Make "unlock_buffer()" use the optimized "smp_mb__after_clear_bit()".

"clear_buffer_locked()" is just a nicer name for a clear-bit, and
on x86 (and others) we don't need any expensive memory barriers
after that.

fs/buffer.c

@@ -76,7 +76,7 @@ EXPORT_SYMBOL(__lock_buffer);
 void fastcall unlock_buffer(struct buffer_head *bh)
 {
 	clear_buffer_locked(bh);
-	smp_mb();
+	smp_mb__after_clear_bit();
 	wake_up_bit(&bh->b_state, BH_Lock);
 }