bytes.  */
+	.align	4
+L(copy_LE_8):
+	bne	cr6,4f
+
+	/* Though we could've used lfd/stfd here, they are still
+	slow for unaligned cases.  */
+
+	lwz	6,0(4)
+	lwz	7,4(4)
+	stw	6,0(3)
+	stw	7,4(3)
+
+	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	addi	1,1,32
+	blr
+
+	.align	4
+4:	/* Copies 4~7 bytes.  */
+	bf	29,2b
+
+	lwz	6,0(4)
+	stw	6,0(3)
+	bf	30,5f
+	lhz	7,4(4)
+	sth	7,4(3)
+	bf	31,0f
+	lbz	8,6(4)
+	stb	8,6(3)
+
+	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	addi	1,1,32
+	blr
+
+	.align	4
+5:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,4(4)
+	stb	6,4(3)
+
+0:	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	addi	1,1,32
+	blr
+
+	/* Handle copies of 32+ bytes where DST is aligned (to quadword) but
+	SRC is not. Use aligned quadword loads from SRC, shifted to realign
+	the data, allowing for aligned DST stores.  */
+	.align	4
+L(copy_GE_32_unaligned):
+	andi.	11,3,15	      /* Check alignment of DST.  */
+	clrlwi	0,0,28	      /* Number of bytes until the 1st
+				 quadword of DST.  */
+	srwi	9,5,4	      /* Number of full quadwords remaining.  */
+
+	beq	L(copy_GE_32_unaligned_cont)
+
+	/* SRC is not quadword aligned, get it aligned.  */
+
+	mtcrf	0x01,0
+	subf	31,0,5
+
+	/* Vector instructions work best when proper alignment (16-bytes)
+	is present.  Move 0~15 bytes as needed to get DST quadword-aligned.  */
+1:	/* Copy 1 byte.  */
+	bf	31,2f
+
+	lbz	6,0(12)
+	addi	12,12,1
+	stb	6,0(3)
+	addi	3,3,1
+2:	/* Copy 2 bytes.  */
+	bf		30,4f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+4:	/* Copy 4 bytes.  */
+	bf	29,8f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+8:	/* Copy 8 bytes.  */
+	bf	28,0f
+
+	lfd	6,0(12)
+	addi	12,12,8
+	stfd	6,0(3)
+	addi	3,3,8
+0:
+	clrlwi	10,12,28      /* Check alignment of SRC.  */
+	srwi	9,31,4	      /* Number of full quadwords remaining.  */
+
+	/* The proper alignment is present, it is OK to copy the bytes now.  */
+L(copy_GE_32_unaligned_cont):
+
+	/* Setup two indexes to speed up the indexed vector operations.  */
+	clrlwi	11,31,28
+	li	6,16	      /* Index for 16-bytes offsets.  */
+	li	7,32	      /* Index for 32-bytes offsets.  */
+	cmplwi	cr1,11,0
+	srwi	8,31,5	      /* Setup the loop counter.  */
+	mr	10,3
+	mr	11,12
+	mtcrf	0x01,9
+	cmplwi	cr6,9,1
+	lvsl	5,0,12
+	lvx	3,0,12
+	bf	31,L(setup_unaligned_loop)
+
+	/* Copy another 16 bytes to align to 32-bytes due to the loop .  */
+	lvx	4,12,6
+	vperm	6,3,4,5
+	addi	11,12,16
+	addi	10,3,16
+	stvx	6,0,3
+	vor	3,4,4
+
+L(setup_unaligned_loop):
+	mtctr	8
+	ble	cr6,L(end_unaligned_loop)
+
+	/* Copy 32 bytes at a time using vector instructions.  */
+	.align	4
+L(unaligned_loop):
+
+	/* Note: vr6/vr10 may contain data that was already copied,
+	but in order to get proper alignment, we may have to copy
+	some portions again. This is faster than having unaligned
+	vector instructions though.  */
+
+	lvx	4,11,6	      /* vr4 = r11+16.  */
+	vperm	6,3,4,5	      /* Merge the correctly-aligned portions
+				 of vr3/vr4 into vr6.  */
+	lvx	3,11,7	      /* vr3 = r11+32.  */
+	vperm	10,4,3,5      /* Merge the correctly-aligned portions
+				 of vr3/vr4 into vr10.  */
+	addi	11,11,32
+	stvx	6,0,10
+	stvx	10,10,6
+	addi	10,10,32
+
+	bdnz	L(unaligned_loop)
+
+	.align	4
+L(end_unaligned_loop):
+
+	/* Check for tail bytes.  */
+	clrrwi	0,31,4
+	mtcrf	0x01,31
+	beq	cr1,0f
+
+	add	3,3,0
+	add	12,12,0
+
+	/*  We have 1~15 tail bytes to copy, and DST is quadword aligned.  */
+8:	/* Copy 8 bytes.  */
+	bf	28,4f
+
+	lwz	6,0(12)
+	lwz	7,4(12)
+	addi	12,12,8
+	stw	6,0(3)
+	stw	7,4(3)
+	addi	3,3,8
+4:	/* Copy 4 bytes.  */
+	bf	29,2f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+2:	/* Copy 2~3 bytes.  */
+	bf	30,1f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+1:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,0(12)
+	stb	6,0(3)
+0:	/* Return DST + LEN pointer.  */
+	add	3,30,5
+	lwz	30,20(1)
+	lwz	31,24(1)
+	addi	1,1,32
+	blr
+
+END (BP_SYM (__mempcpy))
+libc_hidden_def (BP_SYM (__mempcpy))
+weak_alias (BP_SYM (__mempcpy), BP_SYM (mempcpy))
+libc_hidden_builtin_def (mempcpy)
diff --git a/libc/sysdeps/powerpc/powerpc64/a2/memcpy.S b/libc/sysdeps/powerpc/powerpc64/a2/memcpy.S
new file mode 100644
index 000000000..ac95171aa
--- /dev/null
+++ b/libc/sysdeps/powerpc/powerpc64/a2/memcpy.S
@@ -0,0 +1,501 @@
+/* Optimized memcpy implementation for PowerPC A2.
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   Contributed by Michael Brutman <brutman@us.ibm.com>.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+#define PREFETCH_AHEAD 4        /* no cache lines SRC prefetching ahead  */
+#define ZERO_AHEAD 2            /* no cache lines DST zeroing ahead  */
+
+	.section        ".toc","aw"
+.LC0:
+	.tc __cache_line_size[TC],__cache_line_size
+	.section        ".text"
+	.align 2
+
+
+	.machine  a2
+EALIGN (BP_SYM (memcpy), 5, 0)
+	CALL_MCOUNT 3
+
+	dcbt    0,r4            /* Prefetch ONE SRC cacheline  */
+	cmpldi  cr1,r5,16       /* is size < 16 ?  */
+	mr      r6,r3           /* Copy dest reg to r6; */
+	blt+    cr1,L(shortcopy)
+
+
+	/* Big copy (16 bytes or more)
+
+	   Figure out how far to the nearest quadword boundary, or if we are
+	   on one already.  Also get the cache line size.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	*/
+
+	neg     r8,r3           /* LS 4 bits = # bytes to 8-byte dest bdry  */
+	ld      r9,.LC0@toc(r2) /* Get cache line size (part 1) */
+	clrldi  r8,r8,64-4      /* align to 16byte boundary  */
+	sub     r7,r4,r3        /* compute offset to src from dest */
+	lwz     r9,0(r9)        /* Get cache line size (part 2) */
+	cmpldi  cr0,r8,0        /* Were we aligned on a 16 byte bdy? */
+	addi    r10,r9,-1       /* Cache line mask */
+	beq+    L(dst_aligned)
+
+
+
+	/* Destination is not aligned on quadword boundary.  Get us to one.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	   r7 - offset to src from dest
+	   r8 - number of bytes to quadword boundary
+	*/
+
+	mtcrf   0x01,r8         /* put #bytes to boundary into cr7  */
+	subf    r5,r8,r5        /* adjust remaining len */
+
+	bf      cr7*4+3,1f
+	lbzx    r0,r7,r6        /* copy 1 byte addr */
+	stb     r0,0(r6)
+	addi    r6,r6,1
+1:
+	bf      cr7*4+2,2f
+	lhzx    r0,r7,r6        /* copy 2 byte addr */
+	sth     r0,0(r6)
+	addi    r6,r6,2
+2:
+	bf      cr7*4+1,4f
+	lwzx    r0,r7,r6        /* copy 4 byte addr */
+	stw     r0,0(r6)
+	addi    r6,r6,4
+4:
+	bf      cr7*4+0,8f
+	ldx     r0,r7,r6        /* copy 8 byte addr */
+	std     r0,0(r6)
+	addi    r6,r6,8
+8:
+	add     r4,r7,r6        /* update src addr */
+
+
+
+	/* Dest is quadword aligned now.
+
+	   Lots of decisions to make.  If we are copying less than a cache
+	   line we won't be here long.  If we are not on a cache line
+	   boundary we need to get there.  And then we need to figure out
+	   how many cache lines ahead to pre-touch.
+
+	   r3 - return value (always)
+	   r4 - current source addr
+	   r5 - copy length
+	   r6 - current dest addr
+	*/
+
+
+	.align 4
+L(dst_aligned):
+
+
+	cmpd   cr5,r5,r10       /* Less than a cacheline to go? */
+
+	neg     r7,r6           /* How far to next cacheline bdy? */
+
+	addi    r6,r6,-8        /* prepare for stdu  */
+	cmpdi   cr0,r9,128
+	addi    r4,r4,-8        /* prepare for ldu  */
+
+
+	ble+    cr5,L(lessthancacheline)
+
+	beq-    cr0,L(big_lines) /* 128 byte line code */
+
+
+
+	/* More than a cacheline left to go, and using 64 byte cachelines */
+
+	clrldi  r7,r7,64-6      /* How far to next cacheline bdy? */
+
+	cmpldi  cr6,r7,0        /* Are we on a cacheline bdy already? */
+
+	/* Reduce total len by what it takes to get to the next cache line */
+	subf    r5,r7,r5
+	srdi    r7,r7,4         /* How many qws to get to the line bdy? */
+
+	/* How many full cache lines to copy after getting to a line bdy? */
+	srdi    r10,r5,6
+
+	cmpldi  r10,0           /* If no full cache lines to copy ... */
+	li      r11,0           /* number cachelines to copy with prefetch  */
+	beq     L(nocacheprefetch)
+
+
+	/* We are here because we have at least one full cache line to copy,
+	   and therefore some pre-touching to do. */
+
+	cmpldi  r10,PREFETCH_AHEAD
+	li      r12,64+8        /* prefetch distance  */
+	ble     L(lessthanmaxprefetch)
+
+	/* We can only do so much pre-fetching.  R11 will have the count of
+	   lines left to prefetch after the initial batch of prefetches
+	   are executed. */
+
+	subi    r11,r10,PREFETCH_AHEAD
+	li      r10,PREFETCH_AHEAD
+
+L(lessthanmaxprefetch):
+	mtctr   r10
+
+	/* At this point r10/ctr hold the number of lines to prefetch in this
+	   initial batch, and r11 holds any remainder. */
+
+L(prefetchSRC):
+	dcbt    r12,r4
+	addi    r12,r12,64
+	bdnz    L(prefetchSRC)
+
+
+	/* Prefetching is done, or was not needed.
+
+	   cr6 - are we on a cacheline boundary already?
+	   r7  - number of quadwords to the next cacheline boundary
+	*/
+
+L(nocacheprefetch):
+	mtctr   r7
+
+	cmpldi  cr1,r5,64   /* Less than a cache line to copy? */
+
+	/* How many bytes are left after we copy whatever full
+	   cache lines we can get? */
+	clrldi  r5,r5,64-6
+
+	beq     cr6,L(cachelinealigned)
+
+
+	/* Copy quadwords up to the next cacheline boundary */
+
+L(aligntocacheline):
+	ld      r9,0x08(r4)
+	ld      r7,0x10(r4)
+	addi    r4,r4,0x10
+	std     r9,0x08(r6)
+	stdu    r7,0x10(r6)
+	bdnz    L(aligntocacheline)
+
+
+	.align 4
+L(cachelinealigned):            /* copy while cache lines  */
+
+	blt-    cr1,L(lessthancacheline) /* size <64  */
+
+L(outerloop):
+	cmpdi   r11,0
+	mtctr   r11
+	beq-    L(endloop)
+
+	li      r11,64*ZERO_AHEAD +8    /* DCBZ dist  */
+
+	.align  4
+	/* Copy whole cachelines, optimized by prefetching SRC cacheline  */
+L(loop):                        /* Copy aligned body  */
+	dcbt    r12,r4          /* PREFETCH SOURCE some cache lines ahead  */
+	ld      r9, 0x08(r4)
+	dcbz    r11,r6
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	addi    r4, r4,0x40
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	stdu    r0, 0x40(r6)
+
+	bdnz    L(loop)
+
+
+L(endloop):
+	cmpdi   r10,0
+	beq-    L(endloop2)
+	mtctr   r10
+
+L(loop2):                       /* Copy aligned body  */
+	ld      r9, 0x08(r4)
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	addi    r4, r4,0x40
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	stdu    r0, 0x40(r6)
+
+	bdnz    L(loop2)
+L(endloop2):
+
+
+	.align 4
+L(lessthancacheline):           /* Was there less than cache to do ?  */
+	cmpldi  cr0,r5,16
+	srdi    r7,r5,4         /* divide size by 16  */
+	blt-    L(do_lt16)
+	mtctr   r7
+
+L(copy_remaining):
+	ld      r8,0x08(r4)
+	ld      r7,0x10(r4)
+	addi    r4,r4,0x10
+	std     r8,0x08(r6)
+	stdu    r7,0x10(r6)
+	bdnz    L(copy_remaining)
+
+L(do_lt16):                     /* less than 16 ?  */
+	cmpldi  cr0,r5,0        /* copy remaining bytes (0-15)  */
+	beqlr+                  /* no rest to copy  */
+	addi    r4,r4,8
+	addi    r6,r6,8
+
+L(shortcopy):                   /* SIMPLE COPY to handle size =< 15 bytes  */
+	mtcrf   0x01,r5
+	sub     r7,r4,r6
+	bf-     cr7*4+0,8f
+	ldx     r0,r7,r6        /* copy 8 byte  */
+	std     r0,0(r6)
+	addi    r6,r6,8
+8:
+	bf      cr7*4+1,4f
+	lwzx    r0,r7,r6        /* copy 4 byte  */
+	stw     r0,0(r6)
+	addi    r6,r6,4
+4:
+	bf      cr7*4+2,2f
+	lhzx    r0,r7,r6        /* copy 2 byte  */
+	sth     r0,0(r6)
+	addi    r6,r6,2
+2:
+	bf      cr7*4+3,1f
+	lbzx    r0,r7,r6        /* copy 1 byte  */
+	stb     r0,0(r6)
+1:
+	blr
+
+
+
+
+
+	/* Similar to above, but for use with 128 byte lines. */
+
+
+L(big_lines):
+
+	clrldi  r7,r7,64-7      /* How far to next cacheline bdy? */
+
+	cmpldi  cr6,r7,0        /* Are we on a cacheline bdy already? */
+
+	/* Reduce total len by what it takes to get to the next cache line */
+	subf    r5,r7,r5
+	srdi    r7,r7,4         /* How many qws to get to the line bdy? */
+
+	/* How many full cache lines to copy after getting to a line bdy? */
+	srdi    r10,r5,7
+
+	cmpldi  r10,0           /* If no full cache lines to copy ... */
+	li      r11,0           /* number cachelines to copy with prefetch  */
+	beq     L(nocacheprefetch_128)
+
+
+	/* We are here because we have at least one full cache line to copy,
+	   and therefore some pre-touching to do. */
+
+	cmpldi  r10,PREFETCH_AHEAD
+	li      r12,128+8       /* prefetch distance  */
+	ble     L(lessthanmaxprefetch_128)
+
+	/* We can only do so much pre-fetching.  R11 will have the count of
+	   lines left to prefetch after the initial batch of prefetches
+	   are executed. */
+
+	subi    r11,r10,PREFETCH_AHEAD
+	li      r10,PREFETCH_AHEAD
+
+L(lessthanmaxprefetch_128):
+	mtctr   r10
+
+	/* At this point r10/ctr hold the number of lines to prefetch in this
+	   initial batch, and r11 holds any remainder. */
+
+L(prefetchSRC_128):
+	dcbt    r12,r4
+	addi    r12,r12,128
+	bdnz    L(prefetchSRC_128)
+
+
+	/* Prefetching is done, or was not needed.
+
+	   cr6 - are we on a cacheline boundary already?
+	   r7  - number of quadwords to the next cacheline boundary
+	*/
+
+L(nocacheprefetch_128):
+	mtctr   r7
+
+	cmpldi  cr1,r5,128  /* Less than a cache line to copy? */
+
+	/* How many bytes are left after we copy whatever full
+	   cache lines we can get? */
+	clrldi  r5,r5,64-7
+
+	beq     cr6,L(cachelinealigned_128)
+
+
+	/* Copy quadwords up to the next cacheline boundary */
+
+L(aligntocacheline_128):
+	ld      r9,0x08(r4)
+	ld      r7,0x10(r4)
+	addi    r4,r4,0x10
+	std     r9,0x08(r6)
+	stdu    r7,0x10(r6)
+	bdnz    L(aligntocacheline_128)
+
+
+L(cachelinealigned_128):        /* copy while cache lines  */
+
+	blt-    cr1,L(lessthancacheline) /* size <128  */
+
+L(outerloop_128):
+	cmpdi   r11,0
+	mtctr   r11
+	beq-    L(endloop_128)
+
+	li      r11,128*ZERO_AHEAD +8    /* DCBZ dist  */
+
+	.align  4
+	/* Copy whole cachelines, optimized by prefetching SRC cacheline  */
+L(loop_128):                    /* Copy aligned body  */
+	dcbt    r12,r4          /* PREFETCH SOURCE some cache lines ahead  */
+	ld      r9, 0x08(r4)
+	dcbz    r11,r6
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	std     r0, 0x40(r6)
+	ld      r9, 0x48(r4)
+	ld      r7, 0x50(r4)
+	ld      r8, 0x58(r4)
+	ld      r0, 0x60(r4)
+	std     r9, 0x48(r6)
+	std     r7, 0x50(r6)
+	std     r8, 0x58(r6)
+	std     r0, 0x60(r6)
+	ld      r9, 0x68(r4)
+	ld      r7, 0x70(r4)
+	ld      r8, 0x78(r4)
+	ld      r0, 0x80(r4)
+	addi    r4, r4,0x80
+	std     r9, 0x68(r6)
+	std     r7, 0x70(r6)
+	std     r8, 0x78(r6)
+	stdu    r0, 0x80(r6)
+
+	bdnz    L(loop_128)
+
+
+L(endloop_128):
+	cmpdi   r10,0
+	beq-    L(endloop2_128)
+	mtctr   r10
+
+L(loop2_128):                       /* Copy aligned body  */
+	ld      r9, 0x08(r4)
+	ld      r7, 0x10(r4)
+	ld      r8, 0x18(r4)
+	ld      r0, 0x20(r4)
+	std     r9, 0x08(r6)
+	std     r7, 0x10(r6)
+	std     r8, 0x18(r6)
+	std     r0, 0x20(r6)
+	ld      r9, 0x28(r4)
+	ld      r7, 0x30(r4)
+	ld      r8, 0x38(r4)
+	ld      r0, 0x40(r4)
+	std     r9, 0x28(r6)
+	std     r7, 0x30(r6)
+	std     r8, 0x38(r6)
+	std     r0, 0x40(r6)
+	ld      r9, 0x48(r4)
+	ld      r7, 0x50(r4)
+	ld      r8, 0x58(r4)
+	ld      r0, 0x60(r4)
+	std     r9, 0x48(r6)
+	std     r7, 0x50(r6)
+	std     r8, 0x58(r6)
+	std     r0, 0x60(r6)
+	ld      r9, 0x68(r4)
+	ld      r7, 0x70(r4)
+	ld      r8, 0x78(r4)
+	ld      r0, 0x80(r4)
+	addi    r4, r4,0x80
+	std     r9, 0x68(r6)
+	std     r7, 0x70(r6)
+	std     r8, 0x78(r6)
+	stdu    r0, 0x80(r6)
+
+	bdnz    L(loop2_128)
+L(endloop2_128):
+
+	b       L(lessthancacheline)
+
+
+END_GEN_TB (BP_SYM (memcpy),TB_TOCLESS)
+libc_hidden_builtin_def (memcpy)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S b/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S
new file mode 100644
index 000000000..09c08ab51
--- /dev/null
+++ b/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S
@@ -0,0 +1,458 @@
+/* Optimized mempcpy implementation for POWER7.
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   Contributed by Luis Machado <luisgpm@br.ibm.com>.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA
+   02110-1301 USA.  */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+
+/* __ptr_t [r3] __mempcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
+    Returns 'dst' + 'len'.  */
+
+	.machine  power7
+EALIGN (BP_SYM (__mempcpy), 5, 0)
+	CALL_MCOUNT 3
+
+	cmpldi	cr1,5,31
+	neg	0,3
+	std	3,-16(1)
+	std	31,-8(1)
+	cfi_offset(31,-8)
+	ble	cr1,L(copy_LT_32)   /* If move < 32 bytes use short move
+				       code.  */
+
+	andi.	11,3,7	      /* Check alignment of DST.  */
+
+
+	clrldi	10,4,61	      /* Check alignment of SRC.  */
+	cmpld	cr6,10,11     /* SRC and DST alignments match?  */
+	mr	12,4
+	mr	31,5
+	bne	cr6,L(copy_GE_32_unaligned)
+
+	srdi	9,5,3	      /* Number of full quadwords remaining.  */
+
+	beq	L(copy_GE_32_aligned_cont)
+
+	clrldi	0,0,61
+	mtcrf	0x01,0
+	subf	31,0,5
+
+	/* Get the SRC aligned to 8 bytes.  */
+
+1:	bf	31,2f
+	lbz	6,0(12)
+	addi	12,12,1
+	stb	6,0(3)
+	addi	3,3,1
+2:	bf	30,4f
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+4:	bf	29,0f
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+0:
+	clrldi	10,12,61      /* Check alignment of SRC again.  */
+	srdi	9,31,3	      /* Number of full doublewords remaining.  */
+
+L(copy_GE_32_aligned_cont):
+
+	clrldi	11,31,61
+	mtcrf	0x01,9
+
+	srdi	8,31,5
+	cmpldi	cr1,9,4
+	cmpldi	cr6,11,0
+	mr	11,12
+
+	/* Copy 1~3 doublewords so the main loop starts
+	at a multiple of 32 bytes.  */
+
+	bf	30,1f
+	ld	6,0(12)
+	ld	7,8(12)
+	addi	11,12,16
+	mtctr	8
+	std	6,0(3)
+	std	7,8(3)
+	addi	10,3,16
+	bf	31,4f
+	ld	0,16(12)
+	std	0,16(3)
+	blt	cr1,3f
+	addi	11,12,24
+	addi	10,3,24
+	b	4f
+
+	.align	4
+1:	/* Copy 1 doubleword and set the counter.  */
+	mr	10,3
+	mtctr	8
+	bf	31,4f
+	ld	6,0(12)
+	addi	11,12,8
+	std	6,0(3)
+	addi	10,3,8
+
+	/* Main aligned copy loop. Copies 32-bytes at a time.  */
+	.align	4
+4:
+	ld	6,0(11)
+	ld	7,8(11)
+	ld	8,16(11)
+	ld	0,24(11)
+	addi	11,11,32
+
+	std	6,0(10)
+	std	7,8(10)
+	std	8,16(10)
+	std	0,24(10)
+	addi	10,10,32
+	bdnz	4b
+3:
+
+	/* Check for tail bytes.  */
+	rldicr	0,31,0,60
+	mtcrf	0x01,31
+	beq	cr6,0f
+
+.L9:
+	add	3,3,0
+	add	12,12,0
+
+	/*  At this point we have a tail of 0-7 bytes and we know that the
+	destination is doubleword-aligned.  */
+4:	/* Copy 4 bytes.  */
+	bf	29,2f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+2:	/* Copy 2 bytes.  */
+	bf	30,1f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+1:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,0(12)
+	stb	6,0(3)
+0:	/* Return DST + LEN pointer.  */
+	ld	31,-8(1)
+	ld	3,-16(1)
+	add	3,3,5
+	blr
+
+	/* Handle copies of 0~31 bytes.  */
+	.align	4
+L(copy_LT_32):
+	cmpldi	cr6,5,8
+	mr	12,4
+	mtcrf	0x01,5
+	ble	cr6,L(copy_LE_8)
+
+	/* At least 9 bytes to go.  */
+	neg	8,4
+	clrrdi	11,4,2
+	andi.	0,8,3
+	cmpldi	cr1,5,16
+	mr	10,5
+	beq	L(copy_LT_32_aligned)
+
+	/* Force 4-bytes alignment for SRC.  */
+	mtocrf  0x01,0
+	subf	10,0,5
+2:	bf	30,1f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+1:	bf	31,L(end_4bytes_alignment)
+
+	lbz	6,0(12)
+	addi	12,12,1
+	stb	6,0(3)
+	addi	3,3,1
+
+	.align	4
+L(end_4bytes_alignment):
+	cmpldi	cr1,10,16
+	mtcrf	0x01,10
+
+L(copy_LT_32_aligned):
+	/* At least 6 bytes to go, and SRC is word-aligned.  */
+	blt	cr1,8f
+
+	/* Copy 16 bytes.  */
+	lwz	6,0(12)
+	lwz	7,4(12)
+	stw	6,0(3)
+	lwz	8,8(12)
+	stw	7,4(3)
+	lwz	6,12(12)
+	addi	12,12,16
+	stw	8,8(3)
+	stw	6,12(3)
+	addi	3,3,16
+8:	/* Copy 8 bytes.  */
+	bf	28,4f
+
+	lwz	6,0(12)
+	lwz	7,4(12)
+	addi	12,12,8
+	stw	6,0(3)
+	stw	7,4(3)
+	addi	3,3,8
+4:	/* Copy 4 bytes.  */
+	bf	29,2f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+2:	/* Copy 2-3 bytes.  */
+	bf	30,1f
+
+	lhz	6,0(12)
+	sth	6,0(3)
+	bf	31,0f
+	lbz	7,2(12)
+	stb	7,2(3)
+	ld	3,-16(1)
+	add	3,3,5
+	blr
+
+	.align	4
+1:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,0(12)
+	stb	6,0(3)
+0:	/* Return DST + LEN pointer.  */
+	ld	3,-16(1)
+	add	3,3,5
+	blr
+
+	/* Handles copies of 0~8 bytes.  */
+	.align	4
+L(copy_LE_8):
+	bne	cr6,4f
+
+	/* Though we could've used ld/std here, they are still
+	slow for unaligned cases.  */
+
+	lwz	6,0(4)
+	lwz	7,4(4)
+	stw	6,0(3)
+	stw	7,4(3)
+	ld	3,-16(1)      /* Return DST + LEN pointer.  */
+	add	3,3,5
+	blr
+
+	.align	4
+4:	/* Copies 4~7 bytes.  */
+	bf	29,2b
+
+	lwz	6,0(4)
+	stw	6,0(3)
+	bf	30,5f
+	lhz	7,4(4)
+	sth	7,4(3)
+	bf	31,0f
+	lbz	8,6(4)
+	stb	8,6(3)
+	ld	3,-16(1)
+	add	3,3,5
+	blr
+
+	.align	4
+5:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,4(4)
+	stb	6,4(3)
+
+0:	/* Return DST + LEN pointer.  */
+	ld	3,-16(1)
+	add	3,3,5
+	blr
+
+	/* Handle copies of 32+ bytes where DST is aligned (to quadword) but
+	SRC is not.  Use aligned quadword loads from SRC, shifted to realign
+	the data, allowing for aligned DST stores.  */
+	.align	4
+L(copy_GE_32_unaligned):
+	clrldi	0,0,60	      /* Number of bytes until the 1st
+				 quadword.  */
+	andi.	11,3,15	      /* Check alignment of DST (against
+				 quadwords).  */
+	srdi	9,5,4	      /* Number of full quadwords remaining.  */
+
+	beq	L(copy_GE_32_unaligned_cont)
+
+	/* SRC is not quadword aligned, get it aligned.  */
+
+	mtcrf	0x01,0
+	subf	31,0,5
+
+	/* Vector instructions work best when proper alignment (16-bytes)
+	is present.  Move 0~15 bytes as needed to get DST quadword-aligned.  */
+1:	/* Copy 1 byte.  */
+	bf	31,2f
+
+	lbz	6,0(12)
+	addi	12,12,1
+	stb	6,0(3)
+	addi	3,3,1
+2:	/* Copy 2 bytes.  */
+	bf	30,4f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+4:	/* Copy 4 bytes.  */
+	bf	29,8f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+8:	/* Copy 8 bytes.  */
+	bf	28,0f
+
+	ld	6,0(12)
+	addi	12,12,8
+	std	6,0(3)
+	addi	3,3,8
+0:
+	clrldi	10,12,60      /* Check alignment of SRC.  */
+	srdi	9,31,4	      /* Number of full quadwords remaining.  */
+
+	/* The proper alignment is present, it is OK to copy the bytes now.  */
+L(copy_GE_32_unaligned_cont):
+
+	/* Setup two indexes to speed up the indexed vector operations.  */
+	clrldi	11,31,60
+	li	6,16	      /* Index for 16-bytes offsets.  */
+	li	7,32	      /* Index for 32-bytes offsets.  */
+	cmpldi	cr1,11,0
+	srdi	8,31,5	      /* Setup the loop counter.  */
+	mr	10,3
+	mr	11,12
+	mtcrf	0x01,9
+	cmpldi	cr6,9,1
+	lvsl	5,0,12
+	lvx	3,0,12
+	bf	31,L(setup_unaligned_loop)
+
+	/* Copy another 16 bytes to align to 32-bytes due to the loop .  */
+	lvx	4,12,6
+	vperm	6,3,4,5
+	addi	11,12,16
+	addi	10,3,16
+	stvx	6,0,3
+	vor	3,4,4
+
+L(setup_unaligned_loop):
+	mtctr	8
+	ble	cr6,L(end_unaligned_loop)
+
+	/* Copy 32 bytes at a time using vector instructions.  */
+	.align	4
+L(unaligned_loop):
+
+	/* Note: vr6/vr10 may contain data that was already copied,
+	but in order to get proper alignment, we may have to copy
+	some portions again. This is faster than having unaligned
+	vector instructions though.  */
+
+	lvx	4,11,6	      /* vr4 = r11+16.  */
+	vperm	6,3,4,5	      /* Merge the correctly-aligned portions
+				 of vr3/vr4 into vr6.  */
+	lvx	3,11,7	      /* vr3 = r11+32.  */
+	vperm	10,4,3,5      /* Merge the correctly-aligned portions
+				 of vr3/vr4 into vr10.  */
+	addi	11,11,32
+	stvx	6,0,10
+	stvx	10,10,6
+	addi	10,10,32
+
+	bdnz	L(unaligned_loop)
+
+	.align	4
+L(end_unaligned_loop):
+
+	/* Check for tail bytes.  */
+	rldicr	0,31,0,59
+	mtcrf	0x01,31
+	beq	cr1,0f
+
+	add	3,3,0
+	add	12,12,0
+
+	/*  We have 1~15 tail bytes to copy, and DST is quadword aligned.  */
+8:	/* Copy 8 bytes.  */
+	bf	28,4f
+
+	lwz	6,0(12)
+	lwz	7,4(12)
+	addi	12,12,8
+	stw	6,0(3)
+	stw	7,4(3)
+	addi	3,3,8
+4:	/* Copy 4 bytes.  */
+	bf	29,2f
+
+	lwz	6,0(12)
+	addi	12,12,4
+	stw	6,0(3)
+	addi	3,3,4
+2:	/* Copy 2~3 bytes.  */
+	bf	30,1f
+
+	lhz	6,0(12)
+	addi	12,12,2
+	sth	6,0(3)
+	addi	3,3,2
+1:	/* Copy 1 byte.  */
+	bf	31,0f
+
+	lbz	6,0(12)
+	stb	6,0(3)
+0:	/* Return DST + LEN pointer.  */
+	ld	31,-8(1)
+	ld	3,-16(1)
+	add	3,3,5
+	blr
+
+END_GEN_TB (BP_SYM (__mempcpy),TB_TOCLESS)
+libc_hidden_def (BP_SYM (__mempcpy))
+weak_alias (BP_SYM (__mempcpy), BP_SYM (mempcpy))
+libc_hidden_builtin_def (mempcpy)
diff --git a/libc/sysdeps/unix/sysv/linux/powerpc/powerpc32/a2/Implies b/libc/sysdeps/unix/sysv/linux/powerpc/powerpc32/a2/Implies
new file mode 100644
index 000000000..6d72414e5
--- /dev/null
+++ b/libc/sysdeps/unix/sysv/linux/powerpc/powerpc32/a2/Implies
@@ -0,0 +1,2 @@
+powerpc/powerpc32/a2/fpu
+powerpc/powerpc32/a2
diff --git a/libc/sysdeps/unix/sysv/linux/powerpc/powerpc64/a2/Implies b/libc/sysdeps/unix/sysv/linux/powerpc/powerpc64/a2/Implies
new file mode 100644
index 000000000..39b19e9c1
--- /dev/null
+++ b/libc/sysdeps/unix/sysv/linux/powerpc/powerpc64/a2/Implies
@@ -0,0 +1,2 @@
+powerpc/powerpc64/a2/fpu
+powerpc/powerpc64/a2
diff --git a/libc/sysdeps/unix/sysv/linux/ttyname.c b/libc/sysdeps/unix/sysv/linux/ttyname.c
index 69af6adc6..6cec3a901 100644
--- a/libc/sysdeps/unix/sysv/linux/ttyname.c
+++ b/libc/sysdeps/unix/sysv/linux/ttyname.c
@@ -1,4 +1,5 @@
-/* Copyright (C) 1991,92,93,1996-2002,2006,2009 Free Software Foundation, Inc.
+/* Copyright (C) 1991-1993,1996-2002,2006,2009,2010
+   Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -131,6 +132,9 @@ ttyname (int fd)
   if (__builtin_expect (__tcgetattr (fd, &term) < 0, 0))
     return NULL;
 
+  if (__fxstat64 (_STAT_VER, fd, &st) < 0)
+    return NULL;
+
   /* We try using the /proc filesystem.  */
   *_fitoa_word (fd, __stpcpy (procname, "/proc/self/fd/"), 10, 0) = '\0';
 
@@ -161,13 +165,32 @@ ttyname (int fd)
     {
       if ((size_t) len >= buflen)
 	return NULL;
+
+#define UNREACHABLE_LEN strlen ("(unreachable)")
+      if (len > UNREACHABLE_LEN
+	  && memcmp (ttyname_buf, "(unreachable)", UNREACHABLE_LEN) == 0)
+	{
+	  memmove (ttyname_buf, ttyname_buf + UNREACHABLE_LEN,
+		   len - UNREACHABLE_LEN);
+	  len -= UNREACHABLE_LEN;
+	}
+
       /* readlink need not terminate the string.  */
       ttyname_buf[len] = '\0';
-      return ttyname_buf;
-    }
 
-  if (__fxstat64 (_STAT_VER, fd, &st) < 0)
-    return NULL;
+      /* Verify readlink result, fall back on iterating through devices.  */
+      if (ttyname_buf[0] == '/'
+	  && __xstat64 (_STAT_VER, ttyname_buf, &st1) == 0
+#ifdef _STATBUF_ST_RDEV
+	  && S_ISCHR (st1.st_mode)
+	  && st1.st_rdev == st.st_rdev
+#else
+	  && st1.st_ino == st.st_ino
+	  && st1.st_dev == st.st_dev
+#endif
+	  )
+	return ttyname_buf;
+    }
 
   if (__xstat64 (_STAT_VER, "/dev/pts", &st1) == 0 && S_ISDIR (st1.st_mode))
     {
diff --git a/libc/sysdeps/unix/sysv/linux/ttyname_r.c b/libc/sysdeps/unix/sysv/linux/ttyname_r.c
index cef8624dc..2fa750347 100644
--- a/libc/sysdeps/unix/sysv/linux/ttyname_r.c
+++ b/libc/sysdeps/unix/sysv/linux/ttyname_r.c
@@ -1,4 +1,5 @@
-/* Copyright (C) 1991,92,93,1995-2001,2003,2006 Free Software Foundation, Inc.
+/* Copyright (C) 1991-1993,1995-2001,2003,2006,2010
+   Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -122,6 +123,9 @@ __ttyname_r (int fd, char *buf, size_t buflen)
   if (__builtin_expect (__tcgetattr (fd, &term) < 0, 0))
     return errno;
 
+  if (__fxstat64 (_STAT_VER, fd, &st) < 0)
+    return errno;
+
   /* We try using the /proc filesystem.  */
   *_fitoa_word (fd, __stpcpy (procname, "/proc/self/fd/"), 10, 0) = '\0';
 
@@ -145,12 +149,30 @@ __ttyname_r (int fd, char *buf, size_t buflen)
 #endif
 			, 1))
     {
+#define UNREACHABLE_LEN strlen ("(unreachable)")
+      if (ret > UNREACHABLE_LEN
+	  && memcmp (buf, "(unreachable)", UNREACHABLE_LEN) == 0)
+	{
+	  memmove (buf, buf + UNREACHABLE_LEN, ret - UNREACHABLE_LEN);
+	  ret -= UNREACHABLE_LEN;
+	}
+
+      /* readlink need not terminate the string.  */
       buf[ret] = '\0';
-      return 0;
-    }
 
-  if (__fxstat64 (_STAT_VER, fd, &st) < 0)
-    return errno;
+      /* Verify readlink result, fall back on iterating through devices.  */
+      if (buf[0] == '/'
+	  && __xstat64 (_STAT_VER, buf, &st1) == 0
+#ifdef _STATBUF_ST_RDEV
+	  && S_ISCHR (st1.st_mode)
+	  && st1.st_rdev == st.st_rdev
+#else
+	  && st1.st_ino == st.st_ino
+	  && st1.st_dev == st.st_dev
+#endif
+	  )
+	return 0;
+    }
 
   /* Prepare the result buffer.  */
   memcpy (buf, "/dev/pts/", sizeof ("/dev/pts/"));
diff --git a/libc/sysdeps/x86_64/cacheinfo.c b/libc/sysdeps/x86_64/cacheinfo.c
index 54220379e..eae54e725 100644
--- a/libc/sysdeps/x86_64/cacheinfo.c
+++ b/libc/sysdeps/x86_64/cacheinfo.c
@@ -455,13 +455,21 @@ __cache_sysconf (int name)
 
 
 /* Data cache size for use in memory and string routines, typically
-   L1 size.  */
+   L1 size, rounded to multiple of 256 bytes.  */
 long int __x86_64_data_cache_size_half attribute_hidden = 32 * 1024 / 2;
 long int __x86_64_data_cache_size attribute_hidden = 32 * 1024;
+/* Similar to __x86_64_data_cache_size_half, but not rounded.  */
+long int __x86_64_raw_data_cache_size_half attribute_hidden = 32 * 1024 / 2;
+/* Similar to __x86_64_data_cache_size, but not rounded.  */
+long int __x86_64_raw_data_cache_size attribute_hidden = 32 * 1024;
 /* Shared cache size for use in memory and string routines, typically
-   L2 or L3 size.  */
+   L2 or L3 size, rounded to multiple of 256 bytes.  */
 long int __x86_64_shared_cache_size_half attribute_hidden = 1024 * 1024 / 2;
 long int __x86_64_shared_cache_size attribute_hidden = 1024 * 1024;
+/* Similar to __x86_64_shared_cache_size_half, but not rounded.  */
+long int __x86_64_raw_shared_cache_size_half attribute_hidden = 1024 * 1024 / 2;
+/* Similar to __x86_64_shared_cache_size, but not rounded.  */
+long int __x86_64_raw_shared_cache_size attribute_hidden = 1024 * 1024;
 
 #ifndef DISABLE_PREFETCHW
 /* PREFETCHW support flag for use in memory and string routines.  */
@@ -661,12 +669,20 @@ init_cacheinfo (void)
 
   if (data > 0)
     {
+      __x86_64_raw_data_cache_size_half = data / 2;
+      __x86_64_raw_data_cache_size = data;
+      /* Round data cache size to multiple of 256 bytes.  */
+      data = data & ~255L;
       __x86_64_data_cache_size_half = data / 2;
       __x86_64_data_cache_size = data;
     }
 
   if (shared > 0)
     {
+      __x86_64_raw_shared_cache_size_half = shared / 2;
+      __x86_64_raw_shared_cache_size = shared;
+      /* Round shared cache size to multiple of 256 bytes.  */
+      shared = shared & ~255L;
       __x86_64_shared_cache_size_half = shared / 2;
       __x86_64_shared_cache_size = shared;
     }
diff --git a/libc/sysdeps/x86_64/multiarch/strchr.S b/libc/sysdeps/x86_64/multiarch/strchr.S
index 27eead985..71845a35f 100644
--- a/libc/sysdeps/x86_64/multiarch/strchr.S
+++ b/libc/sysdeps/x86_64/multiarch/strchr.S
@@ -1,5 +1,5 @@
 /* strchr with SSE4.2
-   Copyright (C) 2009 Free Software Foundation, Inc.
+   Copyright (C) 2009, 2010 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -87,13 +87,13 @@ __strchr_sse42:
 	pxor	%xmm2, %xmm2
 	movd	%esi, %xmm1
 	movl	%edi, %ecx
+	pshufb  %xmm2, %xmm1
 	andl	$15, %ecx
 	movq	%rdi, %r8
 	je	L(aligned_start)
 
 /* Handle unaligned string.  */
 	andq	$-16, %r8
-	pshufb  %xmm2, %xmm1
 	movdqa	(%r8), %xmm0
 	pcmpeqb	 %xmm0, %xmm2
 	pcmpeqb	 %xmm1, %xmm0