sync snd_nxt [1]: new fields to replace comparisons between write_seq, snd_nxt

required because write_seq will no longer be used to assign final seq# for pkts

include/linux/tcp.h

@@ -248,6 +248,8 @@ struct tcp_sock {
   u32 shared_dummy_out; /* number of dummies since last sync on this variable */
   u32 shared_dummy_lsndtime; /* timestamp of last dummy sent by hypervisor */
   u32 loss_prior_cwnd; /* original value of snd_cwnd before entering loss */
+  u64 gen_out; /* #bytes generated in tcp_sendmsg */
+  u64 xmit_out; /* #bytes generated in tcp_sendmsg and xmitted in tcp_write_xmit */
 #endif
 
 	u16	urg_data;	/* Saved octet of OOB data and control flags */

include/net/tcp.h

@@ -46,6 +46,10 @@
 #include <linux/seq_file.h>
 #include <linux/memcontrol.h>
 
+#if defined(CONFIG_PACER_TCP)
+#include "../../drivers/net/ethernet/broadcom/bnx2x/sme/ptcp_hooks.h"
+#endif
+
 extern struct inet_hashinfo tcp_hashinfo;
 
 extern struct percpu_counter tcp_orphan_count;
@@ -1769,7 +1773,16 @@ static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
 static inline bool tcp_stream_memory_free(const struct sock *sk)
 {
 	const struct tcp_sock *tp = tcp_sk(sk);
+#if defined(CONFIG_XEN_SME) && defined(CONFIG_PACER_TCP)
+  u32 notsent_bytes = 0;
+  if (ptcp_is_paced_flow((struct sock *) sk) == 0) {
+    notsent_bytes = (u32) (tp->xmit_out - tp->gen_out);
+  } else {
+    notsent_bytes = tp->write_seq - tp->snd_nxt;
+  }
+#else
 	u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
+#endif
 
 	return notsent_bytes < tcp_notsent_lowat(tp);
 }

net/ipv4/tcp.c

@@ -1249,6 +1249,7 @@ new_segment:
 				skb->ip_summed = CHECKSUM_PARTIAL;
 
 #if defined(CONFIG_XEN_SME) && defined(CONFIG_PACER_TCP)
+      tp->gen_out += PTCP_MTU;
       ptcp_ret = ptcp_rxtx_sync_shared_seq(sk, skb, PTCP_MTU);
       if (!ptcp_ret) {
         ptcp_skb_entail(sk, skb);

net/ipv4/tcp_minisocks.c

@@ -497,6 +497,8 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
     newtp->shared_seq = newtp->write_seq;
     newtp->shared_dummy_out = 0;
     newtp->shared_dummy_lsndtime = newtp->lsndtime;
+    newtp->gen_out = 0;
+    newtp->xmit_out = 0;
 #endif
 
 		newtp->rx_opt.saw_tstamp = 0;

net/ipv4/tcp_output.c

@@ -97,6 +97,7 @@ static void tcp_event_new_data_sent(struct sock *sk, const struct sk_buff *skb)
 	tp->packets_out += tcp_skb_pcount(skb);
 #else
   update_var_atomic(&tp->packets_out, tcp_skb_pcount(skb));
+  tp->xmit_out += PTCP_MTU;
 #if 0
   do {
     curr_packets_out = tp->packets_out;
@@ -2119,9 +2120,18 @@ static int tcp_mtu_probe(struct sock *sk)
 		return -1;
 	}
 
+#if defined(CONFIG_XEN_SME) && defined(CONFIG_PACER_TCP)
+  if (ptcp_is_paced_flow(sk) == 0) {
+    if (tp->gen_out - tp->xmit_out < size_needed)
+      return -1;
+  } else {
+#endif
 	/* Have enough data in the send queue to probe? */
 	if (tp->write_seq - tp->snd_nxt < size_needed)
 		return -1;
+#if defined(CONFIG_XEN_SME) && defined(CONFIG_PACER_TCP)
+  }
+#endif
 
 	if (tp->snd_wnd < size_needed)
 		return -1;

net/ipv4/tcp_rate.c

@@ -173,6 +173,16 @@ void tcp_rate_check_app_limited(struct sock *sk)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 
+#if defined(CONFIG_XEN_SME) && defined(CONFIG_PACER_TCP)
+  if (ptcp_is_paced_flow(sk) == 0) {
+    if (tp->gen_out - tp->xmit_out < tp->mss_cache &&
+        sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1) &&
+        tcp_packets_in_flight(tp) < tp->snd_cwnd &&
+        tp->lost_out <= tp->retrans_out)
+      tp->app_limited =
+        (tp->delivered + tcp_packets_in_flight(tp)) ? : 1;
+  } else {
+#endif
 	if (/* We have less than one packet to send. */
 	    tp->write_seq - tp->snd_nxt < tp->mss_cache &&
 	    /* Nothing in sending host's qdisc queues or NIC tx queue. */
@@ -183,4 +193,7 @@ void tcp_rate_check_app_limited(struct sock *sk)
 	    tp->lost_out <= tp->retrans_out)
 		tp->app_limited =
 			(tp->delivered + tcp_packets_in_flight(tp)) ? : 1;
+#if defined(CONFIG_XEN_SME) && defined(CONFIG_PACER_TCP)
+  }
+#endif
 }