aha/net/ipv4
Eric Dumazet 81c3d5470e [INET]: speedup inet (tcp/dccp) lookups
Arnaldo and I agreed it could be applied now, because I have other
pending patches depending on this one (Thank you Arnaldo)

(The other important patch moves skc_refcnt in a separate cache line,
so that the SMP/NUMA performance doesnt suffer from cache line ping pongs)

1) First some performance data :
--------------------------------

tcp_v4_rcv() wastes a *lot* of time in __inet_lookup_established()

The most time critical code is :

sk_for_each(sk, node, &head->chain) {
     if (INET_MATCH(sk, acookie, saddr, daddr, ports, dif))
         goto hit; /* You sunk my battleship! */
}

The sk_for_each() does use prefetch() hints but only the begining of
"struct sock" is prefetched.

As INET_MATCH first comparison uses inet_sk(__sk)->daddr, wich is far
away from the begining of "struct sock", it has to bring into CPU
cache cold cache line. Each iteration has to use at least 2 cache
lines.

This can be problematic if some chains are very long.

2) The goal
-----------

The idea I had is to change things so that INET_MATCH() may return
FALSE in 99% of cases only using the data already in the CPU cache,
using one cache line per iteration.

3) Description of the patch
---------------------------

Adds a new 'unsigned int skc_hash' field in 'struct sock_common',
filling a 32 bits hole on 64 bits platform.

struct sock_common {
	unsigned short		skc_family;
	volatile unsigned char	skc_state;
	unsigned char		skc_reuse;
	int			skc_bound_dev_if;
	struct hlist_node	skc_node;
	struct hlist_node	skc_bind_node;
	atomic_t		skc_refcnt;
+	unsigned int		skc_hash;
	struct proto		*skc_prot;
};

Store in this 32 bits field the full hash, not masked by (ehash_size -
1) Using this full hash as the first comparison done in INET_MATCH
permits us immediatly skip the element without touching a second cache
line in case of a miss.

Suppress the sk_hashent/tw_hashent fields since skc_hash (aliased to
sk_hash and tw_hash) already contains the slot number if we mask with
(ehash_size - 1)

File include/net/inet_hashtables.h

64 bits platforms :
#define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
     (((__sk)->sk_hash == (__hash))
     ((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie))   &&  \
     ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports))   &&  \
     (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))

32bits platforms:
#define TCP_IPV4_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
     (((__sk)->sk_hash == (__hash))                 &&  \
     (inet_sk(__sk)->daddr          == (__saddr))   &&  \
     (inet_sk(__sk)->rcv_saddr      == (__daddr))   &&  \
     (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))


- Adds a prefetch(head->chain.first) in 
__inet_lookup_established()/__tcp_v4_check_established() and 
__inet6_lookup_established()/__tcp_v6_check_established() and 
__dccp_v4_check_established() to bring into cache the first element of the 
list, before the {read|write}_lock(&head->lock);

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2005-10-03 14:13:38 -07:00
..
ipvs [IPVS]: ip_vs_ftp breaks connections using persistence 2005-09-14 21:08:51 -07:00
netfilter [NET]: Fix packet timestamping. 2005-10-03 13:57:23 -07:00
af_inet.c [IPV4] fib_trie: fix proc interface 2005-09-09 13:35:42 -07:00
ah4.c [CRYPTO]: crypto_free_tfm() callers no longer need to check for NULL 2005-09-01 17:44:29 -07:00
arp.c [NET]: Store skb->timestamp as offset to a base timestamp 2005-08-29 15:58:24 -07:00
datagram.c [NET]: Fix sparse warnings 2005-08-29 16:01:32 -07:00
devinet.c [NETLINK]: Convert netlink users to use group numbers instead of bitmasks 2005-08-29 16:00:54 -07:00
esp4.c [CRYPTO]: crypto_free_tfm() callers no longer need to check for NULL 2005-09-01 17:44:29 -07:00
fib_frontend.c [NETLINK]: Add "groups" argument to netlink_kernel_create 2005-08-29 16:01:11 -07:00
fib_hash.c [NET]: use __read_mostly on kmem_cache_t , DEFINE_SNMP_STAT pointers 2005-08-29 16:11:18 -07:00
fib_lookup.h [IPV4]: Prepare FIB core for RCU. 2005-08-29 16:08:31 -07:00
fib_rules.c [NETLINK]: Correctly set NLM_F_MULTI without checking the pid 2005-06-18 22:54:12 -07:00
fib_semantics.c [IPV4]: Prepare FIB core for RCU. 2005-08-29 16:08:31 -07:00
fib_trie.c [FIB_TRIE]: message cleanup 2005-09-21 00:15:39 -07:00
icmp.c [NET]: use __read_mostly on kmem_cache_t , DEFINE_SNMP_STAT pointers 2005-08-29 16:11:18 -07:00
igmp.c [MCAST]: Fix MCAST_EXCLUDE line dupes 2005-09-14 20:53:42 -07:00
inet_connection_sock.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
inet_diag.c [NETLINK]: Add "groups" argument to netlink_kernel_create 2005-08-29 16:01:11 -07:00
inet_hashtables.c [NET]: Introduce inet_connection_sock 2005-08-29 15:43:19 -07:00
inet_timewait_sock.c [INET]: speedup inet (tcp/dccp) lookups 2005-10-03 14:13:38 -07:00
inetpeer.c [PATCH] timer initialization cleanup: DEFINE_TIMER 2005-09-09 14:03:48 -07:00
ip_forward.c [IPV4]: Remove some dead code from ip_forward() 2005-08-29 16:03:06 -07:00
ip_fragment.c [IPV4]: Reassembly trim not clearing CHECKSUM_HW 2005-09-06 15:51:48 -07:00
ip_gre.c [NET]: fix oops after tunnel module unload 2005-07-30 17:46:44 -07:00
ip_input.c [NET]: use __read_mostly on kmem_cache_t , DEFINE_SNMP_STAT pointers 2005-08-29 16:11:18 -07:00
ip_options.c [IP]: Introduce ip_options_get_from_user 2005-08-29 16:01:39 -07:00
ip_output.c [IPV4]: ip_finish_output() can be inlined 2005-08-29 16:03:10 -07:00
ip_sockglue.c [IP]: Introduce ip_options_get_from_user 2005-08-29 16:01:39 -07:00
ipcomp.c [CRYPTO]: crypto_free_tfm() callers no longer need to check for NULL 2005-09-01 17:44:29 -07:00
ipconfig.c [NET]: fix-up schedule_timeout() usage 2005-09-12 14:15:34 -07:00
ipip.c [NET]: fix oops after tunnel module unload 2005-07-30 17:46:44 -07:00
ipmr.c [NET]: use __read_mostly on kmem_cache_t , DEFINE_SNMP_STAT pointers 2005-08-29 16:11:18 -07:00
Kconfig [INET_DIAG]: Move the tcp_diag interface to the proper place 2005-08-29 15:57:54 -07:00
Makefile [INET_DIAG]: Move the tcp_diag interface to the proper place 2005-08-29 15:57:54 -07:00
multipath.c Linux-2.6.12-rc2 2005-04-16 15:20:36 -07:00
multipath_drr.c [IPV4]: possible cleanups 2005-08-29 15:33:20 -07:00
multipath_random.c [IPV4]: Multipath modules need a license to prevent kernel tainting. 2005-06-13 14:29:06 -07:00
multipath_rr.c [IPV4]: Multipath modules need a license to prevent kernel tainting. 2005-06-13 14:29:06 -07:00
multipath_wrandom.c [IPV4]: Multipath modules need a license to prevent kernel tainting. 2005-06-13 14:29:06 -07:00
netfilter.c [NETFILTER]: Move reroute-after-queue code up to the nf_queue layer. 2005-08-29 15:36:19 -07:00
proc.c [NET]: Fix sparse warnings 2005-08-29 16:01:32 -07:00
protocol.c [TCP]: Move the tcp sock states to net/tcp_states.h 2005-08-29 15:41:54 -07:00
raw.c [PATCH] raw_sendmsg DoS on 2.6 2005-09-19 18:45:42 -07:00
route.c [IPV4]: Fix refcount damaging in net/ipv4/route.c 2005-09-08 13:34:47 -07:00
syncookies.c [NET]: Fix sparse warnings 2005-08-29 16:01:32 -07:00
sysctl_net_ipv4.c [NET]: Fix sparse warnings 2005-08-29 16:01:32 -07:00
tcp.c [TCP]: Fix TCP_OFF() bug check introduced by previous change. 2005-09-05 18:55:48 -07:00
tcp_bic.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_cong.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_diag.c [INET_DIAG]: Move the tcp_diag interface to the proper place 2005-08-29 15:57:54 -07:00
tcp_highspeed.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_htcp.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_hybla.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_input.c [TCP]: Don't over-clamp window in tcp_clamp_window() 2005-09-29 17:17:15 -07:00
tcp_ipv4.c [INET]: speedup inet (tcp/dccp) lookups 2005-10-03 14:13:38 -07:00
tcp_minisocks.c [TCP]: Set default congestion control correctly for incoming connections. 2005-09-21 00:19:46 -07:00
tcp_output.c [TCP]: Revert 6b251858d3 2005-09-29 17:07:20 -07:00
tcp_scalable.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_timer.c [ICSK]: Move TCP congestion avoidance members to icsk 2005-08-29 15:56:18 -07:00
tcp_vegas.c [INET_DIAG]: Rename tcp_diag.[ch] to inet_diag.[ch] 2005-08-29 15:57:48 -07:00
tcp_westwood.c [INET_DIAG]: Rename tcp_diag.[ch] to inet_diag.[ch] 2005-08-29 15:57:48 -07:00
udp.c [IPV4] udp: trim forgets about CHECKSUM_HW 2005-09-08 12:32:21 -07:00
xfrm4_input.c Linux-2.6.12-rc2 2005-04-16 15:20:36 -07:00
xfrm4_output.c [IPSEC]: Add XFRM_STATE_NOPMTUDISC flag 2005-06-20 13:21:43 -07:00
xfrm4_policy.c [IPSEC]: Store idev entries 2005-05-03 16:27:10 -07:00
xfrm4_state.c [IPV4]: possible cleanups 2005-08-29 15:33:20 -07:00
xfrm4_tunnel.c [NET]: Make ipip/ip6_tunnel independant of XFRM 2005-07-19 14:03:34 -07:00