tcp: switch TCP TS option (RFC 7323) to 1ms clock

TCP Timestamps option is defined in RFC 7323

Traditionally on linux, it has been tied to the internal
'jiffies' variable, because it had been a cheap and good enough
generator.

For TCP flows on the Internet, 1 ms resolution would be much better
than 4ms or 10ms (HZ=250 or HZ=100 respectively)

For TCP flows in the DC, Google has used usec resolution for more
than two years with great success [1]

Receive size autotuning (DRS) is indeed more precise and converges
faster to optimal window size.

This patch converts tp->tcp_mstamp to a plain u64 value storing
a 1 usec TCP clock.

This choice will allow us to upstream the 1 usec TS option as
discussed in IETF 97.

[1] https://www.ietf.org/proceedings/97/slides/slides-97-tcpm-tcp-options-for-low-latency-00.pdf

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 62 insertions, 81 deletions

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index bfc7892f6c33..7c0cb2ce8b01 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -506,66 +506,6 @@ typedef unsigned int sk_buff_data_t;
 typedef unsigned char *sk_buff_data_t;
 #endif
 
-/**
- * struct skb_mstamp - multi resolution time stamps
- * @stamp_us: timestamp in us resolution
- * @stamp_jiffies: timestamp in jiffies
- */
-struct skb_mstamp {
-	union {
-		u64		v64;
-		struct {
-			u32	stamp_us;
-			u32	stamp_jiffies;
-		};
-	};
-};
-
-/**
- * skb_mstamp_get - get current timestamp
- * @cl: place to store timestamps
- */
-static inline void skb_mstamp_get(struct skb_mstamp *cl)
-{
-	u64 val = local_clock();
-
-	do_div(val, NSEC_PER_USEC);
-	cl->stamp_us = (u32)val;
-	cl->stamp_jiffies = (u32)jiffies;
-}
-
-/**
- * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
- * @t1: pointer to newest sample
- * @t0: pointer to oldest sample
- */
-static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
-				      const struct skb_mstamp *t0)
-{
-	s32 delta_us = t1->stamp_us - t0->stamp_us;
-	u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
-
-	/* If delta_us is negative, this might be because interval is too big,
-	 * or local_clock() drift is too big : fallback using jiffies.
-	 */
-	if (delta_us <= 0 ||
-	    delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
-
-		delta_us = jiffies_to_usecs(delta_jiffies);
-
-	return delta_us;
-}
-
-static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
-				    const struct skb_mstamp *t0)
-{
-	s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
-
-	if (!diff)
-		diff = t1->stamp_us - t0->stamp_us;
-	return diff > 0;
-}
-
 /** 
  *	struct sk_buff - socket buffer
  *	@next: Next buffer in list
@@ -646,7 +586,7 @@ struct sk_buff {
 
 			union {
 				ktime_t		tstamp;
-				struct skb_mstamp skb_mstamp;
+				u64		skb_mstamp;
 			};
 		};
 		struct rb_node	rbnode; /* used in netem & tcp stack */
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 22854f028434..542ca1ae02c4 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -123,7 +123,7 @@ struct tcp_request_sock_ops;
 struct tcp_request_sock {
 	struct inet_request_sock 	req;
 	const struct tcp_request_sock_ops *af_specific;
-	struct skb_mstamp		snt_synack; /* first SYNACK sent time */
+	u64				snt_synack; /* first SYNACK sent time */
 	bool				tfo_listener;
 	u32				txhash;
 	u32				rcv_isn;
@@ -211,7 +211,7 @@ struct tcp_sock {
 
 	/* Information of the most recently (s)acked skb */
 	struct tcp_rack {
-		struct skb_mstamp mstamp; /* (Re)sent time of the skb */
+		u64 mstamp; /* (Re)sent time of the skb */
 		u32 rtt_us;  /* Associated RTT */
 		u32 end_seq; /* Ending TCP sequence of the skb */
 		u8 advanced; /* mstamp advanced since last lost marking */
@@ -240,7 +240,7 @@ struct tcp_sock {
 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP retransmit. */
 
 /* RTT measurement */
-	struct skb_mstamp tcp_mstamp; /* most recent packet received/sent */
+	u64	tcp_mstamp;	/* most recent packet received/sent */
 	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	*/
@@ -280,8 +280,8 @@ struct tcp_sock {
 	u32	delivered;	/* Total data packets delivered incl. rexmits */
 	u32	lost;		/* Total data packets lost incl. rexmits */
 	u32	app_limited;	/* limited until "delivered" reaches this val */
-	struct skb_mstamp first_tx_mstamp;  /* start of window send phase */
-	struct skb_mstamp delivered_mstamp; /* time we reached "delivered" */
+	u64	first_tx_mstamp;  /* start of window send phase */
+	u64	delivered_mstamp; /* time we reached "delivered" */
 	u32	rate_delivered;    /* saved rate sample: packets delivered */
 	u32	rate_interval_us;  /* saved rate sample: time elapsed */
 
@@ -335,16 +335,16 @@ struct tcp_sock {
 
 /* Receiver side RTT estimation */
 	struct {
-		u32		rtt_us;
-		u32		seq;
-		struct skb_mstamp time;
+		u32	rtt_us;
+		u32	seq;
+		u64	time;
 	} rcv_rtt_est;
 
 /* Receiver queue space */
 	struct {
-		int		space;
-		u32		seq;
-		struct skb_mstamp time;
+		int	space;
+		u32	seq;
+		u64	time;
 	} rcvq_space;
 
 /* TCP-specific MTU probe information. */
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 5b2932b8363f..82462db97183 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -519,7 +519,7 @@ static inline u32 tcp_cookie_time(void)
 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
 			      u16 *mssp);
 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
-__u32 cookie_init_timestamp(struct request_sock *req);
+u64 cookie_init_timestamp(struct request_sock *req);
 bool cookie_timestamp_decode(struct tcp_options_received *opt);
 bool cookie_ecn_ok(const struct tcp_options_received *opt,
 		   const struct net *net, const struct dst_entry *dst);
@@ -706,14 +706,55 @@ void tcp_send_window_probe(struct sock *sk);
  */
 #define tcp_jiffies32 ((u32)jiffies)
 
-/* Generator for TCP TS option (RFC 7323)
- * Currently tied to 'jiffies' but will soon be driven by 1 ms clock.
+/*
+ * Deliver a 32bit value for TCP timestamp option (RFC 7323)
+ * It is no longer tied to jiffies, but to 1 ms clock.
+ * Note: double check if you want to use tcp_jiffies32 instead of this.
+ */
+#define TCP_TS_HZ	1000
+
+static inline u64 tcp_clock_ns(void)
+{
+	return local_clock();
+}
+
+static inline u64 tcp_clock_us(void)
+{
+	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
+}
+
+/* This should only be used in contexts where tp->tcp_mstamp is up to date */
+static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
+{
+	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
+}
+
+/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
+static inline u32 tcp_time_stamp_raw(void)
+{
+	return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
+}
+
+
+/* Refresh 1us clock of a TCP socket,
+ * ensuring monotically increasing values.
  */
-#define tcp_time_stamp		((__u32)(jiffies))
+static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
+{
+	u64 val = tcp_clock_us();
+
+	if (val > tp->tcp_mstamp)
+		tp->tcp_mstamp = val;
+}
+
+static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
+{
+	return max_t(s64, t1 - t0, 0);
+}
 
 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
 {
-	return skb->skb_mstamp.stamp_jiffies;
+	return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
 }
 
 
@@ -778,9 +819,9 @@ struct tcp_skb_cb {
 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
 			__u32 delivered;
 			/* start of send pipeline phase */
-			struct skb_mstamp first_tx_mstamp;
+			u64 first_tx_mstamp;
 			/* when we reached the "delivered" count */
-			struct skb_mstamp delivered_mstamp;
+			u64 delivered_mstamp;
 		} tx;   /* only used for outgoing skbs */
 		union {
 			struct inet_skb_parm	h4;
@@ -896,7 +937,7 @@ struct ack_sample {
  * A sample is invalid if "delivered" or "interval_us" is negative.
  */
 struct rate_sample {
-	struct	skb_mstamp prior_mstamp; /* starting timestamp for interval */
+	u64  prior_mstamp; /* starting timestamp for interval */
 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
 	s32  delivered;		/* number of packets delivered over interval */
 	long interval_us;	/* time for tp->delivered to incr "delivered" */
@@ -1862,7 +1903,7 @@ void tcp_init(void);
 /* tcp_recovery.c */
 extern void tcp_rack_mark_lost(struct sock *sk);
 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
-			     const struct skb_mstamp *xmit_time);
+			     u64 xmit_time);
 extern void tcp_rack_reo_timeout(struct sock *sk);
 
 /*