[INET]: Uninline the __inet_hash function.
authorPavel Emelyanov <xemul@openvz.org>
Thu, 20 Dec 2007 23:31:33 +0000 (15:31 -0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 28 Jan 2008 22:59:26 +0000 (14:59 -0800)
This one is used in quite many places in the networking code and
seems to big to be inline.

After the patch net/ipv4/build-in.o loses ~650 bytes:
add/remove: 2/0 grow/shrink: 0/5 up/down: 461/-1114 (-653)
function                                     old     new   delta
__inet_hash_nolisten                           -     282    +282
__inet_hash                                    -     179    +179
tcp_sacktag_write_queue                     2255    2254      -1
__inet_lookup_listener                       284     274     -10
tcp_v4_syn_recv_sock                         755     493    -262
tcp_v4_hash                                  389      35    -354
inet_hash_connect                           1086     599    -487

This version addresses the issue pointed by Eric, that
while being inline this function was optimized by gcc
in respect to the 'listen_possible' argument.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/inet_hashtables.h
net/dccp/ipv4.c
net/ipv4/inet_hashtables.c
net/ipv4/tcp_ipv4.c

index 37f6cb11212723f7f2094a16b6cec92bbe9ef670..982d487751f9172d0fd333d66ffa978f8d7786cf 100644 (file)
@@ -264,37 +264,14 @@ static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
                wake_up(&hashinfo->lhash_wait);
 }
 
-static inline void __inet_hash(struct inet_hashinfo *hashinfo,
-                              struct sock *sk, const int listen_possible)
-{
-       struct hlist_head *list;
-       rwlock_t *lock;
-
-       BUG_TRAP(sk_unhashed(sk));
-       if (listen_possible && sk->sk_state == TCP_LISTEN) {
-               list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
-               lock = &hashinfo->lhash_lock;
-               inet_listen_wlock(hashinfo);
-       } else {
-               struct inet_ehash_bucket *head;
-               sk->sk_hash = inet_sk_ehashfn(sk);
-               head = inet_ehash_bucket(hashinfo, sk->sk_hash);
-               list = &head->chain;
-               lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
-               write_lock(lock);
-       }
-       __sk_add_node(sk, list);
-       sock_prot_inc_use(sk->sk_prot);
-       write_unlock(lock);
-       if (listen_possible && sk->sk_state == TCP_LISTEN)
-               wake_up(&hashinfo->lhash_wait);
-}
+extern void __inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk);
+extern void __inet_hash_nolisten(struct inet_hashinfo *hinfo, struct sock *sk);
 
 static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk)
 {
        if (sk->sk_state != TCP_CLOSE) {
                local_bh_disable();
-               __inet_hash(hashinfo, sk, 1);
+               __inet_hash(hashinfo, sk);
                local_bh_enable();
        }
 }
index 02fc91ce2505b1eb1260b5ad2d6c03c3a4876196..f450df2fc86374d72b88c4e9860b1020e3c2a8bf 100644 (file)
@@ -408,7 +408,7 @@ struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 
        dccp_sync_mss(newsk, dst_mtu(dst));
 
-       __inet_hash(&dccp_hashinfo, newsk, 0);
+       __inet_hash_nolisten(&dccp_hashinfo, newsk);
        __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 
        return newsk;
index 67704da04fc4ca98d88eeaae7db3e53d1a6157aa..8dfd5a691e5f2ae3ee85bf508f3a9692e6df10ce 100644 (file)
@@ -267,6 +267,48 @@ static inline u32 inet_sk_port_offset(const struct sock *sk)
                                          inet->dport);
 }
 
+void __inet_hash_nolisten(struct inet_hashinfo *hashinfo, struct sock *sk)
+{
+       struct hlist_head *list;
+       rwlock_t *lock;
+       struct inet_ehash_bucket *head;
+
+       BUG_TRAP(sk_unhashed(sk));
+
+       sk->sk_hash = inet_sk_ehashfn(sk);
+       head = inet_ehash_bucket(hashinfo, sk->sk_hash);
+       list = &head->chain;
+       lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
+
+       write_lock(lock);
+       __sk_add_node(sk, list);
+       sock_prot_inc_use(sk->sk_prot);
+       write_unlock(lock);
+}
+EXPORT_SYMBOL_GPL(__inet_hash_nolisten);
+
+void __inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk)
+{
+       struct hlist_head *list;
+       rwlock_t *lock;
+
+       if (sk->sk_state != TCP_LISTEN) {
+               __inet_hash_nolisten(hashinfo, sk);
+               return;
+       }
+
+       BUG_TRAP(sk_unhashed(sk));
+       list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
+       lock = &hashinfo->lhash_lock;
+
+       inet_listen_wlock(hashinfo);
+       __sk_add_node(sk, list);
+       sock_prot_inc_use(sk->sk_prot);
+       write_unlock(lock);
+       wake_up(&hashinfo->lhash_wait);
+}
+EXPORT_SYMBOL_GPL(__inet_hash);
+
 /*
  * Bind a port for a connect operation and hash it.
  */
@@ -334,7 +376,7 @@ ok:
                inet_bind_hash(sk, tb, port);
                if (sk_unhashed(sk)) {
                        inet_sk(sk)->sport = htons(port);
-                       __inet_hash(hinfo, sk, 0);
+                       __inet_hash_nolisten(hinfo, sk);
                }
                spin_unlock(&head->lock);
 
@@ -351,7 +393,7 @@ ok:
        tb  = inet_csk(sk)->icsk_bind_hash;
        spin_lock_bh(&head->lock);
        if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
-               __inet_hash(hinfo, sk, 0);
+               __inet_hash_nolisten(hinfo, sk);
                spin_unlock_bh(&head->lock);
                return 0;
        } else {
index 652c32368cccb6cb590c2c4eec96b77026288d0b..fc9bdd8b7dcf0db4d7e95a757aa62ae72cd157dc 100644 (file)
@@ -1478,7 +1478,7 @@ struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
        }
 #endif
 
-       __inet_hash(&tcp_hashinfo, newsk, 0);
+       __inet_hash_nolisten(&tcp_hashinfo, newsk);
        __inet_inherit_port(&tcp_hashinfo, sk, newsk);
 
        return newsk;