to #29722367

On some specific hardware with 128 bytes LLC cacheline, tk_core may
cause false sharing problem. We can align it to 128 bytes so that
it won't be affected by other global variables.

This change will make a bit waste on cache utilization but get good
number of performance improvement. So for both 64 and 128 bytes aligned
LLC cacheline, we adjust tk_core memory layout to avoid potential cacheline
contention.
Signed-off-by: NPeng Wang <rocking@linux.alibaba.com>
Acked-by: NShanpei Chen <shanpeic@linux.alibaba.com>
Reviewed-by: NShile Zhang <shile.zhang@linux.alibaba.com>