ath9k: Reorganize code in xmit.c
authorSujith <Sujith.Manoharan@atheros.com>
Fri, 16 Jan 2009 16:08:42 +0000 (21:38 +0530)
committerJohn W. Linville <linville@tuxdriver.com>
Thu, 29 Jan 2009 21:00:39 +0000 (16:00 -0500)
This patch starts cleaning up all the crufty code in transmission path,
grouping functions into logical blocks.

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
drivers/net/wireless/ath9k/core.h
drivers/net/wireless/ath9k/rc.c
drivers/net/wireless/ath9k/xmit.c

index b687ae9c9e175fc71fac83cc4ea1256a833de88d..4eb21a7f0d80a25958e1a23285a0c309ac97a3ef 100644 (file)
@@ -490,23 +490,17 @@ void ath_tx_draintxq(struct ath_softc *sc,
                     struct ath_txq *txq, bool retry_tx);
 void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an);
 void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an);
-void ath_tx_node_free(struct ath_softc *sc, struct ath_node *an);
 void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq);
 int ath_tx_init(struct ath_softc *sc, int nbufs);
 int ath_tx_cleanup(struct ath_softc *sc);
-int ath_tx_get_qnum(struct ath_softc *sc, int qtype, int haltype);
 struct ath_txq *ath_test_get_txq(struct ath_softc *sc, struct sk_buff *skb);
 int ath_txq_update(struct ath_softc *sc, int qnum,
                   struct ath9k_tx_queue_info *q);
 int ath_tx_start(struct ath_softc *sc, struct sk_buff *skb,
                 struct ath_tx_control *txctl);
 void ath_tx_tasklet(struct ath_softc *sc);
-u32 ath_txq_depth(struct ath_softc *sc, int qnum);
-u32 ath_txq_aggr_depth(struct ath_softc *sc, int qnum);
 void ath_tx_cabq(struct ath_softc *sc, struct sk_buff *skb);
-void ath_tx_resume_tid(struct ath_softc *sc, struct ath_atx_tid *tid);
 bool ath_tx_aggr_check(struct ath_softc *sc, struct ath_node *an, u8 tidno);
-void ath_tx_aggr_teardown(struct ath_softc *sc,        struct ath_node *an, u8 tidno);
 int ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
                      u16 tid, u16 *ssn);
 int ath_tx_aggr_stop(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid);
index 68bda2309ceebbca8f923362ac78d6af2d44a830..46172b5f492c618ff64f50634e239e772d1ffdac 100644 (file)
@@ -23,7 +23,7 @@ static struct ath_rate_table ar5416_11na_ratetable = {
                { VALID, VALID, WLAN_RC_PHY_OFDM, 6000, /* 6 Mb */
                        5400, 0x0b, 0x00, 12,
                        0, 2, 1, 0, 0, 0, 0, 0 },
-               { VALID,        VALID, WLAN_RC_PHY_OFDM, 9000, /* 9 Mb */
+               { VALID, VALID, WLAN_RC_PHY_OFDM, 9000, /* 9 Mb */
                        7800,  0x0f, 0x00, 18,
                        0, 3, 1, 1, 1, 1, 1, 0 },
                { VALID, VALID, WLAN_RC_PHY_OFDM, 12000, /* 12 Mb */
index 522078d809318570b39864e40077cbfc607ba043..5e6e5cf9b67f1fc524296b24a0b7595d8556c64f 100644 (file)
@@ -55,94 +55,19 @@ static u32 bits_per_symbol[][2] = {
 
 #define IS_HT_RATE(_rate)     ((_rate) & 0x80)
 
-/*
- * Insert a chain of ath_buf (descriptors) on a txq and
- * assume the descriptors are already chained together by caller.
- * NB: must be called with txq lock held
- */
-
+static void ath_tx_send_normal(struct ath_softc *sc, struct ath_txq *txq,
+                              struct ath_atx_tid *tid,
+                              struct list_head *bf_head);
+static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
+                               struct list_head *bf_q,
+                               int txok, int sendbar);
 static void ath_tx_txqaddbuf(struct ath_softc *sc, struct ath_txq *txq,
-                            struct list_head *head)
-{
-       struct ath_hal *ah = sc->sc_ah;
-       struct ath_buf *bf;
-
-       /*
-        * Insert the frame on the outbound list and
-        * pass it on to the hardware.
-        */
-
-       if (list_empty(head))
-               return;
-
-       bf = list_first_entry(head, struct ath_buf, list);
-
-       list_splice_tail_init(head, &txq->axq_q);
-       txq->axq_depth++;
-       txq->axq_totalqueued++;
-       txq->axq_linkbuf = list_entry(txq->axq_q.prev, struct ath_buf, list);
-
-       DPRINTF(sc, ATH_DBG_QUEUE,
-               "qnum: %d, txq depth: %d\n", txq->axq_qnum, txq->axq_depth);
-
-       if (txq->axq_link == NULL) {
-               ath9k_hw_puttxbuf(ah, txq->axq_qnum, bf->bf_daddr);
-               DPRINTF(sc, ATH_DBG_XMIT,
-                       "TXDP[%u] = %llx (%p)\n",
-                       txq->axq_qnum, ito64(bf->bf_daddr), bf->bf_desc);
-       } else {
-               *txq->axq_link = bf->bf_daddr;
-               DPRINTF(sc, ATH_DBG_XMIT, "link[%u] (%p)=%llx (%p)\n",
-                       txq->axq_qnum, txq->axq_link,
-                       ito64(bf->bf_daddr), bf->bf_desc);
-       }
-       txq->axq_link = &(bf->bf_lastbf->bf_desc->ds_link);
-       ath9k_hw_txstart(ah, txq->axq_qnum);
-}
-
-static void ath_tx_complete(struct ath_softc *sc, struct sk_buff *skb,
-                           struct ath_xmit_status *tx_status)
-{
-       struct ieee80211_hw *hw = sc->hw;
-       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
-       struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
-       int hdrlen, padsize;
-
-       DPRINTF(sc, ATH_DBG_XMIT, "TX complete: skb: %p\n", skb);
-
-       if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK ||
-           tx_info->flags & IEEE80211_TX_STAT_TX_FILTERED) {
-               kfree(tx_info_priv);
-               tx_info->rate_driver_data[0] = NULL;
-       }
-
-       if (tx_status->flags & ATH_TX_BAR) {
-               tx_info->flags |= IEEE80211_TX_STAT_AMPDU_NO_BACK;
-               tx_status->flags &= ~ATH_TX_BAR;
-       }
+                            struct list_head *head);
+static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf);
 
-       if (!(tx_status->flags & (ATH_TX_ERROR | ATH_TX_XRETRY))) {
-               /* Frame was ACKed */
-               tx_info->flags |= IEEE80211_TX_STAT_ACK;
-       }
-
-       tx_info->status.rates[0].count = tx_status->retries + 1;
-
-       hdrlen = ieee80211_get_hdrlen_from_skb(skb);
-       padsize = hdrlen & 3;
-       if (padsize && hdrlen >= 24) {
-               /*
-                * Remove MAC header padding before giving the frame back to
-                * mac80211.
-                */
-               memmove(skb->data + padsize, skb->data, hdrlen);
-               skb_pull(skb, padsize);
-       }
-
-       ieee80211_tx_status(hw, skb);
-}
-
-/* Check if it's okay to send out aggregates */
+/*********************/
+/* Aggregation logic */
+/*********************/
 
 static int ath_aggr_query(struct ath_softc *sc, struct ath_node *an, u8 tidno)
 {
@@ -156,984 +81,820 @@ static int ath_aggr_query(struct ath_softc *sc, struct ath_node *an, u8 tidno)
                return 0;
 }
 
-static void ath_get_beaconconfig(struct ath_softc *sc, int if_id,
-                                struct ath_beacon_config *conf)
+static void ath_tx_queue_tid(struct ath_txq *txq, struct ath_atx_tid *tid)
 {
-       struct ieee80211_hw *hw = sc->hw;
+       struct ath_atx_ac *ac = tid->ac;
 
-       /* fill in beacon config data */
+       if (tid->paused)
+               return;
 
-       conf->beacon_interval = hw->conf.beacon_int;
-       conf->listen_interval = 100;
-       conf->dtim_count = 1;
-       conf->bmiss_timeout = ATH_DEFAULT_BMISS_LIMIT * conf->listen_interval;
-}
+       if (tid->sched)
+               return;
 
-/* Calculate Atheros packet type from IEEE80211 packet header */
+       tid->sched = true;
+       list_add_tail(&tid->list, &ac->tid_q);
 
-static enum ath9k_pkt_type get_hw_packet_type(struct sk_buff *skb)
-{
-       struct ieee80211_hdr *hdr;
-       enum ath9k_pkt_type htype;
-       __le16 fc;
+       if (ac->sched)
+               return;
 
-       hdr = (struct ieee80211_hdr *)skb->data;
-       fc = hdr->frame_control;
+       ac->sched = true;
+       list_add_tail(&ac->list, &txq->axq_acq);
+}
 
-       if (ieee80211_is_beacon(fc))
-               htype = ATH9K_PKT_TYPE_BEACON;
-       else if (ieee80211_is_probe_resp(fc))
-               htype = ATH9K_PKT_TYPE_PROBE_RESP;
-       else if (ieee80211_is_atim(fc))
-               htype = ATH9K_PKT_TYPE_ATIM;
-       else if (ieee80211_is_pspoll(fc))
-               htype = ATH9K_PKT_TYPE_PSPOLL;
-       else
-               htype = ATH9K_PKT_TYPE_NORMAL;
+static void ath_tx_pause_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
+{
+       struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
 
-       return htype;
+       spin_lock_bh(&txq->axq_lock);
+       tid->paused++;
+       spin_unlock_bh(&txq->axq_lock);
 }
 
-static bool is_pae(struct sk_buff *skb)
+static void ath_tx_resume_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
 {
-       struct ieee80211_hdr *hdr;
-       __le16 fc;
-
-       hdr = (struct ieee80211_hdr *)skb->data;
-       fc = hdr->frame_control;
+       struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
 
-       if (ieee80211_is_data(fc)) {
-               if (ieee80211_is_nullfunc(fc) ||
-                   /* Port Access Entity (IEEE 802.1X) */
-                   (skb->protocol == cpu_to_be16(ETH_P_PAE))) {
-                       return true;
-               }
-       }
+       ASSERT(tid->paused > 0);
+       spin_lock_bh(&txq->axq_lock);
 
-       return false;
-}
+       tid->paused--;
 
-static int get_hw_crypto_keytype(struct sk_buff *skb)
-{
-       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+       if (tid->paused > 0)
+               goto unlock;
 
-       if (tx_info->control.hw_key) {
-               if (tx_info->control.hw_key->alg == ALG_WEP)
-                       return ATH9K_KEY_TYPE_WEP;
-               else if (tx_info->control.hw_key->alg == ALG_TKIP)
-                       return ATH9K_KEY_TYPE_TKIP;
-               else if (tx_info->control.hw_key->alg == ALG_CCMP)
-                       return ATH9K_KEY_TYPE_AES;
-       }
+       if (list_empty(&tid->buf_q))
+               goto unlock;
 
-       return ATH9K_KEY_TYPE_CLEAR;
+       ath_tx_queue_tid(txq, tid);
+       ath_txq_schedule(sc, txq);
+unlock:
+       spin_unlock_bh(&txq->axq_lock);
 }
 
-/* Called only when tx aggregation is enabled and HT is supported */
-
-static void assign_aggr_tid_seqno(struct sk_buff *skb,
-                                 struct ath_buf *bf)
+static void ath_tx_flush_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
 {
-       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
-       struct ieee80211_hdr *hdr;
-       struct ath_node *an;
-       struct ath_atx_tid *tid;
-       __le16 fc;
-       u8 *qc;
+       struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
+       struct ath_buf *bf;
+       struct list_head bf_head;
+       INIT_LIST_HEAD(&bf_head);
 
-       if (!tx_info->control.sta)
-               return;
+       ASSERT(tid->paused > 0);
+       spin_lock_bh(&txq->axq_lock);
 
-       an = (struct ath_node *)tx_info->control.sta->drv_priv;
-       hdr = (struct ieee80211_hdr *)skb->data;
-       fc = hdr->frame_control;
+       tid->paused--;
 
-       /* Get tidno */
+       if (tid->paused > 0) {
+               spin_unlock_bh(&txq->axq_lock);
+               return;
+       }
 
-       if (ieee80211_is_data_qos(fc)) {
-               qc = ieee80211_get_qos_ctl(hdr);
-               bf->bf_tidno = qc[0] & 0xf;
+       while (!list_empty(&tid->buf_q)) {
+               bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
+               ASSERT(!bf_isretried(bf));
+               list_cut_position(&bf_head, &tid->buf_q, &bf->bf_lastfrm->list);
+               ath_tx_send_normal(sc, txq, tid, &bf_head);
        }
 
-       /* Get seqno */
-       /* For HT capable stations, we save tidno for later use.
-        * We also override seqno set by upper layer with the one
-        * in tx aggregation state.
-        *
-        * If fragmentation is on, the sequence number is
-        * not overridden, since it has been
-        * incremented by the fragmentation routine.
-        *
-        * FIXME: check if the fragmentation threshold exceeds
-        * IEEE80211 max.
-        */
-       tid = ATH_AN_2_TID(an, bf->bf_tidno);
-       hdr->seq_ctrl = cpu_to_le16(tid->seq_next <<
-                       IEEE80211_SEQ_SEQ_SHIFT);
-       bf->bf_seqno = tid->seq_next;
-       INCR(tid->seq_next, IEEE80211_SEQ_MAX);
+       spin_unlock_bh(&txq->axq_lock);
 }
 
-static int setup_tx_flags(struct ath_softc *sc, struct sk_buff *skb,
-                         struct ath_txq *txq)
+static void ath_tx_update_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
+                             int seqno)
 {
-       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
-       int flags = 0;
+       int index, cindex;
 
-       flags |= ATH9K_TXDESC_CLRDMASK; /* needed for crypto errors */
-       flags |= ATH9K_TXDESC_INTREQ;
+       index  = ATH_BA_INDEX(tid->seq_start, seqno);
+       cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
 
-       if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK)
-               flags |= ATH9K_TXDESC_NOACK;
-       if (tx_info->control.rates[0].flags & IEEE80211_TX_RC_USE_RTS_CTS)
-               flags |= ATH9K_TXDESC_RTSENA;
+       tid->tx_buf[cindex] = NULL;
 
-       return flags;
+       while (tid->baw_head != tid->baw_tail && !tid->tx_buf[tid->baw_head]) {
+               INCR(tid->seq_start, IEEE80211_SEQ_MAX);
+               INCR(tid->baw_head, ATH_TID_MAX_BUFS);
+       }
 }
 
-static struct ath_buf *ath_tx_get_buffer(struct ath_softc *sc)
+static void ath_tx_addto_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
+                            struct ath_buf *bf)
 {
-       struct ath_buf *bf = NULL;
+       int index, cindex;
 
-       spin_lock_bh(&sc->tx.txbuflock);
+       if (bf_isretried(bf))
+               return;
 
-       if (unlikely(list_empty(&sc->tx.txbuf))) {
-               spin_unlock_bh(&sc->tx.txbuflock);
-               return NULL;
-       }
+       index  = ATH_BA_INDEX(tid->seq_start, bf->bf_seqno);
+       cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
 
-       bf = list_first_entry(&sc->tx.txbuf, struct ath_buf, list);
-       list_del(&bf->list);
-
-       spin_unlock_bh(&sc->tx.txbuflock);
-
-       return bf;
-}
-
-/* To complete a chain of buffers associated a frame */
-
-static void ath_tx_complete_buf(struct ath_softc *sc,
-                               struct ath_buf *bf,
-                               struct list_head *bf_q,
-                               int txok, int sendbar)
-{
-       struct sk_buff *skb = bf->bf_mpdu;
-       struct ath_xmit_status tx_status;
-       unsigned long flags;
-
-       /*
-        * Set retry information.
-        * NB: Don't use the information in the descriptor, because the frame
-        * could be software retried.
-        */
-       tx_status.retries = bf->bf_retries;
-       tx_status.flags = 0;
-
-       if (sendbar)
-               tx_status.flags = ATH_TX_BAR;
-
-       if (!txok) {
-               tx_status.flags |= ATH_TX_ERROR;
+       ASSERT(tid->tx_buf[cindex] == NULL);
+       tid->tx_buf[cindex] = bf;
 
-               if (bf_isxretried(bf))
-                       tx_status.flags |= ATH_TX_XRETRY;
+       if (index >= ((tid->baw_tail - tid->baw_head) &
+               (ATH_TID_MAX_BUFS - 1))) {
+               tid->baw_tail = cindex;
+               INCR(tid->baw_tail, ATH_TID_MAX_BUFS);
        }
-
-       /* Unmap this frame */
-       dma_unmap_single(sc->dev, bf->bf_dmacontext, skb->len, DMA_TO_DEVICE);
-
-       /* complete this frame */
-       ath_tx_complete(sc, skb, &tx_status);
-
-       /*
-        * Return the list of ath_buf of this mpdu to free queue
-        */
-       spin_lock_irqsave(&sc->tx.txbuflock, flags);
-       list_splice_tail_init(bf_q, &sc->tx.txbuf);
-       spin_unlock_irqrestore(&sc->tx.txbuflock, flags);
 }
 
 /*
- * queue up a dest/ac pair for tx scheduling
- * NB: must be called with txq lock held
+ * TODO: For frame(s) that are in the retry state, we will reuse the
+ * sequence number(s) without setting the retry bit. The
+ * alternative is to give up on these and BAR the receiver's window
+ * forward.
  */
+static void ath_tid_drain(struct ath_softc *sc, struct ath_txq *txq,
+                         struct ath_atx_tid *tid)
 
-static void ath_tx_queue_tid(struct ath_txq *txq, struct ath_atx_tid *tid)
 {
-       struct ath_atx_ac *ac = tid->ac;
-
-       /*
-        * if tid is paused, hold off
-        */
-       if (tid->paused)
-               return;
-
-       /*
-        * add tid to ac atmost once
-        */
-       if (tid->sched)
-               return;
-
-       tid->sched = true;
-       list_add_tail(&tid->list, &ac->tid_q);
-
-       /*
-        * add node ac to txq atmost once
-        */
-       if (ac->sched)
-               return;
-
-       ac->sched = true;
-       list_add_tail(&ac->list, &txq->axq_acq);
-}
+       struct ath_buf *bf;
+       struct list_head bf_head;
+       INIT_LIST_HEAD(&bf_head);
 
-/* pause a tid */
+       for (;;) {
+               if (list_empty(&tid->buf_q))
+                       break;
+               bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
 
-static void ath_tx_pause_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
-{
-       struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
+               list_cut_position(&bf_head, &tid->buf_q, &bf->bf_lastfrm->list);
 
-       spin_lock_bh(&txq->axq_lock);
+               if (bf_isretried(bf))
+                       ath_tx_update_baw(sc, tid, bf->bf_seqno);
 
-       tid->paused++;
+               spin_unlock(&txq->axq_lock);
+               ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
+               spin_lock(&txq->axq_lock);
+       }
 
-       spin_unlock_bh(&txq->axq_lock);
+       tid->seq_next = tid->seq_start;
+       tid->baw_tail = tid->baw_head;
 }
 
-/* resume a tid and schedule aggregate */
-
-void ath_tx_resume_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
+static void ath_tx_set_retry(struct ath_softc *sc, struct ath_buf *bf)
 {
-       struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
-
-       ASSERT(tid->paused > 0);
-       spin_lock_bh(&txq->axq_lock);
-
-       tid->paused--;
-
-       if (tid->paused > 0)
-               goto unlock;
+       struct sk_buff *skb;
+       struct ieee80211_hdr *hdr;
 
-       if (list_empty(&tid->buf_q))
-               goto unlock;
+       bf->bf_state.bf_type |= BUF_RETRY;
+       bf->bf_retries++;
 
-       /*
-        * Add this TID to scheduler and try to send out aggregates
-        */
-       ath_tx_queue_tid(txq, tid);
-       ath_txq_schedule(sc, txq);
-unlock:
-       spin_unlock_bh(&txq->axq_lock);
+       skb = bf->bf_mpdu;
+       hdr = (struct ieee80211_hdr *)skb->data;
+       hdr->frame_control |= cpu_to_le16(IEEE80211_FCTL_RETRY);
 }
 
-/* Compute the number of bad frames */
-
-static int ath_tx_num_badfrms(struct ath_softc *sc, struct ath_buf *bf,
-                             int txok)
+static void ath_tx_complete_aggr_rifs(struct ath_softc *sc, struct ath_txq *txq,
+                                     struct ath_buf *bf, struct list_head *bf_q,
+                                     int txok)
 {
+       struct ath_node *an = NULL;
+       struct sk_buff *skb;
+       struct ieee80211_tx_info *tx_info;
+       struct ath_atx_tid *tid = NULL;
        struct ath_buf *bf_last = bf->bf_lastbf;
        struct ath_desc *ds = bf_last->bf_desc;
+       struct ath_buf *bf_next, *bf_lastq = NULL;
+       struct list_head bf_head, bf_pending;
        u16 seq_st = 0;
        u32 ba[WME_BA_BMP_SIZE >> 5];
-       int ba_index;
-       int nbad = 0;
-       int isaggr = 0;
+       int isaggr, txfail, txpending, sendbar = 0, needreset = 0;
 
-       if (ds->ds_txstat.ts_flags == ATH9K_TX_SW_ABORTED)
-               return 0;
+       skb = (struct sk_buff *)bf->bf_mpdu;
+       tx_info = IEEE80211_SKB_CB(skb);
 
-       isaggr = bf_isaggr(bf);
-       if (isaggr) {
-               seq_st = ATH_DS_BA_SEQ(ds);
-               memcpy(ba, ATH_DS_BA_BITMAP(ds), WME_BA_BMP_SIZE >> 3);
+       if (tx_info->control.sta) {
+               an = (struct ath_node *)tx_info->control.sta->drv_priv;
+               tid = ATH_AN_2_TID(an, bf->bf_tidno);
        }
 
-       while (bf) {
-               ba_index = ATH_BA_INDEX(seq_st, bf->bf_seqno);
-               if (!txok || (isaggr && !ATH_BA_ISSET(ba, ba_index)))
-                       nbad++;
+       isaggr = bf_isaggr(bf);
+       if (isaggr) {
+               if (txok) {
+                       if (ATH_DS_TX_BA(ds)) {
+                               seq_st = ATH_DS_BA_SEQ(ds);
+                               memcpy(ba, ATH_DS_BA_BITMAP(ds),
+                                      WME_BA_BMP_SIZE >> 3);
+                       } else {
+                               memset(ba, 0, WME_BA_BMP_SIZE >> 3);
 
-               bf = bf->bf_next;
+                               /*
+                                * AR5416 can become deaf/mute when BA
+                                * issue happens. Chip needs to be reset.
+                                * But AP code may have sychronization issues
+                                * when perform internal reset in this routine.
+                                * Only enable reset in STA mode for now.
+                                */
+                               if (sc->sc_ah->ah_opmode ==
+                                           NL80211_IFTYPE_STATION)
+                                       needreset = 1;
+                       }
+               } else {
+                       memset(ba, 0, WME_BA_BMP_SIZE >> 3);
+               }
        }
 
-       return nbad;
-}
+       INIT_LIST_HEAD(&bf_pending);
+       INIT_LIST_HEAD(&bf_head);
 
-static void ath_tx_set_retry(struct ath_softc *sc, struct ath_buf *bf)
-{
-       struct sk_buff *skb;
-       struct ieee80211_hdr *hdr;
+       while (bf) {
+               txfail = txpending = 0;
+               bf_next = bf->bf_next;
 
-       bf->bf_state.bf_type |= BUF_RETRY;
-       bf->bf_retries++;
+               if (ATH_BA_ISSET(ba, ATH_BA_INDEX(seq_st, bf->bf_seqno))) {
+                       /* transmit completion, subframe is
+                        * acked by block ack */
+               } else if (!isaggr && txok) {
+                       /* transmit completion */
+               } else {
 
-       skb = bf->bf_mpdu;
-       hdr = (struct ieee80211_hdr *)skb->data;
-       hdr->frame_control |= cpu_to_le16(IEEE80211_FCTL_RETRY);
-}
+                       if (!(tid->state & AGGR_CLEANUP) &&
+                           ds->ds_txstat.ts_flags != ATH9K_TX_SW_ABORTED) {
+                               if (bf->bf_retries < ATH_MAX_SW_RETRIES) {
+                                       ath_tx_set_retry(sc, bf);
+                                       txpending = 1;
+                               } else {
+                                       bf->bf_state.bf_type |= BUF_XRETRY;
+                                       txfail = 1;
+                                       sendbar = 1;
+                               }
+                       } else {
+                               /*
+                                * cleanup in progress, just fail
+                                * the un-acked sub-frames
+                                */
+                               txfail = 1;
+                       }
+               }
 
-/* Update block ack window */
+               if (bf_next == NULL) {
+                       ASSERT(bf->bf_lastfrm == bf_last);
+                       if (!list_empty(bf_q)) {
+                               bf_lastq = list_entry(bf_q->prev,
+                                       struct ath_buf, list);
+                               list_cut_position(&bf_head,
+                                       bf_q, &bf_lastq->list);
+                       } else {
+                               INIT_LIST_HEAD(&bf_head);
+                       }
+               } else {
+                       ASSERT(!list_empty(bf_q));
+                       list_cut_position(&bf_head,
+                               bf_q, &bf->bf_lastfrm->list);
+               }
 
-static void ath_tx_update_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
-                             int seqno)
-{
-       int index, cindex;
+               if (!txpending) {
+                       /*
+                        * complete the acked-ones/xretried ones; update
+                        * block-ack window
+                        */
+                       spin_lock_bh(&txq->axq_lock);
+                       ath_tx_update_baw(sc, tid, bf->bf_seqno);
+                       spin_unlock_bh(&txq->axq_lock);
 
-       index  = ATH_BA_INDEX(tid->seq_start, seqno);
-       cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
+                       ath_tx_complete_buf(sc, bf, &bf_head, !txfail, sendbar);
+               } else {
+                       /*
+                        * retry the un-acked ones
+                        */
+                       if (bf->bf_next == NULL &&
+                           bf_last->bf_status & ATH_BUFSTATUS_STALE) {
+                               struct ath_buf *tbf;
 
-       tid->tx_buf[cindex] = NULL;
+                               /* allocate new descriptor */
+                               spin_lock_bh(&sc->tx.txbuflock);
+                               ASSERT(!list_empty((&sc->tx.txbuf)));
+                               tbf = list_first_entry(&sc->tx.txbuf,
+                                               struct ath_buf, list);
+                               list_del(&tbf->list);
+                               spin_unlock_bh(&sc->tx.txbuflock);
 
-       while (tid->baw_head != tid->baw_tail && !tid->tx_buf[tid->baw_head]) {
-               INCR(tid->seq_start, IEEE80211_SEQ_MAX);
-               INCR(tid->baw_head, ATH_TID_MAX_BUFS);
+                               ATH_TXBUF_RESET(tbf);
+
+                               /* copy descriptor content */
+                               tbf->bf_mpdu = bf_last->bf_mpdu;
+                               tbf->bf_buf_addr = bf_last->bf_buf_addr;
+                               *(tbf->bf_desc) = *(bf_last->bf_desc);
+
+                               /* link it to the frame */
+                               if (bf_lastq) {
+                                       bf_lastq->bf_desc->ds_link =
+                                               tbf->bf_daddr;
+                                       bf->bf_lastfrm = tbf;
+                                       ath9k_hw_cleartxdesc(sc->sc_ah,
+                                               bf->bf_lastfrm->bf_desc);
+                               } else {
+                                       tbf->bf_state = bf_last->bf_state;
+                                       tbf->bf_lastfrm = tbf;
+                                       ath9k_hw_cleartxdesc(sc->sc_ah,
+                                               tbf->bf_lastfrm->bf_desc);
+
+                                       /* copy the DMA context */
+                                       tbf->bf_dmacontext =
+                                               bf_last->bf_dmacontext;
+                               }
+                               list_add_tail(&tbf->list, &bf_head);
+                       } else {
+                               /*
+                                * Clear descriptor status words for
+                                * software retry
+                                */
+                               ath9k_hw_cleartxdesc(sc->sc_ah,
+                                                    bf->bf_lastfrm->bf_desc);
+                       }
+
+                       /*
+                        * Put this buffer to the temporary pending
+                        * queue to retain ordering
+                        */
+                       list_splice_tail_init(&bf_head, &bf_pending);
+               }
+
+               bf = bf_next;
        }
-}
 
-/*
- * ath_pkt_dur - compute packet duration (NB: not NAV)
- *
- * rix - rate index
- * pktlen - total bytes (delims + data + fcs + pads + pad delims)
- * width  - 0 for 20 MHz, 1 for 40 MHz
- * half_gi - to use 4us v/s 3.6 us for symbol time
- */
-static u32 ath_pkt_duration(struct ath_softc *sc, u8 rix, struct ath_buf *bf,
-                           int width, int half_gi, bool shortPreamble)
-{
-       struct ath_rate_table *rate_table = sc->cur_rate_table;
-       u32 nbits, nsymbits, duration, nsymbols;
-       u8 rc;
-       int streams, pktlen;
+       if (tid->state & AGGR_CLEANUP) {
+               /* check to see if we're done with cleaning the h/w queue */
+               spin_lock_bh(&txq->axq_lock);
 
-       pktlen = bf_isaggr(bf) ? bf->bf_al : bf->bf_frmlen;
-       rc = rate_table->info[rix].ratecode;
+               if (tid->baw_head == tid->baw_tail) {
+                       tid->state &= ~AGGR_ADDBA_COMPLETE;
+                       tid->addba_exchangeattempts = 0;
+                       spin_unlock_bh(&txq->axq_lock);
 
-       /* for legacy rates, use old function to compute packet duration */
-       if (!IS_HT_RATE(rc))
-               return ath9k_hw_computetxtime(sc->sc_ah, rate_table, pktlen,
-                                             rix, shortPreamble);
+                       tid->state &= ~AGGR_CLEANUP;
 
-       /* find number of symbols: PLCP + data */
-       nbits = (pktlen << 3) + OFDM_PLCP_BITS;
-       nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
-       nsymbols = (nbits + nsymbits - 1) / nsymbits;
+                       /* send buffered frames as singles */
+                       ath_tx_flush_tid(sc, tid);
+               } else
+                       spin_unlock_bh(&txq->axq_lock);
 
-       if (!half_gi)
-               duration = SYMBOL_TIME(nsymbols);
-       else
-               duration = SYMBOL_TIME_HALFGI(nsymbols);
+               return;
+       }
 
-       /* addup duration for legacy/ht training and signal fields */
-       streams = HT_RC_2_STREAMS(rc);
-       duration += L_STF + L_LTF + L_SIG + HT_SIG + HT_STF + HT_LTF(streams);
+       /*
+        * prepend un-acked frames to the beginning of the pending frame queue
+        */
+       if (!list_empty(&bf_pending)) {
+               spin_lock_bh(&txq->axq_lock);
+               list_splice(&bf_pending, &tid->buf_q);
+               ath_tx_queue_tid(txq, tid);
+               spin_unlock_bh(&txq->axq_lock);
+       }
 
-       return duration;
-}
+       if (needreset)
+               ath_reset(sc, false);
 
-/* Rate module function to set rate related fields in tx descriptor */
+       return;
+}
 
-static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf)
+static u32 ath_lookup_rate(struct ath_softc *sc, struct ath_buf *bf,
+                          struct ath_atx_tid *tid)
 {
-       struct ath_hal *ah = sc->sc_ah;
-       struct ath_rate_table *rt;
-       struct ath_desc *ds = bf->bf_desc;
-       struct ath_desc *lastds = bf->bf_lastbf->bf_desc;
-       struct ath9k_11n_rate_series series[4];
+       struct ath_rate_table *rate_table = sc->cur_rate_table;
        struct sk_buff *skb;
        struct ieee80211_tx_info *tx_info;
        struct ieee80211_tx_rate *rates;
-       struct ieee80211_hdr *hdr;
-       struct ieee80211_hw *hw = sc->hw;
-       int i, flags, rtsctsena = 0, enable_g_protection = 0;
-       u32 ctsduration = 0;
-       u8 rix = 0, cix, ctsrate = 0;
-       __le16 fc;
-
-       memset(series, 0, sizeof(struct ath9k_11n_rate_series) * 4);
+       struct ath_tx_info_priv *tx_info_priv;
+       u32 max_4ms_framelen, frame_length;
+       u16 aggr_limit, legacy = 0, maxampdu;
+       int i;
 
        skb = (struct sk_buff *)bf->bf_mpdu;
-       hdr = (struct ieee80211_hdr *)skb->data;
-       fc = hdr->frame_control;
        tx_info = IEEE80211_SKB_CB(skb);
        rates = tx_info->control.rates;
+       tx_info_priv =
+               (struct ath_tx_info_priv *)tx_info->rate_driver_data[0];
 
-       if (ieee80211_has_morefrags(fc) ||
-           (le16_to_cpu(hdr->seq_ctrl) & IEEE80211_SCTL_FRAG)) {
-               rates[1].count = rates[2].count = rates[3].count = 0;
-               rates[1].idx = rates[2].idx = rates[3].idx = 0;
-               rates[0].count = ATH_TXMAXTRY;
-       }
+       /*
+        * Find the lowest frame length among the rate series that will have a
+        * 4ms transmit duration.
+        * TODO - TXOP limit needs to be considered.
+        */
+       max_4ms_framelen = ATH_AMPDU_LIMIT_MAX;
 
-       /* get the cix for the lowest valid rix */
-       rt = sc->cur_rate_table;
-       for (i = 3; i >= 0; i--) {
-               if (rates[i].count && (rates[i].idx >= 0)) {
-                       rix = rates[i].idx;
-                       break;
+       for (i = 0; i < 4; i++) {
+               if (rates[i].count) {
+                       if (!WLAN_RC_PHY_HT(rate_table->info[rates[i].idx].phy)) {
+                               legacy = 1;
+                               break;
+                       }
+
+                       frame_length =
+                               rate_table->info[rates[i].idx].max_4ms_framelen;
+                       max_4ms_framelen = min(max_4ms_framelen, frame_length);
                }
        }
 
-       flags = (bf->bf_flags & (ATH9K_TXDESC_RTSENA | ATH9K_TXDESC_CTSENA));
-       cix = rt->info[rix].ctrl_rate;
-
-       /* All protection frames are transmited at 2Mb/s for 802.11g,
-        * otherwise we transmit them at 1Mb/s */
-       if (hw->conf.channel->band == IEEE80211_BAND_2GHZ &&
-         !conf_is_ht(&hw->conf))
-               enable_g_protection = 1;
-
        /*
-        * If 802.11g protection is enabled, determine whether to use RTS/CTS or
-        * just CTS.  Note that this is only done for OFDM/HT unicast frames.
+        * limit aggregate size by the minimum rate if rate selected is
+        * not a probe rate, if rate selected is a probe rate then
+        * avoid aggregation of this packet.
         */
-       if (sc->sc_protmode != PROT_M_NONE && !(bf->bf_flags & ATH9K_TXDESC_NOACK)
-           && (rt->info[rix].phy == WLAN_RC_PHY_OFDM ||
-               WLAN_RC_PHY_HT(rt->info[rix].phy))) {
-               if (sc->sc_protmode == PROT_M_RTSCTS)
-                       flags = ATH9K_TXDESC_RTSENA;
-               else if (sc->sc_protmode == PROT_M_CTSONLY)
-                       flags = ATH9K_TXDESC_CTSENA;
+       if (tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE || legacy)
+               return 0;
 
-               cix = rt->info[enable_g_protection].ctrl_rate;
-               rtsctsena = 1;
-       }
+       aggr_limit = min(max_4ms_framelen,
+               (u32)ATH_AMPDU_LIMIT_DEFAULT);
 
-       /* For 11n, the default behavior is to enable RTS for hw retried frames.
-        * We enable the global flag here and let rate series flags determine
-        * which rates will actually use RTS.
+       /*
+        * h/w can accept aggregates upto 16 bit lengths (65535).
+        * The IE, however can hold upto 65536, which shows up here
+        * as zero. Ignore 65536 since we  are constrained by hw.
         */
-       if ((ah->ah_caps.hw_caps & ATH9K_HW_CAP_HT) && bf_isdata(bf)) {
-               /* 802.11g protection not needed, use our default behavior */
-               if (!rtsctsena)
-                       flags = ATH9K_TXDESC_RTSENA;
-       }
+       maxampdu = tid->an->maxampdu;
+       if (maxampdu)
+               aggr_limit = min(aggr_limit, maxampdu);
 
-       /* Set protection if aggregate protection on */
-       if (sc->sc_config.ath_aggr_prot &&
-           (!bf_isaggr(bf) || (bf_isaggr(bf) && bf->bf_al < 8192))) {
-               flags = ATH9K_TXDESC_RTSENA;
-               cix = rt->info[enable_g_protection].ctrl_rate;
-               rtsctsena = 1;
-       }
+       return aggr_limit;
+}
 
-       /* For AR5416 - RTS cannot be followed by a frame larger than 8K */
-       if (bf_isaggr(bf) && (bf->bf_al > ah->ah_caps.rts_aggr_limit))
-               flags &= ~(ATH9K_TXDESC_RTSENA);
+/*
+ * returns the number of delimiters to be added to
+ * meet the minimum required mpdudensity.
+ * caller should make sure that the rate is  HT rate .
+ */
+static int ath_compute_num_delims(struct ath_softc *sc, struct ath_atx_tid *tid,
+                                 struct ath_buf *bf, u16 frmlen)
+{
+       struct ath_rate_table *rt = sc->cur_rate_table;
+       struct sk_buff *skb = bf->bf_mpdu;
+       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+       u32 nsymbits, nsymbols, mpdudensity;
+       u16 minlen;
+       u8 rc, flags, rix;
+       int width, half_gi, ndelim, mindelim;
+
+       /* Select standard number of delimiters based on frame length alone */
+       ndelim = ATH_AGGR_GET_NDELIM(frmlen);
 
        /*
-        * CTS transmit rate is derived from the transmit rate by looking in the
-        * h/w rate table.  We must also factor in whether or not a short
-        * preamble is to be used. NB: cix is set above where RTS/CTS is enabled
+        * If encryption enabled, hardware requires some more padding between
+        * subframes.
+        * TODO - this could be improved to be dependent on the rate.
+        *      The hardware can keep up at lower rates, but not higher rates
         */
-       ctsrate = rt->info[cix].ratecode |
-               (bf_isshpreamble(bf) ? rt->info[cix].short_preamble : 0);
+       if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR)
+               ndelim += ATH_AGGR_ENCRYPTDELIM;
 
-       for (i = 0; i < 4; i++) {
-               if (!rates[i].count || (rates[i].idx < 0))
-                       continue;
+       /*
+        * Convert desired mpdu density from microeconds to bytes based
+        * on highest rate in rate series (i.e. first rate) to determine
+        * required minimum length for subframe. Take into account
+        * whether high rate is 20 or 40Mhz and half or full GI.
+        */
+       mpdudensity = tid->an->mpdudensity;
 
-               rix = rates[i].idx;
+       /*
+        * If there is no mpdu density restriction, no further calculation
+        * is needed.
+        */
+       if (mpdudensity == 0)
+               return ndelim;
 
-               series[i].Rate = rt->info[rix].ratecode |
-                       (bf_isshpreamble(bf) ? rt->info[rix].short_preamble : 0);
-
-               series[i].Tries = rates[i].count;
+       rix = tx_info->control.rates[0].idx;
+       flags = tx_info->control.rates[0].flags;
+       rc = rt->info[rix].ratecode;
+       width = (flags & IEEE80211_TX_RC_40_MHZ_WIDTH) ? 1 : 0;
+       half_gi = (flags & IEEE80211_TX_RC_SHORT_GI) ? 1 : 0;
 
-               series[i].RateFlags = (
-                       (rates[i].flags & IEEE80211_TX_RC_USE_RTS_CTS) ?
-                               ATH9K_RATESERIES_RTS_CTS : 0) |
-                       ((rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH) ?
-                               ATH9K_RATESERIES_2040 : 0) |
-                       ((rates[i].flags & IEEE80211_TX_RC_SHORT_GI) ?
-                               ATH9K_RATESERIES_HALFGI : 0);
+       if (half_gi)
+               nsymbols = NUM_SYMBOLS_PER_USEC_HALFGI(mpdudensity);
+       else
+               nsymbols = NUM_SYMBOLS_PER_USEC(mpdudensity);
 
-               series[i].PktDuration = ath_pkt_duration(sc, rix, bf,
-                        (rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH) != 0,
-                        (rates[i].flags & IEEE80211_TX_RC_SHORT_GI),
-                        bf_isshpreamble(bf));
+       if (nsymbols == 0)
+               nsymbols = 1;
 
-               series[i].ChSel = sc->sc_tx_chainmask;
+       nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
+       minlen = (nsymbols * nsymbits) / BITS_PER_BYTE;
 
-               if (rtsctsena)
-                       series[i].RateFlags |= ATH9K_RATESERIES_RTS_CTS;
+       /* Is frame shorter than required minimum length? */
+       if (frmlen < minlen) {
+               /* Get the minimum number of delimiters required. */
+               mindelim = (minlen - frmlen) / ATH_AGGR_DELIM_SZ;
+               ndelim = max(mindelim, ndelim);
        }
 
-       /* set dur_update_en for l-sig computation except for PS-Poll frames */
-       ath9k_hw_set11n_ratescenario(ah, ds, lastds, !bf_ispspoll(bf),
-                                    ctsrate, ctsduration,
-                                    series, 4, flags);
-
-       if (sc->sc_config.ath_aggr_prot && flags)
-               ath9k_hw_set11n_burstduration(ah, ds, 8192);
+       return ndelim;
 }
 
-/*
- * Function to send a normal HT (non-AMPDU) frame
- * NB: must be called with txq lock held
- */
-static int ath_tx_send_normal(struct ath_softc *sc,
-                             struct ath_txq *txq,
-                             struct ath_atx_tid *tid,
-                             struct list_head *bf_head)
+static enum ATH_AGGR_STATUS ath_tx_form_aggr(struct ath_softc *sc,
+                    struct ath_atx_tid *tid, struct list_head *bf_q,
+                    struct ath_buf **bf_last, struct aggr_rifs_param *param,
+                    int *prev_frames)
 {
-       struct ath_buf *bf;
+#define PADBYTES(_len) ((4 - ((_len) % 4)) % 4)
+       struct ath_buf *bf, *tbf, *bf_first, *bf_prev = NULL;
+       struct list_head bf_head;
+       int rl = 0, 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;
+       int prev_al = 0;
+       INIT_LIST_HEAD(&bf_head);
 
-       BUG_ON(list_empty(bf_head));
+       BUG_ON(list_empty(&tid->buf_q));
 
-       bf = list_first_entry(bf_head, struct ath_buf, list);
-       bf->bf_state.bf_type &= ~BUF_AMPDU; /* regular HT frame */
+       bf_first = list_first_entry(&tid->buf_q, struct ath_buf, list);
 
-       /* update starting sequence number for subsequent ADDBA request */
-       INCR(tid->seq_start, IEEE80211_SEQ_MAX);
+       do {
+               bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
 
-       /* Queue to h/w without aggregation */
-       bf->bf_nframes = 1;
-       bf->bf_lastbf = bf->bf_lastfrm; /* one single frame */
-       ath_buf_set_rate(sc, bf);
-       ath_tx_txqaddbuf(sc, txq, bf_head);
+               /*
+                * do not step over block-ack window
+                */
+               if (!BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno)) {
+                       status = ATH_AGGR_BAW_CLOSED;
+                       break;
+               }
 
-       return 0;
-}
+               if (!rl) {
+                       aggr_limit = ath_lookup_rate(sc, bf, tid);
+                       rl = 1;
+               }
 
-/* flush tid's software queue and send frames as non-ampdu's */
+               /*
+                * do not exceed aggregation limit
+                */
+               al_delta = ATH_AGGR_DELIM_SZ + bf->bf_frmlen;
 
-static void ath_tx_flush_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
-{
-       struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
-       struct ath_buf *bf;
-       struct list_head bf_head;
-       INIT_LIST_HEAD(&bf_head);
+               if (nframes && (aggr_limit <
+                       (al + bpad + al_delta + prev_al))) {
+                       status = ATH_AGGR_LIMITED;
+                       break;
+               }
 
-       ASSERT(tid->paused > 0);
-       spin_lock_bh(&txq->axq_lock);
+               /*
+                * do not exceed subframe limit
+                */
+               if ((nframes + *prev_frames) >=
+                   min((int)h_baw, ATH_AMPDU_SUBFRAME_DEFAULT)) {
+                       status = ATH_AGGR_LIMITED;
+                       break;
+               }
 
-       tid->paused--;
+               /*
+                * add padding for previous frame to aggregation length
+                */
+               al += bpad + al_delta;
 
-       if (tid->paused > 0) {
-               spin_unlock_bh(&txq->axq_lock);
-               return;
-       }
+               /*
+                * Get the delimiters needed to meet the MPDU
+                * density for this node.
+                */
+               ndelim = ath_compute_num_delims(sc, tid, bf_first, bf->bf_frmlen);
 
-       while (!list_empty(&tid->buf_q)) {
-               bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
-               ASSERT(!bf_isretried(bf));
-               list_cut_position(&bf_head, &tid->buf_q, &bf->bf_lastfrm->list);
-               ath_tx_send_normal(sc, txq, tid, &bf_head);
-       }
+               bpad = PADBYTES(al_delta) + (ndelim << 2);
 
-       spin_unlock_bh(&txq->axq_lock);
-}
+               bf->bf_next = NULL;
+               bf->bf_lastfrm->bf_desc->ds_link = 0;
 
-/* Completion routine of an aggregate */
+               /*
+                * this packet is part of an aggregate
+                * - remove all descriptors belonging to this frame from
+                *   software queue
+                * - add it to block ack window
+                * - set up descriptors for aggregation
+                */
+               list_cut_position(&bf_head, &tid->buf_q, &bf->bf_lastfrm->list);
+               ath_tx_addto_baw(sc, tid, bf);
 
-static void ath_tx_complete_aggr_rifs(struct ath_softc *sc,
-                                     struct ath_txq *txq,
-                                     struct ath_buf *bf,
-                                     struct list_head *bf_q,
-                                     int txok)
-{
-       struct ath_node *an = NULL;
-       struct sk_buff *skb;
-       struct ieee80211_tx_info *tx_info;
-       struct ath_atx_tid *tid = NULL;
-       struct ath_buf *bf_last = bf->bf_lastbf;
-       struct ath_desc *ds = bf_last->bf_desc;
-       struct ath_buf *bf_next, *bf_lastq = NULL;
-       struct list_head bf_head, bf_pending;
-       u16 seq_st = 0;
-       u32 ba[WME_BA_BMP_SIZE >> 5];
-       int isaggr, txfail, txpending, sendbar = 0, needreset = 0;
+               list_for_each_entry(tbf, &bf_head, list) {
+                       ath9k_hw_set11n_aggr_middle(sc->sc_ah,
+                               tbf->bf_desc, ndelim);
+               }
 
-       skb = (struct sk_buff *)bf->bf_mpdu;
-       tx_info = IEEE80211_SKB_CB(skb);
+               /*
+                * link buffers of this frame to the aggregate
+                */
+               list_splice_tail_init(&bf_head, bf_q);
+               nframes++;
 
-       if (tx_info->control.sta) {
-               an = (struct ath_node *)tx_info->control.sta->drv_priv;
-               tid = ATH_AN_2_TID(an, bf->bf_tidno);
-       }
+               if (bf_prev) {
+                       bf_prev->bf_next = bf;
+                       bf_prev->bf_lastfrm->bf_desc->ds_link = bf->bf_daddr;
+               }
+               bf_prev = bf;
 
-       isaggr = bf_isaggr(bf);
-       if (isaggr) {
-               if (txok) {
-                       if (ATH_DS_TX_BA(ds)) {
-                               /*
-                                * extract starting sequence and
-                                * block-ack bitmap
-                                */
-                               seq_st = ATH_DS_BA_SEQ(ds);
-                               memcpy(ba,
-                                       ATH_DS_BA_BITMAP(ds),
-                                       WME_BA_BMP_SIZE >> 3);
-                       } else {
-                               memset(ba, 0, WME_BA_BMP_SIZE >> 3);
+       } while (!list_empty(&tid->buf_q));
 
-                               /*
-                                * AR5416 can become deaf/mute when BA
-                                * issue happens. Chip needs to be reset.
-                                * But AP code may have sychronization issues
-                                * when perform internal reset in this routine.
-                                * Only enable reset in STA mode for now.
-                                */
-                               if (sc->sc_ah->ah_opmode ==
-                                           NL80211_IFTYPE_STATION)
-                                       needreset = 1;
-                       }
-               } else {
-                       memset(ba, 0, WME_BA_BMP_SIZE >> 3);
-               }
-       }
+       bf_first->bf_al = al;
+       bf_first->bf_nframes = nframes;
+       *bf_last = bf_prev;
+       return status;
+#undef PADBYTES
+}
 
-       INIT_LIST_HEAD(&bf_pending);
-       INIT_LIST_HEAD(&bf_head);
+static void ath_tx_sched_aggr(struct ath_softc *sc, struct ath_txq *txq,
+                             struct ath_atx_tid *tid)
+{
+       struct ath_buf *bf, *tbf, *bf_last, *bf_lastaggr = NULL;
+       enum ATH_AGGR_STATUS status;
+       struct list_head bf_q;
+       struct aggr_rifs_param param = {0, 0, 0, 0, NULL};
+       int prev_frames = 0;
 
-       while (bf) {
-               txfail = txpending = 0;
-               bf_next = bf->bf_next;
+       do {
+               if (list_empty(&tid->buf_q))
+                       return;
 
-               if (ATH_BA_ISSET(ba, ATH_BA_INDEX(seq_st, bf->bf_seqno))) {
-                       /* transmit completion, subframe is
-                        * acked by block ack */
-               } else if (!isaggr && txok) {
-                       /* transmit completion */
-               } else {
+               INIT_LIST_HEAD(&bf_q);
+
+               status = ath_tx_form_aggr(sc, tid, &bf_q, &bf_lastaggr, &param,
+                                         &prev_frames);
 
-                       if (!(tid->state & AGGR_CLEANUP) &&
-                           ds->ds_txstat.ts_flags != ATH9K_TX_SW_ABORTED) {
-                               if (bf->bf_retries < ATH_MAX_SW_RETRIES) {
-                                       ath_tx_set_retry(sc, bf);
-                                       txpending = 1;
-                               } else {
-                                       bf->bf_state.bf_type |= BUF_XRETRY;
-                                       txfail = 1;
-                                       sendbar = 1;
-                               }
-                       } else {
-                               /*
-                                * cleanup in progress, just fail
-                                * the un-acked sub-frames
-                                */
-                               txfail = 1;
-                       }
-               }
                /*
-                * Remove ath_buf's of this sub-frame from aggregate queue.
+                * no frames picked up to be aggregated; block-ack
+                * window is not open
                 */
-               if (bf_next == NULL) {  /* last subframe in the aggregate */
-                       ASSERT(bf->bf_lastfrm == bf_last);
-
-                       /*
-                        * The last descriptor of the last sub frame could be
-                        * a holding descriptor for h/w. If that's the case,
-                        * bf->bf_lastfrm won't be in the bf_q.
-                        * Make sure we handle bf_q properly here.
-                        */
+               if (list_empty(&bf_q))
+                       break;
 
-                       if (!list_empty(bf_q)) {
-                               bf_lastq = list_entry(bf_q->prev,
-                                       struct ath_buf, list);
-                               list_cut_position(&bf_head,
-                                       bf_q, &bf_lastq->list);
-                       } else {
-                               /*
-                                * XXX: if the last subframe only has one
-                                * descriptor which is also being used as
-                                * a holding descriptor. Then the ath_buf
-                                * is not in the bf_q at all.
-                                */
-                               INIT_LIST_HEAD(&bf_head);
-                       }
-               } else {
-                       ASSERT(!list_empty(bf_q));
-                       list_cut_position(&bf_head,
-                               bf_q, &bf->bf_lastfrm->list);
-               }
+               bf = list_first_entry(&bf_q, struct ath_buf, list);
+               bf_last = list_entry(bf_q.prev, struct ath_buf, list);
+               bf->bf_lastbf = bf_last;
 
-               if (!txpending) {
-                       /*
-                        * complete the acked-ones/xretried ones; update
-                        * block-ack window
-                        */
-                       spin_lock_bh(&txq->axq_lock);
-                       ath_tx_update_baw(sc, tid, bf->bf_seqno);
-                       spin_unlock_bh(&txq->axq_lock);
+               /*
+                * if only one frame, send as non-aggregate
+                */
+               if (bf->bf_nframes == 1) {
+                       ASSERT(bf->bf_lastfrm == bf_last);
 
-                       /* complete this sub-frame */
-                       ath_tx_complete_buf(sc, bf, &bf_head, !txfail, sendbar);
-               } else {
-                       /*
-                        * retry the un-acked ones
-                        */
+                       bf->bf_state.bf_type &= ~BUF_AGGR;
                        /*
-                        * XXX: if the last descriptor is holding descriptor,
-                        * in order to requeue the frame to software queue, we
-                        * need to allocate a new descriptor and
-                        * copy the content of holding descriptor to it.
+                        * clear aggr bits for every descriptor
+                        * XXX TODO: is there a way to optimize it?
                         */
-                       if (bf->bf_next == NULL &&
-                           bf_last->bf_status & ATH_BUFSTATUS_STALE) {
-                               struct ath_buf *tbf;
+                       list_for_each_entry(tbf, &bf_q, list) {
+                               ath9k_hw_clr11n_aggr(sc->sc_ah, tbf->bf_desc);
+                       }
 
-                               /* allocate new descriptor */
-                               spin_lock_bh(&sc->tx.txbuflock);
-                               ASSERT(!list_empty((&sc->tx.txbuf)));
-                               tbf = list_first_entry(&sc->tx.txbuf,
-                                               struct ath_buf, list);
-                               list_del(&tbf->list);
-                               spin_unlock_bh(&sc->tx.txbuflock);
+                       ath_buf_set_rate(sc, bf);
+                       ath_tx_txqaddbuf(sc, txq, &bf_q);
+                       continue;
+               }
 
-                               ATH_TXBUF_RESET(tbf);
+               /*
+                * setup first desc with rate and aggr info
+                */
+               bf->bf_state.bf_type |= BUF_AGGR;
+               ath_buf_set_rate(sc, bf);
+               ath9k_hw_set11n_aggr_first(sc->sc_ah, bf->bf_desc, bf->bf_al);
 
-                               /* copy descriptor content */
-                               tbf->bf_mpdu = bf_last->bf_mpdu;
-                               tbf->bf_buf_addr = bf_last->bf_buf_addr;
-                               *(tbf->bf_desc) = *(bf_last->bf_desc);
+               /*
+                * anchor last frame of aggregate correctly
+                */
+               ASSERT(bf_lastaggr);
+               ASSERT(bf_lastaggr->bf_lastfrm == bf_last);
+               tbf = bf_lastaggr;
+               ath9k_hw_set11n_aggr_last(sc->sc_ah, tbf->bf_desc);
 
-                               /* link it to the frame */
-                               if (bf_lastq) {
-                                       bf_lastq->bf_desc->ds_link =
-                                               tbf->bf_daddr;
-                                       bf->bf_lastfrm = tbf;
-                                       ath9k_hw_cleartxdesc(sc->sc_ah,
-                                               bf->bf_lastfrm->bf_desc);
-                               } else {
-                                       tbf->bf_state = bf_last->bf_state;
-                                       tbf->bf_lastfrm = tbf;
-                                       ath9k_hw_cleartxdesc(sc->sc_ah,
-                                               tbf->bf_lastfrm->bf_desc);
+               /* XXX: We don't enter into this loop, consider removing this */
+               while (!list_empty(&bf_q) && !list_is_last(&tbf->list, &bf_q)) {
+                       tbf = list_entry(tbf->list.next, struct ath_buf, list);
+                       ath9k_hw_set11n_aggr_last(sc->sc_ah, tbf->bf_desc);
+               }
 
-                                       /* copy the DMA context */
-                                       tbf->bf_dmacontext =
-                                               bf_last->bf_dmacontext;
-                               }
-                               list_add_tail(&tbf->list, &bf_head);
-                       } else {
-                               /*
-                                * Clear descriptor status words for
-                                * software retry
-                                */
-                               ath9k_hw_cleartxdesc(sc->sc_ah,
-                                                    bf->bf_lastfrm->bf_desc);
-                       }
+               txq->axq_aggr_depth++;
 
-                       /*
-                        * Put this buffer to the temporary pending
-                        * queue to retain ordering
-                        */
-                       list_splice_tail_init(&bf_head, &bf_pending);
-               }
+               /*
+                * Normal aggregate, queue to hardware
+                */
+               ath_tx_txqaddbuf(sc, txq, &bf_q);
 
-               bf = bf_next;
+       } while (txq->axq_depth < ATH_AGGR_MIN_QDEPTH &&
+                status != ATH_AGGR_BAW_CLOSED);
+}
+
+int ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
+                     u16 tid, u16 *ssn)
+{
+       struct ath_atx_tid *txtid;
+       struct ath_node *an;
+
+       an = (struct ath_node *)sta->drv_priv;
+
+       if (sc->sc_flags & SC_OP_TXAGGR) {
+               txtid = ATH_AN_2_TID(an, tid);
+               txtid->state |= AGGR_ADDBA_PROGRESS;
+               ath_tx_pause_tid(sc, txtid);
        }
 
-       if (tid->state & AGGR_CLEANUP) {
-               /* check to see if we're done with cleaning the h/w queue */
-               spin_lock_bh(&txq->axq_lock);
+       return 0;
+}
 
-               if (tid->baw_head == tid->baw_tail) {
-                       tid->state &= ~AGGR_ADDBA_COMPLETE;
-                       tid->addba_exchangeattempts = 0;
-                       spin_unlock_bh(&txq->axq_lock);
+int ath_tx_aggr_stop(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
+{
+       struct ath_node *an = (struct ath_node *)sta->drv_priv;
+       struct ath_atx_tid *txtid = ATH_AN_2_TID(an, tid);
+       struct ath_txq *txq = &sc->tx.txq[txtid->ac->qnum];
+       struct ath_buf *bf;
+       struct list_head bf_head;
+       INIT_LIST_HEAD(&bf_head);
 
-                       tid->state &= ~AGGR_CLEANUP;
+       if (txtid->state & AGGR_CLEANUP)
+               return 0;
 
-                       /* send buffered frames as singles */
-                       ath_tx_flush_tid(sc, tid);
-               } else
-                       spin_unlock_bh(&txq->axq_lock);
+       if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
+               txtid->addba_exchangeattempts = 0;
+               return 0;
+       }
 
-               return;
+       ath_tx_pause_tid(sc, txtid);
+
+       /* drop all software retried frames and mark this TID */
+       spin_lock_bh(&txq->axq_lock);
+       while (!list_empty(&txtid->buf_q)) {
+               bf = list_first_entry(&txtid->buf_q, struct ath_buf, list);
+               if (!bf_isretried(bf)) {
+                       /*
+                        * NB: it's based on the assumption that
+                        * software retried frame will always stay
+                        * at the head of software queue.
+                        */
+                       break;
+               }
+               list_cut_position(&bf_head,
+                       &txtid->buf_q, &bf->bf_lastfrm->list);
+               ath_tx_update_baw(sc, txtid, bf->bf_seqno);
+               ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
        }
 
-       /*
-        * prepend un-acked frames to the beginning of the pending frame queue
-        */
-       if (!list_empty(&bf_pending)) {
-               spin_lock_bh(&txq->axq_lock);
-               /* Note: we _prepend_, we _do_not_ at to
-                * the end of the queue ! */
-               list_splice(&bf_pending, &tid->buf_q);
-               ath_tx_queue_tid(txq, tid);
+       if (txtid->baw_head != txtid->baw_tail) {
+               spin_unlock_bh(&txq->axq_lock);
+               txtid->state |= AGGR_CLEANUP;
+       } else {
+               txtid->state &= ~AGGR_ADDBA_COMPLETE;
+               txtid->addba_exchangeattempts = 0;
                spin_unlock_bh(&txq->axq_lock);
+               ath_tx_flush_tid(sc, txtid);
        }
 
-       if (needreset)
-               ath_reset(sc, false);
+       return 0;
+}
 
-       return;
+void ath_tx_aggr_resume(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
+{
+       struct ath_atx_tid *txtid;
+       struct ath_node *an;
+
+       an = (struct ath_node *)sta->drv_priv;
+
+       if (sc->sc_flags & SC_OP_TXAGGR) {
+               txtid = ATH_AN_2_TID(an, tid);
+               txtid->baw_size =
+                       IEEE80211_MIN_AMPDU_BUF << sta->ht_cap.ampdu_factor;
+               txtid->state |= AGGR_ADDBA_COMPLETE;
+               txtid->state &= ~AGGR_ADDBA_PROGRESS;
+               ath_tx_resume_tid(sc, txtid);
+       }
 }
 
-static void ath_tx_rc_status(struct ath_buf *bf, struct ath_desc *ds, int nbad)
+bool ath_tx_aggr_check(struct ath_softc *sc, struct ath_node *an, u8 tidno)
 {
-       struct sk_buff *skb = (struct sk_buff *)bf->bf_mpdu;
-       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
-       struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+       struct ath_atx_tid *txtid;
 
-       tx_info_priv->update_rc = false;
-       if (ds->ds_txstat.ts_status & ATH9K_TXERR_FILT)
-               tx_info->flags |= IEEE80211_TX_STAT_TX_FILTERED;
+       if (!(sc->sc_flags & SC_OP_TXAGGR))
+               return false;
 
-       if ((ds->ds_txstat.ts_status & ATH9K_TXERR_FILT) == 0 &&
-           (bf->bf_flags & ATH9K_TXDESC_NOACK) == 0) {
-               if (bf_isdata(bf)) {
-                       memcpy(&tx_info_priv->tx, &ds->ds_txstat,
-                              sizeof(tx_info_priv->tx));
-                       tx_info_priv->n_frames = bf->bf_nframes;
-                       tx_info_priv->n_bad_frames = nbad;
-                       tx_info_priv->update_rc = true;
+       txtid = ATH_AN_2_TID(an, tidno);
+
+       if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
+               if (!(txtid->state & AGGR_ADDBA_PROGRESS) &&
+                   (txtid->addba_exchangeattempts < ADDBA_EXCHANGE_ATTEMPTS)) {
+                       txtid->addba_exchangeattempts++;
+                       return true;
                }
        }
+
+       return false;
 }
 
-/* Process completed xmit descriptors from the specified queue */
+/********************/
+/* Queue Management */
+/********************/
 
-static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
+static u32 ath_txq_depth(struct ath_softc *sc, int qnum)
+{
+       return sc->tx.txq[qnum].axq_depth;
+}
+
+static void ath_tx_stopdma(struct ath_softc *sc, struct ath_txq *txq)
 {
        struct ath_hal *ah = sc->sc_ah;
-       struct ath_buf *bf, *lastbf, *bf_held = NULL;
-       struct list_head bf_head;
-       struct ath_desc *ds;
-       int txok, nbad = 0;
-       int status;
+       (void) ath9k_hw_stoptxdma(ah, txq->axq_qnum);
+}
 
-       DPRINTF(sc, ATH_DBG_QUEUE, "tx queue %d (%x), link %p\n",
-               txq->axq_qnum, ath9k_hw_gettxbuf(sc->sc_ah, txq->axq_qnum),
-               txq->axq_link);
-
-       for (;;) {
-               spin_lock_bh(&txq->axq_lock);
-               if (list_empty(&txq->axq_q)) {
-                       txq->axq_link = NULL;
-                       txq->axq_linkbuf = NULL;
-                       spin_unlock_bh(&txq->axq_lock);
-                       break;
-               }
-               bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
-
-               /*
-                * There is a race condition that a BH gets scheduled
-                * after sw writes TxE and before hw re-load the last
-                * descriptor to get the newly chained one.
-                * Software must keep the last DONE descriptor as a
-                * holding descriptor - software does so by marking
-                * it with the STALE flag.
-                */
-               bf_held = NULL;
-               if (bf->bf_status & ATH_BUFSTATUS_STALE) {
-                       bf_held = bf;
-                       if (list_is_last(&bf_held->list, &txq->axq_q)) {
-                               /* FIXME:
-                                * The holding descriptor is the last
-                                * descriptor in queue. It's safe to remove
-                                * the last holding descriptor in BH context.
-                                */
-                               spin_unlock_bh(&txq->axq_lock);
-                               break;
-                       } else {
-                               /* Lets work with the next buffer now */
-                               bf = list_entry(bf_held->list.next,
-                                       struct ath_buf, list);
-                       }
-               }
-
-               lastbf = bf->bf_lastbf;
-               ds = lastbf->bf_desc;    /* NB: last decriptor */
-
-               status = ath9k_hw_txprocdesc(ah, ds);
-               if (status == -EINPROGRESS) {
-                       spin_unlock_bh(&txq->axq_lock);
-                       break;
-               }
-               if (bf->bf_desc == txq->axq_lastdsWithCTS)
-                       txq->axq_lastdsWithCTS = NULL;
-               if (ds == txq->axq_gatingds)
-                       txq->axq_gatingds = NULL;
-
-               /*
-                * Remove ath_buf's of the same transmit unit from txq,
-                * however leave the last descriptor back as the holding
-                * descriptor for hw.
-                */
-               lastbf->bf_status |= ATH_BUFSTATUS_STALE;
-               INIT_LIST_HEAD(&bf_head);
-
-               if (!list_is_singular(&lastbf->list))
-                       list_cut_position(&bf_head,
-                               &txq->axq_q, lastbf->list.prev);
-
-               txq->axq_depth--;
-
-               if (bf_isaggr(bf))
-                       txq->axq_aggr_depth--;
-
-               txok = (ds->ds_txstat.ts_status == 0);
-
-               spin_unlock_bh(&txq->axq_lock);
-
-               if (bf_held) {
-                       list_del(&bf_held->list);
-                       spin_lock_bh(&sc->tx.txbuflock);
-                       list_add_tail(&bf_held->list, &sc->tx.txbuf);
-                       spin_unlock_bh(&sc->tx.txbuflock);
-               }
-
-               if (!bf_isampdu(bf)) {
-                       /*
-                        * This frame is sent out as a single frame.
-                        * Use hardware retry status for this frame.
-                        */
-                       bf->bf_retries = ds->ds_txstat.ts_longretry;
-                       if (ds->ds_txstat.ts_status & ATH9K_TXERR_XRETRY)
-                               bf->bf_state.bf_type |= BUF_XRETRY;
-                       nbad = 0;
-               } else {
-                       nbad = ath_tx_num_badfrms(sc, bf, txok);
-               }
-
-               ath_tx_rc_status(bf, ds, nbad);
-
-               /*
-                * Complete this transmit unit
-                */
-               if (bf_isampdu(bf))
-                       ath_tx_complete_aggr_rifs(sc, txq, bf, &bf_head, txok);
-               else
-                       ath_tx_complete_buf(sc, bf, &bf_head, txok, 0);
-
-               /* Wake up mac80211 queue */
-
-               spin_lock_bh(&txq->axq_lock);
-               if (txq->stopped && ath_txq_depth(sc, txq->axq_qnum) <=
-                               (ATH_TXBUF - 20)) {
-                       int qnum;
-                       qnum = ath_get_mac80211_qnum(txq->axq_qnum, sc);
-                       if (qnum != -1) {
-                               ieee80211_wake_queue(sc->hw, qnum);
-                               txq->stopped = 0;
-                       }
-
-               }
-
-               /*
-                * schedule any pending packets if aggregation is enabled
-                */
-               if (sc->sc_flags & SC_OP_TXAGGR)
-                       ath_txq_schedule(sc, txq);
-               spin_unlock_bh(&txq->axq_lock);
-       }
-}
-
-static void ath_tx_stopdma(struct ath_softc *sc, struct ath_txq *txq)
+static void ath_get_beaconconfig(struct ath_softc *sc, int if_id,
+                                struct ath_beacon_config *conf)
 {
-       struct ath_hal *ah = sc->sc_ah;
+       struct ieee80211_hw *hw = sc->hw;
 
-       (void) ath9k_hw_stoptxdma(ah, txq->axq_qnum);
-       DPRINTF(sc, ATH_DBG_XMIT, "tx queue [%u] %x, link %p\n",
-               txq->axq_qnum, ath9k_hw_gettxbuf(ah, txq->axq_qnum),
-               txq->axq_link);
-}
+       /* fill in beacon config data */
 
-/* Drain only the data queues */
+       conf->beacon_interval = hw->conf.beacon_int;
+       conf->listen_interval = 100;
+       conf->dtim_count = 1;
+       conf->bmiss_timeout = ATH_DEFAULT_BMISS_LIMIT * conf->listen_interval;
+}
 
 static void ath_drain_txdataq(struct ath_softc *sc, bool retry_tx)
 {
@@ -1144,8 +905,6 @@ static void ath_drain_txdataq(struct ath_softc *sc, bool retry_tx)
                for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
                        if (ATH_TXQ_SETUP(sc, i)) {
                                ath_tx_stopdma(sc, &sc->tx.txq[i]);
-                               /* The TxDMA may not really be stopped.
-                                * Double check the hal tx pending count */
                                npend += ath9k_hw_numtxpending(ah,
                                                       sc->tx.txq[i].axq_qnum);
                        }
@@ -1154,7 +913,7 @@ static void ath_drain_txdataq(struct ath_softc *sc, bool retry_tx)
 
        if (npend) {
                int r;
-               /* TxDMA not stopped, reset the hal */
+
                DPRINTF(sc, ATH_DBG_XMIT, "Unable to stop TxDMA. Reset HAL!\n");
 
                spin_lock_bh(&sc->sc_resetlock);
@@ -1172,485 +931,751 @@ static void ath_drain_txdataq(struct ath_softc *sc, bool retry_tx)
        }
 }
 
-/* Add a sub-frame to block ack window */
-
-static void ath_tx_addto_baw(struct ath_softc *sc,
-                            struct ath_atx_tid *tid,
-                            struct ath_buf *bf)
+static void ath_txq_drain_pending_buffers(struct ath_softc *sc,
+                                         struct ath_txq *txq)
 {
-       int index, cindex;
-
-       if (bf_isretried(bf))
-               return;
-
-       index  = ATH_BA_INDEX(tid->seq_start, bf->bf_seqno);
-       cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
-
-       ASSERT(tid->tx_buf[cindex] == NULL);
-       tid->tx_buf[cindex] = bf;
+       struct ath_atx_ac *ac, *ac_tmp;
+       struct ath_atx_tid *tid, *tid_tmp;
 
-       if (index >= ((tid->baw_tail - tid->baw_head) &
-               (ATH_TID_MAX_BUFS - 1))) {
-               tid->baw_tail = cindex;
-               INCR(tid->baw_tail, ATH_TID_MAX_BUFS);
+       list_for_each_entry_safe(ac, ac_tmp, &txq->axq_acq, list) {
+               list_del(&ac->list);
+               ac->sched = false;
+               list_for_each_entry_safe(tid, tid_tmp, &ac->tid_q, list) {
+                       list_del(&tid->list);
+                       tid->sched = false;
+                       ath_tid_drain(sc, txq, tid);
+               }
        }
 }
 
-/*
- * Function to send an A-MPDU
- * NB: must be called with txq lock held
- */
-static int ath_tx_send_ampdu(struct ath_softc *sc,
-                            struct ath_atx_tid *tid,
-                            struct list_head *bf_head,
-                            struct ath_tx_control *txctl)
+struct ath_txq *ath_txq_setup(struct ath_softc *sc, int qtype, int subtype)
 {
-       struct ath_buf *bf;
-
-       BUG_ON(list_empty(bf_head));
+       struct ath_hal *ah = sc->sc_ah;
+       struct ath9k_tx_queue_info qi;
+       int qnum;
 
-       bf = list_first_entry(bf_head, struct ath_buf, list);
-       bf->bf_state.bf_type |= BUF_AMPDU;
+       memset(&qi, 0, sizeof(qi));
+       qi.tqi_subtype = subtype;
+       qi.tqi_aifs = ATH9K_TXQ_USEDEFAULT;
+       qi.tqi_cwmin = ATH9K_TXQ_USEDEFAULT;
+       qi.tqi_cwmax = ATH9K_TXQ_USEDEFAULT;
+       qi.tqi_physCompBuf = 0;
 
        /*
-        * 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
+        * Enable interrupts only for EOL and DESC conditions.
+        * We mark tx descriptors to receive a DESC interrupt
+        * when a tx queue gets deep; otherwise waiting for the
+        * EOL to reap descriptors.  Note that this is done to
+        * reduce interrupt load and this only defers reaping
+        * descriptors, never transmitting frames.  Aside from
+        * reducing interrupts this also permits more concurrency.
+        * The only potential downside is if the tx queue backs
+        * up in which case the top half of the kernel may backup
+        * due to a lack of tx descriptors.
+        *
+        * The UAPSD queue is an exception, since we take a desc-
+        * based intr on the EOSP frames.
         */
-       if (!list_empty(&tid->buf_q) || tid->paused ||
-           !BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno) ||
-           txctl->txq->axq_depth >= ATH_AGGR_MIN_QDEPTH) {
+       if (qtype == ATH9K_TX_QUEUE_UAPSD)
+               qi.tqi_qflags = TXQ_FLAG_TXDESCINT_ENABLE;
+       else
+               qi.tqi_qflags = TXQ_FLAG_TXEOLINT_ENABLE |
+                       TXQ_FLAG_TXDESCINT_ENABLE;
+       qnum = ath9k_hw_setuptxqueue(ah, qtype, &qi);
+       if (qnum == -1) {
                /*
-                * Add this frame to software queue for scheduling later
-                * for aggregation.
+                * NB: don't print a message, this happens
+                * normally on parts with too few tx queues
                 */
-               list_splice_tail_init(bf_head, &tid->buf_q);
-               ath_tx_queue_tid(txctl->txq, tid);
-               return 0;
+               return NULL;
        }
+       if (qnum >= ARRAY_SIZE(sc->tx.txq)) {
+               DPRINTF(sc, ATH_DBG_FATAL,
+                       "qnum %u out of range, max %u!\n",
+                       qnum, (unsigned int)ARRAY_SIZE(sc->tx.txq));
+               ath9k_hw_releasetxqueue(ah, qnum);
+               return NULL;
+       }
+       if (!ATH_TXQ_SETUP(sc, qnum)) {
+               struct ath_txq *txq = &sc->tx.txq[qnum];
 
-       /* Add sub-frame to BAW */
-       ath_tx_addto_baw(sc, tid, bf);
-
-       /* Queue to h/w without aggregation */
-       bf->bf_nframes = 1;
-       bf->bf_lastbf = bf->bf_lastfrm; /* one single frame */
-       ath_buf_set_rate(sc, bf);
-       ath_tx_txqaddbuf(sc, txctl->txq, bf_head);
-
-       return 0;
+               txq->axq_qnum = qnum;
+               txq->axq_link = NULL;
+               INIT_LIST_HEAD(&txq->axq_q);
+               INIT_LIST_HEAD(&txq->axq_acq);
+               spin_lock_init(&txq->axq_lock);
+               txq->axq_depth = 0;
+               txq->axq_aggr_depth = 0;
+               txq->axq_totalqueued = 0;
+               txq->axq_linkbuf = NULL;
+               sc->tx.txqsetup |= 1<<qnum;
+       }
+       return &sc->tx.txq[qnum];
 }
 
-/*
- * looks up the rate
- * returns aggr limit based on lowest of the rates
- */
-static u32 ath_lookup_rate(struct ath_softc *sc,
-                          struct ath_buf *bf,
-                          struct ath_atx_tid *tid)
+static int ath_tx_get_qnum(struct ath_softc *sc, int qtype, int haltype)
 {
-       struct ath_rate_table *rate_table = sc->cur_rate_table;
-       struct sk_buff *skb;
-       struct ieee80211_tx_info *tx_info;
-       struct ieee80211_tx_rate *rates;
-       struct ath_tx_info_priv *tx_info_priv;
-       u32 max_4ms_framelen, frame_length;
-       u16 aggr_limit, legacy = 0, maxampdu;
-       int i;
+       int qnum;
 
-       skb = (struct sk_buff *)bf->bf_mpdu;
-       tx_info = IEEE80211_SKB_CB(skb);
-       rates = tx_info->control.rates;
-       tx_info_priv =
-               (struct ath_tx_info_priv *)tx_info->rate_driver_data[0];
+       switch (qtype) {
+       case ATH9K_TX_QUEUE_DATA:
+               if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
+                       DPRINTF(sc, ATH_DBG_FATAL,
+                               "HAL AC %u out of range, max %zu!\n",
+                               haltype, ARRAY_SIZE(sc->tx.hwq_map));
+                       return -1;
+               }
+               qnum = sc->tx.hwq_map[haltype];
+               break;
+       case ATH9K_TX_QUEUE_BEACON:
+               qnum = sc->beacon.beaconq;
+               break;
+       case ATH9K_TX_QUEUE_CAB:
+               qnum = sc->beacon.cabq->axq_qnum;
+               break;
+       default:
+               qnum = -1;
+       }
+       return qnum;
+}
 
-       /*
-        * Find the lowest frame length among the rate series that will have a
-        * 4ms transmit duration.
-        * TODO - TXOP limit needs to be considered.
-        */
-       max_4ms_framelen = ATH_AMPDU_LIMIT_MAX;
+struct ath_txq *ath_test_get_txq(struct ath_softc *sc, struct sk_buff *skb)
+{
+       struct ath_txq *txq = NULL;
+       int qnum;
 
-       for (i = 0; i < 4; i++) {
-               if (rates[i].count) {
-                       if (!WLAN_RC_PHY_HT(rate_table->info[rates[i].idx].phy)) {
-                               legacy = 1;
-                               break;
-                       }
+       qnum = ath_get_hal_qnum(skb_get_queue_mapping(skb), sc);
+       txq = &sc->tx.txq[qnum];
 
-                       frame_length =
-                               rate_table->info[rates[i].idx].max_4ms_framelen;
-                       max_4ms_framelen = min(max_4ms_framelen, frame_length);
-               }
+       spin_lock_bh(&txq->axq_lock);
+
+       if (txq->axq_depth >= (ATH_TXBUF - 20)) {
+               DPRINTF(sc, ATH_DBG_FATAL,
+                       "TX queue: %d is full, depth: %d\n",
+                       qnum, txq->axq_depth);
+               ieee80211_stop_queue(sc->hw, skb_get_queue_mapping(skb));
+               txq->stopped = 1;
+               spin_unlock_bh(&txq->axq_lock);
+               return NULL;
        }
 
-       /*
-        * limit aggregate size by the minimum rate if rate selected is
-        * not a probe rate, if rate selected is a probe rate then
-        * avoid aggregation of this packet.
-        */
-       if (tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE || legacy)
+       spin_unlock_bh(&txq->axq_lock);
+
+       return txq;
+}
+
+int ath_txq_update(struct ath_softc *sc, int qnum,
+                  struct ath9k_tx_queue_info *qinfo)
+{
+       struct ath_hal *ah = sc->sc_ah;
+       int error = 0;
+       struct ath9k_tx_queue_info qi;
+
+       if (qnum == sc->beacon.beaconq) {
+               /*
+                * XXX: for beacon queue, we just save the parameter.
+                * It will be picked up by ath_beaconq_config when
+                * it's necessary.
+                */
+               sc->beacon.beacon_qi = *qinfo;
                return 0;
+       }
 
-       aggr_limit = min(max_4ms_framelen,
-               (u32)ATH_AMPDU_LIMIT_DEFAULT);
+       ASSERT(sc->tx.txq[qnum].axq_qnum == qnum);
+
+       ath9k_hw_get_txq_props(ah, qnum, &qi);
+       qi.tqi_aifs = qinfo->tqi_aifs;
+       qi.tqi_cwmin = qinfo->tqi_cwmin;
+       qi.tqi_cwmax = qinfo->tqi_cwmax;
+       qi.tqi_burstTime = qinfo->tqi_burstTime;
+       qi.tqi_readyTime = qinfo->tqi_readyTime;
+
+       if (!ath9k_hw_set_txq_props(ah, qnum, &qi)) {
+               DPRINTF(sc, ATH_DBG_FATAL,
+                       "Unable to update hardware queue %u!\n", qnum);
+               error = -EIO;
+       } else {
+               ath9k_hw_resettxqueue(ah, qnum);
+       }
+
+       return error;
+}
+
+int ath_cabq_update(struct ath_softc *sc)
+{
+       struct ath9k_tx_queue_info qi;
+       int qnum = sc->beacon.cabq->axq_qnum;
+       struct ath_beacon_config conf;
 
+       ath9k_hw_get_txq_props(sc->sc_ah, qnum, &qi);
        /*
-        * h/w can accept aggregates upto 16 bit lengths (65535).
-        * The IE, however can hold upto 65536, which shows up here
-        * as zero. Ignore 65536 since we  are constrained by hw.
+        * Ensure the readytime % is within the bounds.
         */
-       maxampdu = tid->an->maxampdu;
-       if (maxampdu)
-               aggr_limit = min(aggr_limit, maxampdu);
+       if (sc->sc_config.cabqReadytime < ATH9K_READY_TIME_LO_BOUND)
+               sc->sc_config.cabqReadytime = ATH9K_READY_TIME_LO_BOUND;
+       else if (sc->sc_config.cabqReadytime > ATH9K_READY_TIME_HI_BOUND)
+               sc->sc_config.cabqReadytime = ATH9K_READY_TIME_HI_BOUND;
 
-       return aggr_limit;
+       ath_get_beaconconfig(sc, ATH_IF_ID_ANY, &conf);
+       qi.tqi_readyTime =
+               (conf.beacon_interval * sc->sc_config.cabqReadytime) / 100;
+       ath_txq_update(sc, qnum, &qi);
+
+       return 0;
 }
 
-/*
- * returns the number of delimiters to be added to
- * meet the minimum required mpdudensity.
- * caller should make sure that the rate is  HT rate .
- */
-static int ath_compute_num_delims(struct ath_softc *sc,
-                                 struct ath_atx_tid *tid,
-                                 struct ath_buf *bf,
-                                 u16 frmlen)
+void ath_tx_draintxq(struct ath_softc *sc, struct ath_txq *txq, bool retry_tx)
 {
-       struct ath_rate_table *rt = sc->cur_rate_table;
-       struct sk_buff *skb = bf->bf_mpdu;
-       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
-       u32 nsymbits, nsymbols, mpdudensity;
-       u16 minlen;
-       u8 rc, flags, rix;
-       int width, half_gi, ndelim, mindelim;
+       struct ath_buf *bf, *lastbf;
+       struct list_head bf_head;
 
-       /* Select standard number of delimiters based on frame length alone */
-       ndelim = ATH_AGGR_GET_NDELIM(frmlen);
+       INIT_LIST_HEAD(&bf_head);
 
        /*
-        * If encryption enabled, hardware requires some more padding between
-        * subframes.
-        * TODO - this could be improved to be dependent on the rate.
-        *      The hardware can keep up at lower rates, but not higher rates
+        * NB: this assumes output has been stopped and
+        *     we do not need to block ath_tx_tasklet
         */
-       if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR)
-               ndelim += ATH_AGGR_ENCRYPTDELIM;
+       for (;;) {
+               spin_lock_bh(&txq->axq_lock);
 
-       /*
-        * Convert desired mpdu density from microeconds to bytes based
-        * on highest rate in rate series (i.e. first rate) to determine
-        * required minimum length for subframe. Take into account
-        * whether high rate is 20 or 40Mhz and half or full GI.
-        */
-       mpdudensity = tid->an->mpdudensity;
+               if (list_empty(&txq->axq_q)) {
+                       txq->axq_link = NULL;
+                       txq->axq_linkbuf = NULL;
+                       spin_unlock_bh(&txq->axq_lock);
+                       break;
+               }
 
-       /*
-        * If there is no mpdu density restriction, no further calculation
-        * is needed.
-        */
-       if (mpdudensity == 0)
-               return ndelim;
+               bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
 
-       rix = tx_info->control.rates[0].idx;
-       flags = tx_info->control.rates[0].flags;
-       rc = rt->info[rix].ratecode;
-       width = (flags & IEEE80211_TX_RC_40_MHZ_WIDTH) ? 1 : 0;
-       half_gi = (flags & IEEE80211_TX_RC_SHORT_GI) ? 1 : 0;
+               if (bf->bf_status & ATH_BUFSTATUS_STALE) {
+                       list_del(&bf->list);
+                       spin_unlock_bh(&txq->axq_lock);
 
-       if (half_gi)
-               nsymbols = NUM_SYMBOLS_PER_USEC_HALFGI(mpdudensity);
-       else
-               nsymbols = NUM_SYMBOLS_PER_USEC(mpdudensity);
+                       spin_lock_bh(&sc->tx.txbuflock);
+                       list_add_tail(&bf->list, &sc->tx.txbuf);
+                       spin_unlock_bh(&sc->tx.txbuflock);
+                       continue;
+               }
 
-       if (nsymbols == 0)
-               nsymbols = 1;
+               lastbf = bf->bf_lastbf;
+               if (!retry_tx)
+                       lastbf->bf_desc->ds_txstat.ts_flags =
+                               ATH9K_TX_SW_ABORTED;
 
-       nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
-       minlen = (nsymbols * nsymbits) / BITS_PER_BYTE;
+               /* remove ath_buf's of the same mpdu from txq */
+               list_cut_position(&bf_head, &txq->axq_q, &lastbf->list);
+               txq->axq_depth--;
 
-       /* Is frame shorter than required minimum length? */
-       if (frmlen < minlen) {
-               /* Get the minimum number of delimiters required. */
-               mindelim = (minlen - frmlen) / ATH_AGGR_DELIM_SZ;
-               ndelim = max(mindelim, ndelim);
+               spin_unlock_bh(&txq->axq_lock);
+
+               if (bf_isampdu(bf))
+                       ath_tx_complete_aggr_rifs(sc, txq, bf, &bf_head, 0);
+               else
+                       ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
        }
 
-       return ndelim;
+       /* flush any pending frames if aggregation is enabled */
+       if (sc->sc_flags & SC_OP_TXAGGR) {
+               if (!retry_tx) {
+                       spin_lock_bh(&txq->axq_lock);
+                       ath_txq_drain_pending_buffers(sc, txq);
+                       spin_unlock_bh(&txq->axq_lock);
+               }
+       }
 }
 
-/*
- * For aggregation from software buffer queue.
- * NB: must be called with txq lock held
- */
-static enum ATH_AGGR_STATUS ath_tx_form_aggr(struct ath_softc *sc,
-                                       struct ath_atx_tid *tid,
-                                       struct list_head *bf_q,
-                                       struct ath_buf **bf_last,
-                                       struct aggr_rifs_param *param,
-                                       int *prev_frames)
+void ath_tx_cleanupq(struct ath_softc *sc, struct ath_txq *txq)
 {
-#define PADBYTES(_len) ((4 - ((_len) % 4)) % 4)
-       struct ath_buf *bf, *tbf, *bf_first, *bf_prev = NULL;
-       struct list_head bf_head;
-       int rl = 0, 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;
-       int prev_al = 0;
-       INIT_LIST_HEAD(&bf_head);
+       ath9k_hw_releasetxqueue(sc->sc_ah, txq->axq_qnum);
+       sc->tx.txqsetup &= ~(1<<txq->axq_qnum);
+}
 
-       BUG_ON(list_empty(&tid->buf_q));
+void ath_draintxq(struct ath_softc *sc, bool retry_tx)
+{
+       if (!(sc->sc_flags & SC_OP_INVALID))
+               (void) ath9k_hw_stoptxdma(sc->sc_ah, sc->beacon.beaconq);
 
-       bf_first = list_first_entry(&tid->buf_q, struct ath_buf, list);
+       ath_drain_txdataq(sc, retry_tx);
+}
 
-       do {
-               bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
+void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq)
+{
+       struct ath_atx_ac *ac;
+       struct ath_atx_tid *tid;
 
-               /*
-                * do not step over block-ack window
-                */
-               if (!BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno)) {
-                       status = ATH_AGGR_BAW_CLOSED;
-                       break;
-               }
-
-               if (!rl) {
-                       aggr_limit = ath_lookup_rate(sc, bf, tid);
-                       rl = 1;
-               }
-
-               /*
-                * do not exceed aggregation limit
-                */
-               al_delta = ATH_AGGR_DELIM_SZ + bf->bf_frmlen;
-
-               if (nframes && (aggr_limit <
-                       (al + bpad + al_delta + prev_al))) {
-                       status = ATH_AGGR_LIMITED;
-                       break;
-               }
+       if (list_empty(&txq->axq_acq))
+               return;
 
-               /*
-                * do not exceed subframe limit
-                */
-               if ((nframes + *prev_frames) >=
-                   min((int)h_baw, ATH_AMPDU_SUBFRAME_DEFAULT)) {
-                       status = ATH_AGGR_LIMITED;
-                       break;
-               }
+       ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac, list);
+       list_del(&ac->list);
+       ac->sched = false;
 
-               /*
-                * add padding for previous frame to aggregation length
-                */
-               al += bpad + al_delta;
+       do {
+               if (list_empty(&ac->tid_q))
+                       return;
 
-               /*
-                * Get the delimiters needed to meet the MPDU
-                * density for this node.
-                */
-               ndelim = ath_compute_num_delims(sc, tid, bf_first, bf->bf_frmlen);
+               tid = list_first_entry(&ac->tid_q, struct ath_atx_tid, list);
+               list_del(&tid->list);
+               tid->sched = false;
 
-               bpad = PADBYTES(al_delta) + (ndelim << 2);
+               if (tid->paused)
+                       continue;
 
-               bf->bf_next = NULL;
-               bf->bf_lastfrm->bf_desc->ds_link = 0;
+               if ((txq->axq_depth % 2) == 0)
+                       ath_tx_sched_aggr(sc, txq, tid);
 
                /*
-                * this packet is part of an aggregate
-                * - remove all descriptors belonging to this frame from
-                *   software queue
-                * - add it to block ack window
-                * - set up descriptors for aggregation
+                * add tid to round-robin queue if more frames
+                * are pending for the tid
                 */
-               list_cut_position(&bf_head, &tid->buf_q, &bf->bf_lastfrm->list);
-               ath_tx_addto_baw(sc, tid, bf);
-
-               list_for_each_entry(tbf, &bf_head, list) {
-                       ath9k_hw_set11n_aggr_middle(sc->sc_ah,
-                               tbf->bf_desc, ndelim);
-               }
+               if (!list_empty(&tid->buf_q))
+                       ath_tx_queue_tid(txq, tid);
 
-               /*
-                * link buffers of this frame to the aggregate
-                */
-               list_splice_tail_init(&bf_head, bf_q);
-               nframes++;
+               break;
+       } while (!list_empty(&ac->tid_q));
 
-               if (bf_prev) {
-                       bf_prev->bf_next = bf;
-                       bf_prev->bf_lastfrm->bf_desc->ds_link = bf->bf_daddr;
+       if (!list_empty(&ac->tid_q)) {
+               if (!ac->sched) {
+                       ac->sched = true;
+                       list_add_tail(&ac->list, &txq->axq_acq);
                }
-               bf_prev = bf;
+       }
+}
 
-#ifdef AGGR_NOSHORT
-               /*
-                * terminate aggregation on a small packet boundary
-                */
-               if (bf->bf_frmlen < ATH_AGGR_MINPLEN) {
-                       status = ATH_AGGR_SHORTPKT;
-                       break;
-               }
-#endif
-       } while (!list_empty(&tid->buf_q));
+int ath_tx_setup(struct ath_softc *sc, int haltype)
+{
+       struct ath_txq *txq;
 
-       bf_first->bf_al = al;
-       bf_first->bf_nframes = nframes;
-       *bf_last = bf_prev;
-       return status;
-#undef PADBYTES
+       if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
+               DPRINTF(sc, ATH_DBG_FATAL,
+                       "HAL AC %u out of range, max %zu!\n",
+                        haltype, ARRAY_SIZE(sc->tx.hwq_map));
+               return 0;
+       }
+       txq = ath_txq_setup(sc, ATH9K_TX_QUEUE_DATA, haltype);
+       if (txq != NULL) {
+               sc->tx.hwq_map[haltype] = txq->axq_qnum;
+               return 1;
+       } else
+               return 0;
 }
 
+/***********/
+/* TX, DMA */
+/***********/
+
 /*
- * process pending frames possibly doing a-mpdu aggregation
- * NB: must be called with txq lock held
+ * Insert a chain of ath_buf (descriptors) on a txq and
+ * assume the descriptors are already chained together by caller.
  */
-static void ath_tx_sched_aggr(struct ath_softc *sc,
-       struct ath_txq *txq, struct ath_atx_tid *tid)
+static void ath_tx_txqaddbuf(struct ath_softc *sc, struct ath_txq *txq,
+                            struct list_head *head)
 {
-       struct ath_buf *bf, *tbf, *bf_last, *bf_lastaggr = NULL;
-       enum ATH_AGGR_STATUS status;
-       struct list_head bf_q;
-       struct aggr_rifs_param param = {0, 0, 0, 0, NULL};
-       int prev_frames = 0;
-
-       do {
-               if (list_empty(&tid->buf_q))
-                       return;
-
-               INIT_LIST_HEAD(&bf_q);
+       struct ath_hal *ah = sc->sc_ah;
+       struct ath_buf *bf;
 
-               status = ath_tx_form_aggr(sc, tid, &bf_q, &bf_lastaggr, &param,
-                                         &prev_frames);
+       /*
+        * Insert the frame on the outbound list and
+        * pass it on to the hardware.
+        */
 
-               /*
-                * no frames picked up to be aggregated; block-ack
-                * window is not open
-                */
-               if (list_empty(&bf_q))
-                       break;
+       if (list_empty(head))
+               return;
 
-               bf = list_first_entry(&bf_q, struct ath_buf, list);
-               bf_last = list_entry(bf_q.prev, struct ath_buf, list);
-               bf->bf_lastbf = bf_last;
+       bf = list_first_entry(head, struct ath_buf, list);
 
-               /*
-                * if only one frame, send as non-aggregate
-                */
-               if (bf->bf_nframes == 1) {
-                       ASSERT(bf->bf_lastfrm == bf_last);
+       list_splice_tail_init(head, &txq->axq_q);
+       txq->axq_depth++;
+       txq->axq_totalqueued++;
+       txq->axq_linkbuf = list_entry(txq->axq_q.prev, struct ath_buf, list);
 
-                       bf->bf_state.bf_type &= ~BUF_AGGR;
-                       /*
-                        * clear aggr bits for every descriptor
-                        * XXX TODO: is there a way to optimize it?
-                        */
-                       list_for_each_entry(tbf, &bf_q, list) {
-                               ath9k_hw_clr11n_aggr(sc->sc_ah, tbf->bf_desc);
-                       }
+       DPRINTF(sc, ATH_DBG_QUEUE,
+               "qnum: %d, txq depth: %d\n", txq->axq_qnum, txq->axq_depth);
 
-                       ath_buf_set_rate(sc, bf);
-                       ath_tx_txqaddbuf(sc, txq, &bf_q);
-                       continue;
-               }
+       if (txq->axq_link == NULL) {
+               ath9k_hw_puttxbuf(ah, txq->axq_qnum, bf->bf_daddr);
+               DPRINTF(sc, ATH_DBG_XMIT,
+                       "TXDP[%u] = %llx (%p)\n",
+                       txq->axq_qnum, ito64(bf->bf_daddr), bf->bf_desc);
+       } else {
+               *txq->axq_link = bf->bf_daddr;
+               DPRINTF(sc, ATH_DBG_XMIT, "link[%u] (%p)=%llx (%p)\n",
+                       txq->axq_qnum, txq->axq_link,
+                       ito64(bf->bf_daddr), bf->bf_desc);
+       }
+       txq->axq_link = &(bf->bf_lastbf->bf_desc->ds_link);
+       ath9k_hw_txstart(ah, txq->axq_qnum);
+}
 
-               /*
-                * setup first desc with rate and aggr info
-                */
-               bf->bf_state.bf_type |= BUF_AGGR;
-               ath_buf_set_rate(sc, bf);
-               ath9k_hw_set11n_aggr_first(sc->sc_ah, bf->bf_desc, bf->bf_al);
+static struct ath_buf *ath_tx_get_buffer(struct ath_softc *sc)
+{
+       struct ath_buf *bf = NULL;
 
-               /*
-                * anchor last frame of aggregate correctly
-                */
-               ASSERT(bf_lastaggr);
-               ASSERT(bf_lastaggr->bf_lastfrm == bf_last);
-               tbf = bf_lastaggr;
-               ath9k_hw_set11n_aggr_last(sc->sc_ah, tbf->bf_desc);
+       spin_lock_bh(&sc->tx.txbuflock);
 
-               /* XXX: We don't enter into this loop, consider removing this */
-               while (!list_empty(&bf_q) && !list_is_last(&tbf->list, &bf_q)) {
-                       tbf = list_entry(tbf->list.next, struct ath_buf, list);
-                       ath9k_hw_set11n_aggr_last(sc->sc_ah, tbf->bf_desc);
-               }
+       if (unlikely(list_empty(&sc->tx.txbuf))) {
+               spin_unlock_bh(&sc->tx.txbuflock);
+               return NULL;
+       }
 
-               txq->axq_aggr_depth++;
+       bf = list_first_entry(&sc->tx.txbuf, struct ath_buf, list);
+       list_del(&bf->list);
 
-               /*
-                * Normal aggregate, queue to hardware
-                */
-               ath_tx_txqaddbuf(sc, txq, &bf_q);
+       spin_unlock_bh(&sc->tx.txbuflock);
 
-       } while (txq->axq_depth < ATH_AGGR_MIN_QDEPTH &&
-                status != ATH_AGGR_BAW_CLOSED);
+       return bf;
 }
 
-/* Called with txq lock held */
-
-static void ath_tid_drain(struct ath_softc *sc,
-                         struct ath_txq *txq,
-                         struct ath_atx_tid *tid)
-
+static void ath_tx_send_ampdu(struct ath_softc *sc, struct ath_atx_tid *tid,
+                             struct list_head *bf_head,
+                             struct ath_tx_control *txctl)
 {
        struct ath_buf *bf;
-       struct list_head bf_head;
-       INIT_LIST_HEAD(&bf_head);
-
-       for (;;) {
-               if (list_empty(&tid->buf_q))
-                       break;
-               bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
 
-               list_cut_position(&bf_head, &tid->buf_q, &bf->bf_lastfrm->list);
+       BUG_ON(list_empty(bf_head));
 
-               /* update baw for software retried frame */
-               if (bf_isretried(bf))
-                       ath_tx_update_baw(sc, tid, bf->bf_seqno);
+       bf = list_first_entry(bf_head, struct ath_buf, list);
+       bf->bf_state.bf_type |= BUF_AMPDU;
 
+       /*
+        * 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 (!list_empty(&tid->buf_q) || tid->paused ||
+           !BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno) ||
+           txctl->txq->axq_depth >= ATH_AGGR_MIN_QDEPTH) {
                /*
-                * do not indicate packets while holding txq spinlock.
-                * unlock is intentional here
+                * Add this frame to software queue for scheduling later
+                * for aggregation.
                 */
-               spin_unlock(&txq->axq_lock);
+               list_splice_tail_init(bf_head, &tid->buf_q);
+               ath_tx_queue_tid(txctl->txq, tid);
+               return;
+       }
+
+       /* Add sub-frame to BAW */
+       ath_tx_addto_baw(sc, tid, bf);
+
+       /* Queue to h/w without aggregation */
+       bf->bf_nframes = 1;
+       bf->bf_lastbf = bf->bf_lastfrm; /* one single frame */
+       ath_buf_set_rate(sc, bf);
+       ath_tx_txqaddbuf(sc, txctl->txq, bf_head);
+
+       return;
+}
+
+static void ath_tx_send_normal(struct ath_softc *sc, struct ath_txq *txq,
+                              struct ath_atx_tid *tid,
+                              struct list_head *bf_head)
+{
+       struct ath_buf *bf;
+
+       BUG_ON(list_empty(bf_head));
+
+       bf = list_first_entry(bf_head, struct ath_buf, list);
+       bf->bf_state.bf_type &= ~BUF_AMPDU;
+
+       /* update starting sequence number for subsequent ADDBA request */
+       INCR(tid->seq_start, IEEE80211_SEQ_MAX);
+
+       bf->bf_nframes = 1;
+       bf->bf_lastbf = bf->bf_lastfrm;
+       ath_buf_set_rate(sc, bf);
+       ath_tx_txqaddbuf(sc, txq, bf_head);
+}
+
+static enum ath9k_pkt_type get_hw_packet_type(struct sk_buff *skb)
+{
+       struct ieee80211_hdr *hdr;
+       enum ath9k_pkt_type htype;
+       __le16 fc;
+
+       hdr = (struct ieee80211_hdr *)skb->data;
+       fc = hdr->frame_control;
+
+       if (ieee80211_is_beacon(fc))
+               htype = ATH9K_PKT_TYPE_BEACON;
+       else if (ieee80211_is_probe_resp(fc))
+               htype = ATH9K_PKT_TYPE_PROBE_RESP;
+       else if (ieee80211_is_atim(fc))
+               htype = ATH9K_PKT_TYPE_ATIM;
+       else if (ieee80211_is_pspoll(fc))
+               htype = ATH9K_PKT_TYPE_PSPOLL;
+       else
+               htype = ATH9K_PKT_TYPE_NORMAL;
+
+       return htype;
+}
+
+static bool is_pae(struct sk_buff *skb)
+{
+       struct ieee80211_hdr *hdr;
+       __le16 fc;
+
+       hdr = (struct ieee80211_hdr *)skb->data;
+       fc = hdr->frame_control;
+
+       if (ieee80211_is_data(fc)) {
+               if (ieee80211_is_nullfunc(fc) ||
+                   /* Port Access Entity (IEEE 802.1X) */
+                   (skb->protocol == cpu_to_be16(ETH_P_PAE))) {
+                       return true;
+               }
+       }
+
+       return false;
+}
+
+static int get_hw_crypto_keytype(struct sk_buff *skb)
+{
+       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+
+       if (tx_info->control.hw_key) {
+               if (tx_info->control.hw_key->alg == ALG_WEP)
+                       return ATH9K_KEY_TYPE_WEP;
+               else if (tx_info->control.hw_key->alg == ALG_TKIP)
+                       return ATH9K_KEY_TYPE_TKIP;
+               else if (tx_info->control.hw_key->alg == ALG_CCMP)
+                       return ATH9K_KEY_TYPE_AES;
+       }
+
+       return ATH9K_KEY_TYPE_CLEAR;
+}
+
+static void assign_aggr_tid_seqno(struct sk_buff *skb,
+                                 struct ath_buf *bf)
+{
+       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+       struct ieee80211_hdr *hdr;
+       struct ath_node *an;
+       struct ath_atx_tid *tid;
+       __le16 fc;
+       u8 *qc;
+
+       if (!tx_info->control.sta)
+               return;
+
+       an = (struct ath_node *)tx_info->control.sta->drv_priv;
+       hdr = (struct ieee80211_hdr *)skb->data;
+       fc = hdr->frame_control;
+
+       if (ieee80211_is_data_qos(fc)) {
+               qc = ieee80211_get_qos_ctl(hdr);
+               bf->bf_tidno = qc[0] & 0xf;
+       }
+
+       /*
+        * For HT capable stations, we save tidno for later use.
+        * We also override seqno set by upper layer with the one
+        * in tx aggregation state.
+        *
+        * If fragmentation is on, the sequence number is
+        * not overridden, since it has been
+        * incremented by the fragmentation routine.
+        *
+        * FIXME: check if the fragmentation threshold exceeds
+        * IEEE80211 max.
+        */
+       tid = ATH_AN_2_TID(an, bf->bf_tidno);
+       hdr->seq_ctrl = cpu_to_le16(tid->seq_next <<
+                       IEEE80211_SEQ_SEQ_SHIFT);
+       bf->bf_seqno = tid->seq_next;
+       INCR(tid->seq_next, IEEE80211_SEQ_MAX);
+}
+
+static int setup_tx_flags(struct ath_softc *sc, struct sk_buff *skb,
+                         struct ath_txq *txq)
+{
+       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+       int flags = 0;
+
+       flags |= ATH9K_TXDESC_CLRDMASK; /* needed for crypto errors */
+       flags |= ATH9K_TXDESC_INTREQ;
+
+       if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK)
+               flags |= ATH9K_TXDESC_NOACK;
+       if (tx_info->control.rates[0].flags & IEEE80211_TX_RC_USE_RTS_CTS)
+               flags |= ATH9K_TXDESC_RTSENA;
+
+       return flags;
+}
+
+/*
+ * rix - rate index
+ * pktlen - total bytes (delims + data + fcs + pads + pad delims)
+ * width  - 0 for 20 MHz, 1 for 40 MHz
+ * half_gi - to use 4us v/s 3.6 us for symbol time
+ */
+static u32 ath_pkt_duration(struct ath_softc *sc, u8 rix, struct ath_buf *bf,
+                           int width, int half_gi, bool shortPreamble)
+{
+       struct ath_rate_table *rate_table = sc->cur_rate_table;
+       u32 nbits, nsymbits, duration, nsymbols;
+       u8 rc;
+       int streams, pktlen;
+
+       pktlen = bf_isaggr(bf) ? bf->bf_al : bf->bf_frmlen;
+       rc = rate_table->info[rix].ratecode;
+
+       /* for legacy rates, use old function to compute packet duration */
+       if (!IS_HT_RATE(rc))
+               return ath9k_hw_computetxtime(sc->sc_ah, rate_table, pktlen,
+                                             rix, shortPreamble);
+
+       /* find number of symbols: PLCP + data */
+       nbits = (pktlen << 3) + OFDM_PLCP_BITS;
+       nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
+       nsymbols = (nbits + nsymbits - 1) / nsymbits;
+
+       if (!half_gi)
+               duration = SYMBOL_TIME(nsymbols);
+       else
+               duration = SYMBOL_TIME_HALFGI(nsymbols);
+
+       /* addup duration for legacy/ht training and signal fields */
+       streams = HT_RC_2_STREAMS(rc);
+       duration += L_STF + L_LTF + L_SIG + HT_SIG + HT_STF + HT_LTF(streams);
+
+       return duration;
+}
+
+static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf)
+{
+       struct ath_hal *ah = sc->sc_ah;
+       struct ath_rate_table *rt;
+       struct ath_desc *ds = bf->bf_desc;
+       struct ath_desc *lastds = bf->bf_lastbf->bf_desc;
+       struct ath9k_11n_rate_series series[4];
+       struct sk_buff *skb;
+       struct ieee80211_tx_info *tx_info;
+       struct ieee80211_tx_rate *rates;
+       struct ieee80211_hdr *hdr;
+       struct ieee80211_hw *hw = sc->hw;
+       int i, flags, rtsctsena = 0, enable_g_protection = 0;
+       u32 ctsduration = 0;
+       u8 rix = 0, cix, ctsrate = 0;
+       __le16 fc;
+
+       memset(series, 0, sizeof(struct ath9k_11n_rate_series) * 4);
+
+       skb = (struct sk_buff *)bf->bf_mpdu;
+       hdr = (struct ieee80211_hdr *)skb->data;
+       fc = hdr->frame_control;
+       tx_info = IEEE80211_SKB_CB(skb);
+       rates = tx_info->control.rates;
+
+       if (ieee80211_has_morefrags(fc) ||
+           (le16_to_cpu(hdr->seq_ctrl) & IEEE80211_SCTL_FRAG)) {
+               rates[1].count = rates[2].count = rates[3].count = 0;
+               rates[1].idx = rates[2].idx = rates[3].idx = 0;
+               rates[0].count = ATH_TXMAXTRY;
+       }
+
+       /* get the cix for the lowest valid rix */
+       rt = sc->cur_rate_table;
+       for (i = 3; i >= 0; i--) {
+               if (rates[i].count && (rates[i].idx >= 0)) {
+                       rix = rates[i].idx;
+                       break;
+               }
+       }
+
+       flags = (bf->bf_flags & (ATH9K_TXDESC_RTSENA | ATH9K_TXDESC_CTSENA));
+       cix = rt->info[rix].ctrl_rate;
+
+       /* All protection frames are transmited at 2Mb/s for 802.11g,
+        * otherwise we transmit them at 1Mb/s */
+       if (hw->conf.channel->band == IEEE80211_BAND_2GHZ &&
+         !conf_is_ht(&hw->conf))
+               enable_g_protection = 1;
+
+       /*
+        * If 802.11g protection is enabled, determine whether to use RTS/CTS or
+        * just CTS.  Note that this is only done for OFDM/HT unicast frames.
+        */
+       if (sc->sc_protmode != PROT_M_NONE && !(bf->bf_flags & ATH9K_TXDESC_NOACK)
+           && (rt->info[rix].phy == WLAN_RC_PHY_OFDM ||
+               WLAN_RC_PHY_HT(rt->info[rix].phy))) {
+               if (sc->sc_protmode == PROT_M_RTSCTS)
+                       flags = ATH9K_TXDESC_RTSENA;
+               else if (sc->sc_protmode == PROT_M_CTSONLY)
+                       flags = ATH9K_TXDESC_CTSENA;
+
+               cix = rt->info[enable_g_protection].ctrl_rate;
+               rtsctsena = 1;
+       }
+
+       /* For 11n, the default behavior is to enable RTS for hw retried frames.
+        * We enable the global flag here and let rate series flags determine
+        * which rates will actually use RTS.
+        */
+       if ((ah->ah_caps.hw_caps & ATH9K_HW_CAP_HT) && bf_isdata(bf)) {
+               /* 802.11g protection not needed, use our default behavior */
+               if (!rtsctsena)
+                       flags = ATH9K_TXDESC_RTSENA;
+       }
+
+       /* Set protection if aggregate protection on */
+       if (sc->sc_config.ath_aggr_prot &&
+           (!bf_isaggr(bf) || (bf_isaggr(bf) && bf->bf_al < 8192))) {
+               flags = ATH9K_TXDESC_RTSENA;
+               cix = rt->info[enable_g_protection].ctrl_rate;
+               rtsctsena = 1;
+       }
+
+       /* For AR5416 - RTS cannot be followed by a frame larger than 8K */
+       if (bf_isaggr(bf) && (bf->bf_al > ah->ah_caps.rts_aggr_limit))
+               flags &= ~(ATH9K_TXDESC_RTSENA);
+
+       /*
+        * CTS transmit rate is derived from the transmit rate by looking in the
+        * h/w rate table.  We must also factor in whether or not a short
+        * preamble is to be used. NB: cix is set above where RTS/CTS is enabled
+        */
+       ctsrate = rt->info[cix].ratecode |
+               (bf_isshpreamble(bf) ? rt->info[cix].short_preamble : 0);
+
+       for (i = 0; i < 4; i++) {
+               if (!rates[i].count || (rates[i].idx < 0))
+                       continue;
+
+               rix = rates[i].idx;
+
+               series[i].Rate = rt->info[rix].ratecode |
+                       (bf_isshpreamble(bf) ? rt->info[rix].short_preamble : 0);
+
+               series[i].Tries = rates[i].count;
+
+               series[i].RateFlags = (
+                       (rates[i].flags & IEEE80211_TX_RC_USE_RTS_CTS) ?
+                               ATH9K_RATESERIES_RTS_CTS : 0) |
+                       ((rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH) ?
+                               ATH9K_RATESERIES_2040 : 0) |
+                       ((rates[i].flags & IEEE80211_TX_RC_SHORT_GI) ?
+                               ATH9K_RATESERIES_HALFGI : 0);
+
+               series[i].PktDuration = ath_pkt_duration(sc, rix, bf,
+                        (rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH) != 0,
+                        (rates[i].flags & IEEE80211_TX_RC_SHORT_GI),
+                        bf_isshpreamble(bf));
 
-               /* complete this sub-frame */
-               ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
+               series[i].ChSel = sc->sc_tx_chainmask;
 
-               spin_lock(&txq->axq_lock);
+               if (rtsctsena)
+                       series[i].RateFlags |= ATH9K_RATESERIES_RTS_CTS;
        }
 
-       /*
-        * TODO: For frame(s) that are in the retry state, we will reuse the
-        * sequence number(s) without setting the retry bit. The
-        * alternative is to give up on these and BAR the receiver's window
-        * forward.
-        */
-       tid->seq_next = tid->seq_start;
-       tid->baw_tail = tid->baw_head;
-}
-
-/*
- * Drain all pending buffers
- * NB: must be called with txq lock held
- */
-static void ath_txq_drain_pending_buffers(struct ath_softc *sc,
-                                         struct ath_txq *txq)
-{
-       struct ath_atx_ac *ac, *ac_tmp;
-       struct ath_atx_tid *tid, *tid_tmp;
+       /* set dur_update_en for l-sig computation except for PS-Poll frames */
+       ath9k_hw_set11n_ratescenario(ah, ds, lastds, !bf_ispspoll(bf),
+                                    ctsrate, ctsduration,
+                                    series, 4, flags);
 
-       list_for_each_entry_safe(ac, ac_tmp, &txq->axq_acq, list) {
-               list_del(&ac->list);
-               ac->sched = false;
-               list_for_each_entry_safe(tid, tid_tmp, &ac->tid_q, list) {
-                       list_del(&tid->list);
-                       tid->sched = false;
-                       ath_tid_drain(sc, txq, tid);
-               }
-       }
+       if (sc->sc_config.ath_aggr_prot && flags)
+               ath9k_hw_set11n_burstduration(ah, ds, 8192);
 }
 
 static int ath_tx_setup_buffer(struct ath_softc *sc, struct ath_buf *bf,
@@ -1672,8 +1697,6 @@ static int ath_tx_setup_buffer(struct ath_softc *sc, struct ath_buf *bf,
 
        ATH_TXBUF_RESET(bf);
 
-       /* Frame type */
-
        bf->bf_frmlen = skb->len + FCS_LEN - (hdrlen & 3);
 
        ieee80211_is_data(fc) ?
@@ -1695,8 +1718,6 @@ static int ath_tx_setup_buffer(struct ath_softc *sc, struct ath_buf *bf,
 
        bf->bf_flags = setup_tx_flags(sc, skb, txctl->txq);
 
-       /* Crypto */
-
        bf->bf_keytype = get_hw_crypto_keytype(skb);
 
        if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR) {
@@ -1706,12 +1727,9 @@ static int ath_tx_setup_buffer(struct ath_softc *sc, struct ath_buf *bf,
                bf->bf_keyix = ATH9K_TXKEYIX_INVALID;
        }
 
-       /* Assign seqno, tidno */
-
        if (ieee80211_is_data_qos(fc) && (sc->sc_flags & SC_OP_TXAGGR))
                assign_aggr_tid_seqno(skb, bf);
 
-       /* DMA setup */
        bf->bf_mpdu = skb;
 
        bf->bf_dmacontext = dma_map_single(sc->dev, skb->data,
@@ -1745,14 +1763,10 @@ static void ath_tx_start_dma(struct ath_softc *sc, struct ath_buf *bf,
        INIT_LIST_HEAD(&bf_head);
        list_add_tail(&bf->list, &bf_head);
 
-       /* setup descriptor */
-
        ds = bf->bf_desc;
        ds->ds_link = 0;
        ds->ds_data = bf->bf_buf_addr;
 
-       /* Formulate first tx descriptor with tx controls */
-
        ath9k_hw_set11n_txdesc(ah, ds, bf->bf_frmlen, frm_type, MAX_RATE_POWER,
                               bf->bf_keyix, bf->bf_keytype, bf->bf_flags);
 
@@ -1803,8 +1817,6 @@ int ath_tx_start(struct ath_softc *sc, struct sk_buff *skb,
        struct ath_buf *bf;
        int r;
 
-       /* Check if a tx buffer is available */
-
        bf = ath_tx_get_buffer(sc);
        if (!bf) {
                DPRINTF(sc, ATH_DBG_XMIT, "TX buffers are full\n");
@@ -1841,570 +1853,378 @@ int ath_tx_start(struct ath_softc *sc, struct sk_buff *skb,
        return 0;
 }
 
-/* Initialize TX queue and h/w */
-
-int ath_tx_init(struct ath_softc *sc, int nbufs)
-{
-       int error = 0;
-
-       do {
-               spin_lock_init(&sc->tx.txbuflock);
-
-               /* Setup tx descriptors */
-               error = ath_descdma_setup(sc, &sc->tx.txdma, &sc->tx.txbuf,
-                       "tx", nbufs, 1);
-               if (error != 0) {
-                       DPRINTF(sc, ATH_DBG_FATAL,
-                               "Failed to allocate tx descriptors: %d\n",
-                               error);
-                       break;
-               }
-
-               /* XXX allocate beacon state together with vap */
-               error = ath_descdma_setup(sc, &sc->beacon.bdma, &sc->beacon.bbuf,
-                                         "beacon", ATH_BCBUF, 1);
-               if (error != 0) {
-                       DPRINTF(sc, ATH_DBG_FATAL,
-                               "Failed to allocate beacon descriptors: %d\n",
-                               error);
-                       break;
-               }
-
-       } while (0);
-
-       if (error != 0)
-               ath_tx_cleanup(sc);
-
-       return error;
-}
-
-/* Reclaim all tx queue resources */
-
-int ath_tx_cleanup(struct ath_softc *sc)
-{
-       /* cleanup beacon descriptors */
-       if (sc->beacon.bdma.dd_desc_len != 0)
-               ath_descdma_cleanup(sc, &sc->beacon.bdma, &sc->beacon.bbuf);
-
-       /* cleanup tx descriptors */
-       if (sc->tx.txdma.dd_desc_len != 0)
-               ath_descdma_cleanup(sc, &sc->tx.txdma, &sc->tx.txbuf);
-
-       return 0;
-}
-
-/* Setup a h/w transmit queue */
-
-struct ath_txq *ath_txq_setup(struct ath_softc *sc, int qtype, int subtype)
+void ath_tx_cabq(struct ath_softc *sc, struct sk_buff *skb)
 {
-       struct ath_hal *ah = sc->sc_ah;
-       struct ath9k_tx_queue_info qi;
-       int qnum;
+       int hdrlen, padsize;
+       struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+       struct ath_tx_control txctl;
 
-       memset(&qi, 0, sizeof(qi));
-       qi.tqi_subtype = subtype;
-       qi.tqi_aifs = ATH9K_TXQ_USEDEFAULT;
-       qi.tqi_cwmin = ATH9K_TXQ_USEDEFAULT;
-       qi.tqi_cwmax = ATH9K_TXQ_USEDEFAULT;
-       qi.tqi_physCompBuf = 0;
+       memset(&txctl, 0, sizeof(struct ath_tx_control));
 
        /*
-        * Enable interrupts only for EOL and DESC conditions.
-        * We mark tx descriptors to receive a DESC interrupt
-        * when a tx queue gets deep; otherwise waiting for the
-        * EOL to reap descriptors.  Note that this is done to
-        * reduce interrupt load and this only defers reaping
-        * descriptors, never transmitting frames.  Aside from
-        * reducing interrupts this also permits more concurrency.
-        * The only potential downside is if the tx queue backs
-        * up in which case the top half of the kernel may backup
-        * due to a lack of tx descriptors.
-        *
-        * The UAPSD queue is an exception, since we take a desc-
-        * based intr on the EOSP frames.
+        * As a temporary workaround, assign seq# here; this will likely need
+        * to be cleaned up to work better with Beacon transmission and virtual
+        * BSSes.
         */
-       if (qtype == ATH9K_TX_QUEUE_UAPSD)
-               qi.tqi_qflags = TXQ_FLAG_TXDESCINT_ENABLE;
-       else
-               qi.tqi_qflags = TXQ_FLAG_TXEOLINT_ENABLE |
-                       TXQ_FLAG_TXDESCINT_ENABLE;
-       qnum = ath9k_hw_setuptxqueue(ah, qtype, &qi);
-       if (qnum == -1) {
-               /*
-                * NB: don't print a message, this happens
-                * normally on parts with too few tx queues
-                */
-               return NULL;
-       }
-       if (qnum >= ARRAY_SIZE(sc->tx.txq)) {
-               DPRINTF(sc, ATH_DBG_FATAL,
-                       "qnum %u out of range, max %u!\n",
-                       qnum, (unsigned int)ARRAY_SIZE(sc->tx.txq));
-               ath9k_hw_releasetxqueue(ah, qnum);
-               return NULL;
+       if (info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ) {
+               struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
+               if (info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT)
+                       sc->tx.seq_no += 0x10;
+               hdr->seq_ctrl &= cpu_to_le16(IEEE80211_SCTL_FRAG);
+               hdr->seq_ctrl |= cpu_to_le16(sc->tx.seq_no);
        }
-       if (!ATH_TXQ_SETUP(sc, qnum)) {
-               struct ath_txq *txq = &sc->tx.txq[qnum];
 
-               txq->axq_qnum = qnum;
-               txq->axq_link = NULL;
-               INIT_LIST_HEAD(&txq->axq_q);
-               INIT_LIST_HEAD(&txq->axq_acq);
-               spin_lock_init(&txq->axq_lock);
-               txq->axq_depth = 0;
-               txq->axq_aggr_depth = 0;
-               txq->axq_totalqueued = 0;
-               txq->axq_linkbuf = NULL;
-               sc->tx.txqsetup |= 1<<qnum;
+       /* Add the padding after the header if this is not already done */
+       hdrlen = ieee80211_get_hdrlen_from_skb(skb);
+       if (hdrlen & 3) {
+               padsize = hdrlen % 4;
+               if (skb_headroom(skb) < padsize) {
+                       DPRINTF(sc, ATH_DBG_XMIT, "TX CABQ padding failed\n");
+                       dev_kfree_skb_any(skb);
+                       return;
+               }
+               skb_push(skb, padsize);
+               memmove(skb->data, skb->data + padsize, hdrlen);
        }
-       return &sc->tx.txq[qnum];
-}
-
-/* Reclaim resources for a setup queue */
-
-void ath_tx_cleanupq(struct ath_softc *sc, struct ath_txq *txq)
-{
-       ath9k_hw_releasetxqueue(sc->sc_ah, txq->axq_qnum);
-       sc->tx.txqsetup &= ~(1<<txq->axq_qnum);
-}
 
-/*
- * Setup a hardware data transmit queue for the specified
- * access control.  The hal may not support all requested
- * queues in which case it will return a reference to a
- * previously setup queue.  We record the mapping from ac's
- * to h/w queues for use by ath_tx_start and also track
- * the set of h/w queues being used to optimize work in the
- * transmit interrupt handler and related routines.
- */
+       txctl.txq = sc->beacon.cabq;
 
-int ath_tx_setup(struct ath_softc *sc, int haltype)
-{
-       struct ath_txq *txq;
+       DPRINTF(sc, ATH_DBG_XMIT, "transmitting CABQ packet, skb: %p\n", skb);
 
-       if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
-               DPRINTF(sc, ATH_DBG_FATAL,
-                       "HAL AC %u out of range, max %zu!\n",
-                        haltype, ARRAY_SIZE(sc->tx.hwq_map));
-               return 0;
+       if (ath_tx_start(sc, skb, &txctl) != 0) {
+               DPRINTF(sc, ATH_DBG_XMIT, "CABQ TX failed\n");
+               goto exit;
        }
-       txq = ath_txq_setup(sc, ATH9K_TX_QUEUE_DATA, haltype);
-       if (txq != NULL) {
-               sc->tx.hwq_map[haltype] = txq->axq_qnum;
-               return 1;
-       } else
-               return 0;
-}
-
-int ath_tx_get_qnum(struct ath_softc *sc, int qtype, int haltype)
-{
-       int qnum;
 
-       switch (qtype) {
-       case ATH9K_TX_QUEUE_DATA:
-               if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
-                       DPRINTF(sc, ATH_DBG_FATAL,
-                               "HAL AC %u out of range, max %zu!\n",
-                               haltype, ARRAY_SIZE(sc->tx.hwq_map));
-                       return -1;
-               }
-               qnum = sc->tx.hwq_map[haltype];
-               break;
-       case ATH9K_TX_QUEUE_BEACON:
-               qnum = sc->beacon.beaconq;
-               break;
-       case ATH9K_TX_QUEUE_CAB:
-               qnum = sc->beacon.cabq->axq_qnum;
-               break;
-       default:
-               qnum = -1;
-       }
-       return qnum;
+       return;
+exit:
+       dev_kfree_skb_any(skb);
 }
 
-/* Get a transmit queue, if available */
+/*****************/
+/* TX Completion */
+/*****************/
 
-struct ath_txq *ath_test_get_txq(struct ath_softc *sc, struct sk_buff *skb)
+static void ath_tx_complete(struct ath_softc *sc, struct sk_buff *skb,
+                           struct ath_xmit_status *tx_status)
 {
-       struct ath_txq *txq = NULL;
-       int qnum;
+       struct ieee80211_hw *hw = sc->hw;
+       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+       struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
+       int hdrlen, padsize;
 
-       qnum = ath_get_hal_qnum(skb_get_queue_mapping(skb), sc);
-       txq = &sc->tx.txq[qnum];
+       DPRINTF(sc, ATH_DBG_XMIT, "TX complete: skb: %p\n", skb);
 
-       spin_lock_bh(&txq->axq_lock);
+       if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK ||
+           tx_info->flags & IEEE80211_TX_STAT_TX_FILTERED) {
+               kfree(tx_info_priv);
+               tx_info->rate_driver_data[0] = NULL;
+       }
 
-       /* Try to avoid running out of descriptors */
-       if (txq->axq_depth >= (ATH_TXBUF - 20)) {
-               DPRINTF(sc, ATH_DBG_FATAL,
-                       "TX queue: %d is full, depth: %d\n",
-                       qnum, txq->axq_depth);
-               ieee80211_stop_queue(sc->hw, skb_get_queue_mapping(skb));
-               txq->stopped = 1;
-               spin_unlock_bh(&txq->axq_lock);
-               return NULL;
+       if (tx_status->flags & ATH_TX_BAR) {
+               tx_info->flags |= IEEE80211_TX_STAT_AMPDU_NO_BACK;
+               tx_status->flags &= ~ATH_TX_BAR;
+       }
+
+       if (!(tx_status->flags & (ATH_TX_ERROR | ATH_TX_XRETRY))) {
+               /* Frame was ACKed */
+               tx_info->flags |= IEEE80211_TX_STAT_ACK;
        }
 
-       spin_unlock_bh(&txq->axq_lock);
+       tx_info->status.rates[0].count = tx_status->retries + 1;
 
-       return txq;
-}
+       hdrlen = ieee80211_get_hdrlen_from_skb(skb);
+       padsize = hdrlen & 3;
+       if (padsize && hdrlen >= 24) {
+               /*
+                * Remove MAC header padding before giving the frame back to
+                * mac80211.
+                */
+               memmove(skb->data + padsize, skb->data, hdrlen);
+               skb_pull(skb, padsize);
+       }
 
-/* Update parameters for a transmit queue */
+       ieee80211_tx_status(hw, skb);
+}
 
-int ath_txq_update(struct ath_softc *sc, int qnum,
-                  struct ath9k_tx_queue_info *qinfo)
+static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
+                               struct list_head *bf_q,
+                               int txok, int sendbar)
 {
-       struct ath_hal *ah = sc->sc_ah;
-       int error = 0;
-       struct ath9k_tx_queue_info qi;
+       struct sk_buff *skb = bf->bf_mpdu;
+       struct ath_xmit_status tx_status;
+       unsigned long flags;
 
-       if (qnum == sc->beacon.beaconq) {
-               /*
-                * XXX: for beacon queue, we just save the parameter.
-                * It will be picked up by ath_beaconq_config when
-                * it's necessary.
-                */
-               sc->beacon.beacon_qi = *qinfo;
-               return 0;
-       }
+       /*
+        * Set retry information.
+        * NB: Don't use the information in the descriptor, because the frame
+        * could be software retried.
+        */
+       tx_status.retries = bf->bf_retries;
+       tx_status.flags = 0;
 
-       ASSERT(sc->tx.txq[qnum].axq_qnum == qnum);
+       if (sendbar)
+               tx_status.flags = ATH_TX_BAR;
 
-       ath9k_hw_get_txq_props(ah, qnum, &qi);
-       qi.tqi_aifs = qinfo->tqi_aifs;
-       qi.tqi_cwmin = qinfo->tqi_cwmin;
-       qi.tqi_cwmax = qinfo->tqi_cwmax;
-       qi.tqi_burstTime = qinfo->tqi_burstTime;
-       qi.tqi_readyTime = qinfo->tqi_readyTime;
+       if (!txok) {
+               tx_status.flags |= ATH_TX_ERROR;
 
-       if (!ath9k_hw_set_txq_props(ah, qnum, &qi)) {
-               DPRINTF(sc, ATH_DBG_FATAL,
-                       "Unable to update hardware queue %u!\n", qnum);
-               error = -EIO;
-       } else {
-               ath9k_hw_resettxqueue(ah, qnum); /* push to h/w */
+               if (bf_isxretried(bf))
+                       tx_status.flags |= ATH_TX_XRETRY;
        }
 
-       return error;
+       dma_unmap_single(sc->dev, bf->bf_dmacontext, skb->len, DMA_TO_DEVICE);
+       ath_tx_complete(sc, skb, &tx_status);
+
+       /*
+        * Return the list of ath_buf of this mpdu to free queue
+        */
+       spin_lock_irqsave(&sc->tx.txbuflock, flags);
+       list_splice_tail_init(bf_q, &sc->tx.txbuf);
+       spin_unlock_irqrestore(&sc->tx.txbuflock, flags);
 }
 
-int ath_cabq_update(struct ath_softc *sc)
+static int ath_tx_num_badfrms(struct ath_softc *sc, struct ath_buf *bf,
+                             int txok)
 {
-       struct ath9k_tx_queue_info qi;
-       int qnum = sc->beacon.cabq->axq_qnum;
-       struct ath_beacon_config conf;
+       struct ath_buf *bf_last = bf->bf_lastbf;
+       struct ath_desc *ds = bf_last->bf_desc;
+       u16 seq_st = 0;
+       u32 ba[WME_BA_BMP_SIZE >> 5];
+       int ba_index;
+       int nbad = 0;
+       int isaggr = 0;
 
-       ath9k_hw_get_txq_props(sc->sc_ah, qnum, &qi);
-       /*
-        * Ensure the readytime % is within the bounds.
-        */
-       if (sc->sc_config.cabqReadytime < ATH9K_READY_TIME_LO_BOUND)
-               sc->sc_config.cabqReadytime = ATH9K_READY_TIME_LO_BOUND;
-       else if (sc->sc_config.cabqReadytime > ATH9K_READY_TIME_HI_BOUND)
-               sc->sc_config.cabqReadytime = ATH9K_READY_TIME_HI_BOUND;
+       if (ds->ds_txstat.ts_flags == ATH9K_TX_SW_ABORTED)
+               return 0;
 
-       ath_get_beaconconfig(sc, ATH_IF_ID_ANY, &conf);
-       qi.tqi_readyTime =
-               (conf.beacon_interval * sc->sc_config.cabqReadytime) / 100;
-       ath_txq_update(sc, qnum, &qi);
+       isaggr = bf_isaggr(bf);
+       if (isaggr) {
+               seq_st = ATH_DS_BA_SEQ(ds);
+               memcpy(ba, ATH_DS_BA_BITMAP(ds), WME_BA_BMP_SIZE >> 3);
+       }
 
-       return 0;
-}
+       while (bf) {
+               ba_index = ATH_BA_INDEX(seq_st, bf->bf_seqno);
+               if (!txok || (isaggr && !ATH_BA_ISSET(ba, ba_index)))
+                       nbad++;
+
+               bf = bf->bf_next;
+       }
 
-/* Deferred processing of transmit interrupt */
+       return nbad;
+}
 
-void ath_tx_tasklet(struct ath_softc *sc)
+static void ath_tx_rc_status(struct ath_buf *bf, struct ath_desc *ds, int nbad)
 {
-       int i;
-       u32 qcumask = ((1 << ATH9K_NUM_TX_QUEUES) - 1);
+       struct sk_buff *skb = (struct sk_buff *)bf->bf_mpdu;
+       struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
+       struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
 
-       ath9k_hw_gettxintrtxqs(sc->sc_ah, &qcumask);
+       tx_info_priv->update_rc = false;
+       if (ds->ds_txstat.ts_status & ATH9K_TXERR_FILT)
+               tx_info->flags |= IEEE80211_TX_STAT_TX_FILTERED;
 
-       /*
-        * Process each active queue.
-        */
-       for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
-               if (ATH_TXQ_SETUP(sc, i) && (qcumask & (1 << i)))
-                       ath_tx_processq(sc, &sc->tx.txq[i]);
+       if ((ds->ds_txstat.ts_status & ATH9K_TXERR_FILT) == 0 &&
+           (bf->bf_flags & ATH9K_TXDESC_NOACK) == 0) {
+               if (bf_isdata(bf)) {
+                       memcpy(&tx_info_priv->tx, &ds->ds_txstat,
+                              sizeof(tx_info_priv->tx));
+                       tx_info_priv->n_frames = bf->bf_nframes;
+                       tx_info_priv->n_bad_frames = nbad;
+                       tx_info_priv->update_rc = true;
+               }
        }
 }
 
-void ath_tx_draintxq(struct ath_softc *sc,
-       struct ath_txq *txq, bool retry_tx)
+static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
 {
-       struct ath_buf *bf, *lastbf;
+       struct ath_hal *ah = sc->sc_ah;
+       struct ath_buf *bf, *lastbf, *bf_held = NULL;
        struct list_head bf_head;
+       struct ath_desc *ds;
+       int txok, nbad = 0;
+       int status;
 
-       INIT_LIST_HEAD(&bf_head);
+       DPRINTF(sc, ATH_DBG_QUEUE, "tx queue %d (%x), link %p\n",
+               txq->axq_qnum, ath9k_hw_gettxbuf(sc->sc_ah, txq->axq_qnum),
+               txq->axq_link);
 
-       /*
-        * NB: this assumes output has been stopped and
-        *     we do not need to block ath_tx_tasklet
-        */
        for (;;) {
                spin_lock_bh(&txq->axq_lock);
-
                if (list_empty(&txq->axq_q)) {
                        txq->axq_link = NULL;
                        txq->axq_linkbuf = NULL;
                        spin_unlock_bh(&txq->axq_lock);
                        break;
                }
-
                bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
 
+               /*
+                * There is a race condition that a BH gets scheduled
+                * after sw writes TxE and before hw re-load the last
+                * descriptor to get the newly chained one.
+                * Software must keep the last DONE descriptor as a
+                * holding descriptor - software does so by marking
+                * it with the STALE flag.
+                */
+               bf_held = NULL;
                if (bf->bf_status & ATH_BUFSTATUS_STALE) {
-                       list_del(&bf->list);
-                       spin_unlock_bh(&txq->axq_lock);
-
-                       spin_lock_bh(&sc->tx.txbuflock);
-                       list_add_tail(&bf->list, &sc->tx.txbuf);
-                       spin_unlock_bh(&sc->tx.txbuflock);
-                       continue;
+                       bf_held = bf;
+                       if (list_is_last(&bf_held->list, &txq->axq_q)) {
+                               /* FIXME:
+                                * The holding descriptor is the last
+                                * descriptor in queue. It's safe to remove
+                                * the last holding descriptor in BH context.
+                                */
+                               spin_unlock_bh(&txq->axq_lock);
+                               break;
+                       } else {
+                               bf = list_entry(bf_held->list.next,
+                                       struct ath_buf, list);
+                       }
                }
 
                lastbf = bf->bf_lastbf;
-               if (!retry_tx)
-                       lastbf->bf_desc->ds_txstat.ts_flags =
-                               ATH9K_TX_SW_ABORTED;
-
-               /* remove ath_buf's of the same mpdu from txq */
-               list_cut_position(&bf_head, &txq->axq_q, &lastbf->list);
-               txq->axq_depth--;
-
-               spin_unlock_bh(&txq->axq_lock);
-
-               if (bf_isampdu(bf))
-                       ath_tx_complete_aggr_rifs(sc, txq, bf, &bf_head, 0);
-               else
-                       ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
-       }
+               ds = lastbf->bf_desc;
 
-       /* flush any pending frames if aggregation is enabled */
-       if (sc->sc_flags & SC_OP_TXAGGR) {
-               if (!retry_tx) {
-                       spin_lock_bh(&txq->axq_lock);
-                       ath_txq_drain_pending_buffers(sc, txq);
+               status = ath9k_hw_txprocdesc(ah, ds);
+               if (status == -EINPROGRESS) {
                        spin_unlock_bh(&txq->axq_lock);
+                       break;
                }
-       }
-}
-
-/* Drain the transmit queues and reclaim resources */
-
-void ath_draintxq(struct ath_softc *sc, bool retry_tx)
-{
-       /* stop beacon queue. The beacon will be freed when
-        * we go to INIT state */
-       if (!(sc->sc_flags & SC_OP_INVALID)) {
-               (void) ath9k_hw_stoptxdma(sc->sc_ah, sc->beacon.beaconq);
-               DPRINTF(sc, ATH_DBG_XMIT, "beacon queue %x\n",
-                       ath9k_hw_gettxbuf(sc->sc_ah, sc->beacon.beaconq));
-       }
-
-       ath_drain_txdataq(sc, retry_tx);
-}
-
-u32 ath_txq_depth(struct ath_softc *sc, int qnum)
-{
-       return sc->tx.txq[qnum].axq_depth;
-}
-
-u32 ath_txq_aggr_depth(struct ath_softc *sc, int qnum)
-{
-       return sc->tx.txq[qnum].axq_aggr_depth;
-}
-
-bool ath_tx_aggr_check(struct ath_softc *sc, struct ath_node *an, u8 tidno)
-{
-       struct ath_atx_tid *txtid;
-
-       if (!(sc->sc_flags & SC_OP_TXAGGR))
-               return false;
-
-       txtid = ATH_AN_2_TID(an, tidno);
-
-       if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
-               if (!(txtid->state & AGGR_ADDBA_PROGRESS) &&
-                   (txtid->addba_exchangeattempts < ADDBA_EXCHANGE_ATTEMPTS)) {
-                       txtid->addba_exchangeattempts++;
-                       return true;
-               }
-       }
+               if (bf->bf_desc == txq->axq_lastdsWithCTS)
+                       txq->axq_lastdsWithCTS = NULL;
+               if (ds == txq->axq_gatingds)
+                       txq->axq_gatingds = NULL;
 
-       return false;
-}
+               /*
+                * Remove ath_buf's of the same transmit unit from txq,
+                * however leave the last descriptor back as the holding
+                * descriptor for hw.
+                */
+               lastbf->bf_status |= ATH_BUFSTATUS_STALE;
+               INIT_LIST_HEAD(&bf_head);
 
-/* Start TX aggregation */
+               if (!list_is_singular(&lastbf->list))
+                       list_cut_position(&bf_head,
+                               &txq->axq_q, lastbf->list.prev);
 
-int ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
-                     u16 tid, u16 *ssn)
-{
-       struct ath_atx_tid *txtid;
-       struct ath_node *an;
+               txq->axq_depth--;
 
-       an = (struct ath_node *)sta->drv_priv;
+               if (bf_isaggr(bf))
+                       txq->axq_aggr_depth--;
 
-       if (sc->sc_flags & SC_OP_TXAGGR) {
-               txtid = ATH_AN_2_TID(an, tid);
-               txtid->state |= AGGR_ADDBA_PROGRESS;
-               ath_tx_pause_tid(sc, txtid);
-       }
+               txok = (ds->ds_txstat.ts_status == 0);
 
-       return 0;
-}
+               spin_unlock_bh(&txq->axq_lock);
 
-/* Stop tx aggregation */
+               if (bf_held) {
+                       list_del(&bf_held->list);
+                       spin_lock_bh(&sc->tx.txbuflock);
+                       list_add_tail(&bf_held->list, &sc->tx.txbuf);
+                       spin_unlock_bh(&sc->tx.txbuflock);
+               }
 
-int ath_tx_aggr_stop(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
-{
-       struct ath_node *an = (struct ath_node *)sta->drv_priv;
+               if (!bf_isampdu(bf)) {
+                       /*
+                        * This frame is sent out as a single frame.
+                        * Use hardware retry status for this frame.
+                        */
+                       bf->bf_retries = ds->ds_txstat.ts_longretry;
+                       if (ds->ds_txstat.ts_status & ATH9K_TXERR_XRETRY)
+                               bf->bf_state.bf_type |= BUF_XRETRY;
+                       nbad = 0;
+               } else {
+                       nbad = ath_tx_num_badfrms(sc, bf, txok);
+               }
 
-       ath_tx_aggr_teardown(sc, an, tid);
-       return 0;
-}
+               ath_tx_rc_status(bf, ds, nbad);
 
-/* Resume tx aggregation */
+               if (bf_isampdu(bf))
+                       ath_tx_complete_aggr_rifs(sc, txq, bf, &bf_head, txok);
+               else
+                       ath_tx_complete_buf(sc, bf, &bf_head, txok, 0);
 
-void ath_tx_aggr_resume(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
-{
-       struct ath_atx_tid *txtid;
-       struct ath_node *an;
+               spin_lock_bh(&txq->axq_lock);
+               if (txq->stopped && ath_txq_depth(sc, txq->axq_qnum) <=
+                               (ATH_TXBUF - 20)) {
+                       int qnum;
+                       qnum = ath_get_mac80211_qnum(txq->axq_qnum, sc);
+                       if (qnum != -1) {
+                               ieee80211_wake_queue(sc->hw, qnum);
+                               txq->stopped = 0;
+                       }
 
-       an = (struct ath_node *)sta->drv_priv;
+               }
 
-       if (sc->sc_flags & SC_OP_TXAGGR) {
-               txtid = ATH_AN_2_TID(an, tid);
-               txtid->baw_size =
-                       IEEE80211_MIN_AMPDU_BUF << sta->ht_cap.ampdu_factor;
-               txtid->state |= AGGR_ADDBA_COMPLETE;
-               txtid->state &= ~AGGR_ADDBA_PROGRESS;
-               ath_tx_resume_tid(sc, txtid);
+               if (sc->sc_flags & SC_OP_TXAGGR)
+                       ath_txq_schedule(sc, txq);
+               spin_unlock_bh(&txq->axq_lock);
        }
 }
 
-/*
- * Performs transmit side cleanup when TID changes from aggregated to
- * unaggregated.
- * - Pause the TID and mark cleanup in progress
- * - Discard all retry frames from the s/w queue.
- */
 
-void ath_tx_aggr_teardown(struct ath_softc *sc, struct ath_node *an, u8 tid)
+void ath_tx_tasklet(struct ath_softc *sc)
 {
-       struct ath_atx_tid *txtid = ATH_AN_2_TID(an, tid);
-       struct ath_txq *txq = &sc->tx.txq[txtid->ac->qnum];
-       struct ath_buf *bf;
-       struct list_head bf_head;
-       INIT_LIST_HEAD(&bf_head);
-
-       if (txtid->state & AGGR_CLEANUP) /* cleanup is in progress */
-               return;
-
-       if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
-               txtid->addba_exchangeattempts = 0;
-               return;
-       }
-
-       /* TID must be paused first */
-       ath_tx_pause_tid(sc, txtid);
-
-       /* drop all software retried frames and mark this TID */
-       spin_lock_bh(&txq->axq_lock);
-       while (!list_empty(&txtid->buf_q)) {
-               bf = list_first_entry(&txtid->buf_q, struct ath_buf, list);
-               if (!bf_isretried(bf)) {
-                       /*
-                        * NB: it's based on the assumption that
-                        * software retried frame will always stay
-                        * at the head of software queue.
-                        */
-                       break;
-               }
-               list_cut_position(&bf_head,
-                       &txtid->buf_q, &bf->bf_lastfrm->list);
-               ath_tx_update_baw(sc, txtid, bf->bf_seqno);
+       int i;
+       u32 qcumask = ((1 << ATH9K_NUM_TX_QUEUES) - 1);
 
-               /* complete this sub-frame */
-               ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
-       }
+       ath9k_hw_gettxintrtxqs(sc->sc_ah, &qcumask);
 
-       if (txtid->baw_head != txtid->baw_tail) {
-               spin_unlock_bh(&txq->axq_lock);
-               txtid->state |= AGGR_CLEANUP;
-       } else {
-               txtid->state &= ~AGGR_ADDBA_COMPLETE;
-               txtid->addba_exchangeattempts = 0;
-               spin_unlock_bh(&txq->axq_lock);
-               ath_tx_flush_tid(sc, txtid);
+       for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
+               if (ATH_TXQ_SETUP(sc, i) && (qcumask & (1 << i)))
+                       ath_tx_processq(sc, &sc->tx.txq[i]);
        }
 }
 
-/*
- * Tx scheduling logic
- * NB: must be called with txq lock held
- */
+/*****************/
+/* Init, Cleanup */
+/*****************/
 
-void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq)
+int ath_tx_init(struct ath_softc *sc, int nbufs)
 {
-       struct ath_atx_ac *ac;
-       struct ath_atx_tid *tid;
-
-       /* nothing to schedule */
-       if (list_empty(&txq->axq_acq))
-               return;
-       /*
-        * get the first node/ac pair on the queue
-        */
-       ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac, list);
-       list_del(&ac->list);
-       ac->sched = false;
+       int error = 0;
 
-       /*
-        * process a single tid per destination
-        */
        do {
-               /* nothing to schedule */
-               if (list_empty(&ac->tid_q))
-                       return;
-
-               tid = list_first_entry(&ac->tid_q, struct ath_atx_tid, list);
-               list_del(&tid->list);
-               tid->sched = false;
+               spin_lock_init(&sc->tx.txbuflock);
 
-               if (tid->paused)    /* check next tid to keep h/w busy */
-                       continue;
+               error = ath_descdma_setup(sc, &sc->tx.txdma, &sc->tx.txbuf,
+                       "tx", nbufs, 1);
+               if (error != 0) {
+                       DPRINTF(sc, ATH_DBG_FATAL,
+                               "Failed to allocate tx descriptors: %d\n",
+                               error);
+                       break;
+               }
 
-               if ((txq->axq_depth % 2) == 0)
-                       ath_tx_sched_aggr(sc, txq, tid);
+               error = ath_descdma_setup(sc, &sc->beacon.bdma, &sc->beacon.bbuf,
+                                         "beacon", ATH_BCBUF, 1);
+               if (error != 0) {
+                       DPRINTF(sc, ATH_DBG_FATAL,
+                               "Failed to allocate beacon descriptors: %d\n",
+                               error);
+                       break;
+               }
 
-               /*
-                * add tid to round-robin queue if more frames
-                * are pending for the tid
-                */
-               if (!list_empty(&tid->buf_q))
-                       ath_tx_queue_tid(txq, tid);
+       } while (0);
 
-               /* only schedule one TID at a time */
-               break;
-       } while (!list_empty(&ac->tid_q));
+       if (error != 0)
+               ath_tx_cleanup(sc);
 
-       /*
-        * schedule AC if more TIDs need processing
-        */
-       if (!list_empty(&ac->tid_q)) {
-               /*
-                * add dest ac to txq if not already added
-                */
-               if (!ac->sched) {
-                       ac->sched = true;
-                       list_add_tail(&ac->list, &txq->axq_acq);
-               }
-       }
+       return error;
 }
 
-/* Initialize per-node transmit state */
+int ath_tx_cleanup(struct ath_softc *sc)
+{
+       if (sc->beacon.bdma.dd_desc_len != 0)
+               ath_descdma_cleanup(sc, &sc->beacon.bdma, &sc->beacon.bbuf);
+
+       if (sc->tx.txdma.dd_desc_len != 0)
+               ath_descdma_cleanup(sc, &sc->tx.txdma, &sc->tx.txbuf);
+
+       return 0;
+}
 
 void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
 {
@@ -2412,9 +2232,6 @@ void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
        struct ath_atx_ac *ac;
        int tidno, acno;
 
-       /*
-        * Init per tid tx state
-        */
        for (tidno = 0, tid = &an->tid[tidno];
             tidno < WME_NUM_TID;
             tidno++, tid++) {
@@ -2424,22 +2241,16 @@ void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
                tid->baw_size  = WME_MAX_BA;
                tid->baw_head  = tid->baw_tail = 0;
                tid->sched     = false;
-               tid->paused = false;
+               tid->paused    = false;
                tid->state &= ~AGGR_CLEANUP;
                INIT_LIST_HEAD(&tid->buf_q);
-
                acno = TID_TO_WME_AC(tidno);
                tid->ac = &an->ac[acno];
-
-               /* ADDBA state */
                tid->state &= ~AGGR_ADDBA_COMPLETE;
                tid->state &= ~AGGR_ADDBA_PROGRESS;
                tid->addba_exchangeattempts = 0;
        }
 
-       /*
-        * Init per ac tx state
-        */
        for (acno = 0, ac = &an->ac[acno];
             acno < WME_NUM_AC; acno++, ac++) {
                ac->sched    = false;
@@ -2466,14 +2277,13 @@ void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
        }
 }
 
-/* Cleanupthe pending buffers for the node. */
-
 void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an)
 {
        int i;
        struct ath_atx_ac *ac, *ac_tmp;
        struct ath_atx_tid *tid, *tid_tmp;
        struct ath_txq *txq;
+
        for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
                if (ATH_TXQ_SETUP(sc, i)) {
                        txq = &sc->tx.txq[i];
@@ -2504,51 +2314,3 @@ void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an)
                }
        }
 }
-
-void ath_tx_cabq(struct ath_softc *sc, struct sk_buff *skb)
-{
-       int hdrlen, padsize;
-       struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
-       struct ath_tx_control txctl;
-
-       memset(&txctl, 0, sizeof(struct ath_tx_control));
-
-       /*
-        * As a temporary workaround, assign seq# here; this will likely need
-        * to be cleaned up to work better with Beacon transmission and virtual
-        * BSSes.
-        */
-       if (info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ) {
-               struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
-               if (info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT)
-                       sc->tx.seq_no += 0x10;
-               hdr->seq_ctrl &= cpu_to_le16(IEEE80211_SCTL_FRAG);
-               hdr->seq_ctrl |= cpu_to_le16(sc->tx.seq_no);
-       }
-
-       /* Add the padding after the header if this is not already done */
-       hdrlen = ieee80211_get_hdrlen_from_skb(skb);
-       if (hdrlen & 3) {
-               padsize = hdrlen % 4;
-               if (skb_headroom(skb) < padsize) {
-                       DPRINTF(sc, ATH_DBG_XMIT, "TX CABQ padding failed\n");
-                       dev_kfree_skb_any(skb);
-                       return;
-               }
-               skb_push(skb, padsize);
-               memmove(skb->data, skb->data + padsize, hdrlen);
-       }
-
-       txctl.txq = sc->beacon.cabq;
-
-       DPRINTF(sc, ATH_DBG_XMIT, "transmitting CABQ packet, skb: %p\n", skb);
-
-       if (ath_tx_start(sc, skb, &txctl) != 0) {
-               DPRINTF(sc, ATH_DBG_XMIT, "CABQ TX failed\n");
-               goto exit;
-       }
-
-       return;
-exit:
-       dev_kfree_skb_any(skb);
-}