Updated to fedora-glibc-20090407T0545

19 files changed, 608 insertions, 476 deletions

diff --git a/sysdeps/posix/preadv.c b/sysdeps/posix/preadv.c
new file mode 100644
index 0000000000..e697604c19
--- /dev/null
+++ b/sysdeps/posix/preadv.c
@@ -0,0 +1,108 @@
+/* Copyright (C) 2009 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
+   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 <errno.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+#include <limits.h>
+#include <stdbool.h>
+#include <sys/param.h>
+#if __WORDSIZE == 64 && !defined PREADV
+/* Hide the preadv64 declaration.  */
+# define preadv64 __redirect_preadv64
+#endif
+#include <sys/uio.h>
+#include <bits/wordsize.h>
+
+#ifndef PREADV
+# define PREADV preadv
+# define PREAD __pread
+# define OFF_T off_t
+#endif
+
+
+static void
+ifree (char **ptrp)
+{
+  free (*ptrp);
+}
+
+
+/* Read data from file descriptor FD at the given position OFFSET
+   without change the file pointer, and put the result in the buffers
+   described by VECTOR, which is a vector of COUNT 'struct iovec's.
+   The buffers are filled in the order specified.  Operates just like
+   'read' (see <unistd.h>) except that data are put in VECTOR instead
+   of a contiguous buffer.  */
+ssize_t
+PREADV (int fd, const struct iovec *vector, int count, OFF_T offset)
+{
+  /* Find the total number of bytes to be read.  */
+  size_t bytes = 0;
+  for (int i = 0; i < count; ++i)
+    {
+      /* Check for ssize_t overflow.  */
+      if (SSIZE_MAX - bytes < vector[i].iov_len)
+	{
+	  __set_errno (EINVAL);
+	  return -1;
+	}
+      bytes += vector[i].iov_len;
+    }
+
+  /* Allocate a temporary buffer to hold the data.  We should normally
+     use alloca since it's faster and does not require synchronization
+     with other threads.  But we cannot if the amount of memory
+     required is too large.  */
+  char *buffer;
+  char *malloced_buffer __attribute__ ((__cleanup__ (ifree))) = NULL;
+  if (__libc_use_alloca (bytes))
+    buffer = (char *) __alloca (bytes);
+  else
+    {
+      malloced_buffer = buffer = (char *) malloc (bytes);
+      if (buffer == NULL)
+	return -1;
+    }
+
+  /* Read the data.  */
+  ssize_t bytes_read = PREAD (fd, buffer, bytes, offset);
+  if (bytes_read <= 0)
+    return -1;
+
+  /* Copy the data from BUFFER into the memory specified by VECTOR.  */
+  bytes = bytes_read;
+  for (int i = 0; i < count; ++i)
+    {
+      size_t copy = MIN (vector[i].iov_len, bytes);
+
+      (void) memcpy ((void *) vector[i].iov_base, (void *) buffer, copy);
+
+      buffer += copy;
+      bytes -= copy;
+      if (bytes == 0)
+	break;
+    }
+
+  return bytes_read;
+}
+#if __WORDSIZE == 64 && defined preadv64
+# undef preadv64
+strong_alias (preadv, preadv64)
+#endif
diff --git a/sysdeps/posix/preadv64.c b/sysdeps/posix/preadv64.c
new file mode 100644
index 0000000000..198622353a
--- /dev/null
+++ b/sysdeps/posix/preadv64.c
@@ -0,0 +1,9 @@
+#include <bits/wordsize.h>
+
+#if __WORDSIZE == 32
+# define PREADV preadv64
+# define PREAD __pread64
+# define OFF_T off64_t
+
+# include "preadv.c"
+#endif
diff --git a/sysdeps/posix/pwritev.c b/sysdeps/posix/pwritev.c
new file mode 100644
index 0000000000..0b6627dc9d
--- /dev/null
+++ b/sysdeps/posix/pwritev.c
@@ -0,0 +1,108 @@
+/* Copyright (C) 2009 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
+   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 <errno.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+#include <limits.h>
+#include <stdbool.h>
+#include <sys/param.h>
+#if __WORDSIZE == 64 && !defined PWRITEV
+/* Hide the pwritev64 declaration.  */
+# define pwritev64 __redirect_pwritev64
+#endif
+#include <sys/uio.h>
+#include <bits/wordsize.h>
+
+#ifndef PWRITEV
+# define PWRITEV pwritev
+# define PWRITE __pwrite
+# define OFF_T off_t
+#endif
+
+
+static void
+ifree (char **ptrp)
+{
+  free (*ptrp);
+}
+
+
+/* Read data from file descriptor FD at the given position OFFSET
+   without change the file pointer, and put the result in the buffers
+   described by VECTOR, which is a vector of COUNT 'struct iovec's.
+   The buffers are filled in the order specified.  Operates just like
+   'read' (see <unistd.h>) except that data are put in VECTOR instead
+   of a contiguous buffer.  */
+ssize_t
+PWRITEV (int fd, const struct iovec *vector, int count, OFF_T offset)
+{
+  /* Find the total number of bytes to be read.  */
+  size_t bytes = 0;
+  for (int i = 0; i < count; ++i)
+    {
+      /* Check for ssize_t overflow.  */
+      if (SSIZE_MAX - bytes < vector[i].iov_len)
+	{
+	  __set_errno (EINVAL);
+	  return -1;
+	}
+      bytes += vector[i].iov_len;
+    }
+
+  /* Allocate a temporary buffer to hold the data.  We should normally
+     use alloca since it's faster and does not require synchronization
+     with other threads.  But we cannot if the amount of memory
+     required is too large.  */
+  char *buffer;
+  char *malloced_buffer __attribute__ ((__cleanup__ (ifree))) = NULL;
+  if (__libc_use_alloca (bytes))
+    buffer = (char *) __alloca (bytes);
+  else
+    {
+      malloced_buffer = buffer = (char *) malloc (bytes);
+      if (buffer == NULL)
+	return -1;
+    }
+
+  /* Read the data.  */
+  ssize_t bytes_read = PWRITE (fd, buffer, bytes, offset);
+  if (bytes_read <= 0)
+    return -1;
+
+  /* Copy the data from BUFFER into the memory specified by VECTOR.  */
+  bytes = bytes_read;
+  for (int i = 0; i < count; ++i)
+    {
+      size_t copy = MIN (vector[i].iov_len, bytes);
+
+      (void) memcpy ((void *) vector[i].iov_base, (void *) buffer, copy);
+
+      buffer += copy;
+      bytes -= copy;
+      if (bytes == 0)
+	break;
+    }
+
+  return bytes_read;
+}
+#if __WORDSIZE == 64 && defined pwritev64
+# undef pwritev64
+strong_alias (pwritev, pwritev64)
+#endif
diff --git a/sysdeps/posix/pwritev64.c b/sysdeps/posix/pwritev64.c
new file mode 100644
index 0000000000..4948d2efee
--- /dev/null
+++ b/sysdeps/posix/pwritev64.c
@@ -0,0 +1,9 @@
+#include <bits/wordsize.h>
+
+#if __WORDSIZE == 32
+# define PWRITEV pwritev64
+# define PWRITE __pwrite64
+# define OFF_T off64_t
+
+# include "pwritev.c"
+#endif
diff --git a/sysdeps/posix/readv.c b/sysdeps/posix/readv.c
index f0e78e6662..50bcc91315 100644
--- a/sysdeps/posix/readv.c
+++ b/sysdeps/posix/readv.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 1991, 1992, 1996, 1997, 2002 Free Software Foundation, Inc.
+/* Copyright (C) 1991,1992,1996,1997,2002,2009 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
@@ -25,24 +25,24 @@
 #include <sys/uio.h>
 #include <errno.h>
 
+
+static void
+ifree (char **ptrp)
+{
+  free (*ptrp);
+}
+
 /* Read data from file descriptor FD, and put the result in the
-   buffers described by VECTOR, which is a vector of COUNT `struct iovec's.
+   buffers described by VECTOR, which is a vector of COUNT 'struct iovec's.
    The buffers are filled in the order specified.
-   Operates just like `read' (see <unistd.h>) except that data are
+   Operates just like 'read' (see <unistd.h>) except that data are
    put in VECTOR instead of a contiguous buffer.  */
 ssize_t
 __libc_readv (int fd, const struct iovec *vector, int count)
 {
-  char *buffer;
-  char *buffer_start;
-  size_t bytes;
-  ssize_t bytes_read;
-  int i;
-  bool use_malloc = false;
-
   /* Find the total number of bytes to be read.  */
-  bytes = 0;
-  for (i = 0; i < count; ++i)
+  size_t bytes = 0;
+  for (int i = 0; i < count; ++i)
     {
       /* Check for ssize_t overflow.  */
       if (SSIZE_MAX - bytes < vector[i].iov_len)
@@ -57,28 +57,25 @@ __libc_readv (int fd, const struct iovec *vector, int count)
      use alloca since it's faster and does not require synchronization
      with other threads.  But we cannot if the amount of memory
      required is too large.  */
+  char *buffer;
+  char *malloced_buffer __attribute__ ((__cleanup__ (ifree))) = NULL;
   if (__libc_use_alloca (bytes))
     buffer = (char *) __alloca (bytes);
   else
     {
-      buffer = (char *) malloc (bytes);
+      malloced_buffer = buffer = (char *) malloc (bytes);
       if (buffer == NULL)
-	/* XXX I don't know whether it is acceptable to try reading
-	   the data in chunks.  Probably not so we just fail here.  */
 	return -1;
-
-      use_malloc = true;
     }
 
   /* Read the data.  */
-  bytes_read = __read (fd, buffer, bytes);
+  ssize_t bytes_read = __read (fd, buffer, bytes);
   if (bytes_read <= 0)
     return -1;
 
   /* Copy the data from BUFFER into the memory specified by VECTOR.  */
   bytes = bytes_read;
-  buffer_start = buffer;
-  for (i = 0; i < count; ++i)
+  for (int i = 0; i < count; ++i)
     {
       size_t copy = MIN (vector[i].iov_len, bytes);
 
@@ -90,9 +87,6 @@ __libc_readv (int fd, const struct iovec *vector, int count)
 	break;
     }
 
-  if (use_malloc)
-    free (buffer_start);
-
   return bytes_read;
 }
 #ifndef __libc_readv
diff --git a/sysdeps/posix/writev.c b/sysdeps/posix/writev.c
index a347cc2eac..203cf9cebd 100644
--- a/sysdeps/posix/writev.c
+++ b/sysdeps/posix/writev.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 1991, 1992, 1996, 1997, 2002 Free Software Foundation, Inc.
+/* Copyright (C) 1991,1992,1996,1997,2002,2009 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
@@ -25,24 +25,25 @@
 #include <sys/uio.h>
 #include <errno.h>
 
+
+static void
+ifree (char **ptrp)
+{
+  free (*ptrp);
+}
+
+
 /* Write data pointed by the buffers described by VECTOR, which
-   is a vector of COUNT `struct iovec's, to file descriptor FD.
+   is a vector of COUNT 'struct iovec's, to file descriptor FD.
    The data is written in the order specified.
-   Operates just like `write' (see <unistd.h>) except that the data
+   Operates just like 'write' (see <unistd.h>) except that the data
    are taken from VECTOR instead of a contiguous buffer.  */
 ssize_t
 __libc_writev (int fd, const struct iovec *vector, int count)
 {
-  char *buffer;
-  register char *bp;
-  size_t bytes, to_copy;
-  ssize_t bytes_written;
-  int i;
-  bool use_malloc = false;
-
   /* Find the total number of bytes to be written.  */
-  bytes = 0;
-  for (i = 0; i < count; ++i)
+  size_t bytes = 0;
+  for (int i = 0; i < count; ++i)
     {
       /* Check for ssize_t overflow.  */
       if (SSIZE_MAX - bytes < vector[i].iov_len)
@@ -57,23 +58,23 @@ __libc_writev (int fd, const struct iovec *vector, int count)
      use alloca since it's faster and does not require synchronization
      with other threads.  But we cannot if the amount of memory
      required is too large.  */
+  char *buffer;
+  char *malloced_buffer __attribute__ ((__cleanup__ (ifree))) = NULL;
   if (__libc_use_alloca (bytes))
     buffer = (char *) __alloca (bytes);
   else
     {
-      buffer = (char *) malloc (bytes);
+      malloced_buffer = buffer = (char *) malloc (bytes);
       if (buffer == NULL)
 	/* XXX I don't know whether it is acceptable to try writing
 	   the data in chunks.  Probably not so we just fail here.  */
 	return -1;
-
-      use_malloc = true;
     }
 
   /* Copy the data into BUFFER.  */
-  to_copy = bytes;
-  bp = buffer;
-  for (i = 0; i < count; ++i)
+  size_t to_copy = bytes;
+  char *bp = buffer;
+  for (int i = 0; i < count; ++i)
     {
       size_t copy = MIN (vector[i].iov_len, to_copy);
 
@@ -84,10 +85,7 @@ __libc_writev (int fd, const struct iovec *vector, int count)
 	break;
     }
 
-  bytes_written = __write (fd, buffer, bytes);
-
-  if (use_malloc)
-    free (buffer);
+  ssize_t bytes_written = __write (fd, buffer, bytes);
 
   return bytes_written;
 }
diff --git a/sysdeps/unix/sysv/linux/bits/socket.h b/sysdeps/unix/sysv/linux/bits/socket.h
index 72c7335ae5..88062e59ad 100644
--- a/sysdeps/unix/sysv/linux/bits/socket.h
+++ b/sysdeps/unix/sysv/linux/bits/socket.h
@@ -95,15 +95,20 @@ enum __socket_type
 #define	PF_ASH		18	/* Ash.  */
 #define	PF_ECONET	19	/* Acorn Econet.  */
 #define	PF_ATMSVC	20	/* ATM SVCs.  */
+#define PF_RDS		21	/* RDS sockets.  */
 #define	PF_SNA		22	/* Linux SNA Project */
 #define	PF_IRDA		23	/* IRDA sockets.  */
 #define	PF_PPPOX	24	/* PPPoX sockets.  */
 #define	PF_WANPIPE	25	/* Wanpipe API sockets.  */
+#define PF_LLC		26	/* Linux LLC.  */
+#define PF_CAN		29	/* Controller Area Network.  */
+#define PF_TIPC		30	/* TIPC sockets.  */
 #define	PF_BLUETOOTH	31	/* Bluetooth sockets.  */
 #define	PF_IUCV		32	/* IUCV sockets.  */
 #define PF_RXRPC	33	/* RxRPC sockets.  */
 #define PF_ISDN		34	/* mISDN sockets.  */
-#define	PF_MAX		35	/* For now..  */
+#define PF_PHONET	35	/* Phonet sockets.  */
+#define	PF_MAX		36	/* For now..  */
 
 /* Address families.  */
 #define	AF_UNSPEC	PF_UNSPEC
@@ -130,14 +135,19 @@ enum __socket_type
 #define	AF_ASH		PF_ASH
 #define	AF_ECONET	PF_ECONET
 #define	AF_ATMSVC	PF_ATMSVC
+#define AF_RDS		PF_RDS
 #define	AF_SNA		PF_SNA
 #define	AF_IRDA		PF_IRDA
 #define	AF_PPPOX	PF_PPPOX
 #define	AF_WANPIPE	PF_WANPIPE
+#define AF_LLC		PF_LLC
+#define AF_CAN		PF_CAN
+#define AF_TIPC		PF_TIPC
 #define	AF_BLUETOOTH	PF_BLUETOOTH
 #define	AF_IUCV		PF_IUCV
 #define AF_RXRPC	PF_RXRPC
 #define AF_ISDN		PF_ISDN
+#define AF_PHONET	PF_PHONET
 #define	AF_MAX		PF_MAX
 
 /* Socket level values.  Others are defined in the appropriate headers.
diff --git a/sysdeps/unix/sysv/linux/kernel-features.h b/sysdeps/unix/sysv/linux/kernel-features.h
index 29afe27711..a640765b5c 100644
--- a/sysdeps/unix/sysv/linux/kernel-features.h
+++ b/sysdeps/unix/sysv/linux/kernel-features.h
@@ -462,6 +462,13 @@
 # define __ASSUME_SET_ROBUST_LIST	1
 #endif
 
+/* Pessimistically assume that 2.6.18 introduced real handling of
+   large numbers of requests to readv and writev and that we don't
+   need a fallback.  It likely worked for much longer.  */
+#if __LINUX_KERNEL_VERSION >= 0x020612
+# define __ASSUME_COMPLETE_READV_WRITEV	1
+#endif
+
 /* Support for PI futexes was added in 2.6.18.  */
 #if __LINUX_KERNEL_VERSION >= 0x020612
 # define __ASSUME_FUTEX_LOCK_PI	1
@@ -522,3 +529,10 @@
 #if __LINUX_KERNEL_VERSION >= 0x02061d
 # define __ASSUME_FUTEX_CLOCK_REALTIME	1
 #endif
+
+/* Support for preadv and pwritev was added in 2.6.30.  */
+#if __LINUX_KERNEL_VERSION >= 0x02061e \
+    && (defined __i386__ || defined __x86_64__)
+# define __ASSUME_PREADV	1
+# define __ASSUME_PWRITEV	1
+#endif
diff --git a/sysdeps/unix/sysv/linux/preadv.c b/sysdeps/unix/sysv/linux/preadv.c
new file mode 100644
index 0000000000..e2f8238596
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/preadv.c
@@ -0,0 +1,88 @@
+/* Copyright (C) 2009 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
+   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 <errno.h>
+#include <stddef.h>
+#include <sys/param.h>
+#if __WORDSIZE == 64
+/* Hide the preadv64 declaration.  */
+# define preadv64 __redirect_preadv64
+#endif
+#include <sys/uio.h>
+
+#include <sysdep-cancel.h>
+#include <sys/syscall.h>
+#include <kernel-features.h>
+
+#ifndef PREADV
+# define PREADV preadv
+# define PREADV_REPLACEMENT __atomic_preadv_replacement
+# define PREAD __pread
+# define OFF_T off_t
+#endif
+
+static ssize_t PREADV_REPLACEMENT (int, __const struct iovec *,
+				   int, OFF_T) internal_function;
+
+
+ssize_t
+PREADV (fd, vector, count, offset)
+     int fd;
+     const struct iovec *vector;
+     int count;
+     OFF_T offset;
+{
+#ifdef __NR_preadv
+  ssize_t result;
+
+  if (SINGLE_THREAD_P)
+    result = INLINE_SYSCALL (preadv, 5, fd, vector, count, offset >> 32,
+			     offset & 0xffffffff);
+  else
+    {
+      int oldtype = LIBC_CANCEL_ASYNC ();
+
+      result = INLINE_SYSCALL (preadv, 5, fd, vector, count, offset >> 32,
+			       offset & 0xffffffff);
+
+      LIBC_CANCEL_RESET (oldtype);
+    }
+# ifdef __ASSUME_PREADV
+  return result;
+# endif
+#endif
+
+#ifndef __ASSUME_PREADV
+# ifdef __NR_preadv
+  if (result >= 0 || errno != ENOSYS)
+    return result;
+# endif
+
+  return PREADV_REPLACEMENT (fd, vector, count, offset);
+#endif
+}
+#if __WORDSIZE == 64
+# undef preadv64
+strong_alias (preadv, preadv64)
+#endif
+
+#ifndef __ASSUME_PREADV
+# undef PREADV
+# define PREADV static internal_function PREADV_REPLACEMENT
+# include <sysdeps/posix/preadv.c>
+#endif
diff --git a/sysdeps/unix/sysv/linux/preadv64.c b/sysdeps/unix/sysv/linux/preadv64.c
new file mode 100644
index 0000000000..739df00257
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/preadv64.c
@@ -0,0 +1,6 @@
+#define PREADV preadv64
+#define PREADV_REPLACEMENT __atomic_preadv64_replacement
+#define PREAD __pread64
+#define OFF_T off64_t
+
+#include "preadv.c"
diff --git a/sysdeps/unix/sysv/linux/pwritev.c b/sysdeps/unix/sysv/linux/pwritev.c
new file mode 100644
index 0000000000..df430ffe46
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/pwritev.c
@@ -0,0 +1,88 @@
+/* Copyright (C) 2009 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
+   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 <errno.h>
+#include <stddef.h>
+#include <sys/param.h>
+#if __WORDSIZE == 64 && !defined PWRITEV
+/* Hide the pwritev64 declaration.  */
+# define pwritev64 __redirect_pwritev64
+#endif
+#include <sys/uio.h>
+
+#include <sysdep-cancel.h>
+#include <sys/syscall.h>
+#include <kernel-features.h>
+
+#ifndef PWRITEV
+# define PWRITEV pwritev
+# define PWRITEV_REPLACEMENT __atomic_pwritev_replacement
+# define PWRITE __pwrite
+# define OFF_T off_t
+#endif
+
+static ssize_t PWRITEV_REPLACEMENT (int, __const struct iovec *,
+				    int, OFF_T) internal_function;
+
+
+ssize_t
+PWRITEV (fd, vector, count, offset)
+     int fd;
+     const struct iovec *vector;
+     int count;
+     OFF_T offset;
+{
+#ifdef __NR_pwritev
+  ssize_t result;
+
+  if (SINGLE_THREAD_P)
+    result = INLINE_SYSCALL (pwritev, 5, fd, vector, count, offset >> 32,
+			     offset & 0xffffffff);
+  else
+    {
+      int oldtype = LIBC_CANCEL_ASYNC ();
+
+      result = INLINE_SYSCALL (pwritev, 5, fd, vector, count, offset >> 32,
+			       offset & 0xffffffff);
+
+      LIBC_CANCEL_RESET (oldtype);
+    }
+# ifdef __ASSUME_PWRITEV
+  return result;
+# endif
+#endif
+
+#ifndef __ASSUME_PWRITEV
+# ifdef __NR_pwritev
+  if (result >= 0 || errno != ENOSYS)
+    return result;
+# endif
+
+  return PWRITEV_REPLACEMENT (fd, vector, count, offset);
+#endif
+}
+#if __WORDSIZE == 64 && defined pwritev64
+# undef pwritev64
+strong_alias (pwritev, pwritev64)
+#endif
+
+#ifndef __ASSUME_PWRITEV
+# undef PWRITEV
+# define PWRITEV static internal_function PWRITEV_REPLACEMENT
+# include <sysdeps/posix/pwritev.c>
+#endif
diff --git a/sysdeps/unix/sysv/linux/pwritev64.c b/sysdeps/unix/sysv/linux/pwritev64.c
new file mode 100644
index 0000000000..1e8168f103
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/pwritev64.c
@@ -0,0 +1,6 @@
+#define PWRITEV pwritev64
+#define PWRITEV_REPLACEMENT __atomic_pwritev64_replacement
+#define PWRITE __pwrite64
+#define OFF_T off64_t
+
+#include "pwritev.c"
diff --git a/sysdeps/unix/sysv/linux/readv.c b/sysdeps/unix/sysv/linux/readv.c
index 250c00a075..bff4a3ff74 100644
--- a/sysdeps/unix/sysv/linux/readv.c
+++ b/sysdeps/unix/sysv/linux/readv.c
@@ -1,5 +1,5 @@
 /* readv supports all Linux kernels >= 2.0.
-   Copyright (C) 1997,1998,2000,2002,2003 Free Software Foundation, Inc.
+   Copyright (C) 1997,1998,2000,2002,2003,2009 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
@@ -25,6 +25,7 @@
 #include <sysdep-cancel.h>
 #include <sys/syscall.h>
 #include <bp-checks.h>
+#include <kernel-features.h>
 
 static ssize_t __atomic_readv_replacement (int, __const struct iovec *,
 					   int) internal_function;
@@ -36,41 +37,38 @@ static ssize_t __atomic_readv_replacement (int, __const struct iovec *,
 #endif
 
 
-/* We should deal with kernel which have a smaller UIO_FASTIOV as well
-   as a very big count.  */
-static ssize_t
-do_readv (int fd, const struct iovec *vector, int count)
-{
-  ssize_t bytes_read;
-
-  bytes_read = INLINE_SYSCALL (readv, 3, fd, CHECK_N (vector, count), count);
-
-  if (bytes_read >= 0 || errno != EINVAL || count <= UIO_FASTIOV)
-    return bytes_read;
-
-  return __atomic_readv_replacement (fd, vector, count);
-}
-
-
 ssize_t
 __libc_readv (fd, vector, count)
      int fd;
      const struct iovec *vector;
      int count;
 {
-  if (SINGLE_THREAD_P)
-    return do_readv (fd, vector, count);
+  ssize_t result;
 
-  int oldtype = LIBC_CANCEL_ASYNC ();
+  if (SINGLE_THREAD_P)
+    result = INLINE_SYSCALL (readv, 3, fd, CHECK_N (vector, count), count);
+  else
+    {
+      int oldtype = LIBC_CANCEL_ASYNC ();
 
-  int result = do_readv (fd, vector, count);
+      result = INLINE_SYSCALL (readv, 3, fd, CHECK_N (vector, count), count);
 
-  LIBC_CANCEL_RESET (oldtype);
+      LIBC_CANCEL_RESET (oldtype);
+    }
 
+#ifdef __ASSUME_COMPLETE_READV_WRITEV
   return result;
+#else
+  if (result >= 0 || errno != EINVAL || count <= UIO_FASTIOV)
+    return result;
+
+  return __atomic_readv_replacement (fd, vector, count);
+#endif
 }
 strong_alias (__libc_readv, __readv)
 weak_alias (__libc_readv, readv)
 
-#define __libc_readv static internal_function __atomic_readv_replacement
-#include <sysdeps/posix/readv.c>
+#ifndef __ASSUME_COMPLETE_READV_WRITEV
+# define __libc_readv static internal_function __atomic_readv_replacement
+# include <sysdeps/posix/readv.c>
+#endif
diff --git a/sysdeps/unix/sysv/linux/sys/eventfd.h b/sysdeps/unix/sysv/linux/sys/eventfd.h
index 7c6cf7196d..d942df4bf9 100644
--- a/sysdeps/unix/sysv/linux/sys/eventfd.h
+++ b/sysdeps/unix/sysv/linux/sys/eventfd.h
@@ -28,6 +28,8 @@ typedef uint64_t eventfd_t;
 /* Flags for signalfd.  */
 enum
   {
+    EFD_SEMAPHORE = 1,
+#define EFD_SEMAPHORE EFD_SEMAPHORE
     EFD_CLOEXEC = 02000000,
 #define EFD_CLOEXEC EFD_CLOEXEC
     EFD_NONBLOCK = 04000
diff --git a/sysdeps/unix/sysv/linux/wordsize-64/preadv64.c b/sysdeps/unix/sysv/linux/wordsize-64/preadv64.c
new file mode 100644
index 0000000000..fd9320cfc7
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/wordsize-64/preadv64.c
@@ -0,0 +1 @@
+/* Empty since the preadv syscall is equivalent.  */
diff --git a/sysdeps/unix/sysv/linux/wordsize-64/pwritev64.c b/sysdeps/unix/sysv/linux/wordsize-64/pwritev64.c
new file mode 100644
index 0000000000..8b72a2928b
--- /dev/null
+++ b/sysdeps/unix/sysv/linux/wordsize-64/pwritev64.c
@@ -0,0 +1 @@
+/* Empty since the pwritev syscall is equivalent.  */
diff --git a/sysdeps/unix/sysv/linux/writev.c b/sysdeps/unix/sysv/linux/writev.c
index 05978665fa..55e915d34d 100644
--- a/sysdeps/unix/sysv/linux/writev.c
+++ b/sysdeps/unix/sysv/linux/writev.c
@@ -1,5 +1,5 @@
 /* writev supports all Linux kernels >= 2.0.
-   Copyright (C) 1997, 1998, 2000, 2002, 2003 Free Software Foundation, Inc.
+   Copyright (C) 1997,1998,2000,2002,2003,2009 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
@@ -25,6 +25,7 @@
 #include <sysdep-cancel.h>
 #include <sys/syscall.h>
 #include <bp-checks.h>
+#include <kernel-features.h>
 
 static ssize_t __atomic_writev_replacement (int, const struct iovec *,
 					    int) internal_function;
@@ -36,40 +37,38 @@ static ssize_t __atomic_writev_replacement (int, const struct iovec *,
 #endif
 
 
-/* We should deal with kernel which have a smaller UIO_FASTIOV as well
-   as a very big count.  */
-static ssize_t
-do_writev (int fd, const struct iovec *vector, int count)
-{
-  ssize_t bytes_written;
-
-  bytes_written = INLINE_SYSCALL (writev, 3, fd, CHECK_N (vector, count), count);
-
-  if (bytes_written >= 0 || errno != EINVAL || count <= UIO_FASTIOV)
-    return bytes_written;
-
-  return __atomic_writev_replacement (fd, vector, count);
-}
-
 ssize_t
 __libc_writev (fd, vector, count)
      int fd;
      const struct iovec *vector;
      int count;
 {
-  if (SINGLE_THREAD_P)
-    return do_writev (fd, vector, count);
+  ssize_t result;
 
-  int oldtype = LIBC_CANCEL_ASYNC ();
+  if (SINGLE_THREAD_P)
+    result = INLINE_SYSCALL (writev, 3, fd, CHECK_N (vector, count), count);
+  else
+    {
+      int oldtype = LIBC_CANCEL_ASYNC ();
 
-  ssize_t result = do_writev (fd, vector, count);
+      result = INLINE_SYSCALL (writev, 3, fd, CHECK_N (vector, count), count);
 
-  LIBC_CANCEL_RESET (oldtype);
+      LIBC_CANCEL_RESET (oldtype);
+    }
 
+#ifdef __ASSUME_COMPLETE_READV_WRITEV
   return result;
+#else
+  if (result >= 0 || errno != EINVAL || count <= UIO_FASTIOV)
+    return result;
+
+  return __atomic_writev_replacement (fd, vector, count);
+#endif
 }
 strong_alias (__libc_writev, __writev)
 weak_alias (__libc_writev, writev)
 
-#define __libc_writev static internal_function __atomic_writev_replacement
-#include <sysdeps/posix/writev.c>
+#ifndef __ASSUME_COMPLETE_READV_WRITEV
+# define __libc_writev static internal_function __atomic_writev_replacement
+# include <sysdeps/posix/writev.c>
+#endif
diff --git a/sysdeps/x86_64/strchr.S b/sysdeps/x86_64/strchr.S
index 8934697972..8833cd0cc4 100644
--- a/sysdeps/x86_64/strchr.S
+++ b/sysdeps/x86_64/strchr.S
@@ -1,6 +1,6 @@
 /* strchr (str, ch) -- Return pointer to first occurrence of CH in STR.
    For AMD x86-64.
-   Copyright (C) 2002, 2005 Free Software Foundation, Inc.
+   Copyright (C) 2009 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
@@ -19,273 +19,52 @@
    02111-1307 USA.  */
 
 #include <sysdep.h>
-#include "asm-syntax.h"
-#include "bp-sym.h"
-#include "bp-asm.h"
 
 
 	.text
-ENTRY (BP_SYM (strchr))
-
-	/* Before we start with the main loop we process single bytes
-	   until the source pointer is aligned.  This has two reasons:
-	   1. aligned 64-bit memory access is faster
-	   and (more important)
-	   2. we process in the main loop 64 bit in one step although
-	      we don't know the end of the string.  But accessing at
-	      8-byte alignment guarantees that we never access illegal
-	      memory if this would not also be done by the trivial
-	      implementation (this is because all processor inherent
-	      boundaries are multiples of 8).  */
-
-	movq	%rdi, %rdx
-	andl	$7, %edx	/* Mask alignment bits  */
-	movq	%rdi, %rax	/* duplicate destination.  */
-	jz	1f		/* aligned => start loop */
-	neg	%edx
-	addl	$8, %edx	/* Align to 8 bytes.  */
-
-	/* Search the first bytes directly.  */
-0:	movb	(%rax), %cl	/* load byte  */
-	cmpb	%cl,%sil	/* compare byte.  */
-	je	6f		/* target found */
-	testb	%cl,%cl		/* is byte NUL? */
-	je	7f		/* yes => return NULL */
-	incq	%rax		/* increment pointer */
-	decl	%edx
-	jnz	0b
-
-
-1:
-	/* At the moment %rsi contains C.  What we need for the
-	   algorithm is C in all bytes of the register.  Avoid
-	   operations on 16 bit words because these require an
-	   prefix byte (and one more cycle).  */
-	/* Populate 8 bit data to full 64-bit.  */
-	movabs	$0x0101010101010101,%r9
-	movzbl	%sil,%edx
-	imul	%rdx,%r9
-
-	movq $0xfefefefefefefeff, %r8 /* Save magic.  */
-
-      /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to
-	 change any of the hole bits of LONGWORD.
-
-	 1) Is this safe?  Will it catch all the zero bytes?
-	 Suppose there is a byte with all zeros.  Any carry bits
-	 propagating from its left will fall into the hole at its
-	 least significant bit and stop.  Since there will be no
-	 carry from its most significant bit, the LSB of the
-	 byte to the left will be unchanged, and the zero will be
-	 detected.
-
-	 2) Is this worthwhile?  Will it ignore everything except
-	 zero bytes?  Suppose every byte of QUARDWORD has a bit set
-	 somewhere.  There will be a carry into bit 8.	If bit 8
-	 is set, this will carry into bit 16.  If bit 8 is clear,
-	 one of bits 9-15 must be set, so there will be a carry
-	 into bit 16.  Similarly, there will be a carry into bit
-	 24 tec..  If one of bits 54-63 is set, there will be a carry
-	 into bit 64 (=carry flag), so all of the hole bits will
-	 be changed.
-
-	 3) But wait!  Aren't we looking for C, not zero?
-	 Good point.  So what we do is XOR LONGWORD with a longword,
-	 each of whose bytes is C.  This turns each byte that is C
-	 into a zero.  */
-
-	.p2align 4
-4:
-	/* Main Loop is unrolled 4 times.  */
-	/* First unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
-				   are now 0 */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found c => return pointer */
-
-	/* The quadword we looked at does not contain the value we're looking
-	   for.  Let's search now whether we have reached the end of the
-	   string.  */
-	xorq %r9, %rcx		/* restore original dword without reload */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 7f			/* highest byte is NUL => return NULL */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 7f			/* found NUL => return NULL */
-
-	/* Second unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
-				   are now 0 */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found c => return pointer */
-
-	/* The quadword we looked at does not contain the value we're looking
-	   for.  Let's search now whether we have reached the end of the
-	   string.  */
-	xorq %r9, %rcx		/* restore original dword without reload */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 7f			/* highest byte is NUL => return NULL */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 7f			/* found NUL => return NULL */
-	/* Third unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
-				   are now 0 */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found c => return pointer */
-
-	/* The quadword we looked at does not contain the value we're looking
-	   for.  Let's search now whether we have reached the end of the
-	   string.  */
-	xorq %r9, %rcx		/* restore original dword without reload */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 7f			/* highest byte is NUL => return NULL */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 7f			/* found NUL => return NULL */
-	/* Fourth unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
-				   are now 0 */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found c => return pointer */
-
-	/* The quadword we looked at does not contain the value we're looking
-	   for.  Let's search now whether we have reached the end of the
-	   string.  */
-	xorq %r9, %rcx		/* restore original dword without reload */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 7f			/* highest byte is NUL => return NULL */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jz 4b			/* no NUL found => restart loop */
-
-
-7:	/* Return NULL.  */
-	xorl %eax, %eax
-	retq
-
-
-	/* We now scan for the byte in which the character was matched.
-	   But we have to take care of the case that a NUL char is
-	   found before this in the dword.  Note that we XORed %rcx
-	   with the byte we're looking for, therefore the tests below look
-	   reversed.  */
-
-
-	.p2align 4		/* Align, it's a jump target.  */
-3:	movq	%r9,%rdx	/* move to %rdx so that we can access bytes */
-	subq	$8,%rax		/* correct pointer increment.  */
-	testb %cl, %cl		/* is first byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %cl		/* is first byte NUL? */
-	je 7b			/* yes => return NULL */
-	incq %rax		/* increment pointer */
-
-	testb %ch, %ch		/* is second byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %ch		/* is second byte NUL? */
-	je 7b			/* yes => return NULL? */
-	incq %rax		/* increment pointer */
-
-	shrq $16, %rcx		/* make upper bytes accessible */
-	testb %cl, %cl		/* is third byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %cl		/* is third byte NUL? */
-	je 7b			/* yes => return NULL */
-	incq %rax		/* increment pointer */
-
-	testb %ch, %ch		/* is fourth byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %ch		/* is fourth byte NUL? */
-	je 7b			/* yes => return NULL? */
-	incq %rax		/* increment pointer */
-
-	shrq $16, %rcx		/* make upper bytes accessible */
-	testb %cl, %cl		/* is fifth byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %cl		/* is fifth byte NUL? */
-	je 7b			/* yes => return NULL */
-	incq %rax		/* increment pointer */
-
-	testb %ch, %ch		/* is sixth byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %ch		/* is sixth byte NUL? */
-	je 7b			/* yes => return NULL? */
-	incq %rax		/* increment pointer */
-
-	shrq $16, %rcx		/* make upper bytes accessible */
-	testb %cl, %cl		/* is seventh byte C? */
-	jz 6f			/* yes => return pointer */
-	cmpb %dl, %cl		/* is seventh byte NUL? */
-	je 7b			/* yes => return NULL */
-
-	/* It must be in the eigth byte and it cannot be NUL.  */
-	incq %rax
-
-6:
-	nop
-	retq
-END (BP_SYM (strchr))
-
-weak_alias (BP_SYM (strchr), BP_SYM (index))
+ENTRY (strchr)
+	movd	%esi, %xmm1
+	movq	%rdi, %rcx
+	punpcklbw %xmm1, %xmm1
+	andq	$~15, %rdi
+	pxor	%xmm2, %xmm2
+	punpcklbw %xmm1, %xmm1
+	orl	$0xffffffff, %esi
+	movdqa	(%rdi), %xmm0
+	pshufd	$0, %xmm1, %xmm1
+	subq	%rdi, %rcx
+	movdqa	%xmm0, %xmm3
+	leaq	16(%rdi), %rdi
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm3
+	shl	%cl, %esi
+	pmovmskb %xmm0, %edx
+	pmovmskb %xmm3, %ecx
+	andl	%esi, %edx
+	andl	%esi, %ecx
+	orl	%edx, %ecx
+	jnz	1f
+
+2:	movdqa	(%rdi), %xmm0
+	leaq	16(%rdi), %rdi
+	movdqa	%xmm0, %xmm3
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm3
+	pmovmskb %xmm0, %edx
+	pmovmskb %xmm3, %ecx
+	orl	%edx, %ecx
+	jz	2b
+
+1:	bsfl	%edx, %edx
+	jz	4f
+	bsfl	%ecx, %ecx
+	leaq	-16(%rdi,%rdx), %rax
+	cmpl	%edx, %ecx
+	je	5f
+4:	xorl	%eax, %eax
+5:	ret
+END (strchr)
+
+weak_alias (strchr, index)
 libc_hidden_builtin_def (strchr)
+
diff --git a/sysdeps/x86_64/strlen.S b/sysdeps/x86_64/strlen.S
index fd950edaaa..b83332c90c 100644
--- a/sysdeps/x86_64/strlen.S
+++ b/sysdeps/x86_64/strlen.S
@@ -1,6 +1,6 @@
 /* strlen(str) -- determine the length of the string STR.
-   Copyright (C) 2002, 2003 Free Software Foundation, Inc.
-   Based on i486 version contributed by Ulrich Drepper <drepper@redhat.com>.
+   Copyright (C) 2009 Free Software Foundation, Inc.
+   Contributed by Ulrich Drepper <drepper@redhat.com>.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -19,121 +19,35 @@
    02111-1307 USA.  */
 
 #include <sysdep.h>
-#include "asm-syntax.h"
-#include "bp-sym.h"
-#include "bp-asm.h"
 
 
-	.text
-ENTRY (strlen)
-	movq %rdi, %rcx		/* Duplicate source pointer. */
-	andl $7, %ecx		/* mask alignment bits */
-	movq %rdi, %rax		/* duplicate destination.  */
-	jz 1f			/* aligned => start loop */
-
-	neg %ecx		/* We need to align to 8 bytes.  */
-	addl $8,%ecx
-	/* Search the first bytes directly.  */
-0:	cmpb $0x0,(%rax)	/* is byte NUL? */
-	je 2f			/* yes => return */
-	incq %rax		/* increment pointer */
-	decl %ecx
-	jnz 0b
-
-1:	movq $0xfefefefefefefeff,%r8 /* Save magic.  */
-
-	.p2align 4		/* Align loop.  */
-4:	/* Main Loop is unrolled 4 times.  */
-	/* First unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found NUL => return pointer */
-
-	/* Second unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found NUL => return pointer */
-
-	/* Third unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jnz 3f			/* found NUL => return pointer */
-
-	/* Fourth unroll.  */
-	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
-	addq $8,%rax		/* adjust pointer for next word */
-	movq %r8, %rdx		/* magic value */
-	addq %rcx, %rdx		/* add the magic value to the word.  We get
-				   carry bits reported for each byte which
-				   is *not* 0 */
-	jnc 3f			/* highest byte is NUL => return pointer */
-	xorq %rcx, %rdx		/* (word+magic)^word */
-	orq %r8, %rdx		/* set all non-carry bits */
-	incq %rdx		/* add 1: if one carry bit was *not* set
-				   the addition will not result in 0.  */
-	jz 4b			/* no NUL found => continue loop */
-
-	.p2align 4		/* Align, it's a jump target.  */
-3:	subq $8,%rax		/* correct pointer increment.  */
-
-	testb %cl, %cl		/* is first byte NUL? */
-	jz 2f			/* yes => return */
-	incq %rax		/* increment pointer */
-
-	testb %ch, %ch		/* is second byte NUL? */
-	jz 2f			/* yes => return */
-	incq %rax		/* increment pointer */
-
-	testl $0x00ff0000, %ecx /* is third byte NUL? */
-	jz 2f			/* yes => return pointer */
-	incq %rax		/* increment pointer */
-
-	testl $0xff000000, %ecx /* is fourth byte NUL? */
-	jz 2f			/* yes => return pointer */
-	incq %rax		/* increment pointer */
-
-	shrq $32, %rcx		/* look at other half.  */
-
-	testb %cl, %cl		/* is first byte NUL? */
-	jz 2f			/* yes => return */
-	incq %rax		/* increment pointer */
-
-	testb %ch, %ch		/* is second byte NUL? */
-	jz 2f			/* yes => return */
-	incq %rax		/* increment pointer */
-
-	testl $0xff0000, %ecx	/* is third byte NUL? */
-	jz 2f			/* yes => return pointer */
-	incq %rax		/* increment pointer */
-2:
-	subq %rdi, %rax		/* compute difference to string start */
+ENTRY(strlen)
+	movq	%rdi, %rcx
+	movq	%rdi, %r8
+	andq	$~15, %rdi
+	pxor	%xmm1, %xmm1
+	orl	$0xffffffff, %esi
+	movdqa	(%rdi), %xmm0
+	subq	%rdi, %rcx
+	leaq	16(%rdi), %rdi
+	pcmpeqb	%xmm1, %xmm0
+	shl	%cl, %esi
+	pmovmskb %xmm0, %edx
+	xorl	%eax, %eax
+	negq	%r8
+	andl	%esi, %edx
+	jnz	1f
+
+2:	movdqa	(%rdi), %xmm0
+	leaq	16(%rdi), %rdi
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %edx
+	testl	%edx, %edx
+	jz	2b
+
+1:	leaq	-16(%rdi,%r8), %rdi
+	bsfl	%edx, %eax
+	addq	%rdi, %rax
 	ret
-END (strlen)
+END(strlen)
 libc_hidden_builtin_def (strlen)