[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ipv4 / tcp_cubic.c

/*
 * TCP CUBIC: Binary Increase Congestion control for TCP v2.3
 * Home page:
 *      http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
 * This is from the implementation of CUBIC TCP in
 * Sangtae Ha, Injong Rhee and Lisong Xu,
 *  "CUBIC: A New TCP-Friendly High-Speed TCP Variant"
 *  in ACM SIGOPS Operating System Review, July 2008.
 * Available from:
 *  http://netsrv.csc.ncsu.edu/export/cubic_a_new_tcp_2008.pdf
 *
 * CUBIC integrates a new slow start algorithm, called HyStart.
 * The details of HyStart are presented in
 *  Sangtae Ha and Injong Rhee,
 *  "Taming the Elephants: New TCP Slow Start", NCSU TechReport 2008.
 * Available from:
 *  http://netsrv.csc.ncsu.edu/export/hystart_techreport_2008.pdf
 *
 * All testing results are available from:
 * http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
 *
 * Unless CUBIC is enabled and congestion window is large
 * this behaves the same as the original Reno.
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/math64.h>
#include <net/tcp.h>

#define BICTCP_BETA_SCALE    1024	/* Scale factor beta calculation
					 * max_cwnd = snd_cwnd * beta
					 */
#define	BICTCP_HZ		10	/* BIC HZ 2^10 = 1024 */

/* Two methods of hybrid slow start */
#define HYSTART_ACK_TRAIN	0x1
#define HYSTART_DELAY		0x2

/* Number of delay samples for detecting the increase of delay */
#define HYSTART_MIN_SAMPLES	8
#define HYSTART_DELAY_MIN	(4U<<3)
#define HYSTART_DELAY_MAX	(16U<<3)
#define HYSTART_DELAY_THRESH(x)	clamp(x, HYSTART_DELAY_MIN, HYSTART_DELAY_MAX)

static int fast_convergence __read_mostly = 1;
static int beta __read_mostly = 717;	/* = 717/1024 (BICTCP_BETA_SCALE) */
static int initial_ssthresh __read_mostly;
static int bic_scale __read_mostly = 41;
static int tcp_friendliness __read_mostly = 1;

static int hystart __read_mostly = 1;
static int hystart_detect __read_mostly = HYSTART_ACK_TRAIN | HYSTART_DELAY;
static int hystart_low_window __read_mostly = 16;
static int hystart_ack_delta __read_mostly = 2;

static u32 cube_rtt_scale __read_mostly;
static u32 beta_scale __read_mostly;
static u64 cube_factor __read_mostly;

/* Note parameters that are used for precomputing scale factors are read-only */
module_param(fast_convergence, int, 0644);
MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
module_param(beta, int, 0644);
MODULE_PARM_DESC(beta, "beta for multiplicative increase");
module_param(initial_ssthresh, int, 0644);
MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
module_param(bic_scale, int, 0444);
MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
module_param(tcp_friendliness, int, 0644);
MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
module_param(hystart, int, 0644);
MODULE_PARM_DESC(hystart, "turn on/off hybrid slow start algorithm");
module_param(hystart_detect, int, 0644);
MODULE_PARM_DESC(hystart_detect, "hyrbrid slow start detection mechanisms"
		 " 1: packet-train 2: delay 3: both packet-train and delay");
module_param(hystart_low_window, int, 0644);
MODULE_PARM_DESC(hystart_low_window, "lower bound cwnd for hybrid slow start");
module_param(hystart_ack_delta, int, 0644);
MODULE_PARM_DESC(hystart_ack_delta, "spacing between ack's indicating train (msecs)");

/* BIC TCP Parameters */
struct bictcp {
	u32	cnt;		/* increase cwnd by 1 after ACKs */
	u32 	last_max_cwnd;	/* last maximum snd_cwnd */
	u32	loss_cwnd;	/* congestion window at last loss */
	u32	last_cwnd;	/* the last snd_cwnd */
	u32	last_time;	/* time when updated last_cwnd */
	u32	bic_origin_point;/* origin point of bic function */
	u32	bic_K;		/* time to origin point from the beginning of the current epoch */
	u32	delay_min;	/* min delay (msec << 3) */
	u32	epoch_start;	/* beginning of an epoch */
	u32	ack_cnt;	/* number of acks */
	u32	tcp_cwnd;	/* estimated tcp cwnd */
#define ACK_RATIO_SHIFT	4
#define ACK_RATIO_LIMIT (32u << ACK_RATIO_SHIFT)
	u16	delayed_ack;	/* estimate the ratio of Packets/ACKs << 4 */
	u8	sample_cnt;	/* number of samples to decide curr_rtt */
	u8	found;		/* the exit point is found? */
	u32	round_start;	/* beginning of each round */
	u32	end_seq;	/* end_seq of the round */
	u32	last_ack;	/* last time when the ACK spacing is close */
	u32	curr_rtt;	/* the minimum rtt of current round */
};

static inline void bictcp_reset(struct bictcp *ca)
{
	ca->cnt = 0;
	ca->last_max_cwnd = 0;
	ca->last_cwnd = 0;
	ca->last_time = 0;
	ca->bic_origin_point = 0;
	ca->bic_K = 0;
	ca->delay_min = 0;
	ca->epoch_start = 0;
	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	ca->ack_cnt = 0;
	ca->tcp_cwnd = 0;
	ca->found = 0;
}

static inline u32 bictcp_clock(void)
{
#if HZ < 1000
	return ktime_to_ms(ktime_get_real());
#else
	return jiffies_to_msecs(jiffies);
#endif
}

static inline void bictcp_hystart_reset(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	ca->round_start = ca->last_ack = bictcp_clock();
	ca->end_seq = tp->snd_nxt;
	ca->curr_rtt = 0;
	ca->sample_cnt = 0;
}

static void bictcp_init(struct sock *sk)
{
	struct bictcp *ca = inet_csk_ca(sk);

	bictcp_reset(ca);
	ca->loss_cwnd = 0;

	if (hystart)
		bictcp_hystart_reset(sk);

	if (!hystart && initial_ssthresh)
		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
}

/* calculate the cubic root of x using a table lookup followed by one
 * Newton-Raphson iteration.
 * Avg err ~= 0.195%
 */
static u32 cubic_root(u64 a)
{
	u32 x, b, shift;
	/*
	 * cbrt(x) MSB values for x MSB values in [0..63].
	 * Precomputed then refined by hand - Willy Tarreau
	 *
	 * For x in [0..63],
	 *   v = cbrt(x << 18) - 1
	 *   cbrt(x) = (v[x] + 10) >> 6
	 */
	static const u8 v[] = {
		/* 0x00 */    0,   54,   54,   54,  118,  118,  118,  118,
		/* 0x08 */  123,  129,  134,  138,  143,  147,  151,  156,
		/* 0x10 */  157,  161,  164,  168,  170,  173,  176,  179,
		/* 0x18 */  181,  185,  187,  190,  192,  194,  197,  199,
		/* 0x20 */  200,  202,  204,  206,  209,  211,  213,  215,
		/* 0x28 */  217,  219,  221,  222,  224,  225,  227,  229,
		/* 0x30 */  231,  232,  234,  236,  237,  239,  240,  242,
		/* 0x38 */  244,  245,  246,  248,  250,  251,  252,  254,
	};

	b = fls64(a);
	if (b < 7) {
		/* a in [0..63] */
		return ((u32)v[(u32)a] + 35) >> 6;
	}

	b = ((b * 84) >> 8) - 1;
	shift = (a >> (b * 3));

	x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;

	/*
	 * Newton-Raphson iteration
	 *                         2
	 * x    = ( 2 * x  +  a / x  ) / 3
	 *  k+1          k         k
	 */
	x = (2 * x + (u32)div64_u64(a, (u64)x * (u64)(x - 1)));
	x = ((x * 341) >> 10);
	return x;
}

/*
 * Compute congestion window to use.
 */
static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
{
	u32 delta, bic_target, max_cnt;
	u64 offs, t;

	ca->ack_cnt++;	/* count the number of ACKs */

	if (ca->last_cwnd == cwnd &&
	    (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
		return;

	ca->last_cwnd = cwnd;
	ca->last_time = tcp_time_stamp;

	if (ca->epoch_start == 0) {
		ca->epoch_start = tcp_time_stamp;	/* record the beginning of an epoch */
		ca->ack_cnt = 1;			/* start counting */
		ca->tcp_cwnd = cwnd;			/* syn with cubic */

		if (ca->last_max_cwnd <= cwnd) {
			ca->bic_K = 0;
			ca->bic_origin_point = cwnd;
		} else {
			/* Compute new K based on
			 * (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
			 */
			ca->bic_K = cubic_root(cube_factor
					       * (ca->last_max_cwnd - cwnd));
			ca->bic_origin_point = ca->last_max_cwnd;
		}
	}

	/* cubic function - calc*/
	/* calculate c * time^3 / rtt,
	 *  while considering overflow in calculation of time^3
	 * (so time^3 is done by using 64 bit)
	 * and without the support of division of 64bit numbers
	 * (so all divisions are done by using 32 bit)
	 *  also NOTE the unit of those veriables
	 *	  time  = (t - K) / 2^bictcp_HZ
	 *	  c = bic_scale >> 10
	 * rtt  = (srtt >> 3) / HZ
	 * !!! The following code does not have overflow problems,
	 * if the cwnd < 1 million packets !!!
	 */

	t = (s32)(tcp_time_stamp - ca->epoch_start);
	t += msecs_to_jiffies(ca->delay_min >> 3);
	/* change the unit from HZ to bictcp_HZ */
	t <<= BICTCP_HZ;
	do_div(t, HZ);

	if (t < ca->bic_K)		/* t - K */
		offs = ca->bic_K - t;
	else
		offs = t - ca->bic_K;

	/* c/rtt * (t-K)^3 */
	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
	if (t < ca->bic_K)                                	/* below origin*/
		bic_target = ca->bic_origin_point - delta;
	else                                                	/* above origin*/
		bic_target = ca->bic_origin_point + delta;

	/* cubic function - calc bictcp_cnt*/
	if (bic_target > cwnd) {
		ca->cnt = cwnd / (bic_target - cwnd);
	} else {
		ca->cnt = 100 * cwnd;              /* very small increment*/
	}

	/*
	 * The initial growth of cubic function may be too conservative
	 * when the available bandwidth is still unknown.
	 */
	if (ca->last_max_cwnd == 0 && ca->cnt > 20)
		ca->cnt = 20;	/* increase cwnd 5% per RTT */

	/* TCP Friendly */
	if (tcp_friendliness) {
		u32 scale = beta_scale;
		delta = (cwnd * scale) >> 3;
		while (ca->ack_cnt > delta) {		/* update tcp cwnd */
			ca->ack_cnt -= delta;
			ca->tcp_cwnd++;
		}

		if (ca->tcp_cwnd > cwnd){	/* if bic is slower than tcp */
			delta = ca->tcp_cwnd - cwnd;
			max_cnt = cwnd / delta;
			if (ca->cnt > max_cnt)
				ca->cnt = max_cnt;
		}
	}

	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	if (ca->cnt == 0)			/* cannot be zero */
		ca->cnt = 1;
}

static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	if (!tcp_is_cwnd_limited(sk, in_flight))
		return;

	if (tp->snd_cwnd <= tp->snd_ssthresh) {
		if (hystart && after(ack, ca->end_seq))
			bictcp_hystart_reset(sk);
		tcp_slow_start(tp);
	} else {
		bictcp_update(ca, tp->snd_cwnd);
		tcp_cong_avoid_ai(tp, ca->cnt);
	}

}

static u32 bictcp_recalc_ssthresh(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	ca->epoch_start = 0;	/* end of epoch */

	/* Wmax and fast convergence */
	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
		ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
			/ (2 * BICTCP_BETA_SCALE);
	else
		ca->last_max_cwnd = tp->snd_cwnd;

	ca->loss_cwnd = tp->snd_cwnd;

	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
}

static u32 bictcp_undo_cwnd(struct sock *sk)
{
	struct bictcp *ca = inet_csk_ca(sk);

	return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
}

static void bictcp_state(struct sock *sk, u8 new_state)
{
	if (new_state == TCP_CA_Loss) {
		bictcp_reset(inet_csk_ca(sk));
		bictcp_hystart_reset(sk);
	}
}

static void hystart_update(struct sock *sk, u32 delay)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	if (!(ca->found & hystart_detect)) {
		u32 now = bictcp_clock();

		/* first detection parameter - ack-train detection */
		if ((s32)(now - ca->last_ack) <= hystart_ack_delta) {
			ca->last_ack = now;
			if ((s32)(now - ca->round_start) > ca->delay_min >> 4)
				ca->found |= HYSTART_ACK_TRAIN;
		}

		/* obtain the minimum delay of more than sampling packets */
		if (ca->sample_cnt < HYSTART_MIN_SAMPLES) {
			if (ca->curr_rtt == 0 || ca->curr_rtt > delay)
				ca->curr_rtt = delay;

			ca->sample_cnt++;
		} else {
			if (ca->curr_rtt > ca->delay_min +
			    HYSTART_DELAY_THRESH(ca->delay_min>>4))
				ca->found |= HYSTART_DELAY;
		}
		/*
		 * Either one of two conditions are met,
		 * we exit from slow start immediately.
		 */
		if (ca->found & hystart_detect)
			tp->snd_ssthresh = tp->snd_cwnd;
	}
}

/* Track delayed acknowledgment ratio using sliding window
 * ratio = (15*ratio + sample) / 16
 */
static void bictcp_acked(struct sock *sk, u32 cnt, s32 rtt_us)
{
	const struct inet_connection_sock *icsk = inet_csk(sk);
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);
	u32 delay;

	if (icsk->icsk_ca_state == TCP_CA_Open) {
		u32 ratio = ca->delayed_ack;

		ratio -= ca->delayed_ack >> ACK_RATIO_SHIFT;
		ratio += cnt;

		ca->delayed_ack = clamp(ratio, 1U, ACK_RATIO_LIMIT);
	}

	/* Some calls are for duplicates without timetamps */
	if (rtt_us < 0)
		return;

	/* Discard delay samples right after fast recovery */
	if (ca->epoch_start && (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
		return;

	delay = (rtt_us << 3) / USEC_PER_MSEC;
	if (delay == 0)
		delay = 1;

	/* first time call or link delay decreases */
	if (ca->delay_min == 0 || ca->delay_min > delay)
		ca->delay_min = delay;

	/* hystart triggers when cwnd is larger than some threshold */
	if (hystart && tp->snd_cwnd <= tp->snd_ssthresh &&
	    tp->snd_cwnd >= hystart_low_window)
		hystart_update(sk, delay);
}

static struct tcp_congestion_ops cubictcp __read_mostly = {
	.init		= bictcp_init,
	.ssthresh	= bictcp_recalc_ssthresh,
	.cong_avoid	= bictcp_cong_avoid,
	.set_state	= bictcp_state,
	.undo_cwnd	= bictcp_undo_cwnd,
	.pkts_acked     = bictcp_acked,
	.owner		= THIS_MODULE,
	.name		= "cubic",
};

static int __init cubictcp_register(void)
{
	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);

	/* Precompute a bunch of the scaling factors that are used per-packet
	 * based on SRTT of 100ms
	 */

	beta_scale = 8*(BICTCP_BETA_SCALE+beta)/ 3 / (BICTCP_BETA_SCALE - beta);

	cube_rtt_scale = (bic_scale * 10);	/* 1024*c/rtt */

	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	 *  so K = cubic_root( (wmax-cwnd)*rtt/c )
	 * the unit of K is bictcp_HZ=2^10, not HZ
	 *
	 *  c = bic_scale >> 10
	 *  rtt = 100ms
	 *
	 * the following code has been designed and tested for
	 * cwnd < 1 million packets
	 * RTT < 100 seconds
	 * HZ < 1,000,00  (corresponding to 10 nano-second)
	 */

	/* 1/c * 2^2*bictcp_HZ * srtt */
	cube_factor = 1ull << (10+3*BICTCP_HZ); /* 2^40 */

	/* 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);
}

static void __exit cubictcp_unregister(void)
{
	tcp_unregister_congestion_control(&cubictcp);
}

module_init(cubictcp_register);
module_exit(cubictcp_unregister);

MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("CUBIC TCP");
MODULE_VERSION("2.3");
Commit	Line	Data
df3271f3	1	/*
ae27e98a	2	* TCP CUBIC: Binary Increase Congestion control for TCP v2.3
6b3d6263 SH	3	* Home page:
6b3d6263 SH	4	* http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
df3271f3	5	* This is from the implementation of CUBIC TCP in
ae27e98a SH	6	* Sangtae Ha, Injong Rhee and Lisong Xu,
	7	* "CUBIC: A New TCP-Friendly High-Speed TCP Variant"
	8	* in ACM SIGOPS Operating System Review, July 2008.
df3271f3	9	* Available from:
ae27e98a SH	10	* http://netsrv.csc.ncsu.edu/export/cubic_a_new_tcp_2008.pdf
	11	*
	12	* CUBIC integrates a new slow start algorithm, called HyStart.
	13	* The details of HyStart are presented in
	14	* Sangtae Ha and Injong Rhee,
	15	* "Taming the Elephants: New TCP Slow Start", NCSU TechReport 2008.
	16	* Available from:
	17	* http://netsrv.csc.ncsu.edu/export/hystart_techreport_2008.pdf
	18	*
	19	* All testing results are available from:
	20	* http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
df3271f3 SH	21	*
	22	* Unless CUBIC is enabled and congestion window is large
	23	* this behaves the same as the original Reno.
	24	*/
	25
df3271f3 SH	26	#include <linux/mm.h>
df3271f3 SH	27	#include <linux/module.h>
6f6d6a1a	28	#include <linux/math64.h>
df3271f3	29	#include <net/tcp.h>
df3271f3 SH	30
	31	#define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
	32	* max_cwnd = snd_cwnd * beta
	33	*/
df3271f3 SH	34	#define BICTCP_HZ 10 /* BIC HZ 2^10 = 1024 */
df3271f3 SH	35
ae27e98a SH	36	/* Two methods of hybrid slow start */
	37	#define HYSTART_ACK_TRAIN 0x1
	38	#define HYSTART_DELAY 0x2
	39
	40	/* Number of delay samples for detecting the increase of delay */
	41	#define HYSTART_MIN_SAMPLES 8
2b4636a5	42	#define HYSTART_DELAY_MIN (4U<<3)
ae27e98a SH	43	#define HYSTART_DELAY_MAX (16U<<3)
	44	#define HYSTART_DELAY_THRESH(x) clamp(x, HYSTART_DELAY_MIN, HYSTART_DELAY_MAX)
	45
59758f44	46	static int fast_convergence __read_mostly = 1;
6b3d6263	47	static int beta __read_mostly = 717; /* = 717/1024 (BICTCP_BETA_SCALE) */
66e1e3b2	48	static int initial_ssthresh __read_mostly;
59758f44 SH	49	static int bic_scale __read_mostly = 41;
59758f44 SH	50	static int tcp_friendliness __read_mostly = 1;
df3271f3	51
ae27e98a SH	52	static int hystart __read_mostly = 1;
	53	static int hystart_detect __read_mostly = HYSTART_ACK_TRAIN \| HYSTART_DELAY;
	54	static int hystart_low_window __read_mostly = 16;
aac46324	55	static int hystart_ack_delta __read_mostly = 2;
ae27e98a	56
59758f44 SH	57	static u32 cube_rtt_scale __read_mostly;
	58	static u32 beta_scale __read_mostly;
	59	static u64 cube_factor __read_mostly;
89b3d9aa SH	60
89b3d9aa SH	61	/* Note parameters that are used for precomputing scale factors are read-only */
df3271f3 SH	62	module_param(fast_convergence, int, 0644);
df3271f3 SH	63	MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
6b3d6263	64	module_param(beta, int, 0644);
df3271f3 SH	65	MODULE_PARM_DESC(beta, "beta for multiplicative increase");
	66	module_param(initial_ssthresh, int, 0644);
	67	MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
89b3d9aa	68	module_param(bic_scale, int, 0444);
df3271f3 SH	69	MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
	70	module_param(tcp_friendliness, int, 0644);
	71	MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
ae27e98a SH	72	module_param(hystart, int, 0644);
	73	MODULE_PARM_DESC(hystart, "turn on/off hybrid slow start algorithm");
	74	module_param(hystart_detect, int, 0644);
	75	MODULE_PARM_DESC(hystart_detect, "hyrbrid slow start detection mechanisms"
	76	" 1: packet-train 2: delay 3: both packet-train and delay");
	77	module_param(hystart_low_window, int, 0644);
	78	MODULE_PARM_DESC(hystart_low_window, "lower bound cwnd for hybrid slow start");
aac46324	79	module_param(hystart_ack_delta, int, 0644);
aac46324	80	MODULE_PARM_DESC(hystart_ack_delta, "spacing between ack's indicating train (msecs)");
df3271f3	81
df3271f3 SH	82	/* BIC TCP Parameters */
	83	struct bictcp {
	84	u32 cnt; /* increase cwnd by 1 after ACKs */
	85	u32 last_max_cwnd; /* last maximum snd_cwnd */
	86	u32 loss_cwnd; /* congestion window at last loss */
	87	u32 last_cwnd; /* the last snd_cwnd */
	88	u32 last_time; /* time when updated last_cwnd */
	89	u32 bic_origin_point;/* origin point of bic function */
	90	u32 bic_K; /* time to origin point from the beginning of the current epoch */
17a6e9f1	91	u32 delay_min; /* min delay (msec << 3) */
df3271f3 SH	92	u32 epoch_start; /* beginning of an epoch */
	93	u32 ack_cnt; /* number of acks */
	94	u32 tcp_cwnd; /* estimated tcp cwnd */
	95	#define ACK_RATIO_SHIFT 4
b9f47a3a	96	#define ACK_RATIO_LIMIT (32u << ACK_RATIO_SHIFT)
ae27e98a SH	97	u16 delayed_ack; /* estimate the ratio of Packets/ACKs << 4 */
	98	u8 sample_cnt; /* number of samples to decide curr_rtt */
	99	u8 found; /* the exit point is found? */
	100	u32 round_start; /* beginning of each round */
	101	u32 end_seq; /* end_seq of the round */
17a6e9f1	102	u32 last_ack; /* last time when the ACK spacing is close */
ae27e98a	103	u32 curr_rtt; /* the minimum rtt of current round */
df3271f3 SH	104	};
	105
	106	static inline void bictcp_reset(struct bictcp *ca)
	107	{
	108	ca->cnt = 0;
	109	ca->last_max_cwnd = 0;
df3271f3 SH	110	ca->last_cwnd = 0;
	111	ca->last_time = 0;
	112	ca->bic_origin_point = 0;
	113	ca->bic_K = 0;
	114	ca->delay_min = 0;
	115	ca->epoch_start = 0;
	116	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	117	ca->ack_cnt = 0;
	118	ca->tcp_cwnd = 0;
ae27e98a SH	119	ca->found = 0;
	120	}
	121
17a6e9f1	122	static inline u32 bictcp_clock(void)
	123	{
	124	#if HZ < 1000
	125	return ktime_to_ms(ktime_get_real());
	126	#else
	127	return jiffies_to_msecs(jiffies);
	128	#endif
	129	}
	130
ae27e98a SH	131	static inline void bictcp_hystart_reset(struct sock *sk)
	132	{
	133	struct tcp_sock *tp = tcp_sk(sk);
	134	struct bictcp *ca = inet_csk_ca(sk);
	135
17a6e9f1	136	ca->round_start = ca->last_ack = bictcp_clock();
ae27e98a SH	137	ca->end_seq = tp->snd_nxt;
	138	ca->curr_rtt = 0;
	139	ca->sample_cnt = 0;
df3271f3 SH	140	}
	141
	142	static void bictcp_init(struct sock *sk)
	143	{
5a45f008 NC	144	struct bictcp *ca = inet_csk_ca(sk);
	145
	146	bictcp_reset(ca);
	147	ca->loss_cwnd = 0;
ae27e98a SH	148
	149	if (hystart)
	150	bictcp_hystart_reset(sk);
	151
	152	if (!hystart && initial_ssthresh)
df3271f3 SH	153	tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
	154	}
	155
7e58886b SH	156	/* calculate the cubic root of x using a table lookup followed by one
	157	* Newton-Raphson iteration.
	158	* Avg err ~= 0.195%
df3271f3	159	*/
9eb2d627	160	static u32 cubic_root(u64 a)
df3271f3	161	{
7e58886b SH	162	u32 x, b, shift;
	163	/*
	164	* cbrt(x) MSB values for x MSB values in [0..63].
	165	* Precomputed then refined by hand - Willy Tarreau
	166	*
	167	* For x in [0..63],
	168	* v = cbrt(x << 18) - 1
	169	* cbrt(x) = (v[x] + 10) >> 6
9eb2d627	170	*/
7e58886b SH	171	static const u8 v[] = {
	172	/* 0x00 */ 0, 54, 54, 54, 118, 118, 118, 118,
	173	/* 0x08 */ 123, 129, 134, 138, 143, 147, 151, 156,
	174	/* 0x10 */ 157, 161, 164, 168, 170, 173, 176, 179,
	175	/* 0x18 */ 181, 185, 187, 190, 192, 194, 197, 199,
	176	/* 0x20 */ 200, 202, 204, 206, 209, 211, 213, 215,
	177	/* 0x28 */ 217, 219, 221, 222, 224, 225, 227, 229,
	178	/* 0x30 */ 231, 232, 234, 236, 237, 239, 240, 242,
	179	/* 0x38 */ 244, 245, 246, 248, 250, 251, 252, 254,
	180	};
	181
	182	b = fls64(a);
	183	if (b < 7) {
	184	/* a in [0..63] */
	185	return ((u32)v[(u32)a] + 35) >> 6;
	186	}
	187
	188	b = ((b * 84) >> 8) - 1;
	189	shift = (a >> (b * 3));
	190
	191	x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;
	192
	193	/*
	194	* Newton-Raphson iteration
	195	* 2
	196	* x = ( 2 * x + a / x ) / 3
	197	* k+1 k k
	198	*/
6f6d6a1a	199	x = (2 * x + (u32)div64_u64(a, (u64)x * (u64)(x - 1)));
7e58886b	200	x = ((x * 341) >> 10);
9eb2d627	201	return x;
df3271f3 SH	202	}
df3271f3 SH	203
df3271f3 SH	204	/*
	205	* Compute congestion window to use.
	206	*/
	207	static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
	208	{
16f03331 ED	209	u32 delta, bic_target, max_cnt;
16f03331 ED	210	u64 offs, t;
df3271f3 SH	211
	212	ca->ack_cnt++; /* count the number of ACKs */
	213
	214	if (ca->last_cwnd == cwnd &&
	215	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	216	return;
	217
	218	ca->last_cwnd = cwnd;
	219	ca->last_time = tcp_time_stamp;
	220
df3271f3 SH	221	if (ca->epoch_start == 0) {
	222	ca->epoch_start = tcp_time_stamp; /* record the beginning of an epoch */
	223	ca->ack_cnt = 1; /* start counting */
	224	ca->tcp_cwnd = cwnd; /* syn with cubic */
	225
	226	if (ca->last_max_cwnd <= cwnd) {
	227	ca->bic_K = 0;
	228	ca->bic_origin_point = cwnd;
	229	} else {
89b3d9aa SH	230	/* Compute new K based on
	231	* (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
	232	*/
	233	ca->bic_K = cubic_root(cube_factor
	234	* (ca->last_max_cwnd - cwnd));
df3271f3 SH	235	ca->bic_origin_point = ca->last_max_cwnd;
	236	}
	237	}
	238
e905a9ed YH	239	/* cubic function - calc*/
	240	/* calculate c * time^3 / rtt,
	241	* while considering overflow in calculation of time^3
89b3d9aa	242	* (so time^3 is done by using 64 bit)
df3271f3	243	* and without the support of division of 64bit numbers
89b3d9aa	244	* (so all divisions are done by using 32 bit)
e905a9ed YH	245	* also NOTE the unit of those veriables
	246	* time = (t - K) / 2^bictcp_HZ
	247	* c = bic_scale >> 10
df3271f3 SH	248	* rtt = (srtt >> 3) / HZ
	249	* !!! The following code does not have overflow problems,
	250	* if the cwnd < 1 million packets !!!
e905a9ed	251	*/
df3271f3	252
16f03331 ED	253	t = (s32)(tcp_time_stamp - ca->epoch_start);
16f03331 ED	254	t += msecs_to_jiffies(ca->delay_min >> 3);
df3271f3	255	/* change the unit from HZ to bictcp_HZ */
16f03331 ED	256	t <<= BICTCP_HZ;
16f03331 ED	257	do_div(t, HZ);
df3271f3	258
e905a9ed	259	if (t < ca->bic_K) /* t - K */
89b3d9aa	260	offs = ca->bic_K - t;
e905a9ed YH	261	else
e905a9ed YH	262	offs = t - ca->bic_K;
df3271f3	263
89b3d9aa SH	264	/* c/rtt * (t-K)^3 */
89b3d9aa SH	265	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
e905a9ed YH	266	if (t < ca->bic_K) /* below origin*/
	267	bic_target = ca->bic_origin_point - delta;
	268	else /* above origin*/
	269	bic_target = ca->bic_origin_point + delta;
df3271f3	270
e905a9ed YH	271	/* cubic function - calc bictcp_cnt*/
e905a9ed YH	272	if (bic_target > cwnd) {
df3271f3	273	ca->cnt = cwnd / (bic_target - cwnd);
e905a9ed YH	274	} else {
	275	ca->cnt = 100 * cwnd; /* very small increment*/
	276	}
df3271f3	277
b5ccd073 SH	278	/*
	279	* The initial growth of cubic function may be too conservative
	280	* when the available bandwidth is still unknown.
	281	*/
5a45f008	282	if (ca->last_max_cwnd == 0 && ca->cnt > 20)
b5ccd073 SH	283	ca->cnt = 20; /* increase cwnd 5% per RTT */
b5ccd073 SH	284
df3271f3 SH	285	/* TCP Friendly */
df3271f3 SH	286	if (tcp_friendliness) {
89b3d9aa SH	287	u32 scale = beta_scale;
89b3d9aa SH	288	delta = (cwnd * scale) >> 3;
e905a9ed YH	289	while (ca->ack_cnt > delta) { /* update tcp cwnd */
	290	ca->ack_cnt -= delta;
	291	ca->tcp_cwnd++;
df3271f3 SH	292	}
	293
	294	if (ca->tcp_cwnd > cwnd){ /* if bic is slower than tcp */
89b3d9aa SH	295	delta = ca->tcp_cwnd - cwnd;
89b3d9aa SH	296	max_cnt = cwnd / delta;
df3271f3 SH	297	if (ca->cnt > max_cnt)
	298	ca->cnt = max_cnt;
	299	}
e905a9ed	300	}
df3271f3 SH	301
	302	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	303	if (ca->cnt == 0) /* cannot be zero */
	304	ca->cnt = 1;
	305	}
	306
c3a05c60	307	static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
df3271f3 SH	308	{
	309	struct tcp_sock *tp = tcp_sk(sk);
	310	struct bictcp *ca = inet_csk_ca(sk);
	311
df3271f3 SH	312	if (!tcp_is_cwnd_limited(sk, in_flight))
	313	return;
	314
ae27e98a SH	315	if (tp->snd_cwnd <= tp->snd_ssthresh) {
	316	if (hystart && after(ack, ca->end_seq))
	317	bictcp_hystart_reset(sk);
df3271f3	318	tcp_slow_start(tp);
ae27e98a	319	} else {
df3271f3	320	bictcp_update(ca, tp->snd_cwnd);
758ce5c8	321	tcp_cong_avoid_ai(tp, ca->cnt);
df3271f3 SH	322	}
	323
	324	}
	325
	326	static u32 bictcp_recalc_ssthresh(struct sock *sk)
	327	{
	328	const struct tcp_sock *tp = tcp_sk(sk);
	329	struct bictcp *ca = inet_csk_ca(sk);
	330
	331	ca->epoch_start = 0; /* end of epoch */
	332
	333	/* Wmax and fast convergence */
	334	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
	335	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
	336	/ (2 * BICTCP_BETA_SCALE);
	337	else
	338	ca->last_max_cwnd = tp->snd_cwnd;
	339
	340	ca->loss_cwnd = tp->snd_cwnd;
	341
	342	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
	343	}
	344
	345	static u32 bictcp_undo_cwnd(struct sock *sk)
	346	{
	347	struct bictcp *ca = inet_csk_ca(sk);
	348
5a45f008	349	return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
df3271f3 SH	350	}
df3271f3 SH	351
df3271f3 SH	352	static void bictcp_state(struct sock *sk, u8 new_state)
df3271f3 SH	353	{
ae27e98a	354	if (new_state == TCP_CA_Loss) {
df3271f3	355	bictcp_reset(inet_csk_ca(sk));
ae27e98a SH	356	bictcp_hystart_reset(sk);
	357	}
	358	}
	359
	360	static void hystart_update(struct sock *sk, u32 delay)
	361	{
	362	struct tcp_sock *tp = tcp_sk(sk);
	363	struct bictcp *ca = inet_csk_ca(sk);
	364
	365	if (!(ca->found & hystart_detect)) {
17a6e9f1	366	u32 now = bictcp_clock();
ae27e98a SH	367
ae27e98a SH	368	/* first detection parameter - ack-train detection */
17a6e9f1	369	if ((s32)(now - ca->last_ack) <= hystart_ack_delta) {
	370	ca->last_ack = now;
	371	if ((s32)(now - ca->round_start) > ca->delay_min >> 4)
ae27e98a SH	372	ca->found \|= HYSTART_ACK_TRAIN;
	373	}
	374
	375	/* obtain the minimum delay of more than sampling packets */
	376	if (ca->sample_cnt < HYSTART_MIN_SAMPLES) {
	377	if (ca->curr_rtt == 0 \|\| ca->curr_rtt > delay)
	378	ca->curr_rtt = delay;
	379
	380	ca->sample_cnt++;
	381	} else {
	382	if (ca->curr_rtt > ca->delay_min +
	383	HYSTART_DELAY_THRESH(ca->delay_min>>4))
	384	ca->found \|= HYSTART_DELAY;
	385	}
	386	/*
	387	* Either one of two conditions are met,
	388	* we exit from slow start immediately.
	389	*/
	390	if (ca->found & hystart_detect)
	391	tp->snd_ssthresh = tp->snd_cwnd;
	392	}
df3271f3 SH	393	}
	394
	395	/* Track delayed acknowledgment ratio using sliding window
	396	* ratio = (15*ratio + sample) / 16
	397	*/
30cfd0ba	398	static void bictcp_acked(struct sock *sk, u32 cnt, s32 rtt_us)
df3271f3 SH	399	{
df3271f3 SH	400	const struct inet_connection_sock *icsk = inet_csk(sk);
ae27e98a	401	const struct tcp_sock *tp = tcp_sk(sk);
e7d0c885 SH	402	struct bictcp *ca = inet_csk_ca(sk);
e7d0c885 SH	403	u32 delay;
df3271f3	404
35e86941	405	if (icsk->icsk_ca_state == TCP_CA_Open) {
b9f47a3a	406	u32 ratio = ca->delayed_ack;
	407
	408	ratio -= ca->delayed_ack >> ACK_RATIO_SHIFT;
	409	ratio += cnt;
	410
28e00c4c	411	ca->delayed_ack = clamp(ratio, 1U, ACK_RATIO_LIMIT);
df3271f3	412	}
df3271f3	413
e7d0c885 SH	414	/* Some calls are for duplicates without timetamps */
	415	if (rtt_us < 0)
	416	return;
	417
	418	/* Discard delay samples right after fast recovery */
ca02d414	419	if (ca->epoch_start && (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
e7d0c885 SH	420	return;
e7d0c885 SH	421
17a6e9f1	422	delay = (rtt_us << 3) / USEC_PER_MSEC;
e7d0c885 SH	423	if (delay == 0)
	424	delay = 1;
	425
	426	/* first time call or link delay decreases */
	427	if (ca->delay_min == 0 \|\| ca->delay_min > delay)
	428	ca->delay_min = delay;
ae27e98a SH	429
	430	/* hystart triggers when cwnd is larger than some threshold */
	431	if (hystart && tp->snd_cwnd <= tp->snd_ssthresh &&
	432	tp->snd_cwnd >= hystart_low_window)
	433	hystart_update(sk, delay);
e7d0c885	434	}
df3271f3	435
a252bebe	436	static struct tcp_congestion_ops cubictcp __read_mostly = {
df3271f3 SH	437	.init = bictcp_init,
	438	.ssthresh = bictcp_recalc_ssthresh,
	439	.cong_avoid = bictcp_cong_avoid,
	440	.set_state = bictcp_state,
	441	.undo_cwnd = bictcp_undo_cwnd,
df3271f3 SH	442	.pkts_acked = bictcp_acked,
	443	.owner = THIS_MODULE,
	444	.name = "cubic",
	445	};
	446
	447	static int __init cubictcp_register(void)
	448	{
74975d40	449	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
89b3d9aa SH	450
	451	/* Precompute a bunch of the scaling factors that are used per-packet
	452	* based on SRTT of 100ms
	453	*/
	454
	455	beta_scale = 8*(BICTCP_BETA_SCALE+beta)/ 3 / (BICTCP_BETA_SCALE - beta);
	456
22119240	457	cube_rtt_scale = (bic_scale * 10); /* 1024c/rtt /
89b3d9aa SH	458
	459	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	460	* so K = cubic_root( (wmax-cwnd)*rtt/c )
	461	* the unit of K is bictcp_HZ=2^10, not HZ
	462	*
	463	* c = bic_scale >> 10
	464	* rtt = 100ms
	465	*
	466	* the following code has been designed and tested for
	467	* cwnd < 1 million packets
	468	* RTT < 100 seconds
	469	* HZ < 1,000,00 (corresponding to 10 nano-second)
	470	*/
	471
	472	/* 1/c * 2^2bictcp_HZ srtt */
	473	cube_factor = 1ull << (10+3BICTCP_HZ); / 2^40 */
	474
	475	/* divide by bic_scale and by constant Srtt (100ms) */
	476	do_div(cube_factor, bic_scale * 10);
	477
3b585b34	478	/* hystart needs ms clock resolution */
	479	if (hystart && HZ < 1000)
	480	cubictcp.flags \|= TCP_CONG_RTT_STAMP;
	481
df3271f3 SH	482	return tcp_register_congestion_control(&cubictcp);
	483	}
	484
	485	static void __exit cubictcp_unregister(void)
	486	{
	487	tcp_unregister_congestion_control(&cubictcp);
	488	}
	489
	490	module_init(cubictcp_register);
	491	module_exit(cubictcp_unregister);
	492
	493	MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
	494	MODULE_LICENSE("GPL");
	495	MODULE_DESCRIPTION("CUBIC TCP");
ae27e98a	496	MODULE_VERSION("2.3");