[RFC v3] minstrel_ht: new rate control module for 802.11n

[RFC v3] minstrel_ht: new rate control module for 802.11n

[Felix Fietkau]

Here's v3 of minstrel_ht.

Changelog:

v3:
- Add two fixes by Christian Lamparter (division by zero, debugfs mcs 
  rate display)
- Improve throughput a bit more by spreading out slow rate sampling 
  attempts over multiple sampling intervals

v2:
- Use accurate A-MPDU statistics from tx feedback
- Rewrite the sampling algorithm to optimize for good A-MPDU lengths
  and faster throughput recovery
- Implement .rate_update to handle changes in HT capabilities or 
  channel bandwidth
- Remove the private tx flags hack

---
--- a/net/mac80211/Makefile
+++ b/net/mac80211/Makefile
@@ -47,8 +47,8 @@ CFLAGS_driver-trace.o := -I$(src)
 rc80211_pid-y := rc80211_pid_algo.o
 rc80211_pid-$(CONFIG_MAC80211_DEBUGFS) += rc80211_pid_debugfs.o
 
-rc80211_minstrel-y := rc80211_minstrel.o
-rc80211_minstrel-$(CONFIG_MAC80211_DEBUGFS) += rc80211_minstrel_debugfs.o
+rc80211_minstrel-y := rc80211_minstrel.o rc80211_minstrel_ht.o
+rc80211_minstrel-$(CONFIG_MAC80211_DEBUGFS) += rc80211_minstrel_debugfs.o rc80211_minstrel_ht_debugfs.o
 
 mac80211-$(CONFIG_MAC80211_RC_PID) += $(rc80211_pid-y)
 mac80211-$(CONFIG_MAC80211_RC_MINSTREL) += $(rc80211_minstrel-y)
--- a/net/mac80211/main.c
+++ b/net/mac80211/main.c
@@ -714,6 +714,10 @@ static int __init ieee80211_init(void)
 	if (ret)
 		return ret;
 
+	ret = rc80211_minstrel_ht_init();
+	if (ret)
+		goto err_minstrel;
+
 	ret = rc80211_pid_init();
 	if (ret)
 		goto err_pid;
@@ -726,6 +730,8 @@ static int __init ieee80211_init(void)
  err_netdev:
 	rc80211_pid_exit();
  err_pid:
+	rc80211_minstrel_ht_exit();
+ err_minstrel:
 	rc80211_minstrel_exit();
 
 	return ret;
@@ -734,6 +740,7 @@ static int __init ieee80211_init(void)
 static void __exit ieee80211_exit(void)
 {
 	rc80211_pid_exit();
+	rc80211_minstrel_ht_exit();
 	rc80211_minstrel_exit();
 
 	/*
--- a/net/mac80211/rate.h
+++ b/net/mac80211/rate.h
@@ -137,6 +137,8 @@ static inline void rc80211_pid_exit(void
 #ifdef CONFIG_MAC80211_RC_MINSTREL
 extern int rc80211_minstrel_init(void);
 extern void rc80211_minstrel_exit(void);
+extern int rc80211_minstrel_ht_init(void);
+extern void rc80211_minstrel_ht_exit(void);
 #else
 static inline int rc80211_minstrel_init(void)
 {
@@ -145,6 +147,13 @@ static inline int rc80211_minstrel_init(
 static inline void rc80211_minstrel_exit(void)
 {
 }
+static inline int rc80211_minstrel_ht_init(void)
+{
+	return 0;
+}
+static inline void rc80211_minstrel_ht_exit(void)
+{
+}
 #endif
 
 
--- /dev/null
+++ b/net/mac80211/rc80211_minstrel_ht.c
@@ -0,0 +1,809 @@
+/*
+ * Copyright (C) 2010 Felix Fietkau <nbd@openwrt.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/netdevice.h>
+#include <linux/types.h>
+#include <linux/skbuff.h>
+#include <linux/debugfs.h>
+#include <linux/random.h>
+#include <linux/ieee80211.h>
+#include <net/mac80211.h>
+#include "rate.h"
+#include "rc80211_minstrel.h"
+#include "rc80211_minstrel_ht.h"
+
+#define AVG_PKT_SIZE	1200
+#define SAMPLE_COLUMNS	10
+#define EWMA_LEVEL		75
+
+/* Number of bits for an average sized packet */
+#define MCS_NBITS (AVG_PKT_SIZE << 3)
+
+/* Number of symbols for a packet with (bps) bits per symbol */
+#define MCS_NSYMS(bps) ((MCS_NBITS + (bps) - 1) / (bps))
+
+/* Transmission time for a packet containing (syms) symbols */
+#define MCS_SYMBOL_TIME(sgi, syms)					\
+	(sgi ?								\
+	  ((syms) * 18 + 4) / 5 :	/* syms * 3.6 us */		\
+	  (syms) << 2			/* syms * 4 us */		\
+	)
+
+/* Transmit duration for the raw data part of an average sized packet */
+#define MCS_DURATION(streams, sgi, bps) MCS_SYMBOL_TIME(sgi, MCS_NSYMS((streams) * (bps)))
+
+/* MCS rate information for an MCS group */
+#define MCS_GROUP(_streams, _sgi, _ht40) {				\
+	.streams = _streams,						\
+	.flags =							\
+		(_sgi ? IEEE80211_TX_RC_SHORT_GI : 0) |			\
+		(_ht40 ? IEEE80211_TX_RC_40_MHZ_WIDTH : 0),		\
+	.duration = {							\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 54 : 26),		\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 108 : 52),		\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 162 : 78),		\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 216 : 104),	\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 324 : 156),	\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 432 : 208),	\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 486 : 234),	\
+		MCS_DURATION(_streams, _sgi, _ht40 ? 540 : 260)		\
+	}								\
+}
+
+/*
+ * To enable sufficiently targeted rate sampling, MCS rates are divided into
+ * groups, based on the number of streams and flags (HT40, SGI) that they
+ * use.
+ */
+const struct mcs_group minstrel_mcs_groups[] = {
+	MCS_GROUP(1, 0, 0),
+	MCS_GROUP(2, 0, 0),
+#if MINSTREL_MAX_STREAMS >= 3
+	MCS_GROUP(3, 0, 0),
+#endif
+
+	MCS_GROUP(1, 1, 0),
+	MCS_GROUP(2, 1, 0),
+#if MINSTREL_MAX_STREAMS >= 3
+	MCS_GROUP(3, 1, 0),
+#endif
+
+	MCS_GROUP(1, 0, 1),
+	MCS_GROUP(2, 0, 1),
+#if MINSTREL_MAX_STREAMS >= 3
+	MCS_GROUP(3, 0, 1),
+#endif
+
+	MCS_GROUP(1, 1, 1),
+	MCS_GROUP(2, 1, 1),
+#if MINSTREL_MAX_STREAMS >= 3
+	MCS_GROUP(3, 1, 1),
+#endif
+};
+
+static u8 sample_table[SAMPLE_COLUMNS][MCS_GROUP_RATES];
+
+/*
+ * Perform EWMA (Exponentially Weighted Moving Average) calculation
+ */
+static int
+minstrel_ewma(int old, int new, int weight)
+{
+	return (new * (100 - weight) + old * weight) / 100;
+}
+
+/*
+ * Look up an MCS group index based on mac80211 rate information
+ */
+static int
+minstrel_ht_get_group_idx(struct ieee80211_tx_rate *rate)
+{
+	int streams = (rate->idx / MCS_GROUP_RATES) + 1;
+	u32 flags = IEEE80211_TX_RC_SHORT_GI | IEEE80211_TX_RC_40_MHZ_WIDTH;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(minstrel_mcs_groups); i++) {
+		if (minstrel_mcs_groups[i].streams != streams)
+			continue;
+		if (minstrel_mcs_groups[i].flags != (rate->flags & flags))
+			continue;
+
+		return i;
+	}
+
+	WARN_ON(1);
+	return 0;
+}
+
+static inline struct minstrel_rate_stats *
+minstrel_get_ratestats(struct minstrel_ht_sta *mi, int index)
+{
+	return &mi->groups[index / MCS_GROUP_RATES].rates[index % MCS_GROUP_RATES];
+}
+
+
+/*
+ * Recalculate success probabilities and counters for a rate using EWMA
+ */
+static void
+minstrel_calc_rate_ewma(struct minstrel_priv *mp, struct minstrel_rate_stats *mr)
+{
+	if (unlikely(mr->attempts > 0)) {
+		mr->sample_skipped = 0;
+		mr->cur_prob = MINSTREL_FRAC(mr->success, mr->attempts);
+		if (!mr->att_hist)
+			mr->probability = mr->cur_prob;
+		else
+			mr->probability = minstrel_ewma(mr->probability,
+				mr->cur_prob, EWMA_LEVEL);
+		mr->att_hist += mr->attempts;
+		mr->succ_hist += mr->success;
+	} else {
+		mr->sample_skipped++;
+	}
+	mr->last_success = mr->success;
+	mr->last_attempts = mr->attempts;
+	mr->success = 0;
+	mr->attempts = 0;
+}
+
+/*
+ * Calculate throughput based on the average A-MPDU length, taking into account
+ * the expected number of retransmissions and their expected length
+ */
+static void
+minstrel_ht_calc_tp(struct minstrel_priv *mp, struct minstrel_ht_sta *mi,
+                    int group, int rate)
+{
+	struct minstrel_rate_stats *mr;
+	unsigned int usecs;
+
+	mr = &mi->groups[group].rates[rate];
+
+	if (mr->probability < MINSTREL_FRAC(1, 10)) {
+		mr->cur_tp = 0;
+		return;
+	}
+
+	usecs = mi->overhead / MINSTREL_TRUNC(mi->avg_ampdu_len);
+	usecs += minstrel_mcs_groups[group].duration[rate];
+	mr->cur_tp = MINSTREL_TRUNC((1000000 / usecs) * mr->probability);
+}
+
+/*
+ * Update rate statistics and select new primary rates
+ *
+ * Rules for rate selection:
+ *  - max_prob_rate must use only one stream, as a tradeoff between delivery
+ *    probability and throughput during strong fluctuations
+ *  - as long as the max prob rate has a probability of more than 3/4, pick
+ *    higher throughput rates, even if the probablity is a bit lower
+ */
+static void
+minstrel_ht_update_stats(struct minstrel_priv *mp, struct minstrel_ht_sta *mi)
+{
+	struct minstrel_mcs_group_data *mg;
+	struct minstrel_rate_stats *mr;
+	int cur_prob, cur_prob_tp, cur_tp, cur_tp2;
+	int group, i, index;
+
+	mi->sample_slow = 0;
+	mi->sample_count = 0;
+	mi->max_tp_rate = 0;
+	mi->max_tp_rate2 = 0;
+	mi->max_prob_rate = 0;
+
+	for (group = 0; group < ARRAY_SIZE(minstrel_mcs_groups); group++) {
+		cur_prob = 0;
+		cur_prob_tp = 0;
+		cur_tp = 0;
+		cur_tp2 = 0;
+
+		mg = &mi->groups[group];
+		if (!mg->supported)
+			continue;
+
+		mg->max_tp_rate = 0;
+		mg->max_tp_rate2 = 0;
+		mg->max_prob_rate = 0;
+		mi->sample_count++;
+
+		for (i = 0; i < MCS_GROUP_RATES; i++) {
+			if (!(mg->supported & BIT(i)))
+				continue;
+
+			mr = &mg->rates[i];
+			mr->retry_updated = false;
+			index = MCS_GROUP_RATES * group + i;
+			minstrel_calc_rate_ewma(mp, mr);
+			minstrel_ht_calc_tp(mp, mi, group, i);
+
+			if (!mr->cur_tp)
+				continue;
+
+			/* ignore the lowest rate of each single-stream group */
+			if (!i && minstrel_mcs_groups[group].streams == 1)
+				continue;
+
+			if ((mr->cur_tp > cur_prob_tp && mr->probability >
+			     MINSTREL_FRAC(3, 4)) || mr->probability > cur_prob) {
+				mg->max_prob_rate = index;
+				cur_prob = mr->probability;
+			}
+
+			if (mr->cur_tp > cur_tp) {
+				swap(index, mg->max_tp_rate);
+				cur_tp = mr->cur_tp;
+				mr = minstrel_get_ratestats(mi, index);
+			}
+
+			if (index >= mg->max_tp_rate)
+				continue;
+
+			if (mr->cur_tp > cur_tp2) {
+				mg->max_tp_rate2 = index;
+				cur_tp2 = mr->cur_tp;
+			}
+		}
+	}
+
+	/* try to sample up to half of the availble rates during each interval */
+	mi->sample_count *= 4;
+
+	cur_prob = 0;
+	cur_prob_tp = 0;
+	cur_tp = 0;
+	cur_tp2 = 0;
+	for (group = 0; group < ARRAY_SIZE(minstrel_mcs_groups); group++) {
+		mg = &mi->groups[group];
+		if (!mg->supported)
+			continue;
+
+		mr = minstrel_get_ratestats(mi, mg->max_prob_rate);
+		if (cur_prob_tp < mr->cur_tp &&
+		    minstrel_mcs_groups[group].streams == 1) {
+			mi->max_prob_rate = mg->max_prob_rate;
+			cur_prob = mr->cur_prob;
+		}
+
+		mr = minstrel_get_ratestats(mi, mg->max_tp_rate);
+		if (cur_tp < mr->cur_tp) {
+			mi->max_tp_rate = mg->max_tp_rate;
+			cur_tp = mr->cur_tp;
+		}
+
+		mr = minstrel_get_ratestats(mi, mg->max_tp_rate2);
+		if (cur_tp2 < mr->cur_tp) {
+			mi->max_tp_rate2 = mg->max_tp_rate2;
+			cur_tp2 = mr->cur_tp;
+		}
+	}
+
+	mi->stats_update = jiffies;
+}
+
+static bool
+minstrel_ht_txstat_valid(struct ieee80211_tx_rate *rate)
+{
+	if (!rate->count)
+		return false;
+
+	if (rate->idx < 0)
+		return false;
+
+	return !!(rate->flags & IEEE80211_TX_RC_MCS);
+}
+
+static void
+minstrel_next_sample_idx(struct minstrel_ht_sta *mi)
+{
+	struct minstrel_mcs_group_data *mg;
+
+	for (;;) {
+		mi->sample_group++;
+		mi->sample_group %= ARRAY_SIZE(minstrel_mcs_groups);
+		mg = &mi->groups[mi->sample_group];
+
+		if (!mg->supported)
+			continue;
+
+		if (++mg->index > MCS_GROUP_RATES) {
+			mg->index = 0;
+			if (++mg->column > ARRAY_SIZE(sample_table))
+				mg->column = 0;
+		}
+		break;
+	}
+}
+
+static void
+minstrel_downgrade_rate(struct minstrel_ht_sta *mi, int *idx, bool primary)
+{
+	int group, orig_group;
+
+	orig_group = group = *idx / MCS_GROUP_RATES;
+	while (group > 0) {
+		group--;
+
+		if (!mi->groups[group].supported)
+			continue;
+
+		if (minstrel_mcs_groups[group].streams >
+		    minstrel_mcs_groups[orig_group].streams)
+			continue;
+
+		if (primary)
+			*idx = mi->groups[group].max_tp_rate;
+		else
+			*idx = mi->groups[group].max_tp_rate2;
+		break;
+	}
+}
+
+
+static void
+minstrel_ht_tx_status(void *priv, struct ieee80211_supported_band *sband,
+                      struct ieee80211_sta *sta, void *priv_sta,
+                      struct sk_buff *skb)
+{
+	struct minstrel_ht_sta_priv *msp = priv_sta;
+	struct minstrel_ht_sta *mi = &msp->ht;
+	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+	struct ieee80211_tx_rate *ar = info->status.rates;
+	struct minstrel_rate_stats *rate, *rate2;
+	struct minstrel_priv *mp = priv;
+	bool last = false;
+	int group;
+	int i = 0;
+
+	if (!msp->is_ht)
+		return mac80211_minstrel.tx_status(priv, sband, sta, &msp->legacy, skb);
+
+	/* This packet was aggregated but doesn't carry status info */
+	if ((info->flags & IEEE80211_TX_CTL_AMPDU) &&
+	    !(info->flags & IEEE80211_TX_STAT_AMPDU))
+		return;
+
+	if (!info->status.ampdu_len) {
+		info->status.ampdu_ack_len = 1;
+		info->status.ampdu_len = 1;
+	}
+
+	mi->avg_ampdu_len = minstrel_ewma(mi->avg_ampdu_len,
+		MINSTREL_FRAC(info->status.ampdu_len, 1), 90);
+
+	if (!mi->sample_wait && !mi->sample_tries && mi->sample_count > 0) {
+		mi->sample_wait = 4 + MINSTREL_TRUNC(mi->avg_ampdu_len);
+		mi->sample_tries = 3;
+		mi->sample_count--;
+	}
+
+	if (info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE) {
+		mi->sample_packets += info->status.ampdu_len;
+		minstrel_next_sample_idx(mi);
+	}
+
+	for (i = 0; !last; i++) {
+		last = (i == IEEE80211_TX_MAX_RATES - 1) ||
+		       !minstrel_ht_txstat_valid(&ar[i + 1]);
+
+		if (!minstrel_ht_txstat_valid(&ar[i]))
+			break;
+
+		group = minstrel_ht_get_group_idx(&ar[i]);
+		rate = &mi->groups[group].rates[ar[i].idx % 8];
+
+		if (last && (info->flags & IEEE80211_TX_STAT_ACK))
+			rate->success += info->status.ampdu_ack_len;
+
+		rate->attempts += ar[i].count * info->status.ampdu_len;
+	}
+
+	/*
+	 * check for sudden death of spatial multiplexing,
+	 * downgrade to a lower number of streams if necessary.
+	 */
+	rate = minstrel_get_ratestats(mi, mi->max_tp_rate);
+	if (rate->attempts > 30 &&
+	    MINSTREL_FRAC(rate->success, rate->attempts) <
+	    MINSTREL_FRAC(20, 100))
+		minstrel_downgrade_rate(mi, &mi->max_tp_rate, true);
+
+	rate2 = minstrel_get_ratestats(mi, mi->max_tp_rate2);
+	if (rate->attempts > 30 &&
+	    MINSTREL_FRAC(rate->success, rate->attempts) <
+	    MINSTREL_FRAC(20, 100))
+		minstrel_downgrade_rate(mi, &mi->max_tp_rate2, false);
+
+	if (time_after(jiffies, mi->stats_update + (mp->update_interval / 2 * HZ) / 1000))
+		minstrel_ht_update_stats(mp, mi);
+}
+
+static void
+minstrel_calc_retransmit(struct minstrel_priv *mp, struct minstrel_ht_sta *mi,
+                         int index)
+{
+	struct minstrel_rate_stats *mr;
+	const struct mcs_group *group;
+	unsigned int tx_time, tx_time_rtscts, tx_time_data;
+	unsigned int cw = mp->cw_min;
+	unsigned int t_slot = 9; /* FIXME */
+	unsigned int ampdu_len = MINSTREL_TRUNC(mi->avg_ampdu_len);
+
+	mr = minstrel_get_ratestats(mi, index);
+	if (mr->probability < MINSTREL_FRAC(1, 10)) {
+		mr->retry_count = 1;
+		mr->retry_count_rtscts = 1;
+		return;
+	}
+
+	mr->retry_count = 2;
+	mr->retry_count_rtscts = 2;
+	mr->retry_updated = true;
+
+	group = &minstrel_mcs_groups[index / MCS_GROUP_RATES];
+	tx_time_data = group->duration[index % MCS_GROUP_RATES] * ampdu_len;
+	tx_time = 2 * (t_slot + mi->overhead + tx_time_data);
+	tx_time_rtscts = 2 * (t_slot + mi->overhead_rtscts + tx_time_data);
+	do {
+		cw = (cw << 1) | 1;
+		cw = min(cw, mp->cw_max);
+		tx_time += cw + t_slot + mi->overhead;
+		tx_time_rtscts += cw + t_slot + mi->overhead_rtscts;
+		if (tx_time_rtscts < mp->segment_size)
+			mr->retry_count_rtscts++;
+	} while ((tx_time < mp->segment_size) &&
+	         (++mr->retry_count < mp->max_retry));
+}
+
+
+static void
+minstrel_ht_set_rate(struct minstrel_priv *mp, struct minstrel_ht_sta *mi,
+                     struct ieee80211_tx_rate *rate, int index,
+                     struct ieee80211_tx_rate_control *txrc,
+                     bool sample, bool rtscts)
+{
+	const struct mcs_group *group = &minstrel_mcs_groups[index / MCS_GROUP_RATES];
+	struct minstrel_rate_stats *mr;
+
+	mr = minstrel_get_ratestats(mi, index);
+	if (!mr->retry_updated)
+		minstrel_calc_retransmit(mp, mi, index);
+
+	if (mr->probability < MINSTREL_FRAC(20, 100))
+		rate->count = 2;
+	else if (rtscts)
+		rate->count = mr->retry_count_rtscts;
+	else
+		rate->count = mr->retry_count;
+
+	rate->flags = IEEE80211_TX_RC_MCS | group->flags;
+	if (txrc->short_preamble)
+		rate->flags |= IEEE80211_TX_RC_USE_SHORT_PREAMBLE;
+	if (txrc->rts || rtscts)
+		rate->flags |= IEEE80211_TX_RC_USE_RTS_CTS;
+	rate->idx = index % MCS_GROUP_RATES + (group->streams - 1) * MCS_GROUP_RATES;
+}
+
+static inline int
+minstrel_get_duration(int index)
+{
+	const struct mcs_group *group = &minstrel_mcs_groups[index / MCS_GROUP_RATES];
+	return group->duration[index % MCS_GROUP_RATES];
+}
+
+static int
+minstrel_get_sample_rate(struct minstrel_priv *mp, struct minstrel_ht_sta *mi)
+{
+	struct minstrel_rate_stats *mr;
+	struct minstrel_mcs_group_data *mg;
+	int sample_idx = 0;
+
+	if (mi->sample_wait > 0) {
+		mi->sample_wait--;
+		return -1;
+	}
+
+	if (!mi->sample_tries)
+		return -1;
+
+	mi->sample_tries--;
+	mg = &mi->groups[mi->sample_group];
+	sample_idx = sample_table[mg->column][mg->index];
+	mr = &mg->rates[sample_idx];
+	sample_idx += mi->sample_group * MCS_GROUP_RATES;
+
+	/*
+	 * When not using MRR, do not sample if the probability is already
+	 * higher than 95% to avoid wasting airtime
+	 */
+	if (!mp->has_mrr && (mr->probability > MINSTREL_FRAC(95, 100)))
+		goto next;
+
+	/*
+	 * Make sure that lower rates get sampled only occasionally,
+	 * if the link is working perfectly.
+	 */
+	if (minstrel_get_duration(sample_idx) >
+	    minstrel_get_duration(mi->max_tp_rate)) {
+		if (mr->sample_skipped < 10)
+			goto next;
+
+		if (mi->sample_slow++ > 2)
+			goto next;
+	}
+
+	return sample_idx;
+
+next:
+	minstrel_next_sample_idx(mi);
+	return -1;
+}
+
+static void
+minstrel_aggr_check(struct minstrel_priv *mp, struct ieee80211_sta *pubsta, struct sk_buff *skb)
+{
+	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
+	struct sta_info *sta = container_of(pubsta, struct sta_info, sta);
+	u16 tid;
+
+	if (unlikely(!ieee80211_is_data_qos(hdr->frame_control)))
+		return;
+
+	if (unlikely(skb->protocol == cpu_to_be16(ETH_P_PAE)))
+		return;
+
+	tid = *ieee80211_get_qos_ctl(hdr) & IEEE80211_QOS_CTL_TID_MASK;
+	if (likely(sta->ampdu_mlme.tid_state_tx[tid] != HT_AGG_STATE_IDLE))
+		return;
+
+	ieee80211_start_tx_ba_session(pubsta, tid);
+}
+
+static void
+minstrel_ht_get_rate(void *priv, struct ieee80211_sta *sta, void *priv_sta,
+                     struct ieee80211_tx_rate_control *txrc)
+{
+	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(txrc->skb);
+	struct ieee80211_tx_rate *ar = info->status.rates;
+	struct minstrel_ht_sta_priv *msp = priv_sta;
+	struct minstrel_ht_sta *mi = &msp->ht;
+	struct minstrel_priv *mp = priv;
+	int sample_idx;
+
+	if (rate_control_send_low(sta, priv_sta, txrc))
+		return;
+
+	if (!msp->is_ht)
+		return mac80211_minstrel.get_rate(priv, sta, &msp->legacy, txrc);
+
+	minstrel_aggr_check(mp, sta, txrc->skb);
+
+	sample_idx = minstrel_get_sample_rate(mp, mi);
+	if (sample_idx >= 0) {
+		minstrel_ht_set_rate(mp, mi, &ar[0], sample_idx,
+			txrc, true, false);
+		minstrel_ht_set_rate(mp, mi, &ar[1], mi->max_tp_rate,
+			txrc, false, true);
+		info->flags |= IEEE80211_TX_CTL_RATE_CTRL_PROBE;
+	} else {
+		minstrel_ht_set_rate(mp, mi, &ar[0], mi->max_tp_rate,
+			txrc, false, false);
+		minstrel_ht_set_rate(mp, mi, &ar[1], mi->max_tp_rate2,
+			txrc, false, true);
+	}
+	minstrel_ht_set_rate(mp, mi, &ar[2], mi->max_prob_rate, txrc, false, true);
+
+	ar[3].count = 0;
+	ar[3].idx = -1;
+
+	mi->total_packets++;
+
+	/* wraparound */
+	if (mi->total_packets == ~0) {
+		mi->total_packets = 0;
+		mi->sample_packets = 0;
+	}
+}
+
+static void
+minstrel_ht_update_caps(void *priv, struct ieee80211_supported_band *sband,
+                        struct ieee80211_sta *sta, void *priv_sta,
+			enum nl80211_channel_type oper_chan_type)
+{
+	struct minstrel_priv *mp = priv;
+	struct minstrel_ht_sta_priv *msp = priv_sta;
+	struct minstrel_ht_sta *mi = &msp->ht;
+	struct ieee80211_mcs_info *mcs = &sta->ht_cap.mcs;
+	struct ieee80211_local *local = hw_to_local(mp->hw);
+	u16 sta_cap = sta->ht_cap.cap;
+	int ack_dur;
+	int i;
+
+	/* fall back to the old minstrel for legacy stations */
+	if (sta && !sta->ht_cap.ht_supported) {
+		msp->is_ht = false;
+		memset(&msp->legacy, 0, sizeof(msp->legacy));
+		msp->legacy.r = msp->ratelist;
+		msp->legacy.sample_table = msp->sample_table;
+		return mac80211_minstrel.rate_init(priv, sband, sta, &msp->legacy);
+	}
+
+	BUILD_BUG_ON(ARRAY_SIZE(minstrel_mcs_groups) !=
+		MINSTREL_MAX_STREAMS * MINSTREL_STREAM_GROUPS);
+
+	msp->is_ht = true;
+	memset(mi, 0, sizeof(*mi));
+	mi->stats_update = jiffies;
+
+	ack_dur = ieee80211_frame_duration(local, 10, 60, 1, 1);
+	mi->overhead = ieee80211_frame_duration(local, 0, 60, 1, 1) + ack_dur;
+	mi->overhead_rtscts = mi->overhead + 2 * ack_dur;
+
+	mi->avg_ampdu_len = MINSTREL_FRAC(1, 1);
+
+	/* When using MRR, sample more on the first attempt, without delay */
+	if (mp->has_mrr) {
+		mi->sample_count = 16;
+		mi->sample_wait = 0;
+	} else {
+		mi->sample_count = 8;
+		mi->sample_wait = 8;
+	}
+	mi->sample_tries = 4;
+
+	if (oper_chan_type != NL80211_CHAN_HT40MINUS &&
+	    oper_chan_type != NL80211_CHAN_HT40PLUS)
+		sta_cap &= ~IEEE80211_HT_CAP_SUP_WIDTH_20_40;
+
+	for (i = 0; i < ARRAY_SIZE(mi->groups); i++) {
+		u16 req = 0;
+
+		mi->groups[i].supported = 0;
+		if (minstrel_mcs_groups[i].flags & IEEE80211_TX_RC_SHORT_GI) {
+			if (minstrel_mcs_groups[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
+				req |= IEEE80211_HT_CAP_SGI_40;
+			else
+				req |= IEEE80211_HT_CAP_SGI_20;
+		}
+
+		if (minstrel_mcs_groups[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
+			req |= IEEE80211_HT_CAP_SUP_WIDTH_20_40;
+
+		if ((sta_cap & req) != req)
+			continue;
+
+		mi->groups[i].supported =
+			mcs->rx_mask[minstrel_mcs_groups[i].streams - 1];
+	}
+}
+
+static void
+minstrel_ht_rate_init(void *priv, struct ieee80211_supported_band *sband,
+                      struct ieee80211_sta *sta, void *priv_sta)
+{
+	struct minstrel_priv *mp = priv;
+
+	minstrel_ht_update_caps(priv, sband, sta, priv_sta, mp->hw->conf.channel_type);
+}
+
+static void
+minstrel_ht_rate_update(void *priv, struct ieee80211_supported_band *sband,
+                        struct ieee80211_sta *sta, void *priv_sta,
+                        u32 changed, enum nl80211_channel_type oper_chan_type)
+{
+	minstrel_ht_update_caps(priv, sband, sta, priv_sta, oper_chan_type);
+}
+
+static void *
+minstrel_ht_alloc_sta(void *priv, struct ieee80211_sta *sta, gfp_t gfp)
+{
+	struct ieee80211_supported_band *sband;
+	struct minstrel_ht_sta_priv *msp;
+	struct minstrel_priv *mp = priv;
+	struct ieee80211_hw *hw = mp->hw;
+	int max_rates = 0;
+	int i;
+
+	for (i = 0; i < IEEE80211_NUM_BANDS; i++) {
+		sband = hw->wiphy->bands[i];
+		if (sband && sband->n_bitrates > max_rates)
+			max_rates = sband->n_bitrates;
+	}
+
+	msp = kzalloc(sizeof(struct minstrel_ht_sta), gfp);
+	if (!msp)
+		return NULL;
+
+	msp->ratelist = kzalloc(sizeof(struct minstrel_rate) * max_rates, gfp);
+	if (!msp->ratelist)
+		goto error;
+
+	msp->sample_table = kmalloc(SAMPLE_COLUMNS * max_rates, gfp);
+	if (!msp->sample_table)
+		goto error1;
+
+	return msp;
+
+error1:
+	kfree(msp->sample_table);
+error:
+	kfree(msp);
+	return NULL;
+}
+
+static void
+minstrel_ht_free_sta(void *priv, struct ieee80211_sta *sta, void *priv_sta)
+{
+	struct minstrel_ht_sta_priv *msp = priv_sta;
+
+	kfree(msp->sample_table);
+	kfree(msp->ratelist);
+	kfree(msp);
+}
+
+static void *
+minstrel_ht_alloc(struct ieee80211_hw *hw, struct dentry *debugfsdir)
+{
+	return mac80211_minstrel.alloc(hw, debugfsdir);
+}
+
+static void
+minstrel_ht_free(void *priv)
+{
+	mac80211_minstrel.free(priv);
+}
+
+static struct rate_control_ops mac80211_minstrel_ht = {
+	.name = "minstrel_ht",
+	.tx_status = minstrel_ht_tx_status,
+	.get_rate = minstrel_ht_get_rate,
+	.rate_init = minstrel_ht_rate_init,
+	.rate_update = minstrel_ht_rate_update,
+	.alloc_sta = minstrel_ht_alloc_sta,
+	.free_sta = minstrel_ht_free_sta,
+	.alloc = minstrel_ht_alloc,
+	.free = minstrel_ht_free,
+#ifdef CONFIG_MAC80211_DEBUGFS
+	.add_sta_debugfs = minstrel_ht_add_sta_debugfs,
+	.remove_sta_debugfs = minstrel_ht_remove_sta_debugfs,
+#endif
+};
+
+
+static void
+init_sample_table(void)
+{
+	int col, i, new_idx;
+	u8 rnd[MCS_GROUP_RATES];
+
+	memset(sample_table, 0xff, sizeof(sample_table));
+	for (col = 0; col < SAMPLE_COLUMNS; col++) {
+		for (i = 0; i < MCS_GROUP_RATES; i++) {
+			get_random_bytes(rnd, sizeof(rnd));
+			new_idx = (i + rnd[i]) % MCS_GROUP_RATES;
+
+			while (sample_table[col][new_idx] != 0xff)
+				new_idx = (new_idx + 1) % MCS_GROUP_RATES;
+
+			sample_table[col][new_idx] = i;
+		}
+	}
+}
+
+int __init
+rc80211_minstrel_ht_init(void)
+{
+	init_sample_table();
+	return ieee80211_rate_control_register(&mac80211_minstrel_ht);
+}
+
+void
+rc80211_minstrel_ht_exit(void)
+{
+	ieee80211_rate_control_unregister(&mac80211_minstrel_ht);
+}
--- /dev/null
+++ b/net/mac80211/rc80211_minstrel_ht.h
@@ -0,0 +1,121 @@
+/*
+ * Copyright (C) 2010 Felix Fietkau <nbd@openwrt.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __RC_MINSTREL_HT_H
+#define __RC_MINSTREL_HT_H
+
+/*
+ * maximum number of spatial streams to make use of
+ * set this value to 3 once we have drivers that support it
+ */
+#define MINSTREL_MAX_STREAMS	2
+#define MINSTREL_STREAM_GROUPS	4
+
+/* scaled fraction values */
+#define MINSTREL_SCALE	16
+#define MINSTREL_FRAC(val, div) (((val) << MINSTREL_SCALE) / div)
+#define MINSTREL_TRUNC(val) ((val) >> MINSTREL_SCALE)
+
+#define MCS_GROUP_RATES	8
+
+struct mcs_group {
+	u32 flags;
+	unsigned int streams;
+	unsigned int duration[MCS_GROUP_RATES];
+};
+
+struct minstrel_rate_stats {
+	/* current / last sampling period attempts/success counters */
+	unsigned int attempts, last_attempts;
+	unsigned int success, last_success;
+
+	/* total attempts/success counters */
+	u64 att_hist, succ_hist;
+
+	/* current throughput */
+	unsigned int cur_tp;
+
+	/* packet delivery probabilities */
+	unsigned int cur_prob, probability;
+
+	/* maximum retry counts */
+	unsigned int retry_count;
+	unsigned int retry_count_rtscts;
+
+	bool retry_updated;
+	u8 sample_skipped;
+};
+
+struct minstrel_mcs_group_data {
+	u8 index;
+	u8 column;
+
+	/* bitfield of supported MCS rates of this group */
+	u8 supported;
+
+	/* selected primary rates */
+	unsigned int max_tp_rate;
+	unsigned int max_tp_rate2;
+	unsigned int max_prob_rate;
+
+	/* MCS rate statistics */
+	struct minstrel_rate_stats rates[MCS_GROUP_RATES];
+};
+
+struct minstrel_ht_sta {
+	/* ampdu length average (EWMA) */
+	unsigned int avg_ampdu_len;
+
+	/* best throughput rate */
+	unsigned int max_tp_rate;
+
+	/* second best throughput rate */
+	unsigned int max_tp_rate2;
+
+	/* best probability rate */
+	unsigned int max_prob_rate;
+
+	/* time of last status update */
+	unsigned long stats_update;
+
+	/* overhead time in usec for each frame */
+	unsigned int overhead;
+	unsigned int overhead_rtscts;
+
+	unsigned int total_packets;
+	unsigned int sample_packets;
+
+	u8 sample_wait;
+	u8 sample_tries;
+	u8 sample_count;
+	u8 sample_slow;
+
+	/* current MCS group to be sampled */
+	u8 sample_group;
+
+	/* MCS rate group info and statistics */
+	struct minstrel_mcs_group_data groups[MINSTREL_MAX_STREAMS * MINSTREL_STREAM_GROUPS];
+};
+
+struct minstrel_ht_sta_priv {
+	union {
+		struct minstrel_ht_sta ht;
+		struct minstrel_sta_info legacy;
+	};
+#ifdef CONFIG_MAC80211_DEBUGFS
+	struct dentry *dbg_stats;
+#endif
+	void *ratelist;
+	void *sample_table;
+	bool is_ht;
+};
+
+void minstrel_ht_add_sta_debugfs(void *priv, void *priv_sta, struct dentry *dir);
+void minstrel_ht_remove_sta_debugfs(void *priv, void *priv_sta);
+
+#endif
--- /dev/null
+++ b/net/mac80211/rc80211_minstrel_ht_debugfs.c
@@ -0,0 +1,120 @@
+/*
+ * Copyright (C) 2010 Felix Fietkau <nbd@openwrt.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/netdevice.h>
+#include <linux/types.h>
+#include <linux/skbuff.h>
+#include <linux/debugfs.h>
+#include <linux/ieee80211.h>
+#include <net/mac80211.h>
+#include "rc80211_minstrel.h"
+#include "rc80211_minstrel_ht.h"
+
+extern const struct mcs_group minstrel_mcs_groups[];
+
+static int
+minstrel_ht_stats_open(struct inode *inode, struct file *file)
+{
+	struct minstrel_ht_sta_priv *msp = inode->i_private;
+	struct minstrel_ht_sta *mi = &msp->ht;
+	struct minstrel_debugfs_info *ms;
+	unsigned int i, j, tp, prob, eprob;
+	char *p;
+	int ret;
+
+	if (!msp->is_ht) {
+		inode->i_private = &msp->legacy;
+		ret = minstrel_stats_open(inode, file);
+		inode->i_private = msp;
+		return ret;
+	}
+
+	ms = kmalloc(sizeof(*ms) + 8192, GFP_KERNEL);
+	if (!ms)
+		return -ENOMEM;
+
+	file->private_data = ms;
+	p = ms->buf;
+	p += sprintf(p, "type      rate     throughput  ewma prob   this prob  "
+			"this succ/attempt   success    attempts\n");
+	for (i = 0; i < MINSTREL_MAX_STREAMS * MINSTREL_STREAM_GROUPS; i++) {
+		char htmode = '2';
+		char gimode = 'L';
+
+		if (!mi->groups[i].supported)
+			continue;
+
+		if (minstrel_mcs_groups[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
+			htmode = '4';
+		if (minstrel_mcs_groups[i].flags & IEEE80211_TX_RC_SHORT_GI)
+			gimode = 'S';
+
+		for (j = 0; j < MCS_GROUP_RATES; j++) {
+			struct minstrel_rate_stats *mr = &mi->groups[i].rates[j];
+			int idx = i * MCS_GROUP_RATES + j;
+
+			if (!(mi->groups[i].supported & BIT(j)))
+				continue;
+
+			p += sprintf(p, "HT%c0/%cGI ", htmode, gimode);
+
+			*(p++) = (idx == mi->max_tp_rate) ? 'T' : ' ';
+			*(p++) = (idx == mi->max_tp_rate2) ? 't' : ' ';
+			*(p++) = (idx == mi->max_prob_rate) ? 'P' : ' ';
+			p += sprintf(p, "MCS%-2u", (minstrel_mcs_groups[i].streams - 1) *
+					MCS_GROUP_RATES + j);
+
+			tp = mr->cur_tp / 10;
+			prob = MINSTREL_TRUNC(mr->cur_prob * 1000);
+			eprob = MINSTREL_TRUNC(mr->probability * 1000);
+
+			p += sprintf(p, "  %6u.%1u   %6u.%1u   %6u.%1u        "
+					"%3u(%3u)   %8llu    %8llu\n",
+					tp / 10, tp % 10,
+					eprob / 10, eprob % 10,
+					prob / 10, prob % 10,
+					mr->last_success,
+					mr->last_attempts,
+					(unsigned long long)mr->succ_hist,
+					(unsigned long long)mr->att_hist);
+		}
+	}
+	p += sprintf(p, "\nTotal packet count::    ideal %d      "
+			"lookaround %d\n",
+			max(0, (int) mi->total_packets - (int) mi->sample_packets),
+			mi->sample_packets);
+	p += sprintf(p, "Average A-MPDU length: %d.%d\n",
+		MINSTREL_TRUNC(mi->avg_ampdu_len),
+		MINSTREL_TRUNC(mi->avg_ampdu_len * 10) % 10);
+	ms->len = p - ms->buf;
+
+	return 0;
+}
+
+static const struct file_operations minstrel_ht_stat_fops = {
+	.owner = THIS_MODULE,
+	.open = minstrel_ht_stats_open,
+	.read = minstrel_stats_read,
+	.release = minstrel_stats_release,
+};
+
+void
+minstrel_ht_add_sta_debugfs(void *priv, void *priv_sta, struct dentry *dir)
+{
+	struct minstrel_ht_sta_priv *msp = priv_sta;
+
+	msp->dbg_stats = debugfs_create_file("rc_stats", S_IRUGO, dir, msp,
+			&minstrel_ht_stat_fops);
+}
+
+void
+minstrel_ht_remove_sta_debugfs(void *priv, void *priv_sta)
+{
+	struct minstrel_ht_sta_priv *msp = priv_sta;
+
+	debugfs_remove(msp->dbg_stats);
+}

Re: [RFC v3] minstrel_ht: new rate control module for 802.11n

[Duy Nguyen]

Hi Felix, thanks for the awesome patch and great progress on the
802.11n!  I just tried out your patch version 3, and got the following
results.

The results are an average of 10 different iperf runs, Throughputs are
in Mbits/sec
AP: dir-615 (settings: 802.11n mode only, Auto 20/40 Mhz).  This AP
does not seem to allow me to set 40Mhz only
Node interfaces:  D-Link DWA-556 Xtreme N PC


scenario 1: node 1 to node 2

node1-----wireless_link-----AP-----wireless_link-----node2

802.11n Felix's Minstrel    Avg Transfer= 38.6  Avg Bandwidth=32.3
802.11n Atheros Default    Avg Transfer= 36.6  Avg Bandwidth=30.6

scenario 2: node 1 to node 2

node1-----wireless_link-----AP-----wired_link-----node2

802.11n Felix's Minstrel    Avg Transfer= 83.4  Avg Bandwidth=69.8
802.11n Atheros Default    Avg Transfer= 95.2  Avg Bandwidth=79.8

Overall, the results look pretty good.  I am much more interested in
results in scenario 1 but I am going to take a look at scenario 2
again.  Let me know if I can help you test any other scenarios.

Duy


On Sun, Mar 7, 2010 at 10:45 AM, Felix Fietkau <nbd@openwrt.org> wrote:
> Here's v3 of minstrel_ht.
>
> Changelog:
>
> v3:
> - Add two fixes by Christian Lamparter (division by zero, debugfs mcs
>  rate display)
> - Improve throughput a bit more by spreading out slow rate sampling
>  attempts over multiple sampling intervals
>
> v2:
> - Use accurate A-MPDU statistics from tx feedback
> - Rewrite the sampling algorithm to optimize for good A-MPDU lengths
>  and faster throughput recovery
> - Implement .rate_update to handle changes in HT capabilities or
>  channel bandwidth
> - Remove the private tx flags hack

Re: [RFC v3] minstrel_ht: new rate control module for 802.11n

[Duy Nguyen]

After many more runs on scenario 2, it is on par with atheros default.
 Sorry for the earlier hasty reporting.

scenario 2: node 1 to node 2

802.11n Felix's Minstrel    Avg Transfer= 95.5  Avg Bandwidth=80.11
802.11n Atheros Default    Avg Transfer= 95.2  Avg Bandwidth=79.8

On Sun, Mar 7, 2010 at 9:20 PM, Duy Nguyen <dnlove@gmail.com> wrote:
> Hi Felix, thanks for the awesome patch and great progress on the
> 802.11n!  I just tried out your patch version 3, and got the following
> results.
>
> The results are an average of 10 different iperf runs, Throughputs are
> in Mbits/sec
> AP: dir-615 (settings: 802.11n mode only, Auto 20/40 Mhz).  This AP
> does not seem to allow me to set 40Mhz only
> Node interfaces:  D-Link DWA-556 Xtreme N PC
>
>
> scenario 1: node 1 to node 2
>
> node1-----wireless_link-----AP-----wireless_link-----node2
>
> 802.11n Felix's Minstrel    Avg Transfer= 38.6  Avg Bandwidth=32.3
> 802.11n Atheros Default    Avg Transfer= 36.6  Avg Bandwidth=30.6
>
> scenario 2: node 1 to node 2
>
> node1-----wireless_link-----AP-----wired_link-----node2
>
> 802.11n Felix's Minstrel    Avg Transfer= 83.4  Avg Bandwidth=69.8
> 802.11n Atheros Default    Avg Transfer= 95.2  Avg Bandwidth=79.8
>
> Overall, the results look pretty good.  I am much more interested in
> results in scenario 1 but I am going to take a look at scenario 2
> again.  Let me know if I can help you test any other scenarios.
>
> Duy
>
>
> On Sun, Mar 7, 2010 at 10:45 AM, Felix Fietkau <nbd@openwrt.org> wrote:
>> Here's v3 of minstrel_ht.
>>
>> Changelog:
>>
>> v3:
>> - Add two fixes by Christian Lamparter (division by zero, debugfs mcs
>>  rate display)
>> - Improve throughput a bit more by spreading out slow rate sampling
>>  attempts over multiple sampling intervals
>>
>> v2:
>> - Use accurate A-MPDU statistics from tx feedback
>> - Rewrite the sampling algorithm to optimize for good A-MPDU lengths
>>  and faster throughput recovery
>> - Implement .rate_update to handle changes in HT capabilities or
>>  channel bandwidth
>> - Remove the private tx flags hack
>