The reada code from scrub was casting down a u64 to
an unsigned long so it could insert it into a radix tree.
What it really wanted to do was cast down the result of a shift, instead
of casting down the u64.  The bug resulted in trying to insert our
reada struct into the wrong place, which caused soft lockups and other
problems.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
 
        spin_lock(&fs_info->reada_lock);
        ret = radix_tree_insert(&dev->reada_zones,
-                               (unsigned long)zone->end >> PAGE_CACHE_SHIFT,
+                               (unsigned long)(zone->end >> PAGE_CACHE_SHIFT),
                                zone);
        spin_unlock(&fs_info->reada_lock);