path: root/package/mac80211/patches/560-ath9k_tx_queueing_rework.patch

--- a/drivers/net/wireless/ath/ath9k/ath9k.h
+++ b/drivers/net/wireless/ath/ath9k/ath9k.h
@@ -133,7 +133,6 @@ 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
 
 #define IEEE80211_SEQ_SEQ_SHIFT    4
 #define IEEE80211_SEQ_MAX          4096
@@ -208,8 +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 {
@@ -237,6 +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)];
--- 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;
 
--- a/drivers/net/wireless/ath/ath9k/main.c
+++ b/drivers/net/wireless/ath/ath9k/main.c
@@ -1402,9 +1402,6 @@ 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;
--- a/drivers/net/wireless/ath/ath9k/xmit.c
+++ b/drivers/net/wireless/ath/ath9k/xmit.c
@@ -168,6 +168,36 @@ static void ath_txq_skb_done(struct ath_
 	}
 }
 
+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;
+}
+
 static void ath_tx_flush_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
 {
 	struct ath_txq *txq = tid->ac->txq;
@@ -182,7 +212,7 @@ static void ath_tx_flush_tid(struct ath_
 
 	memset(&ts, 0, sizeof(ts));
 
-	while ((skb = __skb_dequeue(&tid->buf_q))) {
+	while ((skb = ath_tid_dequeue(tid))) {
 		fi = get_frame_info(skb);
 		bf = fi->bf;
 
@@ -195,7 +225,7 @@ static void ath_tx_flush_tid(struct ath_
 			}
 		}
 
-		if (fi->retries) {
+		if (fi->baw_tracked) {
 			list_add_tail(&bf->list, &bf_head);
 			ath_tx_update_baw(sc, tid, bf->bf_state.seqno);
 			ath_tx_complete_buf(sc, bf, txq, &bf_head, &ts, 0);
@@ -232,13 +262,16 @@ 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))) {
@@ -266,7 +299,7 @@ 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;
 
@@ -403,7 +436,6 @@ 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;
@@ -440,8 +472,7 @@ 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;
 
@@ -453,7 +484,7 @@ 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);
@@ -489,7 +520,8 @@ 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
@@ -583,7 +615,7 @@ 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);
 
@@ -641,7 +673,7 @@ 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);
 }
 
@@ -815,15 +847,20 @@ 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;
 
@@ -833,12 +870,22 @@ ath_tx_get_tid_subframe(struct ath_softc
 			bf = ath_tx_setup_buffer(sc, txq, tid, skb);
 
 		if (!bf) {
-			__skb_unlink(skb, &tid->buf_q);
+			__skb_unlink(skb, *q);
 			ath_txq_skb_done(sc, txq, skb);
 			ieee80211_free_txskb(sc->hw, skb);
 			continue;
 		}
 
+		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;
 
@@ -852,14 +899,12 @@ 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;
 	}
 
@@ -874,16 +919,17 @@ static enum ATH_AGGR_STATUS ath_tx_form_
 {
 #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;
+	int nframes = 0, ndelim;
 	u16 aggr_limit = 0, al = 0, bpad = 0,
-		al_delta, h_baw = tid->baw_size / 2;
+	    al_delta, h_baw = tid->baw_size / 2;
 	enum ATH_AGGR_STATUS status = ATH_AGGR_DONE;
 	struct ieee80211_tx_info *tx_info;
 	struct ath_frame_info *fi;
 	struct sk_buff *skb;
+	struct sk_buff_head *tid_q;
 
 	do {
-		bf = ath_tx_get_tid_subframe(sc, txq, tid);
+		bf = ath_tx_get_tid_subframe(sc, txq, tid, &tid_q);
 		if (!bf) {
 			status = ATH_AGGR_BAW_CLOSED;
 			break;
@@ -892,33 +938,24 @@ static enum ATH_AGGR_STATUS ath_tx_form_
 		skb = bf->bf_mpdu;
 		fi = get_frame_info(skb);
 
-		if (!bf_first)
+		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) {
+				status = ATH_AGGR_LIMITED;
+				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)
+				break;
 		}
 
 		/* add padding for previous frame to aggregation length */
@@ -936,18 +973,18 @@ 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));
+	} while (ath_tid_has_buffered(tid));
 
 	*aggr_len = al;
 
@@ -1222,7 +1259,7 @@ static void ath_tx_sched_aggr(struct ath
 	int aggr_len;
 
 	do {
-		if (skb_queue_empty(&tid->buf_q))
+		if (!ath_tid_has_buffered(tid))
 			return;
 
 		INIT_LIST_HEAD(&bf_q);
@@ -1301,7 +1338,7 @@ 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_txq_unlock_complete(sc, txq);
 }
@@ -1326,7 +1363,7 @@ 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);
@@ -1358,7 +1395,7 @@ 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);
 		}
@@ -1383,7 +1420,7 @@ 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);
 	}
@@ -1403,6 +1440,7 @@ 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;
 
@@ -1418,15 +1456,15 @@ 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;
@@ -1436,7 +1474,7 @@ void ath9k_release_buffered_frames(struc
 			sent++;
 			TX_STAT_INC(txq->axq_qnum, a_queued_hw);
 
-			if (skb_queue_empty(&tid->buf_q))
+			if (!ath_tid_has_buffered(tid))
 				ieee80211_sta_set_buffered(an->sta, i, false);
 		}
 		ath_txq_unlock_complete(sc, tid->ac->txq);
@@ -1722,7 +1760,7 @@ void ath_txq_schedule(struct ath_softc *
 			 * 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 ||
@@ -1831,7 +1869,7 @@ static void ath_tx_send_ampdu(struct ath
 	 * - seqno is not within block-ack window
 	 * - h/w queue depth exceeds low water mark
 	 */
-	if ((!skb_queue_empty(&tid->buf_q) || tid->paused ||
+	if ((ath_tid_has_buffered(tid) || 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) {
@@ -1859,7 +1897,7 @@ static void ath_tx_send_ampdu(struct ath
 	list_add(&bf->list, &bf_head);
 
 	/* Add sub-frame to BAW */
-	ath_tx_addto_baw(sc, tid, bf->bf_state.seqno);
+	ath_tx_addto_baw(sc, tid, bf);
 
 	/* Queue to h/w without aggregation */
 	TX_STAT_INC(txq->axq_qnum, a_queued_hw);
@@ -2066,7 +2104,6 @@ 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;
 
@@ -2097,9 +2134,7 @@ int ath_tx_start(struct ieee80211_hw *hw
 	}
 
 	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);
+		tid = ath_get_skb_tid(sc, txctl->an, skb);
 
 		WARN_ON(tid->ac->txq != txctl->txq);
 	}
@@ -2372,8 +2407,7 @@ 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);
@@ -2595,6 +2629,7 @@ 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];
 	}
@@ -2602,6 +2637,7 @@ 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);
 	}