[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/bridge/br_fdb.c b/net/bridge/br_fdb.c
index e6c2200..24396b9 100644
--- a/net/bridge/br_fdb.c
+++ b/net/bridge/br_fdb.c
@@ -23,7 +23,7 @@
#include <asm/atomic.h>
#include "br_private.h"
-static kmem_cache_t *br_fdb_cache;
+static kmem_cache_t *br_fdb_cache __read_mostly;
static int fdb_insert(struct net_bridge *br, struct net_bridge_port *source,
const unsigned char *addr);
diff --git a/net/core/flow.c b/net/core/flow.c
index f289570..7e95b39 100644
--- a/net/core/flow.c
+++ b/net/core/flow.c
@@ -42,7 +42,7 @@
#define flow_table(cpu) (per_cpu(flow_tables, cpu))
-static kmem_cache_t *flow_cachep;
+static kmem_cache_t *flow_cachep __read_mostly;
static int flow_lwm, flow_hwm;
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index b853a9b..f80a287 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -68,8 +68,8 @@
#include <asm/uaccess.h>
#include <asm/system.h>
-static kmem_cache_t *skbuff_head_cache;
-static kmem_cache_t *skbuff_fclone_cache;
+static kmem_cache_t *skbuff_head_cache __read_mostly;
+static kmem_cache_t *skbuff_fclone_cache __read_mostly;
struct timeval __read_mostly skb_tv_base;
diff --git a/net/dccp/ccids/ccid3.c b/net/dccp/ccids/ccid3.c
index fe4cc85..cf93b01 100644
--- a/net/dccp/ccids/ccid3.c
+++ b/net/dccp/ccids/ccid3.c
@@ -85,7 +85,7 @@
static struct dccp_tx_hist *ccid3_tx_hist;
static struct dccp_rx_hist *ccid3_rx_hist;
-static kmem_cache_t *ccid3_loss_interval_hist_slab;
+static kmem_cache_t *ccid3_loss_interval_hist_slab __read_mostly;
static inline struct ccid3_loss_interval_hist_entry *
ccid3_loss_interval_hist_entry_new(const unsigned int __nocast prio)
diff --git a/net/dccp/proto.c b/net/dccp/proto.c
index 600dda5..f97e92e 100644
--- a/net/dccp/proto.c
+++ b/net/dccp/proto.c
@@ -39,7 +39,7 @@
#include "ccid.h"
#include "dccp.h"
-DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics);
+DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
atomic_t dccp_orphan_count = ATOMIC_INIT(0);
diff --git a/net/decnet/dn_table.c b/net/decnet/dn_table.c
index 73a8848..eeba56f 100644
--- a/net/decnet/dn_table.c
+++ b/net/decnet/dn_table.c
@@ -79,7 +79,7 @@
static DEFINE_RWLOCK(dn_fib_tables_lock);
struct dn_fib_table *dn_fib_tables[RT_TABLE_MAX + 1];
-static kmem_cache_t *dn_hash_kmem;
+static kmem_cache_t *dn_hash_kmem __read_mostly;
static int dn_fib_hash_zombies;
static inline dn_fib_idx_t dn_hash(dn_fib_key_t key, struct dn_zone *dz)
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 dc806b57..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);
diff --git a/net/ipv6/icmp.c b/net/ipv6/icmp.c
index ff685f2..5176fc6 100644
--- a/net/ipv6/icmp.c
+++ b/net/ipv6/icmp.c
@@ -67,7 +67,7 @@
#include <asm/uaccess.h>
#include <asm/system.h>
-DEFINE_SNMP_STAT(struct icmpv6_mib, icmpv6_statistics);
+DEFINE_SNMP_STAT(struct icmpv6_mib, icmpv6_statistics) __read_mostly;
/*
* The ICMP socket(s). This is the most convenient way to flow control
diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c
index 1b354aa..16af874 100644
--- a/net/ipv6/ip6_fib.c
+++ b/net/ipv6/ip6_fib.c
@@ -49,7 +49,7 @@
struct rt6_statistics rt6_stats;
-static kmem_cache_t * fib6_node_kmem;
+static kmem_cache_t * fib6_node_kmem __read_mostly;
enum fib_walk_state_t
{
diff --git a/net/ipv6/ipv6_sockglue.c b/net/ipv6/ipv6_sockglue.c
index 7516b88..76466af 100644
--- a/net/ipv6/ipv6_sockglue.c
+++ b/net/ipv6/ipv6_sockglue.c
@@ -55,7 +55,7 @@
#include <asm/uaccess.h>
-DEFINE_SNMP_STAT(struct ipstats_mib, ipv6_statistics);
+DEFINE_SNMP_STAT(struct ipstats_mib, ipv6_statistics) __read_mostly;
static struct packet_type ipv6_packet_type = {
.type = __constant_htons(ETH_P_IPV6),
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 67d9a04..390d750 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -59,7 +59,7 @@
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
-DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6);
+DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
/* Grrr, addr_type already calculated by caller, but I don't want
* to add some silly "cookie" argument to this method just for that.
diff --git a/net/ipv6/xfrm6_tunnel.c b/net/ipv6/xfrm6_tunnel.c
index 60c26c8..fbef782 100644
--- a/net/ipv6/xfrm6_tunnel.c
+++ b/net/ipv6/xfrm6_tunnel.c
@@ -79,7 +79,7 @@
#define XFRM6_TUNNEL_SPI_MIN 1
#define XFRM6_TUNNEL_SPI_MAX 0xffffffff
-static kmem_cache_t *xfrm6_tunnel_spi_kmem;
+static kmem_cache_t *xfrm6_tunnel_spi_kmem __read_mostly;
#define XFRM6_TUNNEL_SPI_BYADDR_HSIZE 256
#define XFRM6_TUNNEL_SPI_BYSPI_HSIZE 256
diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c
index 7d8ec65..e7025be 100644
--- a/net/sctp/protocol.c
+++ b/net/sctp/protocol.c
@@ -62,7 +62,7 @@
/* Global data structures. */
struct sctp_globals sctp_globals;
struct proc_dir_entry *proc_net_sctp;
-DEFINE_SNMP_STAT(struct sctp_mib, sctp_statistics);
+DEFINE_SNMP_STAT(struct sctp_mib, sctp_statistics) __read_mostly;
struct idr sctp_assocs_id;
DEFINE_SPINLOCK(sctp_assocs_id_lock);
@@ -78,8 +78,8 @@
static struct sctp_af *sctp_af_v4_specific;
static struct sctp_af *sctp_af_v6_specific;
-kmem_cache_t *sctp_chunk_cachep;
-kmem_cache_t *sctp_bucket_cachep;
+kmem_cache_t *sctp_chunk_cachep __read_mostly;
+kmem_cache_t *sctp_bucket_cachep __read_mostly;
extern int sctp_snmp_proc_init(void);
extern int sctp_snmp_proc_exit(void);
diff --git a/net/socket.c b/net/socket.c
index ce69b78..94fe638 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -274,7 +274,7 @@
#define SOCKFS_MAGIC 0x534F434B
-static kmem_cache_t * sock_inode_cachep;
+static kmem_cache_t * sock_inode_cachep __read_mostly;
static struct inode *sock_alloc_inode(struct super_block *sb)
{
@@ -333,7 +333,7 @@
return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC);
}
-static struct vfsmount *sock_mnt;
+static struct vfsmount *sock_mnt __read_mostly;
static struct file_system_type sock_fs_type = {
.name = "sockfs",
diff --git a/net/sunrpc/rpc_pipe.c b/net/sunrpc/rpc_pipe.c
index 554f224..fe1a73c 100644
--- a/net/sunrpc/rpc_pipe.c
+++ b/net/sunrpc/rpc_pipe.c
@@ -28,13 +28,13 @@
#include <linux/workqueue.h>
#include <linux/sunrpc/rpc_pipe_fs.h>
-static struct vfsmount *rpc_mount;
+static struct vfsmount *rpc_mount __read_mostly;
static int rpc_mount_count;
static struct file_system_type rpc_pipe_fs_type;
-static kmem_cache_t *rpc_inode_cachep;
+static kmem_cache_t *rpc_inode_cachep __read_mostly;
#define RPC_UPCALL_TIMEOUT (30*HZ)
diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index 2d9eb7f..f310403 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -34,10 +34,10 @@
#define RPC_BUFFER_MAXSIZE (2048)
#define RPC_BUFFER_POOLSIZE (8)
#define RPC_TASK_POOLSIZE (8)
-static kmem_cache_t *rpc_task_slabp;
-static kmem_cache_t *rpc_buffer_slabp;
-static mempool_t *rpc_task_mempool;
-static mempool_t *rpc_buffer_mempool;
+static kmem_cache_t *rpc_task_slabp __read_mostly;
+static kmem_cache_t *rpc_buffer_slabp __read_mostly;
+static mempool_t *rpc_task_mempool __read_mostly;
+static mempool_t *rpc_buffer_mempool __read_mostly;
static void __rpc_default_timer(struct rpc_task *task);
static void rpciod_killall(void);
diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
index c58a6f0..2407a70 100644
--- a/net/xfrm/xfrm_input.c
+++ b/net/xfrm/xfrm_input.c
@@ -12,7 +12,7 @@
#include <net/ip.h>
#include <net/xfrm.h>
-static kmem_cache_t *secpath_cachep;
+static kmem_cache_t *secpath_cachep __read_mostly;
void __secpath_destroy(struct sec_path *sp)
{
diff --git a/net/xfrm/xfrm_policy.c b/net/xfrm/xfrm_policy.c
index d65ed86..83c8135 100644
--- a/net/xfrm/xfrm_policy.c
+++ b/net/xfrm/xfrm_policy.c
@@ -37,7 +37,7 @@
static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
-static kmem_cache_t *xfrm_dst_cache;
+static kmem_cache_t *xfrm_dst_cache __read_mostly;
static struct work_struct xfrm_policy_gc_work;
static struct list_head xfrm_policy_gc_list =