ath9k: fold the aggregation rework into 300-pending_work.patch (tracked as upstream submission)

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

git-svn-id: svn://svn.openwrt.org/openwrt/trunk@37716 3c298f89-4303-0410-b956-a3cf2f4a3e73

1 files changed, 970 insertions, 18 deletions

diff --git a/package/kernel/mac80211/patches/300-pending_work.patch b/package/kernel/mac80211/patches/300-pending_work.patch
index 0e5ceb8b92..55449db40f 100644
--- a/package/kernel/mac80211/patches/300-pending_work.patch
+++ b/package/kernel/mac80211/patches/300-pending_work.patch
@@ -278,7 +278,17 @@
  	WLAN_STA_BLOCK_BA,
 --- a/drivers/net/wireless/ath/ath9k/xmit.c
 +++ b/drivers/net/wireless/ath/ath9k/xmit.c
-@@ -146,6 +146,28 @@ static void ath_set_rates(struct ieee802
+@@ -135,6 +135,9 @@ static struct ath_frame_info *get_frame_
+ 
+ static void ath_send_bar(struct ath_atx_tid *tid, u16 seqno)
+ {
++	if (!tid->an->sta)
++		return;
++
+ 	ieee80211_send_bar(tid->an->vif, tid->an->sta->addr, tid->tidno,
+ 			   seqno << IEEE80211_SEQ_SEQ_SHIFT);
+ }
+@@ -146,6 +149,93 @@ static void ath_set_rates(struct ieee802
  			       ARRAY_SIZE(bf->rates));
  }
  
@@ -304,34 +314,780 @@
 +	}
 +}
 +
++static struct ath_atx_tid *
++ath_get_skb_tid(struct ath_softc *sc, struct ath_node *an, struct sk_buff *skb)
++{
++	struct ieee80211_hdr *hdr;
++	u8 tidno = 0;
++
++	hdr = (struct ieee80211_hdr *) skb->data;
++	if (ieee80211_is_data_qos(hdr->frame_control))
++		tidno = ieee80211_get_qos_ctl(hdr)[0];
++
++	tidno &= IEEE80211_QOS_CTL_TID_MASK;
++	return ATH_AN_2_TID(an, tidno);
++}
++
++static bool ath_tid_has_buffered(struct ath_atx_tid *tid)
++{
++	return !skb_queue_empty(&tid->buf_q) || !skb_queue_empty(&tid->retry_q);
++}
++
++static struct sk_buff *ath_tid_dequeue(struct ath_atx_tid *tid)
++{
++	struct sk_buff *skb;
++
++	skb = __skb_dequeue(&tid->retry_q);
++	if (!skb)
++		skb = __skb_dequeue(&tid->buf_q);
++
++	return skb;
++}
++
++/*
++ * ath_tx_tid_change_state:
++ * - clears a-mpdu flag of previous session
++ * - force sequence number allocation to fix next BlockAck Window
++ */
++static void
++ath_tx_tid_change_state(struct ath_softc *sc, struct ath_atx_tid *tid)
++{
++	struct ath_txq *txq = tid->ac->txq;
++	struct ieee80211_tx_info *tx_info;
++	struct sk_buff *skb, *tskb;
++	struct ath_buf *bf;
++	struct ath_frame_info *fi;
++
++	skb_queue_walk_safe(&tid->buf_q, skb, tskb) {
++		fi = get_frame_info(skb);
++		bf = fi->bf;
++
++		tx_info = IEEE80211_SKB_CB(skb);
++		tx_info->flags &= ~IEEE80211_TX_CTL_AMPDU;
++
++		if (bf)
++			continue;
++
++		bf = ath_tx_setup_buffer(sc, txq, tid, skb);
++		if (!bf) {
++			__skb_unlink(skb, &tid->buf_q);
++			ath_txq_skb_done(sc, txq, skb);
++			ieee80211_free_txskb(sc->hw, skb);
++			continue;
++		}
++	}
++
++}
++
  static void ath_tx_flush_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
  {
  	struct ath_txq *txq = tid->ac->txq;
-@@ -167,6 +189,7 @@ static void ath_tx_flush_tid(struct ath_
+@@ -160,27 +250,22 @@ static void ath_tx_flush_tid(struct ath_
+ 
+ 	memset(&ts, 0, sizeof(ts));
+ 
+-	while ((skb = __skb_dequeue(&tid->buf_q))) {
++	while ((skb = __skb_dequeue(&tid->retry_q))) {
+ 		fi = get_frame_info(skb);
+ 		bf = fi->bf;
+-
  		if (!bf) {
+-			bf = ath_tx_setup_buffer(sc, txq, tid, skb);
+-			if (!bf) {
+-				ieee80211_free_txskb(sc->hw, skb);
+-				continue;
+-			}
++			ath_txq_skb_done(sc, txq, skb);
++			ieee80211_free_txskb(sc->hw, skb);
++			continue;
+ 		}
+ 
+-		if (fi->retries) {
+-			list_add_tail(&bf->list, &bf_head);
++		if (fi->baw_tracked) {
+ 			ath_tx_update_baw(sc, tid, bf->bf_state.seqno);
+-			ath_tx_complete_buf(sc, bf, txq, &bf_head, &ts, 0);
+ 			sendbar = true;
+-		} else {
+-			ath_set_rates(tid->an->vif, tid->an->sta, bf);
+-			ath_tx_send_normal(sc, txq, NULL, skb);
+ 		}
++
++		list_add_tail(&bf->list, &bf_head);
++		ath_tx_complete_buf(sc, bf, txq, &bf_head, &ts, 0);
+ 	}
+ 
+ 	if (sendbar) {
+@@ -209,13 +294,16 @@ static void ath_tx_update_baw(struct ath
+ }
+ 
+ static void ath_tx_addto_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
+-			     u16 seqno)
++			     struct ath_buf *bf)
+ {
++	struct ath_frame_info *fi = get_frame_info(bf->bf_mpdu);
++	u16 seqno = bf->bf_state.seqno;
+ 	int index, cindex;
+ 
+ 	index  = ATH_BA_INDEX(tid->seq_start, seqno);
+ 	cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
+ 	__set_bit(cindex, tid->tx_buf);
++	fi->baw_tracked = 1;
+ 
+ 	if (index >= ((tid->baw_tail - tid->baw_head) &
+ 		(ATH_TID_MAX_BUFS - 1))) {
+@@ -243,7 +331,7 @@ static void ath_tid_drain(struct ath_sof
+ 	memset(&ts, 0, sizeof(ts));
+ 	INIT_LIST_HEAD(&bf_head);
+ 
+-	while ((skb = __skb_dequeue(&tid->buf_q))) {
++	while ((skb = ath_tid_dequeue(tid))) {
+ 		fi = get_frame_info(skb);
+ 		bf = fi->bf;
+ 
+@@ -380,7 +468,6 @@ static void ath_tx_complete_aggr(struct 
+ 	struct ieee80211_tx_rate rates[4];
+ 	struct ath_frame_info *fi;
+ 	int nframes;
+-	u8 tidno;
+ 	bool flush = !!(ts->ts_status & ATH9K_TX_FLUSH);
+ 	int i, retries;
+ 	int bar_index = -1;
+@@ -417,8 +504,7 @@ static void ath_tx_complete_aggr(struct 
+ 	}
+ 
+ 	an = (struct ath_node *)sta->drv_priv;
+-	tidno = ieee80211_get_qos_ctl(hdr)[0] & IEEE80211_QOS_CTL_TID_MASK;
+-	tid = ATH_AN_2_TID(an, tidno);
++	tid = ath_get_skb_tid(sc, an, skb);
+ 	seq_first = tid->seq_start;
+ 	isba = ts->ts_flags & ATH9K_TX_BA;
+ 
+@@ -430,7 +516,7 @@ static void ath_tx_complete_aggr(struct 
+ 	 * Only BlockAcks have a TID and therefore normal Acks cannot be
+ 	 * checked
+ 	 */
+-	if (isba && tidno != ts->tid)
++	if (isba && tid->tidno != ts->tid)
+ 		txok = false;
+ 
+ 	isaggr = bf_isaggr(bf);
+@@ -466,7 +552,8 @@ static void ath_tx_complete_aggr(struct 
+ 		tx_info = IEEE80211_SKB_CB(skb);
+ 		fi = get_frame_info(skb);
+ 
+-		if (!BAW_WITHIN(tid->seq_start, tid->baw_size, seqno)) {
++		if (!BAW_WITHIN(tid->seq_start, tid->baw_size, seqno) ||
++		    !tid->active) {
+ 			/*
+ 			 * Outside of the current BlockAck window,
+ 			 * maybe part of a previous session
+@@ -560,7 +647,7 @@ static void ath_tx_complete_aggr(struct 
+ 		if (an->sleeping)
+ 			ieee80211_sta_set_buffered(sta, tid->tidno, true);
+ 
+-		skb_queue_splice(&bf_pending, &tid->buf_q);
++		skb_queue_splice_tail(&bf_pending, &tid->retry_q);
+ 		if (!an->sleeping) {
+ 			ath_tx_queue_tid(txq, tid);
+ 
+@@ -618,7 +705,7 @@ static void ath_tx_process_buffer(struct
+ 	} else
+ 		ath_tx_complete_aggr(sc, txq, bf, bf_head, ts, txok);
+ 
+-	if ((sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_HT) && !flush)
++	if (!flush)
+ 		ath_txq_schedule(sc, txq);
+ }
+ 
+@@ -792,15 +879,20 @@ static int ath_compute_num_delims(struct
+ 
+ static struct ath_buf *
+ ath_tx_get_tid_subframe(struct ath_softc *sc, struct ath_txq *txq,
+-			struct ath_atx_tid *tid)
++			struct ath_atx_tid *tid, struct sk_buff_head **q)
+ {
++	struct ieee80211_tx_info *tx_info;
+ 	struct ath_frame_info *fi;
+ 	struct sk_buff *skb;
+ 	struct ath_buf *bf;
+ 	u16 seqno;
+ 
+ 	while (1) {
+-		skb = skb_peek(&tid->buf_q);
++		*q = &tid->retry_q;
++		if (skb_queue_empty(*q))
++			*q = &tid->buf_q;
++
++		skb = skb_peek(*q);
+ 		if (!skb)
+ 			break;
+ 
+@@ -810,11 +902,22 @@ ath_tx_get_tid_subframe(struct ath_softc
  			bf = ath_tx_setup_buffer(sc, txq, tid, skb);
- 			if (!bf) {
-+				ath_txq_skb_done(sc, txq, skb);
- 				ieee80211_free_txskb(sc->hw, skb);
- 				continue;
- 			}
-@@ -811,6 +834,7 @@ ath_tx_get_tid_subframe(struct ath_softc
  
  		if (!bf) {
- 			__skb_unlink(skb, &tid->buf_q);
+-			__skb_unlink(skb, &tid->buf_q);
++			__skb_unlink(skb, *q);
 +			ath_txq_skb_done(sc, txq, skb);
  			ieee80211_free_txskb(sc->hw, skb);
  			continue;
  		}
-@@ -1824,6 +1848,7 @@ static void ath_tx_send_ampdu(struct ath
  
- 	bf = ath_tx_setup_buffer(sc, txq, tid, skb);
- 	if (!bf) {
-+		ath_txq_skb_done(sc, txq, skb);
- 		ieee80211_free_txskb(sc->hw, skb);
++		bf->bf_next = NULL;
++		bf->bf_lastbf = bf;
++
++		tx_info = IEEE80211_SKB_CB(skb);
++		tx_info->flags &= ~IEEE80211_TX_CTL_CLEAR_PS_FILT;
++		if (!(tx_info->flags & IEEE80211_TX_CTL_AMPDU)) {
++			bf->bf_state.bf_type = 0;
++			return bf;
++		}
++
+ 		bf->bf_state.bf_type = BUF_AMPDU | BUF_AGGR;
+ 		seqno = bf->bf_state.seqno;
+ 
+@@ -828,73 +931,52 @@ ath_tx_get_tid_subframe(struct ath_softc
+ 
+ 			INIT_LIST_HEAD(&bf_head);
+ 			list_add(&bf->list, &bf_head);
+-			__skb_unlink(skb, &tid->buf_q);
++			__skb_unlink(skb, *q);
+ 			ath_tx_update_baw(sc, tid, seqno);
+ 			ath_tx_complete_buf(sc, bf, txq, &bf_head, &ts, 0);
+ 			continue;
+ 		}
+ 
+-		bf->bf_next = NULL;
+-		bf->bf_lastbf = bf;
+ 		return bf;
+ 	}
+ 
+ 	return NULL;
+ }
+ 
+-static enum ATH_AGGR_STATUS ath_tx_form_aggr(struct ath_softc *sc,
+-					     struct ath_txq *txq,
+-					     struct ath_atx_tid *tid,
+-					     struct list_head *bf_q,
+-					     int *aggr_len)
++static bool
++ath_tx_form_aggr(struct ath_softc *sc, struct ath_txq *txq,
++		 struct ath_atx_tid *tid, struct list_head *bf_q,
++		 struct ath_buf *bf_first, struct sk_buff_head *tid_q,
++		 int *aggr_len)
+ {
+ #define PADBYTES(_len) ((4 - ((_len) % 4)) % 4)
+-	struct ath_buf *bf, *bf_first = NULL, *bf_prev = NULL;
+-	int rl = 0, nframes = 0, ndelim, prev_al = 0;
++	struct ath_buf *bf = bf_first, *bf_prev = NULL;
++	int nframes = 0, ndelim;
+ 	u16 aggr_limit = 0, al = 0, bpad = 0,
+-		al_delta, h_baw = tid->baw_size / 2;
+-	enum ATH_AGGR_STATUS status = ATH_AGGR_DONE;
++	    al_delta, h_baw = tid->baw_size / 2;
+ 	struct ieee80211_tx_info *tx_info;
+ 	struct ath_frame_info *fi;
+ 	struct sk_buff *skb;
++	bool closed = false;
+ 
+-	do {
+-		bf = ath_tx_get_tid_subframe(sc, txq, tid);
+-		if (!bf) {
+-			status = ATH_AGGR_BAW_CLOSED;
+-			break;
+-		}
++	bf = bf_first;
++	aggr_limit = ath_lookup_rate(sc, bf, tid);
+ 
++	do {
+ 		skb = bf->bf_mpdu;
+ 		fi = get_frame_info(skb);
+ 
+-		if (!bf_first)
+-			bf_first = bf;
+-
+-		if (!rl) {
+-			ath_set_rates(tid->an->vif, tid->an->sta, bf);
+-			aggr_limit = ath_lookup_rate(sc, bf, tid);
+-			rl = 1;
+-		}
+-
+ 		/* do not exceed aggregation limit */
+ 		al_delta = ATH_AGGR_DELIM_SZ + fi->framelen;
++		if (nframes) {
++			if (aggr_limit < al + bpad + al_delta ||
++			    ath_lookup_legacy(bf) || nframes >= h_baw)
++				break;
+ 
+-		if (nframes &&
+-		    ((aggr_limit < (al + bpad + al_delta + prev_al)) ||
+-		     ath_lookup_legacy(bf))) {
+-			status = ATH_AGGR_LIMITED;
+-			break;
+-		}
+-
+-		tx_info = IEEE80211_SKB_CB(bf->bf_mpdu);
+-		if (nframes && (tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE))
+-			break;
+-
+-		/* do not exceed subframe limit */
+-		if (nframes >= min((int)h_baw, ATH_AMPDU_SUBFRAME_DEFAULT)) {
+-			status = ATH_AGGR_LIMITED;
+-			break;
++			tx_info = IEEE80211_SKB_CB(bf->bf_mpdu);
++			if ((tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE) ||
++			    !(tx_info->flags & IEEE80211_TX_CTL_AMPDU))
++				break;
+ 		}
+ 
+ 		/* add padding for previous frame to aggregation length */
+@@ -912,22 +994,37 @@ static enum ATH_AGGR_STATUS ath_tx_form_
+ 		bf->bf_next = NULL;
+ 
+ 		/* link buffers of this frame to the aggregate */
+-		if (!fi->retries)
+-			ath_tx_addto_baw(sc, tid, bf->bf_state.seqno);
++		if (!fi->baw_tracked)
++			ath_tx_addto_baw(sc, tid, bf);
+ 		bf->bf_state.ndelim = ndelim;
+ 
+-		__skb_unlink(skb, &tid->buf_q);
++		__skb_unlink(skb, tid_q);
+ 		list_add_tail(&bf->list, bf_q);
+ 		if (bf_prev)
+ 			bf_prev->bf_next = bf;
+ 
+ 		bf_prev = bf;
+ 
+-	} while (!skb_queue_empty(&tid->buf_q));
++		bf = ath_tx_get_tid_subframe(sc, txq, tid, &tid_q);
++		if (!bf) {
++			closed = true;
++			break;
++		}
++	} while (ath_tid_has_buffered(tid));
++
++	bf = bf_first;
++	bf->bf_lastbf = bf_prev;
++
++	if (bf == bf_prev) {
++		al = get_frame_info(bf->bf_mpdu)->framelen;
++		bf->bf_state.bf_type = BUF_AMPDU;
++	} else {
++		TX_STAT_INC(txq->axq_qnum, a_aggr);
++	}
+ 
+ 	*aggr_len = al;
+ 
+-	return status;
++	return closed;
+ #undef PADBYTES
+ }
+ 
+@@ -1188,53 +1285,86 @@ static void ath_tx_fill_desc(struct ath_
+ 	}
+ }
+ 
+-static void ath_tx_sched_aggr(struct ath_softc *sc, struct ath_txq *txq,
+-			      struct ath_atx_tid *tid)
++static void
++ath_tx_form_burst(struct ath_softc *sc, struct ath_txq *txq,
++		  struct ath_atx_tid *tid, struct list_head *bf_q,
++		  struct ath_buf *bf_first, struct sk_buff_head *tid_q)
+ {
+-	struct ath_buf *bf;
+-	enum ATH_AGGR_STATUS status;
+-	struct ieee80211_tx_info *tx_info;
+-	struct list_head bf_q;
+-	int aggr_len;
++	struct ath_buf *bf = bf_first, *bf_prev = NULL;
++	struct sk_buff *skb;
++	int nframes = 0;
+ 
+ 	do {
+-		if (skb_queue_empty(&tid->buf_q))
+-			return;
++		struct ieee80211_tx_info *tx_info;
++		skb = bf->bf_mpdu;
+ 
+-		INIT_LIST_HEAD(&bf_q);
++		nframes++;
++		__skb_unlink(skb, tid_q);
++		list_add_tail(&bf->list, bf_q);
++		if (bf_prev)
++			bf_prev->bf_next = bf;
++		bf_prev = bf;
+ 
+-		status = ath_tx_form_aggr(sc, txq, tid, &bf_q, &aggr_len);
++		if (nframes >= 2)
++			break;
+ 
+-		/*
+-		 * no frames picked up to be aggregated;
+-		 * block-ack window is not open.
+-		 */
+-		if (list_empty(&bf_q))
++		bf = ath_tx_get_tid_subframe(sc, txq, tid, &tid_q);
++		if (!bf)
+ 			break;
+ 
+-		bf = list_first_entry(&bf_q, struct ath_buf, list);
+-		bf->bf_lastbf = list_entry(bf_q.prev, struct ath_buf, list);
+ 		tx_info = IEEE80211_SKB_CB(bf->bf_mpdu);
++		if (tx_info->flags & IEEE80211_TX_CTL_AMPDU)
++			break;
+ 
+-		if (tid->ac->clear_ps_filter) {
+-			tid->ac->clear_ps_filter = false;
+-			tx_info->flags |= IEEE80211_TX_CTL_CLEAR_PS_FILT;
+-		} else {
+-			tx_info->flags &= ~IEEE80211_TX_CTL_CLEAR_PS_FILT;
+-		}
++		ath_set_rates(tid->an->vif, tid->an->sta, bf);
++	} while (1);
++}
+ 
+-		/* if only one frame, send as non-aggregate */
+-		if (bf == bf->bf_lastbf) {
+-			aggr_len = get_frame_info(bf->bf_mpdu)->framelen;
+-			bf->bf_state.bf_type = BUF_AMPDU;
+-		} else {
+-			TX_STAT_INC(txq->axq_qnum, a_aggr);
+-		}
++static bool ath_tx_sched_aggr(struct ath_softc *sc, struct ath_txq *txq,
++			      struct ath_atx_tid *tid, bool *stop)
++{
++	struct ath_buf *bf;
++	struct ieee80211_tx_info *tx_info;
++	struct sk_buff_head *tid_q;
++	struct list_head bf_q;
++	int aggr_len = 0;
++	bool aggr, last = true;
++
++	if (!ath_tid_has_buffered(tid))
++		return false;
++
++	INIT_LIST_HEAD(&bf_q);
++
++	bf = ath_tx_get_tid_subframe(sc, txq, tid, &tid_q);
++	if (!bf)
++		return false;
++
++	tx_info = IEEE80211_SKB_CB(bf->bf_mpdu);
++	aggr = !!(tx_info->flags & IEEE80211_TX_CTL_AMPDU);
++	if ((aggr && txq->axq_ampdu_depth >= ATH_AGGR_MIN_QDEPTH) ||
++		(!aggr && txq->axq_depth >= ATH_NON_AGGR_MIN_QDEPTH)) {
++		*stop = true;
++		return false;
++	}
++
++	ath_set_rates(tid->an->vif, tid->an->sta, bf);
++	if (aggr)
++		last = ath_tx_form_aggr(sc, txq, tid, &bf_q, bf,
++					tid_q, &aggr_len);
++	else
++		ath_tx_form_burst(sc, txq, tid, &bf_q, bf, tid_q);
++
++	if (list_empty(&bf_q))
++		return false;
++
++	if (tid->ac->clear_ps_filter || tid->an->no_ps_filter) {
++		tid->ac->clear_ps_filter = false;
++		tx_info->flags |= IEEE80211_TX_CTL_CLEAR_PS_FILT;
++	}
+ 
+-		ath_tx_fill_desc(sc, bf, txq, aggr_len);
+-		ath_tx_txqaddbuf(sc, txq, &bf_q, false);
+-	} while (txq->axq_ampdu_depth < ATH_AGGR_MIN_QDEPTH &&
+-		 status != ATH_AGGR_BAW_CLOSED);
++	ath_tx_fill_desc(sc, bf, txq, aggr_len);
++	ath_tx_txqaddbuf(sc, txq, &bf_q, false);
++	return true;
+ }
+ 
+ int ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
+@@ -1258,6 +1388,9 @@ int ath_tx_aggr_start(struct ath_softc *
+ 		an->mpdudensity = density;
+ 	}
+ 
++	/* force sequence number allocation for pending frames */
++	ath_tx_tid_change_state(sc, txtid);
++
+ 	txtid->active = true;
+ 	txtid->paused = true;
+ 	*ssn = txtid->seq_start = txtid->seq_next;
+@@ -1277,8 +1410,9 @@ void ath_tx_aggr_stop(struct ath_softc *
+ 
+ 	ath_txq_lock(sc, txq);
+ 	txtid->active = false;
+-	txtid->paused = true;
++	txtid->paused = false;
+ 	ath_tx_flush_tid(sc, txtid);
++	ath_tx_tid_change_state(sc, txtid);
+ 	ath_txq_unlock_complete(sc, txq);
+ }
+ 
+@@ -1302,7 +1436,7 @@ void ath_tx_aggr_sleep(struct ieee80211_
+ 
+ 		ath_txq_lock(sc, txq);
+ 
+-		buffered = !skb_queue_empty(&tid->buf_q);
++		buffered = ath_tid_has_buffered(tid);
+ 
+ 		tid->sched = false;
+ 		list_del(&tid->list);
+@@ -1334,7 +1468,7 @@ void ath_tx_aggr_wakeup(struct ath_softc
+ 		ath_txq_lock(sc, txq);
+ 		ac->clear_ps_filter = true;
+ 
+-		if (!skb_queue_empty(&tid->buf_q) && !tid->paused) {
++		if (!tid->paused && ath_tid_has_buffered(tid)) {
+ 			ath_tx_queue_tid(txq, tid);
+ 			ath_txq_schedule(sc, txq);
+ 		}
+@@ -1359,7 +1493,7 @@ void ath_tx_aggr_resume(struct ath_softc
+ 	tid->baw_size = IEEE80211_MIN_AMPDU_BUF << sta->ht_cap.ampdu_factor;
+ 	tid->paused = false;
+ 
+-	if (!skb_queue_empty(&tid->buf_q)) {
++	if (ath_tid_has_buffered(tid)) {
+ 		ath_tx_queue_tid(txq, tid);
+ 		ath_txq_schedule(sc, txq);
+ 	}
+@@ -1379,6 +1513,7 @@ void ath9k_release_buffered_frames(struc
+ 	struct ieee80211_tx_info *info;
+ 	struct list_head bf_q;
+ 	struct ath_buf *bf_tail = NULL, *bf;
++	struct sk_buff_head *tid_q;
+ 	int sent = 0;
+ 	int i;
+ 
+@@ -1394,15 +1529,15 @@ void ath9k_release_buffered_frames(struc
+ 			continue;
+ 
+ 		ath_txq_lock(sc, tid->ac->txq);
+-		while (!skb_queue_empty(&tid->buf_q) && nframes > 0) {
+-			bf = ath_tx_get_tid_subframe(sc, sc->tx.uapsdq, tid);
++		while (nframes > 0) {
++			bf = ath_tx_get_tid_subframe(sc, sc->tx.uapsdq, tid, &tid_q);
+ 			if (!bf)
+ 				break;
+ 
+-			__skb_unlink(bf->bf_mpdu, &tid->buf_q);
++			__skb_unlink(bf->bf_mpdu, tid_q);
+ 			list_add_tail(&bf->list, &bf_q);
+ 			ath_set_rates(tid->an->vif, tid->an->sta, bf);
+-			ath_tx_addto_baw(sc, tid, bf->bf_state.seqno);
++			ath_tx_addto_baw(sc, tid, bf);
+ 			bf->bf_state.bf_type &= ~BUF_AGGR;
+ 			if (bf_tail)
+ 				bf_tail->bf_next = bf;
+@@ -1412,7 +1547,7 @@ void ath9k_release_buffered_frames(struc
+ 			sent++;
+ 			TX_STAT_INC(txq->axq_qnum, a_queued_hw);
+ 
+-			if (skb_queue_empty(&tid->buf_q))
++			if (an->sta && !ath_tid_has_buffered(tid))
+ 				ieee80211_sta_set_buffered(an->sta, i, false);
+ 		}
+ 		ath_txq_unlock_complete(sc, tid->ac->txq);
+@@ -1665,25 +1800,27 @@ void ath_tx_cleanupq(struct ath_softc *s
+  */
+ void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq)
+ {
+-	struct ath_atx_ac *ac, *ac_tmp, *last_ac;
++	struct ath_atx_ac *ac, *last_ac;
+ 	struct ath_atx_tid *tid, *last_tid;
++	bool sent = false;
+ 
+ 	if (test_bit(SC_OP_HW_RESET, &sc->sc_flags) ||
+-	    list_empty(&txq->axq_acq) ||
+-	    txq->axq_ampdu_depth >= ATH_AGGR_MIN_QDEPTH)
++	    list_empty(&txq->axq_acq))
  		return;
+ 
+ 	rcu_read_lock();
+ 
+-	ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac, list);
+ 	last_ac = list_entry(txq->axq_acq.prev, struct ath_atx_ac, list);
++	while (!list_empty(&txq->axq_acq)) {
++		bool stop = false;
+ 
+-	list_for_each_entry_safe(ac, ac_tmp, &txq->axq_acq, list) {
++		ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac, list);
+ 		last_tid = list_entry(ac->tid_q.prev, struct ath_atx_tid, list);
+ 		list_del(&ac->list);
+ 		ac->sched = false;
+ 
+ 		while (!list_empty(&ac->tid_q)) {
++
+ 			tid = list_first_entry(&ac->tid_q, struct ath_atx_tid,
+ 					       list);
+ 			list_del(&tid->list);
+@@ -1692,17 +1829,17 @@ void ath_txq_schedule(struct ath_softc *
+ 			if (tid->paused)
+ 				continue;
+ 
+-			ath_tx_sched_aggr(sc, txq, tid);
++			if (ath_tx_sched_aggr(sc, txq, tid, &stop))
++				sent = true;
+ 
+ 			/*
+ 			 * add tid to round-robin queue if more frames
+ 			 * are pending for the tid
+ 			 */
+-			if (!skb_queue_empty(&tid->buf_q))
++			if (ath_tid_has_buffered(tid))
+ 				ath_tx_queue_tid(txq, tid);
+ 
+-			if (tid == last_tid ||
+-			    txq->axq_ampdu_depth >= ATH_AGGR_MIN_QDEPTH)
++			if (stop || tid == last_tid)
+ 				break;
+ 		}
+ 
+@@ -1711,9 +1848,17 @@ void ath_txq_schedule(struct ath_softc *
+ 			list_add_tail(&ac->list, &txq->axq_acq);
+ 		}
+ 
+-		if (ac == last_ac ||
+-		    txq->axq_ampdu_depth >= ATH_AGGR_MIN_QDEPTH)
++		if (stop)
+ 			break;
++
++		if (ac == last_ac) {
++			if (!sent)
++				break;
++
++			sent = false;
++			last_ac = list_entry(txq->axq_acq.prev,
++					     struct ath_atx_ac, list);
++		}
+ 	}
+ 
+ 	rcu_read_unlock();
+@@ -1792,57 +1937,6 @@ static void ath_tx_txqaddbuf(struct ath_
+ 	}
+ }
+ 
+-static void ath_tx_send_ampdu(struct ath_softc *sc, struct ath_txq *txq,
+-			      struct ath_atx_tid *tid, struct sk_buff *skb,
+-			      struct ath_tx_control *txctl)
+-{
+-	struct ath_frame_info *fi = get_frame_info(skb);
+-	struct list_head bf_head;
+-	struct ath_buf *bf;
+-
+-	/*
+-	 * Do not queue to h/w when any of the following conditions is true:
+-	 * - there are pending frames in software queue
+-	 * - the TID is currently paused for ADDBA/BAR request
+-	 * - seqno is not within block-ack window
+-	 * - h/w queue depth exceeds low water mark
+-	 */
+-	if ((!skb_queue_empty(&tid->buf_q) || tid->paused ||
+-	     !BAW_WITHIN(tid->seq_start, tid->baw_size, tid->seq_next) ||
+-	     txq->axq_ampdu_depth >= ATH_AGGR_MIN_QDEPTH) &&
+-	    txq != sc->tx.uapsdq) {
+-		/*
+-		 * Add this frame to software queue for scheduling later
+-		 * for aggregation.
+-		 */
+-		TX_STAT_INC(txq->axq_qnum, a_queued_sw);
+-		__skb_queue_tail(&tid->buf_q, skb);
+-		if (!txctl->an || !txctl->an->sleeping)
+-			ath_tx_queue_tid(txq, tid);
+-		return;
+-	}
+-
+-	bf = ath_tx_setup_buffer(sc, txq, tid, skb);
+-	if (!bf) {
+-		ieee80211_free_txskb(sc->hw, skb);
+-		return;
+-	}
+-
+-	ath_set_rates(tid->an->vif, tid->an->sta, bf);
+-	bf->bf_state.bf_type = BUF_AMPDU;
+-	INIT_LIST_HEAD(&bf_head);
+-	list_add(&bf->list, &bf_head);
+-
+-	/* Add sub-frame to BAW */
+-	ath_tx_addto_baw(sc, tid, bf->bf_state.seqno);
+-
+-	/* Queue to h/w without aggregation */
+-	TX_STAT_INC(txq->axq_qnum, a_queued_hw);
+-	bf->bf_lastbf = bf;
+-	ath_tx_fill_desc(sc, bf, txq, fi->framelen);
+-	ath_tx_txqaddbuf(sc, txq, &bf_head, false);
+-}
+-
+ static void ath_tx_send_normal(struct ath_softc *sc, struct ath_txq *txq,
+ 			       struct ath_atx_tid *tid, struct sk_buff *skb)
+ {
+@@ -1985,6 +2079,7 @@ static int ath_tx_prepare(struct ieee802
+ 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+ 	struct ieee80211_sta *sta = txctl->sta;
+ 	struct ieee80211_vif *vif = info->control.vif;
++	struct ath_vif *avp;
+ 	struct ath_softc *sc = hw->priv;
+ 	int frmlen = skb->len + FCS_LEN;
+ 	int padpos, padsize;
+@@ -1992,6 +2087,10 @@ static int ath_tx_prepare(struct ieee802
+ 	/* NOTE:  sta can be NULL according to net/mac80211.h */
+ 	if (sta)
+ 		txctl->an = (struct ath_node *)sta->drv_priv;
++	else if (vif && ieee80211_is_data(hdr->frame_control)) {
++		avp = (void *)vif->drv_priv;
++		txctl->an = &avp->mcast_node;
++	}
+ 
+ 	if (info->control.hw_key)
+ 		frmlen += info->control.hw_key->icv_len;
+@@ -2041,7 +2140,6 @@ int ath_tx_start(struct ieee80211_hw *hw
+ 	struct ath_txq *txq = txctl->txq;
+ 	struct ath_atx_tid *tid = NULL;
+ 	struct ath_buf *bf;
+-	u8 tidno;
+ 	int q;
+ 	int ret;
+ 
+@@ -2069,27 +2167,31 @@ int ath_tx_start(struct ieee80211_hw *hw
+ 		ath_txq_unlock(sc, txq);
+ 		txq = sc->tx.uapsdq;
+ 		ath_txq_lock(sc, txq);
+-	}
+-
+-	if (txctl->an && ieee80211_is_data_qos(hdr->frame_control)) {
+-		tidno = ieee80211_get_qos_ctl(hdr)[0] &
+-			IEEE80211_QOS_CTL_TID_MASK;
+-		tid = ATH_AN_2_TID(txctl->an, tidno);
++	} else if (txctl->an &&
++		   ieee80211_is_data_present(hdr->frame_control)) {
++		tid = ath_get_skb_tid(sc, txctl->an, skb);
+ 
+ 		WARN_ON(tid->ac->txq != txctl->txq);
+-	}
+ 
+-	if ((info->flags & IEEE80211_TX_CTL_AMPDU) && tid) {
++		if (info->flags & IEEE80211_TX_CTL_CLEAR_PS_FILT)
++			tid->ac->clear_ps_filter = true;
++
+ 		/*
+-		 * Try aggregation if it's a unicast data frame
+-		 * and the destination is HT capable.
++		 * Add this frame to software queue for scheduling later
++		 * for aggregation.
+ 		 */
+-		ath_tx_send_ampdu(sc, txq, tid, skb, txctl);
++		TX_STAT_INC(txq->axq_qnum, a_queued_sw);
++		__skb_queue_tail(&tid->buf_q, skb);
++		if (!txctl->an->sleeping)
++			ath_tx_queue_tid(txq, tid);
++
++		ath_txq_schedule(sc, txq);
+ 		goto out;
  	}
-@@ -2090,6 +2115,7 @@ int ath_tx_start(struct ieee80211_hw *hw
  
  	bf = ath_tx_setup_buffer(sc, txq, tid, skb);
  	if (!bf) {
@@ -339,7 +1095,7 @@
  		if (txctl->paprd)
  			dev_kfree_skb_any(skb);
  		else
-@@ -2189,7 +2215,7 @@ static void ath_tx_complete(struct ath_s
+@@ -2189,7 +2291,7 @@ static void ath_tx_complete(struct ath_s
  	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
  	struct ath_common *common = ath9k_hw_common(sc->sc_ah);
  	struct ieee80211_hdr * hdr = (struct ieee80211_hdr *)skb->data;
@@ -348,7 +1104,7 @@
  	unsigned long flags;
  
  	ath_dbg(common, XMIT, "TX complete: skb: %p\n", skb);
-@@ -2225,21 +2251,7 @@ static void ath_tx_complete(struct ath_s
+@@ -2225,21 +2327,7 @@ static void ath_tx_complete(struct ath_s
  	spin_unlock_irqrestore(&sc->sc_pm_lock, flags);
  
  	__skb_queue_tail(&txq->complete_q, skb);
@@ -371,9 +1127,84 @@
  }
  
  static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
+@@ -2360,8 +2448,7 @@ static void ath_tx_processq(struct ath_s
+ 
+ 		if (list_empty(&txq->axq_q)) {
+ 			txq->axq_link = NULL;
+-			if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_HT)
+-				ath_txq_schedule(sc, txq);
++			ath_txq_schedule(sc, txq);
+ 			break;
+ 		}
+ 		bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
+@@ -2583,6 +2670,7 @@ void ath_tx_node_init(struct ath_softc *
+ 		tid->paused    = false;
+ 		tid->active	   = false;
+ 		__skb_queue_head_init(&tid->buf_q);
++		__skb_queue_head_init(&tid->retry_q);
+ 		acno = TID_TO_WME_AC(tidno);
+ 		tid->ac = &an->ac[acno];
+ 	}
+@@ -2590,6 +2678,7 @@ void ath_tx_node_init(struct ath_softc *
+ 	for (acno = 0, ac = &an->ac[acno];
+ 	     acno < IEEE80211_NUM_ACS; acno++, ac++) {
+ 		ac->sched    = false;
++		ac->clear_ps_filter = true;
+ 		ac->txq = sc->tx.txq_map[acno];
+ 		INIT_LIST_HEAD(&ac->tid_q);
+ 	}
 --- a/drivers/net/wireless/ath/ath9k/main.c
 +++ b/drivers/net/wireless/ath/ath9k/main.c
-@@ -2094,7 +2094,7 @@ static void ath9k_wow_add_pattern(struct
+@@ -966,6 +966,8 @@ static int ath9k_add_interface(struct ie
+ 	struct ath_softc *sc = hw->priv;
+ 	struct ath_hw *ah = sc->sc_ah;
+ 	struct ath_common *common = ath9k_hw_common(ah);
++	struct ath_vif *avp = (void *)vif->drv_priv;
++	struct ath_node *an = &avp->mcast_node;
+ 
+ 	mutex_lock(&sc->mutex);
+ 
+@@ -979,6 +981,12 @@ static int ath9k_add_interface(struct ie
+ 	if (ath9k_uses_beacons(vif->type))
+ 		ath9k_beacon_assign_slot(sc, vif);
+ 
++	an->sc = sc;
++	an->sta = NULL;
++	an->vif = vif;
++	an->no_ps_filter = true;
++	ath_tx_node_init(sc, an);
++
+ 	mutex_unlock(&sc->mutex);
+ 	return 0;
+ }
+@@ -1016,6 +1024,7 @@ static void ath9k_remove_interface(struc
+ {
+ 	struct ath_softc *sc = hw->priv;
+ 	struct ath_common *common = ath9k_hw_common(sc->sc_ah);
++	struct ath_vif *avp = (void *)vif->drv_priv;
+ 
+ 	ath_dbg(common, CONFIG, "Detach Interface\n");
+ 
+@@ -1030,6 +1039,8 @@ static void ath9k_remove_interface(struc
+ 	ath9k_calculate_summary_state(hw, NULL);
+ 	ath9k_ps_restore(sc);
+ 
++	ath_tx_node_cleanup(sc, &avp->mcast_node);
++
+ 	mutex_unlock(&sc->mutex);
+ }
+ 
+@@ -1374,9 +1385,6 @@ static void ath9k_sta_notify(struct ieee
+ 	struct ath_softc *sc = hw->priv;
+ 	struct ath_node *an = (struct ath_node *) sta->drv_priv;
+ 
+-	if (!sta->ht_cap.ht_supported)
+-		return;
+-
+ 	switch (cmd) {
+ 	case STA_NOTIFY_SLEEP:
+ 		an->sleeping = true;
+@@ -2094,7 +2102,7 @@ static void ath9k_wow_add_pattern(struct
  {
  	struct ath_hw *ah = sc->sc_ah;
  	struct ath9k_wow_pattern *wow_pattern = NULL;
@@ -1402,3 +2233,124 @@
  
  #define	REALTEK_USB_VENQT_READ			0xC0
  #define	REALTEK_USB_VENQT_WRITE			0x40
+--- a/drivers/net/wireless/ath/ath9k/ath9k.h
++++ b/drivers/net/wireless/ath/ath9k/ath9k.h
+@@ -137,7 +137,8 @@ int ath_descdma_setup(struct ath_softc *
+ #define ATH_AGGR_ENCRYPTDELIM      10
+ /* minimum h/w qdepth to be sustained to maximize aggregation */
+ #define ATH_AGGR_MIN_QDEPTH        2
+-#define ATH_AMPDU_SUBFRAME_DEFAULT 32
++/* minimum h/w qdepth for non-aggregated traffic */
++#define ATH_NON_AGGR_MIN_QDEPTH    8
+ 
+ #define IEEE80211_SEQ_SEQ_SHIFT    4
+ #define IEEE80211_SEQ_MAX          4096
+@@ -174,12 +175,6 @@ int ath_descdma_setup(struct ath_softc *
+ 
+ #define ATH_TX_COMPLETE_POLL_INT	1000
+ 
+-enum ATH_AGGR_STATUS {
+-	ATH_AGGR_DONE,
+-	ATH_AGGR_BAW_CLOSED,
+-	ATH_AGGR_LIMITED,
+-};
+-
+ #define ATH_TXFIFO_DEPTH 8
+ struct ath_txq {
+ 	int mac80211_qnum; /* mac80211 queue number, -1 means not mac80211 Q */
+@@ -212,8 +207,9 @@ struct ath_frame_info {
+ 	int framelen;
+ 	enum ath9k_key_type keytype;
+ 	u8 keyix;
+-	u8 retries;
+ 	u8 rtscts_rate;
++	u8 retries : 7;
++	u8 baw_tracked : 1;
+ };
+ 
+ struct ath_buf_state {
+@@ -241,6 +237,7 @@ struct ath_buf {
+ struct ath_atx_tid {
+ 	struct list_head list;
+ 	struct sk_buff_head buf_q;
++	struct sk_buff_head retry_q;
+ 	struct ath_node *an;
+ 	struct ath_atx_ac *ac;
+ 	unsigned long tx_buf[BITS_TO_LONGS(ATH_TID_MAX_BUFS)];
+@@ -268,6 +265,7 @@ struct ath_node {
+ 	u8 mpdudensity;
+ 
+ 	bool sleeping;
++	bool no_ps_filter;
+ 
+ #if defined(CPTCFG_MAC80211_DEBUGFS) && defined(CPTCFG_ATH9K_DEBUGFS)
+ 	struct dentry *node_stat;
+@@ -367,6 +365,7 @@ void ath9k_release_buffered_frames(struc
+ /********/
+ 
+ struct ath_vif {
++	struct ath_node mcast_node;
+ 	int av_bslot;
+ 	bool primary_sta_vif;
+ 	__le64 tsf_adjust; /* TSF adjustment for staggered beacons */
+--- a/drivers/net/wireless/ath/ath9k/debug.c
++++ b/drivers/net/wireless/ath/ath9k/debug.c
+@@ -607,6 +607,28 @@ static ssize_t read_file_xmit(struct fil
+ 	return retval;
+ }
+ 
++static ssize_t print_queue(struct ath_softc *sc, struct ath_txq *txq,
++			   char *buf, ssize_t size)
++{
++	ssize_t len = 0;
++
++	ath_txq_lock(sc, txq);
++
++	len += snprintf(buf + len, size - len, "%s: %d ",
++			"qnum", txq->axq_qnum);
++	len += snprintf(buf + len, size - len, "%s: %2d ",
++			"qdepth", txq->axq_depth);
++	len += snprintf(buf + len, size - len, "%s: %2d ",
++			"ampdu-depth", txq->axq_ampdu_depth);
++	len += snprintf(buf + len, size - len, "%s: %3d ",
++			"pending", txq->pending_frames);
++	len += snprintf(buf + len, size - len, "%s: %d\n",
++			"stopped", txq->stopped);
++
++	ath_txq_unlock(sc, txq);
++	return len;
++}
++
+ static ssize_t read_file_queues(struct file *file, char __user *user_buf,
+ 				size_t count, loff_t *ppos)
+ {
+@@ -624,24 +646,13 @@ static ssize_t read_file_queues(struct f
+ 
+ 	for (i = 0; i < IEEE80211_NUM_ACS; i++) {
+ 		txq = sc->tx.txq_map[i];
+-		len += snprintf(buf + len, size - len, "(%s): ", qname[i]);
+-
+-		ath_txq_lock(sc, txq);
+-
+-		len += snprintf(buf + len, size - len, "%s: %d ",
+-				"qnum", txq->axq_qnum);
+-		len += snprintf(buf + len, size - len, "%s: %2d ",
+-				"qdepth", txq->axq_depth);
+-		len += snprintf(buf + len, size - len, "%s: %2d ",
+-				"ampdu-depth", txq->axq_ampdu_depth);
+-		len += snprintf(buf + len, size - len, "%s: %3d ",
+-				"pending", txq->pending_frames);
+-		len += snprintf(buf + len, size - len, "%s: %d\n",
+-				"stopped", txq->stopped);
+-
+-		ath_txq_unlock(sc, txq);
++		len += snprintf(buf + len, size - len, "(%s):  ", qname[i]);
++		len += print_queue(sc, txq, buf + len, size - len);
+ 	}
+ 
++	len += snprintf(buf + len, size - len, "(CAB): ");
++	len += print_queue(sc, sc->beacon.cabq, buf + len, size - len);
++
+ 	if (len > size)
+ 		len = size;
+