On Sun, 2017-01-08 at 13:54 +0100, Jason A. Donenfeld wrote:
> SHA1 is slower and less secure than SipHash, and so replacing syncookie
> generation with SipHash makes natural sense. Some BSDs have been doing
> this for several years in fact.
>
> The speedup should be similar -- and even more impressive -- to the
> speedup from the sequence number fix in this series.
I confirm a nice speedup under SYNFLOOD.
sha_transform() used to consume ~12 % of cpu cycles, while the
siphash_2u64() only uses ~1.9 %
Depending on the setup, gain is about 9 %
4.48% [kernel] [k] ipt_do_table
4.39% [kernel] [k] fib_table_lookup
3.90% [kernel] [k] __netif_receive_skb_core
3.76% [kernel] [k] fib_rules_lookup
3.15% [kernel] [k] __inet_lookup_established
3.11% [kernel] [k] tcp_conn_request
2.51% [kernel] [k] tcp_v4_rcv
2.42% [kernel] [k] tcp_make_synack
2.22% [kernel] [k] nf_iterate
2.16% [kernel] [k] ip_rcv
1.92% [kernel] [k] siphash_2u64
1.76% [kernel] [k] __ip_route_output_key
1.73% [kernel] [k] mlx4_en_process_rx_cq
1.68% [kernel] [k] memcpy_erms
1.59% [kernel] [k] __alloc_skb
1.49% [kernel] [k] __dev_queue_xmit
1.48% [kernel] [k] kmem_cache_alloc
1.38% [kernel] [k] __local_bh_enable_ip
1.36% [kernel] [k] kmem_cache_free
1.21% [kernel] [k] ___cache_free
1.09% [kernel] [k] __build_skb
1.07% [kernel] [k] inet_reqsk_alloc
1.04% [kernel] [k] kfree
1.04% [kernel] [k] ip_build_and_send_pkt
1.04% [kernel] [k] inet_gro_receive
1.01% [kernel] [k] fib_validate_source
0.98% [kernel] [k] tcp_openreq_init_rwin
0.98% [kernel] [k] inet_csk_route_req
0.97% [kernel] [k] fib_get_table
0.96% [kernel] [k] ip_finish_output2
0.94% [kernel] [k] tcp_v4_do_rcv
0.91% [kernel] [k] ip_local_deliver_finish
0.91% [kernel] [k] netif_skb_features
0.91% [kernel] [k] dev_hard_start_xmit
