Similar to how gso is handled use skb list for skb_bad_tx this is
required with lockless qdiscs because we may have multiple cores
attempting to push skbs into skb_bad_tx concurrently
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>