Skip to contentSkip to conversation

tcp: remove two indirect calls from xmit path

PATCH In review Eric Dumazet 2 participants 4 messages Jun 19, 2020 – Jun 21, 2020

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

cover tcp: remove two indirect calls from xmit path

__tcp_transmit_skb() does two indirect calls per packet, lets get rid of them.

Eric Dumazet (2):
tcp: remove indirect calls for icsk->icsk_af_ops->queue_xmit
tcp: remove indirect calls for icsk->icsk_af_ops->send_check

 include/net/ip.h           |  6 +-----
 include/net/ip6_checksum.h |  9 ---------
 include/net/tcp.h          |  4 ++++
 net/ipv4/ip_output.c       |  6 ++++++
 net/ipv4/tcp_output.c      | 12 ++++++++++--
 net/ipv6/tcp_ipv6.c        |  7 +++++++
 6 files changed, 28 insertions(+), 16 deletions(-)
Signature · 1 line
2.27.0.111.gc72c7da667-goog
#1 lore raw

PATCH 1/2 tcp: remove indirect calls for icsk->icsk_af_ops->queue_xmit

Mitigate RETPOLINE costs in __tcp_transmit_skb()
by using INDIRECT_CALL_INET() wrapper.

---
 include/net/ip.h      | 6 +-----
 include/net/tcp.h     | 1 +
 net/ipv4/ip_output.c  | 6 ++++++
 net/ipv4/tcp_output.c | 7 ++++++-
 4 files changed, 14 insertions(+), 6 deletions(-)
include/net/ip.h +1 -5
index 04ebe7bf54c6a8b21cf7894691b36e564ef1aef1..862c9545833a95e25fdcddafcfc84cd07421835d 100644
@@ -231,11 +231,7 @@ struct sk_buff *ip_make_skb(struct sock *sk, struct flowi4 *fl4,
 			    struct ipcm_cookie *ipc, struct rtable **rtp,
 			    struct inet_cork *cork, unsigned int flags);
 
-static inline int ip_queue_xmit(struct sock *sk, struct sk_buff *skb,
-				struct flowi *fl)
-{
-	return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
-}
+int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl);
 
 static inline struct sk_buff *ip_finish_skb(struct sock *sk, struct flowi4 *fl4)
 {
include/net/tcp.h +1 -0
index 4de9485f73d97d0b44f1e423bcbc8a3598749adc..e5d7e0b099245cf245a5f1c994d164a9fff66124 100644
@@ -25,6 +25,7 @@
 #include <linux/skbuff.h>
 #include <linux/kref.h>
 #include <linux/ktime.h>
+#include <linux/indirect_call_wrapper.h>
 
 #include <net/inet_connection_sock.h>
 #include <net/inet_timewait_sock.h>
net/ipv4/ip_output.c +6 -0
index 090d3097ee15baa87695c530278761fb26534ad5..d946356187eddfb0f81a2f632ee863fafad5f89c 100644
@@ -539,6 +539,12 @@ int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 }
 EXPORT_SYMBOL(__ip_queue_xmit);
 
+int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
+{
+	return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
+}
+EXPORT_SYMBOL(ip_queue_xmit);
+
 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 {
 	to->pkt_type = from->pkt_type;
net/ipv4/tcp_output.c +6 -1
index a50e1990a845a258d4cc6a2a989d09068ea3a973..be1bd37185d82e5827dfc5105ae74cd815ba1877 100644
@@ -1064,6 +1064,9 @@ static void tcp_update_skb_after_send(struct sock *sk, struct sk_buff *skb,
 	list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue);
 }
 
+INDIRECT_CALLABLE_DECLARE(int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl));
+INDIRECT_CALLABLE_DECLARE(int inet6_csk_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl));
+
 /* This routine actually transmits TCP packets queued in by
  * tcp_do_sendmsg().  This is used by both the initial
  * transmission and possible later retransmissions.
@@ -1235,7 +1238,9 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
 
 	tcp_add_tx_delay(skb, tp);
 
-	err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
+	err = INDIRECT_CALL_INET(icsk->icsk_af_ops->queue_xmit,
+				 inet6_csk_xmit, ip_queue_xmit,
+				 sk, skb, &inet->cork.fl);
 
 	if (unlikely(err > 0)) {
 		tcp_enter_cwr(sk);
Signature · 1 line
2.27.0.111.gc72c7da667-goog
#2 lore raw

PATCH 2/2 tcp: remove indirect calls for icsk->icsk_af_ops->send_check

In reply to Eric Dumazet

Mitigate RETPOLINE costs in __tcp_transmit_skb()
by using INDIRECT_CALL_INET() wrapper.

---
 include/net/ip6_checksum.h | 9 ---------
 include/net/tcp.h          | 3 +++
 net/ipv4/tcp_output.c      | 5 ++++-
 net/ipv6/tcp_ipv6.c        | 7 +++++++
 4 files changed, 14 insertions(+), 10 deletions(-)
include/net/ip6_checksum.h +0 -9
index 27ec612cd4a4598dc01e4984e61bc42c6ee5e77b..b3f4eaa88672a2e64ec3fbb3e77a60fe383e59d9 100644
@@ -85,15 +85,6 @@ static inline void tcp_v6_gso_csum_prep(struct sk_buff *skb)
 	th->check = ~tcp_v6_check(0, &ipv6h->saddr, &ipv6h->daddr, 0);
 }
 
-#if IS_ENABLED(CONFIG_IPV6)
-static inline void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
-{
-	struct ipv6_pinfo *np = inet6_sk(sk);
-
-	__tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
-}
-#endif
-
 static inline __sum16 udp_v6_check(int len,
 				   const struct in6_addr *saddr,
 				   const struct in6_addr *daddr,
include/net/tcp.h +3 -0
index e5d7e0b099245cf245a5f1c994d164a9fff66124..cd9cc348dbf9c62efa30d909f23a0ed1b39e4492 100644
@@ -932,6 +932,9 @@ static inline int tcp_v6_sdif(const struct sk_buff *skb)
 #endif
 	return 0;
 }
+
+INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
+
 #endif
 
 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
net/ipv4/tcp_output.c +4 -1
index be1bd37185d82e5827dfc5105ae74cd815ba1877..04b70fe31fa2cdad739ca49447cf4dd7e9d90cce 100644
@@ -1066,6 +1066,7 @@ static void tcp_update_skb_after_send(struct sock *sk, struct sk_buff *skb,
 
 INDIRECT_CALLABLE_DECLARE(int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl));
 INDIRECT_CALLABLE_DECLARE(int inet6_csk_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl));
+INDIRECT_CALLABLE_DECLARE(void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb));
 
 /* This routine actually transmits TCP packets queued in by
  * tcp_do_sendmsg().  This is used by both the initial
@@ -1210,7 +1211,9 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
 	}
 #endif
 
-	icsk->icsk_af_ops->send_check(sk, skb);
+	INDIRECT_CALL_INET(icsk->icsk_af_ops->send_check,
+			   tcp_v6_send_check, tcp_v4_send_check,
+			   sk, skb);
 
 	if (likely(tcb->tcp_flags & TCPHDR_ACK))
 		tcp_event_ack_sent(sk, tcp_skb_pcount(skb), rcv_nxt);
net/ipv6/tcp_ipv6.c +7 -0
index f67d45ff00b468f8c50053f6d2e430a1f5247db5..4502db706f75349b49672a859492a036a8722bb6 100644
@@ -1811,6 +1811,13 @@ static struct timewait_sock_ops tcp6_timewait_sock_ops = {
 	.twsk_destructor = tcp_twsk_destructor,
 };
 
+INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
+{
+	struct ipv6_pinfo *np = inet6_sk(sk);
+
+	__tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
+}
+
 const struct inet_connection_sock_af_ops ipv6_specific = {
 	.queue_xmit	   = inet6_csk_xmit,
 	.send_check	   = tcp_v6_send_check,
Signature · 1 line
2.27.0.111.gc72c7da667-goog