[NET]: use __read_mostly on kmem_cache_t , DEFINE_SNMP_STAT pointers
This patch puts mostly read only data in the right section
(read_mostly), to help sharing of these data between CPUS without
memory ping pongs.
On one of my production machine, tcp_statistics was sitting in a
heavily modified cache line, so *every* SNMP update had to force a
reload.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 5810f9d..bf147f8 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -113,7 +113,7 @@
#include <linux/mroute.h>
#endif
-DEFINE_SNMP_STAT(struct linux_mib, net_statistics);
+DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
extern void ip_mc_drop_socket(struct sock *sk);
diff --git a/net/ipv4/fib_hash.c b/net/ipv4/fib_hash.c
index b10d6bb..2a8c9af 100644
--- a/net/ipv4/fib_hash.c
+++ b/net/ipv4/fib_hash.c
@@ -45,8 +45,8 @@
#include "fib_lookup.h"
-static kmem_cache_t *fn_hash_kmem;
-static kmem_cache_t *fn_alias_kmem;
+static kmem_cache_t *fn_hash_kmem __read_mostly;
+static kmem_cache_t *fn_alias_kmem __read_mostly;
struct fib_node {
struct hlist_node fn_hash;
diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
index ff21748..b2dea4e 100644
--- a/net/ipv4/fib_trie.c
+++ b/net/ipv4/fib_trie.c
@@ -166,7 +166,7 @@
static void tnode_free(struct tnode *tn);
static void trie_dump_seq(struct seq_file *seq, struct trie *t);
-static kmem_cache_t *fn_alias_kmem;
+static kmem_cache_t *fn_alias_kmem __read_mostly;
static struct trie *trie_local = NULL, *trie_main = NULL;
diff --git a/net/ipv4/icmp.c b/net/ipv4/icmp.c
index 25f66b7..24eb56a 100644
--- a/net/ipv4/icmp.c
+++ b/net/ipv4/icmp.c
@@ -114,7 +114,7 @@
/*
* Statistics
*/
-DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics);
+DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
/* An array of errno for error messages from dest unreach. */
/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c
index 4410b9d..f84ba9c 100644
--- a/net/ipv4/inetpeer.c
+++ b/net/ipv4/inetpeer.c
@@ -73,7 +73,7 @@
/* Exported for inet_getid inline function. */
DEFINE_SPINLOCK(inet_peer_idlock);
-static kmem_cache_t *peer_cachep;
+static kmem_cache_t *peer_cachep __read_mostly;
#define node_height(x) x->avl_height
static struct inet_peer peer_fake_node = {
diff --git a/net/ipv4/ip_input.c b/net/ipv4/ip_input.c
index 220a8b5..473d0f2 100644
--- a/net/ipv4/ip_input.c
+++ b/net/ipv4/ip_input.c
@@ -150,7 +150,7 @@
* SNMP management statistics
*/
-DEFINE_SNMP_STAT(struct ipstats_mib, ip_statistics);
+DEFINE_SNMP_STAT(struct ipstats_mib, ip_statistics) __read_mostly;
/*
* Process Router Attention IP option
diff --git a/net/ipv4/ipmr.c b/net/ipv4/ipmr.c
index dc806b5..9dbf590 100644
--- a/net/ipv4/ipmr.c
+++ b/net/ipv4/ipmr.c
@@ -103,7 +103,7 @@
In this case data path is free of exclusive locks at all.
*/
-static kmem_cache_t *mrt_cachep;
+static kmem_cache_t *mrt_cachep __read_mostly;
static int ip_mr_forward(struct sk_buff *skb, struct mfc_cache *cache, int local);
static int ipmr_cache_report(struct sk_buff *pkt, vifi_t vifi, int assert);
diff --git a/net/ipv4/ipvs/ip_vs_conn.c b/net/ipv4/ipvs/ip_vs_conn.c
index d0145a8..e11952e 100644
--- a/net/ipv4/ipvs/ip_vs_conn.c
+++ b/net/ipv4/ipvs/ip_vs_conn.c
@@ -40,7 +40,7 @@
static struct list_head *ip_vs_conn_tab;
/* SLAB cache for IPVS connections */
-static kmem_cache_t *ip_vs_conn_cachep;
+static kmem_cache_t *ip_vs_conn_cachep __read_mostly;
/* counter for current IPVS connections */
static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
diff --git a/net/ipv4/netfilter/ip_conntrack_core.c b/net/ipv4/netfilter/ip_conntrack_core.c
index 285743b..a064860 100644
--- a/net/ipv4/netfilter/ip_conntrack_core.c
+++ b/net/ipv4/netfilter/ip_conntrack_core.c
@@ -70,8 +70,8 @@
unsigned int ip_conntrack_htable_size = 0;
int ip_conntrack_max;
struct list_head *ip_conntrack_hash;
-static kmem_cache_t *ip_conntrack_cachep;
-static kmem_cache_t *ip_conntrack_expect_cachep;
+static kmem_cache_t *ip_conntrack_cachep __read_mostly;
+static kmem_cache_t *ip_conntrack_expect_cachep __read_mostly;
struct ip_conntrack ip_conntrack_untracked;
unsigned int ip_ct_log_invalid;
static LIST_HEAD(unconfirmed);
diff --git a/net/ipv4/netfilter/ipt_hashlimit.c b/net/ipv4/netfilter/ipt_hashlimit.c
index 564b49b..2dd1ccc 100644
--- a/net/ipv4/netfilter/ipt_hashlimit.c
+++ b/net/ipv4/netfilter/ipt_hashlimit.c
@@ -94,7 +94,7 @@
static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */
static DECLARE_MUTEX(hlimit_mutex); /* additional checkentry protection */
static HLIST_HEAD(hashlimit_htables);
-static kmem_cache_t *hashlimit_cachep;
+static kmem_cache_t *hashlimit_cachep __read_mostly;
static inline int dst_cmp(const struct dsthash_ent *ent, struct dsthash_dst *b)
{
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 68626de..02fdda6 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -269,7 +269,7 @@
int sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
-DEFINE_SNMP_STAT(struct tcp_mib, tcp_statistics);
+DEFINE_SNMP_STAT(struct tcp_mib, tcp_statistics) __read_mostly;
atomic_t tcp_orphan_count = ATOMIC_INIT(0);
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 3a5bbbe..e5beca7 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -113,7 +113,7 @@
* Snmp MIB for the UDP layer
*/
-DEFINE_SNMP_STAT(struct udp_mib, udp_statistics);
+DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
struct hlist_head udp_hash[UDP_HTABLE_SIZE];
DEFINE_RWLOCK(udp_hash_lock);