Major overhaul:

o eliminate private state indexed by 802.11 rate codes; use the hal's
  rate tables directly to get the same info
o calculate a mask of operational rates to optimize lookups and checks
  (instead of using for loops and similar)
o optimize size bin operations
o ignore rates marked as "do not use" in the hal phy tables
o fix bug that caused upshifting to break in 11g once the rate dropped
  below 11Mb/s
o add more intelligent multi-rate tx schedules
o add support for 1/2 and 1/4 width channels
o add dev.ath.X.sample_stats sysctl to dump runtime statistics to the console
  (needs to go up to a user app)
o export more tuning knobs via sysctls (still a couple of magic constants)

2 files changed, 569 insertions, 382 deletions

diff --git a/sys/dev/ath/ath_rate/sample/sample.c b/sys/dev/ath/ath_rate/sample/sample.c
index 7d008f3..05d4381 100644
--- a/sys/dev/ath/ath_rate/sample/sample.c
+++ b/sys/dev/ath/ath_rate/sample/sample.c
@@ -62,6 +62,7 @@ __FBSDID("$FreeBSD$");
 #include <net/if.h>
 #include <net/if_media.h>
 #include <net/if_arp.h>
+#include <net/ethernet.h>		/* XXX for ether_sprintf */
 
 #include <net80211/ieee80211_var.h>
 
@@ -100,36 +101,35 @@ __FBSDID("$FreeBSD$");
  * a few different packet sizes independently for each link.
  */
 
-#define STALE_FAILURE_TIMEOUT_MS 10000
-#define MIN_SWITCH_MS 1000
-
 static void	ath_rate_ctl_reset(struct ath_softc *, struct ieee80211_node *);
 
+static const int packet_size_bins[NUM_PACKET_SIZE_BINS] = { 250, 1600 };
+
 static __inline int
 size_to_bin(int size) 
 {
-	int x = 0;
-	for (x = 0; x < NUM_PACKET_SIZE_BINS; x++) {
-		if (size <= packet_size_bins[x]) {
-			return x;
-		}
-	}
+#if NUM_PACKET_SIZE_BINS > 1
+	if (size <= packet_size_bins[0])
+		return 0;
+#endif
+#if NUM_PACKET_SIZE_BINS > 2
+	if (size <= packet_size_bins[1])
+		return 1;
+#endif
+#if NUM_PACKET_SIZE_BINS > 3
+	if (size <= packet_size_bins[2])
+		return 2;
+#endif
+#if NUM_PACKET_SIZE_BINS > 4
+#error "add support for more packet sizes"
+#endif
 	return NUM_PACKET_SIZE_BINS-1;
 }
-static __inline int
-bin_to_size(int index) {
-	return packet_size_bins[index];
-}
 
 static __inline int
-rate_to_ndx(struct sample_node *sn, int rate) {
-	int x = 0;
-	for (x = 0; x < sn->num_rates; x++) {
-		if (sn->rates[x].rate == rate) {
-			return x;
-		}      
-	}
-	return -1;
+bin_to_size(int index)
+{
+	return packet_size_bins[index];
 }
 
 void
@@ -143,336 +143,399 @@ ath_rate_node_cleanup(struct ath_softc *sc, struct ath_node *an)
 {
 }
 
-
 /*
- * returns the ndx with the lowest average_tx_time,
+ * Return the rix with the lowest average_tx_time,
  * or -1 if all the average_tx_times are 0.
  */
-static __inline int best_rate_ndx(struct sample_node *sn, int size_bin, 
-				  int require_acked_before)
+static __inline int
+pick_best_rate(struct sample_node *sn, const HAL_RATE_TABLE *rt,
+    int size_bin, int require_acked_before)
 {
-	int x = 0;
-        int best_rate_ndx = 0;
-        int best_rate_tt = 0;
-        for (x = 0; x < sn->num_rates; x++) {
-		int tt = sn->stats[size_bin][x].average_tx_time;
-		if (tt <= 0 || (require_acked_before && 
-				!sn->stats[size_bin][x].packets_acked)) {
+        int best_rate_rix, best_rate_tt;
+	uint32_t mask;
+	int rix, tt;
+
+        best_rate_rix = 0;
+        best_rate_tt = 0;
+	for (mask = sn->ratemask, rix = 0; mask != 0; mask >>= 1, rix++) {
+		if ((mask & 1) == 0)		/* not a supported rate */
 			continue;
-		}
 
-		/* 9 megabits never works better than 12 */
-		if (sn->rates[x].rate == 18) 
+		tt = sn->stats[size_bin][rix].average_tx_time;
+		if (tt <= 0 ||
+		    (require_acked_before &&
+		     !sn->stats[size_bin][rix].packets_acked))
 			continue;
 
 		/* don't use a bit-rate that has been failing */
-		if (sn->stats[size_bin][x].successive_failures > 3)
+		if (sn->stats[size_bin][rix].successive_failures > 3)
 			continue;
 
-		if (!best_rate_tt || best_rate_tt > tt) {
+		if (best_rate_tt == 0 || tt < best_rate_tt) {
 			best_rate_tt = tt;
-			best_rate_ndx = x;
+			best_rate_rix = rix;
 		}
         }
-        return (best_rate_tt) ? best_rate_ndx : -1;
+        return (best_rate_tt ? best_rate_rix : -1);
 }
 
 /*
- * pick a good "random" bit-rate to sample other than the current one
+ * Pick a good "random" bit-rate to sample other than the current one.
  */
 static __inline int
-pick_sample_ndx(struct sample_node *sn, int size_bin) 
+pick_sample_rate(struct sample_softc *ssc , struct sample_node *sn,
+    const HAL_RATE_TABLE *rt, int size_bin)
 {
-	int x = 0;
-	int current_ndx = 0;
-	unsigned current_tt = 0;
+#define	DOT11RATE(ix)	(rt->info[ix].dot11Rate & IEEE80211_RATE_VAL)
+	int current_rix, rix;
+	unsigned current_tt;
+	uint32_t mask;
 	
-	current_ndx = sn->current_rate[size_bin];
-	if (current_ndx < 0) {
+	current_rix = sn->current_rix[size_bin];
+	if (current_rix < 0) {
 		/* no successes yet, send at the lowest bit-rate */
 		return 0;
 	}
-	
-	current_tt = sn->stats[size_bin][current_ndx].average_tx_time;
-	
-	for (x = 0; x < sn->num_rates; x++) {
-		int ndx = (sn->last_sample_ndx[size_bin]+1+x) % sn->num_rates;
 
-	        /* don't sample the current bit-rate */
-		if (ndx == current_ndx) 
+	current_tt = sn->stats[size_bin][current_rix].average_tx_time;
+
+	rix = sn->last_sample_rix[size_bin]+1;	/* next sample rate */
+	mask = sn->ratemask &~ (1<<current_rix);/* don't sample current rate */
+	while (mask != 0) {
+		if ((mask & (1<<rix)) == 0) {	/* not a supported rate */
+	nextrate:
+			if (++rix >= rt->rateCount)
+				rix = 0;
 			continue;
+		}
 
 		/* this bit-rate is always worse than the current one */
-		if (sn->stats[size_bin][ndx].perfect_tx_time > current_tt) 
-			continue;
+		if (sn->stats[size_bin][rix].perfect_tx_time > current_tt) {
+			mask &= ~(1<<rix);
+			goto nextrate;
+		}
 
 		/* rarely sample bit-rates that fail a lot */
-		if (ticks - sn->stats[size_bin][ndx].last_tx < ((hz * STALE_FAILURE_TIMEOUT_MS)/1000) &&
-		    sn->stats[size_bin][ndx].successive_failures > 3)
-			continue;
-
-		/* don't sample more than 2 indexes higher 
-		 * for rates higher than 11 megabits
-		 */
-		if (sn->rates[ndx].rate > 22 && ndx > current_ndx + 2)
-			continue;
-
-		/* 9 megabits never works better than 12 */
-		if (sn->rates[ndx].rate == 18) 
-			continue;
+		if (sn->stats[size_bin][rix].successive_failures > ssc->max_successive_failures &&
+		    ticks - sn->stats[size_bin][rix].last_tx < ssc->stale_failure_timeout) {
+			mask &= ~(1<<rix);
+			goto nextrate;
+		}
 
-		/* if we're using 11 megabits, only sample up to 12 megabits
-		 */
-		if (sn->rates[current_ndx].rate == 22 && ndx > current_ndx + 1) 
-			continue;
+		/* don't sample more than 2 rates higher for rates > 11M */
+		if (DOT11RATE(rix) > 2*11 && rix > current_rix + 2) {
+			mask &= ~(1<<rix);
+			goto nextrate;
+		}
 
-		sn->last_sample_ndx[size_bin] = ndx;
-		return ndx;
+		sn->last_sample_rix[size_bin] = rix;
+		return rix;
 	}
-	return current_ndx;
+	return current_rix;
+#undef DOT11RATE
 }
 
 void
 ath_rate_findrate(struct ath_softc *sc, struct ath_node *an,
 		  int shortPreamble, size_t frameLen,
-		  u_int8_t *rix, int *try0, u_int8_t *txrate)
+		  u_int8_t *rix0, int *try0, u_int8_t *txrate)
 {
+#define	DOT11RATE(ix)	(rt->info[ix].dot11Rate & IEEE80211_RATE_VAL)
+#define	RATE(ix)	(DOT11RATE(ix) / 2)
 	struct sample_node *sn = ATH_NODE_SAMPLE(an);
 	struct sample_softc *ssc = ATH_SOFTC_SAMPLE(sc);
 	struct ifnet *ifp = sc->sc_ifp;
 	struct ieee80211com *ic = ifp->if_l2com;
-	int ndx, size_bin, mrr, best_ndx, change_rates;
+	const HAL_RATE_TABLE *rt = sc->sc_currates;
+	const int size_bin = size_to_bin(frameLen);
+	int rix, mrr, best_rix, change_rates;
 	unsigned average_tx_time;
 
+	if (sn->static_rix != -1) {
+		rix = sn->static_rix;
+		*try0 = ATH_TXMAXTRY;
+		goto done;
+	}
+
 	mrr = sc->sc_mrretry && !(ic->ic_flags & IEEE80211_F_USEPROT);
-	size_bin = size_to_bin(frameLen);
-	best_ndx = best_rate_ndx(sn, size_bin, !mrr);
 
-	if (best_ndx >= 0) {
-		average_tx_time = sn->stats[size_bin][best_ndx].average_tx_time;
+	best_rix = pick_best_rate(sn, rt, size_bin, !mrr);
+	if (best_rix >= 0) {
+		average_tx_time = sn->stats[size_bin][best_rix].average_tx_time;
 	} else {
 		average_tx_time = 0;
 	}
-	
-	if (sn->static_rate_ndx != -1) {
-		ndx = sn->static_rate_ndx;
-		*try0 = ATH_TXMAXTRY;
-	} else {
-		*try0 = mrr ? 2 : ATH_TXMAXTRY;
-		
-		if (sn->sample_tt[size_bin] < average_tx_time * (sn->packets_since_sample[size_bin]*ssc->ath_sample_rate/100)) {
-			/*
-			 * we want to limit the time measuring the performance
-			 * of other bit-rates to ath_sample_rate% of the
-			 * total transmission time.
-			 */
-			ndx = pick_sample_ndx(sn, size_bin);
-			if (ndx != sn->current_rate[size_bin]) {
-				sn->current_sample_ndx[size_bin] = ndx;
-			} else {
-				sn->current_sample_ndx[size_bin] = -1;
-			}
-			sn->packets_since_sample[size_bin] = 0;
-
+	/*
+	 * Limit the time measuring the performance of other tx
+	 * rates to sample_rate% of the total transmission time.
+	 */
+	if (sn->sample_tt[size_bin] < average_tx_time * (sn->packets_since_sample[size_bin]*ssc->sample_rate/100)) {
+		rix = pick_sample_rate(ssc, sn, rt, size_bin);
+		IEEE80211_NOTE(an->an_node.ni_vap, IEEE80211_MSG_RATECTL,
+		     &an->an_node, "size %u sample rate %d current rate %d",
+		     bin_to_size(size_bin), RATE(rix),
+		     RATE(sn->current_rix[size_bin]));
+		if (rix != sn->current_rix[size_bin]) {
+			sn->current_sample_rix[size_bin] = rix;
 		} else {
-			change_rates = 0;
-			if (!sn->packets_sent[size_bin] || best_ndx == -1) {
-				/* no packet has been sent successfully yet */
-				for (ndx = sn->num_rates-1; ndx > 0; ndx--) {
-					/* 
-					 * pick the highest rate <= 36 Mbps
-					 * that hasn't failed.
-					 */
-					if (sn->rates[ndx].rate <= 72 && 
-					    sn->stats[size_bin][ndx].successive_failures == 0) {
-						break;
-					}
+			sn->current_sample_rix[size_bin] = -1;
+		}
+		sn->packets_since_sample[size_bin] = 0;
+	} else {
+		change_rates = 0;
+		if (!sn->packets_sent[size_bin] || best_rix == -1) {
+			/* no packet has been sent successfully yet */
+			for (rix = rt->rateCount-1; rix > 0; rix--) {
+				if ((sn->ratemask & (1<<rix)) == 0)
+					continue;
+				/* 
+				 * Pick the highest rate <= 36 Mbps
+				 * that hasn't failed.
+				 */
+				if (DOT11RATE(rix) <= 72 && 
+				    sn->stats[size_bin][rix].successive_failures == 0) {
+					break;
 				}
-				change_rates = 1;
-				best_ndx = ndx;
-			} else if (sn->packets_sent[size_bin] < 20) {
-				/* let the bit-rate switch quickly during the first few packets */
-				change_rates = 1;
-			} else if (ticks - ((hz*MIN_SWITCH_MS)/1000) > sn->ticks_since_switch[size_bin]) {
-				/* 2 seconds have gone by */
-				change_rates = 1;
-			} else if (average_tx_time * 2 < sn->stats[size_bin][sn->current_rate[size_bin]].average_tx_time) {
-				/* the current bit-rate is twice as slow as the best one */
-				change_rates = 1;
 			}
+			change_rates = 1;
+			best_rix = rix;
+		} else if (sn->packets_sent[size_bin] < 20) {
+			/* let the bit-rate switch quickly during the first few packets */
+			change_rates = 1;
+		} else if (ticks - ssc->min_switch > sn->ticks_since_switch[size_bin]) {
+			/* min_switch seconds have gone by */
+			change_rates = 1;
+		} else if (2*average_tx_time < sn->stats[size_bin][sn->current_rix[size_bin]].average_tx_time) {
+			/* the current bit-rate is twice as slow as the best one */
+			change_rates = 1;
+		}
 
-			sn->packets_since_sample[size_bin]++;
-			
-			if (change_rates) {
-				if (best_ndx != sn->current_rate[size_bin]) {
-					IEEE80211_NOTE(an->an_node.ni_vap,
-					    IEEE80211_MSG_RATECTL,
-					    &an->an_node,
+		sn->packets_since_sample[size_bin]++;
+		
+		if (change_rates) {
+			if (best_rix != sn->current_rix[size_bin]) {
+				IEEE80211_NOTE(an->an_node.ni_vap,
+				    IEEE80211_MSG_RATECTL,
+				    &an->an_node,
 "%s: size %d switch rate %d (%d/%d) -> %d (%d/%d) after %d packets mrr %d",
-					    __func__,
-					    packet_size_bins[size_bin],
-					    sn->rates[sn->current_rate[size_bin]].rate,
-					    sn->stats[size_bin][sn->current_rate[size_bin]].average_tx_time,
-					    sn->stats[size_bin][sn->current_rate[size_bin]].perfect_tx_time,
-					    sn->rates[best_ndx].rate,
-					    sn->stats[size_bin][best_ndx].average_tx_time,
-					    sn->stats[size_bin][best_ndx].perfect_tx_time,
-					    sn->packets_since_switch[size_bin],
-					    mrr);
-				}
-				sn->packets_since_switch[size_bin] = 0;
-				sn->current_rate[size_bin] = best_ndx;
-				sn->ticks_since_switch[size_bin] = ticks;
-	    			/* 
-	    			 * Set the visible txrate for this node.
-			         */
-				an->an_node.ni_txrate = sn->rates[best_ndx].rate;
+				    __func__,
+				    bin_to_size(size_bin),
+				    RATE(sn->current_rix[size_bin]),
+				    sn->stats[size_bin][sn->current_rix[size_bin]].average_tx_time,
+				    sn->stats[size_bin][sn->current_rix[size_bin]].perfect_tx_time,
+				    RATE(best_rix),
+				    sn->stats[size_bin][best_rix].average_tx_time,
+				    sn->stats[size_bin][best_rix].perfect_tx_time,
+				    sn->packets_since_switch[size_bin],
+				    mrr);
 			}
-			ndx = sn->current_rate[size_bin];
-			sn->packets_since_switch[size_bin]++;
+			sn->packets_since_switch[size_bin] = 0;
+			sn->current_rix[size_bin] = best_rix;
+			sn->ticks_since_switch[size_bin] = ticks;
+			/* 
+			 * Set the visible txrate for this node.
+			 */
+			an->an_node.ni_txrate = DOT11RATE(best_rix);
 		}
+		rix = sn->current_rix[size_bin];
+		sn->packets_since_switch[size_bin]++;
 	}
+	*try0 = mrr ? sn->sched[rix].t0 : ATH_TXMAXTRY;
+done:
+	KASSERT(rix >= 0 && rix < rt->rateCount, ("rix is %d", rix));
 
-	KASSERT(ndx >= 0 && ndx < sn->num_rates, ("ndx is %d", ndx));
-
-	*rix = sn->rates[ndx].rix;
-	if (shortPreamble) {
-		*txrate = sn->rates[ndx].shortPreambleRateCode;
-	} else {
-		*txrate = sn->rates[ndx].rateCode;
-	}
+	*rix0 = rix;
+	*txrate = rt->info[rix].rateCode
+		| (shortPreamble ? rt->info[rix].shortPreamble : 0);
 	sn->packets_sent[size_bin]++;
+#undef DOT11RATE
+#undef RATE
 }
 
+#define A(_r) \
+    (((_r) == 6)   ? 0 : (((_r) == 9)   ? 1 : (((_r) == 12)  ? 2 : \
+    (((_r) == 18)  ? 3 : (((_r) == 24)  ? 4 : (((_r) == 36)  ? 5 : \
+    (((_r) == 48)  ? 6 : (((_r) == 54)  ? 7 : 0))))))))
+static const struct txschedule series_11a[] = {
+	{ 3,A( 6), 3,A(  6), 0,A(  6), 0,A( 6) },	/*   6Mb/s */
+	{ 4,A( 9), 3,A(  6), 4,A(  6), 0,A( 6) },	/*   9Mb/s */
+	{ 4,A(12), 3,A(  6), 4,A(  6), 0,A( 6) },	/*  12Mb/s */
+	{ 4,A(18), 3,A( 12), 4,A(  6), 2,A( 6) },	/*  18Mb/s */
+	{ 4,A(24), 3,A( 18), 4,A( 12), 2,A( 6) },	/*  24Mb/s */
+	{ 4,A(36), 3,A( 24), 4,A( 18), 2,A( 6) },	/*  36Mb/s */
+	{ 4,A(48), 3,A( 36), 4,A( 24), 2,A(12) },	/*  48Mb/s */
+	{ 4,A(54), 3,A( 48), 4,A( 36), 2,A(24) }	/*  54Mb/s */
+};
+#undef A
+
+#define G(_r) \
+    (((_r) == 1)   ? 0 : (((_r) == 2)   ? 1 : (((_r) == 5.5) ? 2 : \
+    (((_r) == 11)  ? 3 : (((_r) == 6)   ? 4 : (((_r) == 9)   ? 5 : \
+    (((_r) == 12)  ? 6 : (((_r) == 18)  ? 7 : (((_r) == 24)  ? 8 : \
+    (((_r) == 36)  ? 9 : (((_r) == 48)  ? 10 : (((_r) == 54)  ? 11 : 0))))))))))))
+static const struct txschedule series_11g[] = {
+	{ 3,G( 1), 3,G(  1), 0,G(  1), 0,G( 1) },	/*   1Mb/s */
+	{ 4,G( 2), 3,G(  1), 4,G(  1), 0,G( 1) },	/*   2Mb/s */
+	{ 4,G(5.5),3,G(  2), 4,G(  1), 2,G( 1) },	/* 5.5Mb/s */
+	{ 4,G(11), 3,G(5.5), 4,G(  2), 2,G( 1) },	/*  11Mb/s */
+	{ 4,G( 6), 3,G(5.5), 4,G(  2), 2,G( 1) },	/*   6Mb/s */
+	{ 4,G( 9), 3,G(  6), 4,G(5.5), 2,G( 1) },	/*   9Mb/s */
+	{ 4,G(12), 3,G( 11), 4,G(5.5), 2,G( 1) },	/*  12Mb/s */
+	{ 4,G(18), 3,G( 12), 4,G( 11), 2,G( 1) },	/*  18Mb/s */
+	{ 4,G(24), 3,G( 18), 4,G( 12), 2,G( 1) },	/*  24Mb/s */
+	{ 4,G(36), 3,G( 24), 4,G( 18), 2,G( 1) },	/*  36Mb/s */
+	{ 4,G(48), 3,G( 36), 4,G( 24), 2,G( 1) },	/*  48Mb/s */
+	{ 4,G(54), 3,G( 48), 4,G( 36), 2,G( 1) }	/*  54Mb/s */
+};
+#undef G
+
+#define H(_r) \
+    (((_r) == 3)   ? 0 : (((_r) == 4.5) ? 1 : (((_r) == 6)  ? 2 : \
+    (((_r) == 9)   ? 3 : (((_r) == 12)  ? 4 : (((_r) == 18) ? 5 : \
+    (((_r) == 24)  ? 6 : (((_r) == 27)  ? 7 : 0))))))))
+static const struct txschedule series_half[] = {
+	{ 3,H( 3), 3,H(  3), 0,H(  3), 0,H( 3) },	/*   3Mb/s */
+	{ 4,H(4.5),3,H(  3), 4,H(  3), 0,H( 3) },	/* 4.5Mb/s */
+	{ 4,H( 6), 3,H(  3), 4,H(  3), 0,H( 3) },	/*   6Mb/s */
+	{ 4,H( 9), 3,H(  6), 4,H(  3), 2,H( 3) },	/*   9Mb/s */
+	{ 4,H(12), 3,H(  9), 4,H(  6), 2,H( 3) },	/*  12Mb/s */
+	{ 4,H(18), 3,H( 12), 4,H(  9), 2,H( 3) },	/*  18Mb/s */
+	{ 4,H(24), 3,H( 18), 4,H( 12), 2,H( 6) },	/*  24Mb/s */
+	{ 4,H(27), 3,H( 24), 4,H( 18), 2,H(12) }	/*  27Mb/s */
+};
+#undef H
+
+#define Q(_r) \
+    (((_r) == 1.5) ? 0 : (((_r) ==2.25) ? 1 : (((_r) == 3)  ? 2 : \
+    (((_r) == 4.5) ? 3 : (((_r) ==  6)  ? 4 : (((_r) == 9)  ? 5 : \
+    (((_r) == 12)  ? 6 : (((_r) == 13.5)? 7 : 0))))))))
+static const struct txschedule series_quarter[] = {
+	{ 3,Q( 1.5),3,Q(1.5), 0,Q(1.5), 0,Q(1.5) },	/* 1.5Mb/s */
+	{ 4,Q(2.25),3,Q(1.5), 4,Q(1.5), 0,Q(1.5) },	/*2.25Mb/s */
+	{ 4,Q(   3),3,Q(1.5), 4,Q(1.5), 0,Q(1.5) },	/*   3Mb/s */
+	{ 4,Q( 4.5),3,Q(  3), 4,Q(1.5), 2,Q(1.5) },	/* 4.5Mb/s */
+	{ 4,Q(   6),3,Q(4.5), 4,Q(  3), 2,Q(1.5) },	/*   6Mb/s */
+	{ 4,Q(   9),3,Q(  6), 4,Q(4.5), 2,Q(1.5) },	/*   9Mb/s */
+	{ 4,Q(  12),3,Q(  9), 4,Q(  6), 2,Q(  3) },	/*  12Mb/s */
+	{ 4,Q(13.5),3,Q( 12), 4,Q(  9), 2,Q(  6) }	/*13.5Mb/s */
+};
+#undef Q
+
 void
 ath_rate_setupxtxdesc(struct ath_softc *sc, struct ath_node *an,
 		      struct ath_desc *ds, int shortPreamble, u_int8_t rix)
 {
 	struct sample_node *sn = ATH_NODE_SAMPLE(an);
-	int rateCode = -1;
-	int frame_size = 0;
-	int size_bin = 0;
-	int ndx = 0;
-
-	size_bin = size_to_bin(frame_size);	// TODO: it's correct that frame_size alway 0 ?
-	ndx = sn->current_rate[size_bin]; /* retry at the current bit-rate */
-	
-	if (!sn->stats[size_bin][ndx].packets_acked) {
-		ndx = 0;  /* use the lowest bit-rate */
-	}
-
-	if (shortPreamble) {
-		rateCode = sn->rates[ndx].shortPreambleRateCode;
-	} else {
-		rateCode = sn->rates[ndx].rateCode;
-	}
-	ath_hal_setupxtxdesc(sc->sc_ah, ds
-			     , rateCode, 3	        /* series 1 */
-			     , sn->rates[0].rateCode, 3	/* series 2 */
-			     , 0, 0	                /* series 3 */
-			     );
+	const struct txschedule *sched = &sn->sched[rix];
+	const HAL_RATE_TABLE *rt = sc->sc_currates;
+	uint8_t rix1, s1code, rix2, s2code, rix3, s3code;
+
+	/* XXX precalculate short preamble tables */
+	rix1 = sched->r1;
+	s1code = rt->info[rix1].rateCode
+	       | (shortPreamble ? rt->info[rix1].shortPreamble : 0);
+	rix2 = sched->r2;
+	s2code = rt->info[rix2].rateCode
+	       | (shortPreamble ? rt->info[rix2].shortPreamble : 0);
+	rix3 = sched->r3;
+	s3code = rt->info[rix3].rateCode
+	       | (shortPreamble ? rt->info[rix3].shortPreamble : 0);
+	ath_hal_setupxtxdesc(sc->sc_ah, ds,
+	    s1code, sched->t1,		/* series 1 */
+	    s2code, sched->t2,		/* series 2 */
+	    s3code, sched->t3);		/* series 3 */
 }
 
 static void
 update_stats(struct ath_softc *sc, struct ath_node *an, 
 		  int frame_size,
-		  int ndx0, int tries0,
-		  int ndx1, int tries1,
-		  int ndx2, int tries2,
-		  int ndx3, int tries3,
+		  int rix0, int tries0,
+		  int rix1, int tries1,
+		  int rix2, int tries2,
+		  int rix3, int tries3,
 		  int short_tries, int tries, int status)
 {
 	struct sample_node *sn = ATH_NODE_SAMPLE(an);
 	struct sample_softc *ssc = ATH_SOFTC_SAMPLE(sc);
-	int tt = 0;
-	int tries_so_far = 0;
-	int size_bin = 0;
-	int size = 0;
-	int rate = 0;
+	const int size_bin = size_to_bin(frame_size);
+	const int size = bin_to_size(size_bin);
+	int tt, tries_so_far;
 
-	size_bin = size_to_bin(frame_size);
-	size = bin_to_size(size_bin);
-
-	if (!(0 <= ndx0 && ndx0 < sn->num_rates))
+	if (!IS_RATE_DEFINED(sn, rix0))
 		return;
-	rate = sn->rates[ndx0].rate;
-
-	tt += calc_usecs_unicast_packet(sc, size, sn->rates[ndx0].rix, 
-					short_tries,
-					MIN(tries0, tries) - 1);
-	tries_so_far += tries0;
-	if (tries1 && tries0 < tries) {
-		if (!(0 <= ndx1 && ndx1 < sn->num_rates))
+	tt = calc_usecs_unicast_packet(sc, size, rix0, short_tries,
+		MIN(tries0, tries) - 1);
+	tries_so_far = tries0;
+
+	if (tries1 && tries_so_far < tries) {
+		if (!IS_RATE_DEFINED(sn, rix1))
 			return;
-		tt += calc_usecs_unicast_packet(sc, size, sn->rates[ndx1].rix, 
-						short_tries,
-						MIN(tries1 + tries_so_far, tries) - tries_so_far - 1);
+		tt += calc_usecs_unicast_packet(sc, size, rix1, short_tries,
+			MIN(tries1 + tries_so_far, tries) - tries_so_far - 1);
+		tries_so_far += tries1;
 	}
-	tries_so_far += tries1;
 
-	if (tries2 && tries0 + tries1 < tries) {
-		if (!(0 <= ndx2 && ndx2 < sn->num_rates))
+	if (tries2 && tries_so_far < tries) {
+		if (!IS_RATE_DEFINED(sn, rix2))
 			return;
-		tt += calc_usecs_unicast_packet(sc, size, sn->rates[ndx2].rix, 
-					       short_tries,
-						MIN(tries2 + tries_so_far, tries) - tries_so_far - 1);
+		tt += calc_usecs_unicast_packet(sc, size, rix2, short_tries,
+			MIN(tries2 + tries_so_far, tries) - tries_so_far - 1);
+		tries_so_far += tries2;
 	}
 
-	tries_so_far += tries2;
-
-	if (tries3 && tries0 + tries1 + tries2 < tries) {
-		if (!(0 <= ndx3 && ndx3 < sn->num_rates))
+	if (tries3 && tries_so_far < tries) {
+		if (!IS_RATE_DEFINED(sn, rix3))
 			return;
-		tt += calc_usecs_unicast_packet(sc, size, sn->rates[ndx3].rix, 
-						short_tries,
-						MIN(tries3 + tries_so_far, tries) - tries_so_far - 1);
+		tt += calc_usecs_unicast_packet(sc, size, rix3, short_tries,
+			MIN(tries3 + tries_so_far, tries) - tries_so_far - 1);
 	}
-	if (sn->stats[size_bin][ndx0].total_packets < (100 / (100 - ssc->ath_smoothing_rate))) {
+
+	if (sn->stats[size_bin][rix0].total_packets < ssc->smoothing_minpackets) {
 		/* just average the first few packets */
-		int avg_tx = sn->stats[size_bin][ndx0].average_tx_time;
-		int packets = sn->stats[size_bin][ndx0].total_packets;
-		sn->stats[size_bin][ndx0].average_tx_time = (tt+(avg_tx*packets))/(packets+1);
+		int avg_tx = sn->stats[size_bin][rix0].average_tx_time;
+		int packets = sn->stats[size_bin][rix0].total_packets;
+		sn->stats[size_bin][rix0].average_tx_time = (tt+(avg_tx*packets))/(packets+1);
 	} else {
 		/* use a ewma */
-		sn->stats[size_bin][ndx0].average_tx_time = 
-			((sn->stats[size_bin][ndx0].average_tx_time * ssc->ath_smoothing_rate) + 
-			 (tt * (100 - ssc->ath_smoothing_rate))) / 100;
+		sn->stats[size_bin][rix0].average_tx_time = 
+			((sn->stats[size_bin][rix0].average_tx_time * ssc->smoothing_rate) + 
+			 (tt * (100 - ssc->smoothing_rate))) / 100;
 	}
 	
-	if (status) {
+	if (status != 0) {
 		int y;
-		sn->stats[size_bin][ndx0].successive_failures++;
+		sn->stats[size_bin][rix0].successive_failures++;
 		for (y = size_bin+1; y < NUM_PACKET_SIZE_BINS; y++) {
-			/* also say larger packets failed since we
-			 * assume if a small packet fails at a lower
+			/*
+			 * Also say larger packets failed since we
+			 * assume if a small packet fails at a
 			 * bit-rate then a larger one will also.
 			 */
-			sn->stats[y][ndx0].successive_failures++;
-			sn->stats[y][ndx0].last_tx = ticks;
-			sn->stats[y][ndx0].tries += tries;
-			sn->stats[y][ndx0].total_packets++;
+			sn->stats[y][rix0].successive_failures++;
+			sn->stats[y][rix0].last_tx = ticks;
+			sn->stats[y][rix0].tries += tries;
+			sn->stats[y][rix0].total_packets++;
 		}
 	} else {
-		sn->stats[size_bin][ndx0].packets_acked++;
-		sn->stats[size_bin][ndx0].successive_failures = 0;
+		sn->stats[size_bin][rix0].packets_acked++;
+		sn->stats[size_bin][rix0].successive_failures = 0;
 	}
-	sn->stats[size_bin][ndx0].tries += tries;
-	sn->stats[size_bin][ndx0].last_tx = ticks;
-	sn->stats[size_bin][ndx0].total_packets++;
-
+	sn->stats[size_bin][rix0].tries += tries;
+	sn->stats[size_bin][rix0].last_tx = ticks;
+	sn->stats[size_bin][rix0].total_packets++;
 
-	if (ndx0 == sn->current_sample_ndx[size_bin]) {
+	if (rix0 == sn->current_sample_rix[size_bin]) {
 		IEEE80211_NOTE(an->an_node.ni_vap, IEEE80211_MSG_RATECTL,
 		   &an->an_node,
 "%s: size %d %s sample rate %d tries (%d/%d) tt %d avg_tt (%d/%d)", 
 		    __func__, 
 		    size,
 		    status ? "FAIL" : "OK",
-		    rate, short_tries, tries, tt, 
-		    sn->stats[size_bin][ndx0].average_tx_time,
-		    sn->stats[size_bin][ndx0].perfect_tx_time);
+		    rix0, short_tries, tries, tt, 
+		    sn->stats[size_bin][rix0].average_tx_time,
+		    sn->stats[size_bin][rix0].perfect_tx_time);
 		sn->sample_tt[size_bin] = tt;
-		sn->current_sample_ndx[size_bin] = -1;
+		sn->current_sample_rix[size_bin] = -1;
 	}
 }
 
@@ -492,19 +555,18 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 	struct sample_node *sn = ATH_NODE_SAMPLE(an);
 	const struct ath_tx_status *ts = &bf->bf_status.ds_txstat;
 	const struct ath_desc *ds0 = &bf->bf_desc[0];
-	int final_rate, short_tries, long_tries, frame_size;
+	int final_rix, short_tries, long_tries, frame_size;
 	const HAL_RATE_TABLE *rt = sc->sc_currates;
 	int mrr;
 
-	final_rate = sc->sc_hwmap[
-	    rt->rateCodeToIndex[ts->ts_rate &~ HAL_TXSTAT_ALTRATE]].ieeerate;
+	final_rix = rt->rateCodeToIndex[ts->ts_rate &~ HAL_TXSTAT_ALTRATE];
 	short_tries = ts->ts_shortretry;
 	long_tries = ts->ts_longretry + 1;
 	frame_size = ds0->ds_ctl0 & 0x0fff; /* low-order 12 bits of ds_ctl0 */
 	if (frame_size == 0)		    /* NB: should not happen */
 		frame_size = 1500;
 
-	if (sn->num_rates <= 0) {
+	if (sn->ratemask == 0) {
 		IEEE80211_NOTE(an->an_node.ni_vap, IEEE80211_MSG_RATECTL,
 		    &an->an_node,
 		    "%s: size %d %s rate/try %d/%d no rates yet", 
@@ -516,9 +578,7 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 	}
 	mrr = sc->sc_mrretry && !(ic->ic_flags & IEEE80211_F_USEPROT);
 	if (!mrr || !(ts->ts_rate & HAL_TXSTAT_ALTRATE)) {
-		int ndx = rate_to_ndx(sn, final_rate);
-
-		if (ndx < 0) {
+		if (!IS_RATE_DEFINED(sn, final_rix)) {
 			badrate(ifp, 0, ts->ts_rate, long_tries, ts->ts_status);
 			return;
 		}
@@ -530,18 +590,18 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 		     __func__,
 		     bin_to_size(size_to_bin(frame_size)),
 		     ts->ts_status ? "FAIL" : "OK",
-		     final_rate, short_tries, long_tries);
+		     final_rix, short_tries, long_tries);
 		update_stats(sc, an, frame_size, 
-			     ndx, long_tries,
+			     final_rix, long_tries,
 			     0, 0,
 			     0, 0,
 			     0, 0,
 			     short_tries, long_tries, ts->ts_status);
 	} else {
-		int hwrate0, rate0, tries0, ndx0;
-		int hwrate1, rate1, tries1, ndx1;
-		int hwrate2, rate2, tries2, ndx2;
-		int hwrate3, rate3, tries3, ndx3;
+		int hwrate0, rix0, tries0;
+		int hwrate1, rix1, tries1;
+		int hwrate2, rix2, tries2;
+		int hwrate3, rix3, tries3;
 		int finalTSIdx = ts->ts_finaltsi;
 
 		/*
@@ -559,30 +619,17 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 			hwrate3 = MS(ds0->ds_ctl3, AR5416_XmitRate3);
 		}
 
-		rate0 = sc->sc_hwmap[rt->rateCodeToIndex[hwrate0]].ieeerate;
+		rix0 = rt->rateCodeToIndex[hwrate0];
 		tries0 = MS(ds0->ds_ctl2, AR_XmitDataTries0);
-		ndx0 = rate_to_ndx(sn, rate0);
 
-		rate1 = sc->sc_hwmap[rt->rateCodeToIndex[hwrate1]].ieeerate;
+		rix1 = rt->rateCodeToIndex[hwrate1];
 		tries1 = MS(ds0->ds_ctl2, AR_XmitDataTries1);
-		ndx1 = rate_to_ndx(sn, rate1);
 
-		rate2 = sc->sc_hwmap[rt->rateCodeToIndex[hwrate2]].ieeerate;
+		rix2 = rt->rateCodeToIndex[hwrate2];
 		tries2 = MS(ds0->ds_ctl2, AR_XmitDataTries2);
-		ndx2 = rate_to_ndx(sn, rate2);
 
-		rate3 = sc->sc_hwmap[rt->rateCodeToIndex[hwrate3]].ieeerate;
+		rix3 = rt->rateCodeToIndex[hwrate3];
 		tries3 = MS(ds0->ds_ctl2, AR_XmitDataTries3);
-		ndx3 = rate_to_ndx(sn, rate3);
-
-		if (tries0 && ndx0 < 0)
-			badrate(ifp, 0, hwrate0, tries0, ts->ts_status);
-		if (tries1 && ndx1 < 0)
-			badrate(ifp, 1, hwrate1, tries1, ts->ts_status);
-		if (tries2 && ndx2 < 0)
-			badrate(ifp, 2, hwrate2, tries2, ts->ts_status);
-		if (tries3 && ndx3 < 0)
-			badrate(ifp, 3, hwrate3, tries3, ts->ts_status);
 
 		IEEE80211_NOTE(an->an_node.ni_vap, IEEE80211_MSG_RATECTL,
 		    &an->an_node,
@@ -592,10 +639,19 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 		     finalTSIdx,
 		     long_tries, 
 		     ts->ts_status ? "FAIL" : "OK",
-		     rate0, tries0,
-		     rate1, tries1,
-		     rate2, tries2,
-		     rate3, tries3);
+		     rix0, tries0,
+		     rix1, tries1,
+		     rix2, tries2,
+		     rix3, tries3);
+
+		if (tries0 && !IS_RATE_DEFINED(sn, rix0))
+			badrate(ifp, 0, hwrate0, tries0, ts->ts_status);
+		if (tries1 && !IS_RATE_DEFINED(sn, rix1))
+			badrate(ifp, 1, hwrate1, tries1, ts->ts_status);
+		if (tries2 && !IS_RATE_DEFINED(sn, rix2))
+			badrate(ifp, 2, hwrate2, tries2, ts->ts_status);
+		if (tries3 && !IS_RATE_DEFINED(sn, rix3))
+			badrate(ifp, 3, hwrate3, tries3, ts->ts_status);
 
 		/*
 		 * NB: series > 0 are not penalized for failure
@@ -606,10 +662,10 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 		 */
 		if (tries0) {
 			update_stats(sc, an, frame_size, 
-				     ndx0, tries0, 
-				     ndx1, tries1, 
-				     ndx2, tries2, 
-				     ndx3, tries3, 
+				     rix0, tries0, 
+				     rix1, tries1, 
+				     rix2, tries2, 
+				     rix3, tries3, 
 				     short_tries, long_tries, 
 				     long_tries > tries0);
 			long_tries -= tries0;
@@ -617,9 +673,9 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 		
 		if (tries1 && finalTSIdx > 0) {
 			update_stats(sc, an, frame_size, 
-				     ndx1, tries1, 
-				     ndx2, tries2, 
-				     ndx3, tries3, 
+				     rix1, tries1, 
+				     rix2, tries2, 
+				     rix3, tries3, 
 				     0, 0, 
 				     short_tries, long_tries, 
 				     ts->ts_status);
@@ -628,8 +684,8 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 
 		if (tries2 && finalTSIdx > 1) {
 			update_stats(sc, an, frame_size, 
-				     ndx2, tries2, 
-				     ndx3, tries3, 
+				     rix2, tries2, 
+				     rix3, tries3, 
 				     0, 0,
 				     0, 0,
 				     short_tries, long_tries, 
@@ -639,7 +695,7 @@ ath_rate_tx_complete(struct ath_softc *sc, struct ath_node *an,
 
 		if (tries3 && finalTSIdx > 2) {
 			update_stats(sc, an, frame_size, 
-				     ndx3, tries3, 
+				     rix3, tries3, 
 				     0, 0,
 				     0, 0,
 				     0, 0,
@@ -656,6 +712,21 @@ ath_rate_newassoc(struct ath_softc *sc, struct ath_node *an, int isnew)
 		ath_rate_ctl_reset(sc, &an->an_node);
 }
 
+static const struct txschedule *mrr_schedules[IEEE80211_MODE_MAX+2] = {
+	NULL,		/* IEEE80211_MODE_AUTO */
+	series_11a,	/* IEEE80211_MODE_11A */
+	series_11g,	/* IEEE80211_MODE_11B */
+	series_11g,	/* IEEE80211_MODE_11G */
+	NULL,		/* IEEE80211_MODE_FH */
+	series_11a,	/* IEEE80211_MODE_TURBO_A */
+	series_11g,	/* IEEE80211_MODE_TURBO_G */
+	series_11a,	/* IEEE80211_MODE_STURBO_A */
+	series_11a,	/* IEEE80211_MODE_11NA */
+	series_11g,	/* IEEE80211_MODE_11NG */
+	series_half,	/* IEEE80211_MODE_HALF */
+	series_quarter,	/* IEEE80211_MODE_QUARTER */
+};
+
 /*
  * Initialize the tables for a node.
  */
@@ -663,14 +734,22 @@ static void
 ath_rate_ctl_reset(struct ath_softc *sc, struct ieee80211_node *ni)
 {
 #define	RATE(_ix)	(ni->ni_rates.rs_rates[(_ix)] & IEEE80211_RATE_VAL)
+#define	DOT11RATE(_ix)	(rt->info[(_ix)].dot11Rate & IEEE80211_RATE_VAL)
 	struct ath_node *an = ATH_NODE(ni);
 	const struct ieee80211_txparam *tp = ni->ni_txparms;
 	struct sample_node *sn = ATH_NODE_SAMPLE(an);
 	const HAL_RATE_TABLE *rt = sc->sc_currates;
-	int x, y, srate;
+	int x, y, srate, rix;
 
 	KASSERT(rt != NULL, ("no rate table, mode %u", sc->sc_curmode));
-        sn->static_rate_ndx = -1;
+
+	KASSERT(sc->sc_curmode < IEEE80211_MODE_MAX+2,
+	    ("curmode %u", sc->sc_curmode));
+	sn->sched = mrr_schedules[sc->sc_curmode];
+	KASSERT(sn->sched != NULL,
+	    ("no mrr schedule for mode %u", sc->sc_curmode));
+
+        sn->static_rix = -1;
 	if (tp != NULL && tp->ucastrate != IEEE80211_FIXED_RATE_NONE) {
 		/*
 		 * A fixed rate is to be used; ic_fixed_rate is the
@@ -689,122 +768,227 @@ ath_rate_ctl_reset(struct ath_softc *sc, struct ieee80211_node *ni)
 		 * can fail.
 		 */
 		if (srate >= 0)
-			sn->static_rate_ndx = srate;
+			sn->static_rix = sc->sc_rixmap[srate];
 	}
 
-	sn->num_rates = ni->ni_rates.rs_nrates;
-        for (x = 0; x < ni->ni_rates.rs_nrates; x++) {
-		sn->rates[x].rate = ni->ni_rates.rs_rates[x] & IEEE80211_RATE_VAL;
-		sn->rates[x].rix = sc->sc_rixmap[sn->rates[x].rate];
-		if (sn->rates[x].rix == 0xff) {
-			IEEE80211_NOTE(ni->ni_vap, IEEE80211_MSG_RATECTL, ni,
-			    "%s: ignore bogus rix at %d", __func__, x);
+	/*
+	 * Construct a bitmask of usable rates.  This has all
+	 * negotiated rates minus those marked by the hal as
+	 * to be ignored for doing rate control.
+	 */
+	sn->ratemask = 0;
+	for (x = 0; x < ni->ni_rates.rs_nrates; x++) {
+		rix = sc->sc_rixmap[RATE(x)];
+		if (rix == 0xff)
 			continue;
-		}
-		sn->rates[x].rateCode = rt->info[sn->rates[x].rix].rateCode;
-		sn->rates[x].shortPreambleRateCode = 
-			rt->info[sn->rates[x].rix].rateCode | 
-			rt->info[sn->rates[x].rix].shortPreamble;
+		/* skip rates marked broken by hal */
+		if (!rt->info[rix].valid)
+			continue;
+		KASSERT(rix < SAMPLE_MAXRATES,
+		    ("rate %u has rix %d", RATE(x), rix));
+		sn->ratemask |= 1<<rix;
 	}
 #ifdef IEEE80211_DEBUG
 	if (ieee80211_msg(ni->ni_vap, IEEE80211_MSG_RATECTL)) {
+		uint32_t mask;
+
 		ieee80211_note(ni->ni_vap, "[%6D] %s: size 1600 rate/tt",
-		    __func__, ni->ni_macaddr, ":");
-		for (x = 0; x < sn->num_rates; x++) {
-			if (sn->rates[x].rix == 0xff)
+		    ni->ni_macaddr, ":", __func__);
+		for (mask = sn->ratemask, rix = 0; mask != 0; mask >>= 1, rix++) {
+			if ((mask & 1) == 0)
 				continue;
-			printf(" %d/%d", sn->rates[x].rate,
-			    calc_usecs_unicast_packet(sc, 1600,
-				sn->rates[x].rix, 0,0));
+			printf(" %d/%d", DOT11RATE(rix) / 2,
+			    calc_usecs_unicast_packet(sc, 1600, rix, 0,0));
 		}
 		printf("\n");
 	}
 #endif
 	for (y = 0; y < NUM_PACKET_SIZE_BINS; y++) {
 		int size = bin_to_size(y);
-		int ndx = 0;
+		uint32_t mask;
+
 		sn->packets_sent[y] = 0;
-		sn->current_sample_ndx[y] = -1;
-		sn->last_sample_ndx[y] = 0;
+		sn->current_sample_rix[y] = -1;
+		sn->last_sample_rix[y] = 0;
+		/* XXX start with first valid rate */
+		sn->current_rix[y] = ffs(sn->ratemask)-1;
 		
-		for (x = 0; x < ni->ni_rates.rs_nrates; x++) {
-			sn->stats[y][x].successive_failures = 0;
-			sn->stats[y][x].tries = 0;
-			sn->stats[y][x].total_packets = 0;
-			sn->stats[y][x].packets_acked = 0;
-			sn->stats[y][x].last_tx = 0;
+		/*
+		 * Initialize the statistics buckets; these are
+		 * indexed by the rate code index.
+		 */
+		for (rix = 0, mask = sn->ratemask; mask != 0; rix++, mask >>= 1) {
+			if ((mask & 1) == 0)		/* not a valid rate */
+				continue;
+			sn->stats[y][rix].successive_failures = 0;
+			sn->stats[y][rix].tries = 0;
+			sn->stats[y][rix].total_packets = 0;
+			sn->stats[y][rix].packets_acked = 0;
+			sn->stats[y][rix].last_tx = 0;
 			
-			sn->stats[y][x].perfect_tx_time = 
-				calc_usecs_unicast_packet(sc, size, 
-							  sn->rates[x].rix,
-							  0, 0);
-			sn->stats[y][x].average_tx_time = sn->stats[y][x].perfect_tx_time;
+			sn->stats[y][rix].perfect_tx_time =
+			    calc_usecs_unicast_packet(sc, size, rix, 0, 0);
+			sn->stats[y][rix].average_tx_time =
+			    sn->stats[y][rix].perfect_tx_time;
 		}
-
-		/* set the initial rate */
-		for (ndx = sn->num_rates-1; ndx > 0; ndx--) {
-			if (sn->rates[ndx].rate <= 72) {
-				break;
-			}
-		}
-		sn->current_rate[y] = ndx;
 	}
-
+#if 0
+	/* XXX 0, num_rates-1 are wrong */
 	IEEE80211_NOTE(ni->ni_vap, IEEE80211_MSG_RATECTL, ni,
 	    "%s: %d rates %d%sMbps (%dus)- %d%sMbps (%dus)", __func__, 
 	    sn->num_rates,
-	    sn->rates[0].rate/2, sn->rates[0].rate % 0x1 ? ".5" : "",
+	    DOT11RATE(0)/2, DOT11RATE(0) % 1 ? ".5" : "",
 	    sn->stats[1][0].perfect_tx_time,
-	    sn->rates[sn->num_rates-1].rate/2,
-		sn->rates[sn->num_rates-1].rate % 0x1 ? ".5" : "",
+	    DOT11RATE(sn->num_rates-1)/2, DOT11RATE(sn->num_rates-1) % 1 ? ".5" : "",
 	    sn->stats[1][sn->num_rates-1].perfect_tx_time
 	);
-
+#endif
 	/* set the visible bit-rate */
-	if (sn->static_rate_ndx != -1)
-		ni->ni_txrate = sn->rates[sn->static_rate_ndx].rate;
+	if (sn->static_rix != -1)
+		ni->ni_txrate = DOT11RATE(sn->static_rix);
 	else
-		ni->ni_txrate = sn->rates[0].rate;
+		ni->ni_txrate = RATE(0);
 #undef RATE
+#undef DOT11RATE
+}
+
+static void
+sample_stats(void *arg, struct ieee80211_node *ni)
+{
+	struct ath_softc *sc = arg;
+	const HAL_RATE_TABLE *rt = sc->sc_currates;
+	struct sample_node *sn = ATH_NODE_SAMPLE(ATH_NODE(ni));
+	uint32_t mask;
+	int rix, y;
+
+	printf("\n[%s] refcnt %d static_rix %d ratemask 0x%x\n",
+	    ether_sprintf(ni->ni_macaddr), ieee80211_node_refcnt(ni),
+	    sn->static_rix, sn->ratemask);
+	for (y = 0; y < NUM_PACKET_SIZE_BINS; y++) {
+		printf("[%4u] cur rix %d since switch: packets %d ticks %u\n",
+		    bin_to_size(y), sn->current_rix[y],
+		    sn->packets_since_switch[y], sn->ticks_since_switch[y]);
+		printf("[%4u] last sample %d cur sample %d packets sent %d\n",
+		    bin_to_size(y), sn->last_sample_rix[y],
+		    sn->current_sample_rix[y], sn->packets_sent[y]);
+		printf("[%4u] packets since sample %d sample tt %u\n",
+		    bin_to_size(y), sn->packets_since_sample[y],
+		    sn->sample_tt[y]);
+	}
+	for (mask = sn->ratemask, rix = 0; mask != 0; mask >>= 1, rix++) {
+		if ((mask & 1) == 0)
+				continue;
+		for (y = 0; y < NUM_PACKET_SIZE_BINS; y++) {
+			if (sn->stats[y][rix].total_packets == 0)
+				continue;
+			printf("[%2u:%4u] %8d:%-8d (%3d%%) T %8d F %4d avg %5u last %u\n",
+			    (rt->info[rix].dot11Rate & IEEE80211_RATE_VAL)/2,
+			    bin_to_size(y),
+			    sn->stats[y][rix].total_packets,
+			    sn->stats[y][rix].packets_acked,
+			    (100*sn->stats[y][rix].packets_acked)/sn->stats[y][rix].total_packets,
+			    sn->stats[y][rix].tries,
+			    sn->stats[y][rix].successive_failures,
+			    sn->stats[y][rix].average_tx_time,
+			    ticks - sn->stats[y][rix].last_tx);
+		}
+	}
+}
+
+static int
+ath_rate_sysctl_stats(SYSCTL_HANDLER_ARGS)
+{
+	struct ath_softc *sc = arg1;
+	struct ifnet *ifp = sc->sc_ifp;
+	struct ieee80211com *ic = ifp->if_l2com;
+	int error, v;
+
+	v = 0;
+	error = sysctl_handle_int(oidp, &v, 0, req);
+	if (error || !req->newptr)
+		return error;
+	ieee80211_iterate_nodes(&ic->ic_sta, sample_stats, sc);
+	return 0;
+}
+
+static int
+ath_rate_sysctl_smoothing_rate(SYSCTL_HANDLER_ARGS)
+{
+	struct sample_softc *ssc = arg1;
+	int rate, error;
+
+	rate = ssc->smoothing_rate;
+	error = sysctl_handle_int(oidp, &rate, 0, req);
+	if (error || !req->newptr)
+		return error;
+	if (!(0 <= rate && rate < 100))
+		return EINVAL;
+	ssc->smoothing_rate = rate;
+	ssc->smoothing_minpackets = 100 / (100 - rate);
+	return 0;
+}
+
+static int
+ath_rate_sysctl_sample_rate(SYSCTL_HANDLER_ARGS)
+{
+	struct sample_softc *ssc = arg1;
+	int rate, error;
+
+	rate = ssc->sample_rate;
+	error = sysctl_handle_int(oidp, &rate, 0, req);
+	if (error || !req->newptr)
+		return error;
+	if (!(2 <= rate && rate <= 100))
+		return EINVAL;
+	ssc->sample_rate = rate;
+	return 0;
 }
 
 static void
-ath_rate_sysctlattach(struct ath_softc *sc, struct sample_softc *osc)
+ath_rate_sysctlattach(struct ath_softc *sc, struct sample_softc *ssc)
 {
 	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->sc_dev);
 	struct sysctl_oid *tree = device_get_sysctl_tree(sc->sc_dev);
 
-	/* XXX bounds check [0..100] */
-	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
-		"smoothing_rate", CTLFLAG_RW, &osc->ath_smoothing_rate, 0,
-		"rate control: retry threshold to credit rate raise (%%)");
-	/* XXX bounds check [2..100] */
-	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
-		"sample_rate", CTLFLAG_RW, &osc->ath_sample_rate,0,
-		"rate control: # good periods before raising rate");
+	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
+	    "smoothing_rate", CTLTYPE_INT | CTLFLAG_RW, ssc, 0,
+	    ath_rate_sysctl_smoothing_rate, "I",
+	    "sample: smoothing rate for avg tx time (%%)");
+	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
+	    "sample_rate", CTLTYPE_INT | CTLFLAG_RW, ssc, 0,
+	    ath_rate_sysctl_sample_rate, "I",
+	    "sample: percent air time devoted to sampling new rates (%%)");
+	/* XXX max_successive_failures, stale_failure_timeout, min_switch */
+	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
+	    "sample_stats", CTLTYPE_INT | CTLFLAG_RW, sc, 0,
+	    ath_rate_sysctl_stats, "I", "sample: print statistics");
 }
 
 struct ath_ratectrl *
 ath_rate_attach(struct ath_softc *sc)
 {
-	struct sample_softc *osc;
+	struct sample_softc *ssc;
 	
-	osc = malloc(sizeof(struct sample_softc), M_DEVBUF, M_NOWAIT|M_ZERO);
-	if (osc == NULL)
+	ssc = malloc(sizeof(struct sample_softc), M_DEVBUF, M_NOWAIT|M_ZERO);
+	if (ssc == NULL)
 		return NULL;
-	osc->arc.arc_space = sizeof(struct sample_node);
-	osc->ath_smoothing_rate = 95;	/* ewma percentage (out of 100) */
-	osc->ath_sample_rate = 10;	/* send a different bit-rate 1/X packets */
-	ath_rate_sysctlattach(sc, osc);
-	return &osc->arc;
+	ssc->arc.arc_space = sizeof(struct sample_node);
+	ssc->smoothing_rate = 95;		/* ewma percentage ([0..99]) */
+	ssc->smoothing_minpackets = 100 / (100 - ssc->smoothing_rate);
+	ssc->sample_rate = 10;			/* %time to try diff tx rates */
+	ssc->max_successive_failures = 3;	/* threshold for rate sampling*/
+	ssc->stale_failure_timeout = 10 * hz;	/* 10 seconds */
+	ssc->min_switch = hz;			/* 1 second */
+	ath_rate_sysctlattach(sc, ssc);
+	return &ssc->arc;
 }
 
 void
 ath_rate_detach(struct ath_ratectrl *arc)
 {
-	struct sample_softc *osc = (struct sample_softc *) arc;
+	struct sample_softc *ssc = (struct sample_softc *) arc;
 	
-	free(osc, M_DEVBUF);
+	free(ssc, M_DEVBUF);
 }
 
 /*
@@ -816,7 +1000,7 @@ sample_modevent(module_t mod, int type, void *unused)
 	switch (type) {
 	case MOD_LOAD:
 		if (bootverbose)
-			printf("ath_rate: version 1.2 <SampleRate bit-rate selection algorithm>\n");
+			printf("ath_rate: version 1.9 <SampleRate bit-rate selection algorithm>\n");
 		return 0;
 	case MOD_UNLOAD:
 		return 0;
diff --git a/sys/dev/ath/ath_rate/sample/sample.h b/sys/dev/ath/ath_rate/sample/sample.h
index 1ea96c4..3baabdf 100644
--- a/sys/dev/ath/ath_rate/sample/sample.h
+++ b/sys/dev/ath/ath_rate/sample/sample.h
@@ -44,20 +44,16 @@
 
 /* per-device state */
 struct sample_softc {
-	struct ath_ratectrl arc;	/* base state */
-	int	ath_smoothing_rate;	/* ewma percentage (out of 100) */
-	int	ath_sample_rate;	/* send a different bit-rate 1/X packets */
+	struct ath_ratectrl arc;	/* base class */
+	int	smoothing_rate;		/* ewma percentage [0..99] */
+	int	smoothing_minpackets;
+	int	sample_rate;		/* %time to try different tx rates */
+	int	max_successive_failures;
+	int	stale_failure_timeout;	/* how long to honor max_successive_failures */
+	int	min_switch;		/* min time between rate changes */
 };
 #define	ATH_SOFTC_SAMPLE(sc)	((struct sample_softc *)sc->sc_rc)
 
-struct rate_info {
-	int rate;
-	int rix;
-	int rateCode;
-	int shortPreambleRateCode;
-};
-
-
 struct rate_stats {	
 	unsigned average_tx_time;
 	int successive_failures;
@@ -68,34 +64,41 @@ struct rate_stats {
 	int last_tx;
 };
 
+struct txschedule {
+	uint8_t	t0, r0;		/* series 0: tries, rate code */
+	uint8_t	t1, r1;		/* series 1: tries, rate code */
+	uint8_t	t2, r2;		/* series 2: tries, rate code */
+	uint8_t	t3, r3;		/* series 3: tries, rate code */
+};
+
 /*
  * for now, we track performance for three different packet
  * size buckets
  */
-#define NUM_PACKET_SIZE_BINS 3
-static int packet_size_bins[NUM_PACKET_SIZE_BINS] = {250, 1600, 3000};
+#define NUM_PACKET_SIZE_BINS 2
 
 /* per-node state */
 struct sample_node {
-	int static_rate_ndx;
-	int num_rates;
+	int static_rix;			/* rate index of fixed tx rate */
+#define	SAMPLE_MAXRATES	32		/* NB: corresponds to hal info[32] */
+	uint32_t ratemask;		/* bit mask of valid rate indices */
+	const struct txschedule *sched;	/* tx schedule table */
 
-	struct rate_info rates[IEEE80211_RATE_MAXSIZE];
-	
-	struct rate_stats stats[NUM_PACKET_SIZE_BINS][IEEE80211_RATE_MAXSIZE];
-	int last_sample_ndx[NUM_PACKET_SIZE_BINS];
+	struct rate_stats stats[NUM_PACKET_SIZE_BINS][SAMPLE_MAXRATES];
+	int last_sample_rix[NUM_PACKET_SIZE_BINS];
 
-	int current_sample_ndx[NUM_PACKET_SIZE_BINS];       
+	int current_sample_rix[NUM_PACKET_SIZE_BINS];       
 	int packets_sent[NUM_PACKET_SIZE_BINS];
 
-	int current_rate[NUM_PACKET_SIZE_BINS];
+	int current_rix[NUM_PACKET_SIZE_BINS];
 	int packets_since_switch[NUM_PACKET_SIZE_BINS];
 	unsigned ticks_since_switch[NUM_PACKET_SIZE_BINS];
 
 	int packets_since_sample[NUM_PACKET_SIZE_BINS];
 	unsigned sample_tt[NUM_PACKET_SIZE_BINS];
 };
-#define	ATH_NODE_SAMPLE(an)	((struct sample_node *)&an[1])
+#define	ATH_NODE_SAMPLE(an)	((struct sample_node *)&(an)[1])
+#define	IS_RATE_DEFINED(sn, rix)	(((sn)->ratemask & (1<<(rix))) != 0)
 
 #ifndef MIN
 #define	MIN(a,b)	((a) < (b) ? (a) : (b))