tcp: switch rtt estimations to usec resolution

Upcoming congestion controls for TCP require usec resolution for RTT
estimations. Millisecond resolution is simply not enough these days.

FQ/pacing in DC environments also require this change for finer control
and removal of bimodal behavior due to the current hack in
tcp_update_pacing_rate() for 'small rtt'

TCP_CONG_RTT_STAMP is no longer needed.

As Julian Anastasov pointed out, we need to keep user compatibility :
tcp_metrics used to export RTT and RTTVAR in msec resolution,
so we added RTT_US and RTTVAR_US. An iproute2 patch is needed
to use the new attributes if provided by the kernel.

In this example ss command displays a srtt of 32 usecs (10Gbit link)

lpk51:~# ./ss -i dst lpk52
Netid  State      Recv-Q Send-Q   Local Address:Port       Peer
Address:Port
tcp    ESTAB      0      1         10.246.11.51:42959
10.246.11.52:64614
         cubic wscale:6,6 rto:201 rtt:0.032/0.001 ato:40 mss:1448
cwnd:10 send
3620.0Mbps pacing_rate 7240.0Mbps unacked:1 rcv_rtt:993 rcv_space:29559

Updated iproute2 ip command displays :

lpk51:~# ./ip tcp_metrics | grep 10.246.11.52
10.246.11.52 age 561.914sec cwnd 10 rtt 274us rttvar 213us source
10.246.11.51

Old binary displays :

lpk51:~# ip tcp_metrics | grep 10.246.11.52
10.246.11.52 age 561.914sec cwnd 10 rtt 250us rttvar 125us source
10.246.11.51

With help from Julian Anastasov, Stephen Hemminger and Yuchung Cheng

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Larry Brakmo <brakmo@google.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/tcp.h

@@ -201,10 +201,10 @@ struct tcp_sock {
 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP retransmit. */
 
 /* RTT measurement */
-	u32	srtt;		/* smoothed round trip time << 3	*/
-	u32	mdev;		/* medium deviation			*/
-	u32	mdev_max;	/* maximal mdev for the last rtt period	*/
-	u32	rttvar;		/* smoothed mdev_max			*/
+	u32	srtt_us;	/* smoothed round trip time << 3 in usecs */
+	u32	mdev_us;	/* medium deviation			*/
+	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
+	u32	rttvar_us;	/* smoothed mdev_max			*/
 	u32	rtt_seq;	/* sequence number to update rttvar	*/
 
 	u32	packets_out;	/* Packets which are "in flight"	*/

include/net/tcp.h

@@ -31,6 +31,7 @@
 #include <linux/crypto.h>
 #include <linux/cryptohash.h>
 #include <linux/kref.h>
+#include <linux/ktime.h>
 
 #include <net/inet_connection_sock.h>
 #include <net/inet_timewait_sock.h>
@@ -478,7 +479,6 @@ int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 			     struct ip_options *opt);
 #ifdef CONFIG_SYN_COOKIES
-#include <linux/ktime.h>
 
 /* Syncookies use a monotonic timer which increments every 64 seconds.
  * This counter is used both as a hash input and partially encoded into
@@ -619,7 +619,7 @@ static inline void tcp_bound_rto(const struct sock *sk)
 
 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
 {
-	return (tp->srtt >> 3) + tp->rttvar;
+	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
 }
 
 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
@@ -656,6 +656,11 @@ static inline u32 tcp_rto_min(struct sock *sk)
 	return rto_min;
 }
 
+static inline u32 tcp_rto_min_us(struct sock *sk)
+{
+	return jiffies_to_usecs(tcp_rto_min(sk));
+}
+
 /* Compute the actual receive window we are currently advertising.
  * Rcv_nxt can be after the window if our peer push more data
  * than the offered window.
@@ -778,7 +783,6 @@ enum tcp_ca_event {
 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
 
 #define TCP_CONG_NON_RESTRICTED 0x1
-#define TCP_CONG_RTT_STAMP	0x2
 
 struct tcp_congestion_ops {
 	struct list_head	list;

include/uapi/linux/tcp_metrics.h

@@ -11,12 +11,15 @@
 #define TCP_METRICS_GENL_VERSION	0x1
 
 enum tcp_metric_index {
-	TCP_METRIC_RTT,
-	TCP_METRIC_RTTVAR,
+	TCP_METRIC_RTT,		/* in ms units */
+	TCP_METRIC_RTTVAR,	/* in ms units */
 	TCP_METRIC_SSTHRESH,
 	TCP_METRIC_CWND,
 	TCP_METRIC_REORDERING,
 
+	TCP_METRIC_RTT_US,	/* in usec units */
+	TCP_METRIC_RTTVAR_US,	/* in usec units */
+
 	/* Always last.  */
 	__TCP_METRIC_MAX,
 };

net/ipv4/tcp.c

@@ -387,7 +387,7 @@ void tcp_init_sock(struct sock *sk)
 	INIT_LIST_HEAD(&tp->tsq_node);
 
 	icsk->icsk_rto = TCP_TIMEOUT_INIT;
-	tp->mdev = TCP_TIMEOUT_INIT;
+	tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
 
 	/* So many TCP implementations out there (incorrectly) count the
 	 * initial SYN frame in their delayed-ACK and congestion control
@@ -2339,7 +2339,7 @@ int tcp_disconnect(struct sock *sk, int flags)
 
 	sk->sk_shutdown = 0;
 	sock_reset_flag(sk, SOCK_DONE);
-	tp->srtt = 0;
+	tp->srtt_us = 0;
 	if ((tp->write_seq += tp->max_window + 2) == 0)
 		tp->write_seq = 1;
 	icsk->icsk_backoff = 0;
@@ -2783,8 +2783,8 @@ void tcp_get_info(const struct sock *sk, struct tcp_info *info)
 
 	info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
 	info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
-	info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
-	info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
+	info->tcpi_rtt = tp->srtt_us >> 3;
+	info->tcpi_rttvar = tp->mdev_us >> 2;
 	info->tcpi_snd_ssthresh = tp->snd_ssthresh;
 	info->tcpi_snd_cwnd = tp->snd_cwnd;
 	info->tcpi_advmss = tp->advmss;

net/ipv4/tcp_cubic.c

@@ -476,10 +476,6 @@ static int __init cubictcp_register(void)
 	/* divide by bic_scale and by constant Srtt (100ms) */
 	do_div(cube_factor, bic_scale * 10);
 
-	/* hystart needs ms clock resolution */
-	if (hystart && HZ < 1000)
-		cubictcp.flags |= TCP_CONG_RTT_STAMP;
-
 	return tcp_register_congestion_control(&cubictcp);
 }
 

net/ipv4/tcp_hybla.c

@@ -21,7 +21,7 @@ struct hybla {
 	u32   rho2;	      /* Rho * Rho, integer part */
 	u32   rho_3ls;	      /* Rho parameter, <<3 */
 	u32   rho2_7ls;	      /* Rho^2, <<7	*/
-	u32   minrtt;	      /* Minimum smoothed round trip time value seen */
+	u32   minrtt_us;      /* Minimum smoothed round trip time value seen */
 };
 
 /* Hybla reference round trip time (default= 1/40 sec = 25 ms), in ms */
@@ -35,7 +35,9 @@ static inline void hybla_recalc_param (struct sock *sk)
 {
 	struct hybla *ca = inet_csk_ca(sk);
 
-	ca->rho_3ls = max_t(u32, tcp_sk(sk)->srtt / msecs_to_jiffies(rtt0), 8);
+	ca->rho_3ls = max_t(u32,
+			    tcp_sk(sk)->srtt_us / (rtt0 * USEC_PER_MSEC),
+			    8U);
 	ca->rho = ca->rho_3ls >> 3;
 	ca->rho2_7ls = (ca->rho_3ls * ca->rho_3ls) << 1;
 	ca->rho2 = ca->rho2_7ls >> 7;
@@ -59,7 +61,7 @@ static void hybla_init(struct sock *sk)
 	hybla_recalc_param(sk);
 
 	/* set minimum rtt as this is the 1st ever seen */
-	ca->minrtt = tp->srtt;
+	ca->minrtt_us = tp->srtt_us;
 	tp->snd_cwnd = ca->rho;
 }
 
@@ -94,9 +96,9 @@ static void hybla_cong_avoid(struct sock *sk, u32 ack, u32 acked,
 	int is_slowstart = 0;
 
 	/*  Recalculate rho only if this srtt is the lowest */
-	if (tp->srtt < ca->minrtt){
+	if (tp->srtt_us < ca->minrtt_us) {
 		hybla_recalc_param(sk);
-		ca->minrtt = tp->srtt;
+		ca->minrtt_us = tp->srtt_us;
 	}
 
 	if (!tcp_is_cwnd_limited(sk, in_flight))

net/ipv4/tcp_illinois.c

@@ -325,7 +325,6 @@ static void tcp_illinois_info(struct sock *sk, u32 ext,
 }
 
 static struct tcp_congestion_ops tcp_illinois __read_mostly = {
-	.flags		= TCP_CONG_RTT_STAMP,
 	.init		= tcp_illinois_init,
 	.ssthresh	= tcp_illinois_ssthresh,
 	.cong_avoid	= tcp_illinois_cong_avoid,