Merge branch 'akpm' (Andrew's patch-bomb)

Merge second batch of patches from Andrew Morton:
 - various misc things
 - core kernel changes to prctl, exit, exec, init, etc.
 - kernel/watchdog.c updates
 - get_maintainer
 - MAINTAINERS
 - the backlight driver queue
 - core bitops code cleanups
 - the led driver queue
 - some core prio_tree work
 - checkpatch udpates
 - largeish crc32 update
 - a new poll() feature for the v4l guys
 - the rtc driver queue
 - fatfs
 - ptrace
 - signals
 - kmod/usermodehelper updates
 - coredump
 - procfs updates

* emailed from Andrew Morton <akpm@linux-foundation.org>: (141 commits)
  seq_file: add seq_set_overflow(), seq_overflow()
  proc-ns: use d_set_d_op() API to set dentry ops in proc_ns_instantiate().
  procfs: speed up /proc/pid/stat, statm
  procfs: add num_to_str() to speed up /proc/stat
  proc: speed up /proc/stat handling
  fs/proc/kcore.c: make get_sparsemem_vmemmap_info() static
  coredump: add VM_NODUMP, MADV_NODUMP, MADV_CLEAR_NODUMP
  coredump: remove VM_ALWAYSDUMP flag
  kmod: make __request_module() killable
  kmod: introduce call_modprobe() helper
  usermodehelper: ____call_usermodehelper() doesn't need do_exit()
  usermodehelper: kill umh_wait, renumber UMH_* constants
  usermodehelper: implement UMH_KILLABLE
  usermodehelper: introduce umh_complete(sub_info)
  usermodehelper: use UMH_WAIT_PROC consistently
  signal: zap_pid_ns_processes: s/SEND_SIG_NOINFO/SEND_SIG_FORCED/
  signal: oom_kill_task: use SEND_SIG_FORCED instead of force_sig()
  signal: cosmetic, s/from_ancestor_ns/force/ in prepare_signal() paths
  signal: give SEND_SIG_FORCED more power to beat SIGNAL_UNKILLABLE
  Hexagon: use set_current_blocked() and block_sigmask()
  ...

1 files changed, 16 insertions, 4 deletions

diff --git a/lib/string.c b/lib/string.c
index dc4a86341f9..3a03782720c 100644
--- a/lib/string.c
+++ b/lib/string.c
@@ -785,12 +785,24 @@ void *memchr_inv(const void *start, int c, size_t bytes)
 	if (bytes <= 16)
 		return check_bytes8(start, value, bytes);
 
-	value64 = value | value << 8 | value << 16 | value << 24;
-	value64 = (value64 & 0xffffffff) | value64 << 32;
-	prefix = 8 - ((unsigned long)start) % 8;
+	value64 = value;
+#if defined(ARCH_HAS_FAST_MULTIPLIER) && BITS_PER_LONG == 64
+	value64 *= 0x0101010101010101;
+#elif defined(ARCH_HAS_FAST_MULTIPLIER)
+	value64 *= 0x01010101;
+	value64 |= value64 << 32;
+#else
+	value64 |= value64 << 8;
+	value64 |= value64 << 16;
+	value64 |= value64 << 32;
+#endif
 
+	prefix = (unsigned long)start % 8;
 	if (prefix) {
-		u8 *r = check_bytes8(start, value, prefix);
+		u8 *r;
+
+		prefix = 8 - prefix;
+		r = check_bytes8(start, value, prefix);
 		if (r)
 			return r;
 		start += prefix;