xref: /linux/drivers/net/wireless/ath/ath9k/xmit.c (revision 2277ab4a1df50e05bc732fe9488d4e902bb8399a)
1 /*
2  * Copyright (c) 2008-2009 Atheros Communications Inc.
3  *
4  * Permission to use, copy, modify, and/or distribute this software for any
5  * purpose with or without fee is hereby granted, provided that the above
6  * copyright notice and this permission notice appear in all copies.
7  *
8  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
15  */
16 
17 #include "ath9k.h"
18 
19 #define BITS_PER_BYTE           8
20 #define OFDM_PLCP_BITS          22
21 #define HT_RC_2_MCS(_rc)        ((_rc) & 0x0f)
22 #define HT_RC_2_STREAMS(_rc)    ((((_rc) & 0x78) >> 3) + 1)
23 #define L_STF                   8
24 #define L_LTF                   8
25 #define L_SIG                   4
26 #define HT_SIG                  8
27 #define HT_STF                  4
28 #define HT_LTF(_ns)             (4 * (_ns))
29 #define SYMBOL_TIME(_ns)        ((_ns) << 2) /* ns * 4 us */
30 #define SYMBOL_TIME_HALFGI(_ns) (((_ns) * 18 + 4) / 5)  /* ns * 3.6 us */
31 #define NUM_SYMBOLS_PER_USEC(_usec) (_usec >> 2)
32 #define NUM_SYMBOLS_PER_USEC_HALFGI(_usec) (((_usec*5)-4)/18)
33 
34 #define OFDM_SIFS_TIME    	    16
35 
36 static u32 bits_per_symbol[][2] = {
37 	/* 20MHz 40MHz */
38 	{    26,   54 },     /*  0: BPSK */
39 	{    52,  108 },     /*  1: QPSK 1/2 */
40 	{    78,  162 },     /*  2: QPSK 3/4 */
41 	{   104,  216 },     /*  3: 16-QAM 1/2 */
42 	{   156,  324 },     /*  4: 16-QAM 3/4 */
43 	{   208,  432 },     /*  5: 64-QAM 2/3 */
44 	{   234,  486 },     /*  6: 64-QAM 3/4 */
45 	{   260,  540 },     /*  7: 64-QAM 5/6 */
46 	{    52,  108 },     /*  8: BPSK */
47 	{   104,  216 },     /*  9: QPSK 1/2 */
48 	{   156,  324 },     /* 10: QPSK 3/4 */
49 	{   208,  432 },     /* 11: 16-QAM 1/2 */
50 	{   312,  648 },     /* 12: 16-QAM 3/4 */
51 	{   416,  864 },     /* 13: 64-QAM 2/3 */
52 	{   468,  972 },     /* 14: 64-QAM 3/4 */
53 	{   520, 1080 },     /* 15: 64-QAM 5/6 */
54 };
55 
56 #define IS_HT_RATE(_rate)     ((_rate) & 0x80)
57 
58 static void ath_tx_send_ht_normal(struct ath_softc *sc, struct ath_txq *txq,
59 				  struct ath_atx_tid *tid,
60 				  struct list_head *bf_head);
61 static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
62 				struct list_head *bf_q,
63 				int txok, int sendbar);
64 static void ath_tx_txqaddbuf(struct ath_softc *sc, struct ath_txq *txq,
65 			     struct list_head *head);
66 static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf);
67 static int ath_tx_num_badfrms(struct ath_softc *sc, struct ath_buf *bf,
68 			      int txok);
69 static void ath_tx_rc_status(struct ath_buf *bf, struct ath_desc *ds,
70 			     int nbad, int txok, bool update_rc);
71 
72 /*********************/
73 /* Aggregation logic */
74 /*********************/
75 
76 static int ath_aggr_query(struct ath_softc *sc, struct ath_node *an, u8 tidno)
77 {
78 	struct ath_atx_tid *tid;
79 	tid = ATH_AN_2_TID(an, tidno);
80 
81 	if (tid->state & AGGR_ADDBA_COMPLETE ||
82 	    tid->state & AGGR_ADDBA_PROGRESS)
83 		return 1;
84 	else
85 		return 0;
86 }
87 
88 static void ath_tx_queue_tid(struct ath_txq *txq, struct ath_atx_tid *tid)
89 {
90 	struct ath_atx_ac *ac = tid->ac;
91 
92 	if (tid->paused)
93 		return;
94 
95 	if (tid->sched)
96 		return;
97 
98 	tid->sched = true;
99 	list_add_tail(&tid->list, &ac->tid_q);
100 
101 	if (ac->sched)
102 		return;
103 
104 	ac->sched = true;
105 	list_add_tail(&ac->list, &txq->axq_acq);
106 }
107 
108 static void ath_tx_pause_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
109 {
110 	struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
111 
112 	spin_lock_bh(&txq->axq_lock);
113 	tid->paused++;
114 	spin_unlock_bh(&txq->axq_lock);
115 }
116 
117 static void ath_tx_resume_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
118 {
119 	struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
120 
121 	ASSERT(tid->paused > 0);
122 	spin_lock_bh(&txq->axq_lock);
123 
124 	tid->paused--;
125 
126 	if (tid->paused > 0)
127 		goto unlock;
128 
129 	if (list_empty(&tid->buf_q))
130 		goto unlock;
131 
132 	ath_tx_queue_tid(txq, tid);
133 	ath_txq_schedule(sc, txq);
134 unlock:
135 	spin_unlock_bh(&txq->axq_lock);
136 }
137 
138 static void ath_tx_flush_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
139 {
140 	struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
141 	struct ath_buf *bf;
142 	struct list_head bf_head;
143 	INIT_LIST_HEAD(&bf_head);
144 
145 	ASSERT(tid->paused > 0);
146 	spin_lock_bh(&txq->axq_lock);
147 
148 	tid->paused--;
149 
150 	if (tid->paused > 0) {
151 		spin_unlock_bh(&txq->axq_lock);
152 		return;
153 	}
154 
155 	while (!list_empty(&tid->buf_q)) {
156 		bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
157 		ASSERT(!bf_isretried(bf));
158 		list_move_tail(&bf->list, &bf_head);
159 		ath_tx_send_ht_normal(sc, txq, tid, &bf_head);
160 	}
161 
162 	spin_unlock_bh(&txq->axq_lock);
163 }
164 
165 static void ath_tx_update_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
166 			      int seqno)
167 {
168 	int index, cindex;
169 
170 	index  = ATH_BA_INDEX(tid->seq_start, seqno);
171 	cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
172 
173 	tid->tx_buf[cindex] = NULL;
174 
175 	while (tid->baw_head != tid->baw_tail && !tid->tx_buf[tid->baw_head]) {
176 		INCR(tid->seq_start, IEEE80211_SEQ_MAX);
177 		INCR(tid->baw_head, ATH_TID_MAX_BUFS);
178 	}
179 }
180 
181 static void ath_tx_addto_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
182 			     struct ath_buf *bf)
183 {
184 	int index, cindex;
185 
186 	if (bf_isretried(bf))
187 		return;
188 
189 	index  = ATH_BA_INDEX(tid->seq_start, bf->bf_seqno);
190 	cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
191 
192 	ASSERT(tid->tx_buf[cindex] == NULL);
193 	tid->tx_buf[cindex] = bf;
194 
195 	if (index >= ((tid->baw_tail - tid->baw_head) &
196 		(ATH_TID_MAX_BUFS - 1))) {
197 		tid->baw_tail = cindex;
198 		INCR(tid->baw_tail, ATH_TID_MAX_BUFS);
199 	}
200 }
201 
202 /*
203  * TODO: For frame(s) that are in the retry state, we will reuse the
204  * sequence number(s) without setting the retry bit. The
205  * alternative is to give up on these and BAR the receiver's window
206  * forward.
207  */
208 static void ath_tid_drain(struct ath_softc *sc, struct ath_txq *txq,
209 			  struct ath_atx_tid *tid)
210 
211 {
212 	struct ath_buf *bf;
213 	struct list_head bf_head;
214 	INIT_LIST_HEAD(&bf_head);
215 
216 	for (;;) {
217 		if (list_empty(&tid->buf_q))
218 			break;
219 
220 		bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
221 		list_move_tail(&bf->list, &bf_head);
222 
223 		if (bf_isretried(bf))
224 			ath_tx_update_baw(sc, tid, bf->bf_seqno);
225 
226 		spin_unlock(&txq->axq_lock);
227 		ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
228 		spin_lock(&txq->axq_lock);
229 	}
230 
231 	tid->seq_next = tid->seq_start;
232 	tid->baw_tail = tid->baw_head;
233 }
234 
235 static void ath_tx_set_retry(struct ath_softc *sc, struct ath_buf *bf)
236 {
237 	struct sk_buff *skb;
238 	struct ieee80211_hdr *hdr;
239 
240 	bf->bf_state.bf_type |= BUF_RETRY;
241 	bf->bf_retries++;
242 
243 	skb = bf->bf_mpdu;
244 	hdr = (struct ieee80211_hdr *)skb->data;
245 	hdr->frame_control |= cpu_to_le16(IEEE80211_FCTL_RETRY);
246 }
247 
248 static struct ath_buf* ath_clone_txbuf(struct ath_softc *sc, struct ath_buf *bf)
249 {
250 	struct ath_buf *tbf;
251 
252 	spin_lock_bh(&sc->tx.txbuflock);
253 	ASSERT(!list_empty((&sc->tx.txbuf)));
254 	tbf = list_first_entry(&sc->tx.txbuf, struct ath_buf, list);
255 	list_del(&tbf->list);
256 	spin_unlock_bh(&sc->tx.txbuflock);
257 
258 	ATH_TXBUF_RESET(tbf);
259 
260 	tbf->bf_mpdu = bf->bf_mpdu;
261 	tbf->bf_buf_addr = bf->bf_buf_addr;
262 	*(tbf->bf_desc) = *(bf->bf_desc);
263 	tbf->bf_state = bf->bf_state;
264 	tbf->bf_dmacontext = bf->bf_dmacontext;
265 
266 	return tbf;
267 }
268 
269 static void ath_tx_complete_aggr(struct ath_softc *sc, struct ath_txq *txq,
270 				 struct ath_buf *bf, struct list_head *bf_q,
271 				 int txok)
272 {
273 	struct ath_node *an = NULL;
274 	struct sk_buff *skb;
275 	struct ieee80211_sta *sta;
276 	struct ieee80211_hdr *hdr;
277 	struct ath_atx_tid *tid = NULL;
278 	struct ath_buf *bf_next, *bf_last = bf->bf_lastbf;
279 	struct ath_desc *ds = bf_last->bf_desc;
280 	struct list_head bf_head, bf_pending;
281 	u16 seq_st = 0, acked_cnt = 0, txfail_cnt = 0;
282 	u32 ba[WME_BA_BMP_SIZE >> 5];
283 	int isaggr, txfail, txpending, sendbar = 0, needreset = 0, nbad = 0;
284 	bool rc_update = true;
285 
286 	skb = bf->bf_mpdu;
287 	hdr = (struct ieee80211_hdr *)skb->data;
288 
289 	rcu_read_lock();
290 
291 	sta = ieee80211_find_sta(sc->hw, hdr->addr1);
292 	if (!sta) {
293 		rcu_read_unlock();
294 		return;
295 	}
296 
297 	an = (struct ath_node *)sta->drv_priv;
298 	tid = ATH_AN_2_TID(an, bf->bf_tidno);
299 
300 	isaggr = bf_isaggr(bf);
301 	memset(ba, 0, WME_BA_BMP_SIZE >> 3);
302 
303 	if (isaggr && txok) {
304 		if (ATH_DS_TX_BA(ds)) {
305 			seq_st = ATH_DS_BA_SEQ(ds);
306 			memcpy(ba, ATH_DS_BA_BITMAP(ds),
307 			       WME_BA_BMP_SIZE >> 3);
308 		} else {
309 			/*
310 			 * AR5416 can become deaf/mute when BA
311 			 * issue happens. Chip needs to be reset.
312 			 * But AP code may have sychronization issues
313 			 * when perform internal reset in this routine.
314 			 * Only enable reset in STA mode for now.
315 			 */
316 			if (sc->sc_ah->opmode == NL80211_IFTYPE_STATION)
317 				needreset = 1;
318 		}
319 	}
320 
321 	INIT_LIST_HEAD(&bf_pending);
322 	INIT_LIST_HEAD(&bf_head);
323 
324 	nbad = ath_tx_num_badfrms(sc, bf, txok);
325 	while (bf) {
326 		txfail = txpending = 0;
327 		bf_next = bf->bf_next;
328 
329 		if (ATH_BA_ISSET(ba, ATH_BA_INDEX(seq_st, bf->bf_seqno))) {
330 			/* transmit completion, subframe is
331 			 * acked by block ack */
332 			acked_cnt++;
333 		} else if (!isaggr && txok) {
334 			/* transmit completion */
335 			acked_cnt++;
336 		} else {
337 			if (!(tid->state & AGGR_CLEANUP) &&
338 			    ds->ds_txstat.ts_flags != ATH9K_TX_SW_ABORTED) {
339 				if (bf->bf_retries < ATH_MAX_SW_RETRIES) {
340 					ath_tx_set_retry(sc, bf);
341 					txpending = 1;
342 				} else {
343 					bf->bf_state.bf_type |= BUF_XRETRY;
344 					txfail = 1;
345 					sendbar = 1;
346 					txfail_cnt++;
347 				}
348 			} else {
349 				/*
350 				 * cleanup in progress, just fail
351 				 * the un-acked sub-frames
352 				 */
353 				txfail = 1;
354 			}
355 		}
356 
357 		if (bf_next == NULL) {
358 			/*
359 			 * Make sure the last desc is reclaimed if it
360 			 * not a holding desc.
361 			 */
362 			if (!bf_last->bf_stale)
363 				list_move_tail(&bf->list, &bf_head);
364 			else
365 				INIT_LIST_HEAD(&bf_head);
366 		} else {
367 			ASSERT(!list_empty(bf_q));
368 			list_move_tail(&bf->list, &bf_head);
369 		}
370 
371 		if (!txpending) {
372 			/*
373 			 * complete the acked-ones/xretried ones; update
374 			 * block-ack window
375 			 */
376 			spin_lock_bh(&txq->axq_lock);
377 			ath_tx_update_baw(sc, tid, bf->bf_seqno);
378 			spin_unlock_bh(&txq->axq_lock);
379 
380 			if (rc_update && (acked_cnt == 1 || txfail_cnt == 1)) {
381 				ath_tx_rc_status(bf, ds, nbad, txok, true);
382 				rc_update = false;
383 			} else {
384 				ath_tx_rc_status(bf, ds, nbad, txok, false);
385 			}
386 
387 			ath_tx_complete_buf(sc, bf, &bf_head, !txfail, sendbar);
388 		} else {
389 			/* retry the un-acked ones */
390 			if (bf->bf_next == NULL && bf_last->bf_stale) {
391 				struct ath_buf *tbf;
392 
393 				tbf = ath_clone_txbuf(sc, bf_last);
394 				ath9k_hw_cleartxdesc(sc->sc_ah, tbf->bf_desc);
395 				list_add_tail(&tbf->list, &bf_head);
396 			} else {
397 				/*
398 				 * Clear descriptor status words for
399 				 * software retry
400 				 */
401 				ath9k_hw_cleartxdesc(sc->sc_ah, bf->bf_desc);
402 			}
403 
404 			/*
405 			 * Put this buffer to the temporary pending
406 			 * queue to retain ordering
407 			 */
408 			list_splice_tail_init(&bf_head, &bf_pending);
409 		}
410 
411 		bf = bf_next;
412 	}
413 
414 	if (tid->state & AGGR_CLEANUP) {
415 		if (tid->baw_head == tid->baw_tail) {
416 			tid->state &= ~AGGR_ADDBA_COMPLETE;
417 			tid->addba_exchangeattempts = 0;
418 			tid->state &= ~AGGR_CLEANUP;
419 
420 			/* send buffered frames as singles */
421 			ath_tx_flush_tid(sc, tid);
422 		}
423 		rcu_read_unlock();
424 		return;
425 	}
426 
427 	/* prepend un-acked frames to the beginning of the pending frame queue */
428 	if (!list_empty(&bf_pending)) {
429 		spin_lock_bh(&txq->axq_lock);
430 		list_splice(&bf_pending, &tid->buf_q);
431 		ath_tx_queue_tid(txq, tid);
432 		spin_unlock_bh(&txq->axq_lock);
433 	}
434 
435 	rcu_read_unlock();
436 
437 	if (needreset)
438 		ath_reset(sc, false);
439 }
440 
441 static u32 ath_lookup_rate(struct ath_softc *sc, struct ath_buf *bf,
442 			   struct ath_atx_tid *tid)
443 {
444 	const struct ath_rate_table *rate_table = sc->cur_rate_table;
445 	struct sk_buff *skb;
446 	struct ieee80211_tx_info *tx_info;
447 	struct ieee80211_tx_rate *rates;
448 	struct ath_tx_info_priv *tx_info_priv;
449 	u32 max_4ms_framelen, frmlen;
450 	u16 aggr_limit, legacy = 0, maxampdu;
451 	int i;
452 
453 	skb = bf->bf_mpdu;
454 	tx_info = IEEE80211_SKB_CB(skb);
455 	rates = tx_info->control.rates;
456 	tx_info_priv = (struct ath_tx_info_priv *)tx_info->rate_driver_data[0];
457 
458 	/*
459 	 * Find the lowest frame length among the rate series that will have a
460 	 * 4ms transmit duration.
461 	 * TODO - TXOP limit needs to be considered.
462 	 */
463 	max_4ms_framelen = ATH_AMPDU_LIMIT_MAX;
464 
465 	for (i = 0; i < 4; i++) {
466 		if (rates[i].count) {
467 			if (!WLAN_RC_PHY_HT(rate_table->info[rates[i].idx].phy)) {
468 				legacy = 1;
469 				break;
470 			}
471 
472 			frmlen = rate_table->info[rates[i].idx].max_4ms_framelen;
473 			max_4ms_framelen = min(max_4ms_framelen, frmlen);
474 		}
475 	}
476 
477 	/*
478 	 * limit aggregate size by the minimum rate if rate selected is
479 	 * not a probe rate, if rate selected is a probe rate then
480 	 * avoid aggregation of this packet.
481 	 */
482 	if (tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE || legacy)
483 		return 0;
484 
485 	aggr_limit = min(max_4ms_framelen, (u32)ATH_AMPDU_LIMIT_DEFAULT);
486 
487 	/*
488 	 * h/w can accept aggregates upto 16 bit lengths (65535).
489 	 * The IE, however can hold upto 65536, which shows up here
490 	 * as zero. Ignore 65536 since we  are constrained by hw.
491 	 */
492 	maxampdu = tid->an->maxampdu;
493 	if (maxampdu)
494 		aggr_limit = min(aggr_limit, maxampdu);
495 
496 	return aggr_limit;
497 }
498 
499 /*
500  * Returns the number of delimiters to be added to
501  * meet the minimum required mpdudensity.
502  * caller should make sure that the rate is HT rate .
503  */
504 static int ath_compute_num_delims(struct ath_softc *sc, struct ath_atx_tid *tid,
505 				  struct ath_buf *bf, u16 frmlen)
506 {
507 	const struct ath_rate_table *rt = sc->cur_rate_table;
508 	struct sk_buff *skb = bf->bf_mpdu;
509 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
510 	u32 nsymbits, nsymbols, mpdudensity;
511 	u16 minlen;
512 	u8 rc, flags, rix;
513 	int width, half_gi, ndelim, mindelim;
514 
515 	/* Select standard number of delimiters based on frame length alone */
516 	ndelim = ATH_AGGR_GET_NDELIM(frmlen);
517 
518 	/*
519 	 * If encryption enabled, hardware requires some more padding between
520 	 * subframes.
521 	 * TODO - this could be improved to be dependent on the rate.
522 	 *      The hardware can keep up at lower rates, but not higher rates
523 	 */
524 	if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR)
525 		ndelim += ATH_AGGR_ENCRYPTDELIM;
526 
527 	/*
528 	 * Convert desired mpdu density from microeconds to bytes based
529 	 * on highest rate in rate series (i.e. first rate) to determine
530 	 * required minimum length for subframe. Take into account
531 	 * whether high rate is 20 or 40Mhz and half or full GI.
532 	 */
533 	mpdudensity = tid->an->mpdudensity;
534 
535 	/*
536 	 * If there is no mpdu density restriction, no further calculation
537 	 * is needed.
538 	 */
539 	if (mpdudensity == 0)
540 		return ndelim;
541 
542 	rix = tx_info->control.rates[0].idx;
543 	flags = tx_info->control.rates[0].flags;
544 	rc = rt->info[rix].ratecode;
545 	width = (flags & IEEE80211_TX_RC_40_MHZ_WIDTH) ? 1 : 0;
546 	half_gi = (flags & IEEE80211_TX_RC_SHORT_GI) ? 1 : 0;
547 
548 	if (half_gi)
549 		nsymbols = NUM_SYMBOLS_PER_USEC_HALFGI(mpdudensity);
550 	else
551 		nsymbols = NUM_SYMBOLS_PER_USEC(mpdudensity);
552 
553 	if (nsymbols == 0)
554 		nsymbols = 1;
555 
556 	nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
557 	minlen = (nsymbols * nsymbits) / BITS_PER_BYTE;
558 
559 	if (frmlen < minlen) {
560 		mindelim = (minlen - frmlen) / ATH_AGGR_DELIM_SZ;
561 		ndelim = max(mindelim, ndelim);
562 	}
563 
564 	return ndelim;
565 }
566 
567 static enum ATH_AGGR_STATUS ath_tx_form_aggr(struct ath_softc *sc,
568 					     struct ath_atx_tid *tid,
569 					     struct list_head *bf_q)
570 {
571 #define PADBYTES(_len) ((4 - ((_len) % 4)) % 4)
572 	struct ath_buf *bf, *bf_first, *bf_prev = NULL;
573 	int rl = 0, nframes = 0, ndelim, prev_al = 0;
574 	u16 aggr_limit = 0, al = 0, bpad = 0,
575 		al_delta, h_baw = tid->baw_size / 2;
576 	enum ATH_AGGR_STATUS status = ATH_AGGR_DONE;
577 
578 	bf_first = list_first_entry(&tid->buf_q, struct ath_buf, list);
579 
580 	do {
581 		bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
582 
583 		/* do not step over block-ack window */
584 		if (!BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno)) {
585 			status = ATH_AGGR_BAW_CLOSED;
586 			break;
587 		}
588 
589 		if (!rl) {
590 			aggr_limit = ath_lookup_rate(sc, bf, tid);
591 			rl = 1;
592 		}
593 
594 		/* do not exceed aggregation limit */
595 		al_delta = ATH_AGGR_DELIM_SZ + bf->bf_frmlen;
596 
597 		if (nframes &&
598 		    (aggr_limit < (al + bpad + al_delta + prev_al))) {
599 			status = ATH_AGGR_LIMITED;
600 			break;
601 		}
602 
603 		/* do not exceed subframe limit */
604 		if (nframes >= min((int)h_baw, ATH_AMPDU_SUBFRAME_DEFAULT)) {
605 			status = ATH_AGGR_LIMITED;
606 			break;
607 		}
608 		nframes++;
609 
610 		/* add padding for previous frame to aggregation length */
611 		al += bpad + al_delta;
612 
613 		/*
614 		 * Get the delimiters needed to meet the MPDU
615 		 * density for this node.
616 		 */
617 		ndelim = ath_compute_num_delims(sc, tid, bf_first, bf->bf_frmlen);
618 		bpad = PADBYTES(al_delta) + (ndelim << 2);
619 
620 		bf->bf_next = NULL;
621 		bf->bf_desc->ds_link = 0;
622 
623 		/* link buffers of this frame to the aggregate */
624 		ath_tx_addto_baw(sc, tid, bf);
625 		ath9k_hw_set11n_aggr_middle(sc->sc_ah, bf->bf_desc, ndelim);
626 		list_move_tail(&bf->list, bf_q);
627 		if (bf_prev) {
628 			bf_prev->bf_next = bf;
629 			bf_prev->bf_desc->ds_link = bf->bf_daddr;
630 		}
631 		bf_prev = bf;
632 	} while (!list_empty(&tid->buf_q));
633 
634 	bf_first->bf_al = al;
635 	bf_first->bf_nframes = nframes;
636 
637 	return status;
638 #undef PADBYTES
639 }
640 
641 static void ath_tx_sched_aggr(struct ath_softc *sc, struct ath_txq *txq,
642 			      struct ath_atx_tid *tid)
643 {
644 	struct ath_buf *bf;
645 	enum ATH_AGGR_STATUS status;
646 	struct list_head bf_q;
647 
648 	do {
649 		if (list_empty(&tid->buf_q))
650 			return;
651 
652 		INIT_LIST_HEAD(&bf_q);
653 
654 		status = ath_tx_form_aggr(sc, tid, &bf_q);
655 
656 		/*
657 		 * no frames picked up to be aggregated;
658 		 * block-ack window is not open.
659 		 */
660 		if (list_empty(&bf_q))
661 			break;
662 
663 		bf = list_first_entry(&bf_q, struct ath_buf, list);
664 		bf->bf_lastbf = list_entry(bf_q.prev, struct ath_buf, list);
665 
666 		/* if only one frame, send as non-aggregate */
667 		if (bf->bf_nframes == 1) {
668 			bf->bf_state.bf_type &= ~BUF_AGGR;
669 			ath9k_hw_clr11n_aggr(sc->sc_ah, bf->bf_desc);
670 			ath_buf_set_rate(sc, bf);
671 			ath_tx_txqaddbuf(sc, txq, &bf_q);
672 			continue;
673 		}
674 
675 		/* setup first desc of aggregate */
676 		bf->bf_state.bf_type |= BUF_AGGR;
677 		ath_buf_set_rate(sc, bf);
678 		ath9k_hw_set11n_aggr_first(sc->sc_ah, bf->bf_desc, bf->bf_al);
679 
680 		/* anchor last desc of aggregate */
681 		ath9k_hw_set11n_aggr_last(sc->sc_ah, bf->bf_lastbf->bf_desc);
682 
683 		txq->axq_aggr_depth++;
684 		ath_tx_txqaddbuf(sc, txq, &bf_q);
685 
686 	} while (txq->axq_depth < ATH_AGGR_MIN_QDEPTH &&
687 		 status != ATH_AGGR_BAW_CLOSED);
688 }
689 
690 int ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
691 		      u16 tid, u16 *ssn)
692 {
693 	struct ath_atx_tid *txtid;
694 	struct ath_node *an;
695 
696 	an = (struct ath_node *)sta->drv_priv;
697 
698 	if (sc->sc_flags & SC_OP_TXAGGR) {
699 		txtid = ATH_AN_2_TID(an, tid);
700 		txtid->state |= AGGR_ADDBA_PROGRESS;
701 		ath_tx_pause_tid(sc, txtid);
702 		*ssn = txtid->seq_start;
703 	}
704 
705 	return 0;
706 }
707 
708 int ath_tx_aggr_stop(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
709 {
710 	struct ath_node *an = (struct ath_node *)sta->drv_priv;
711 	struct ath_atx_tid *txtid = ATH_AN_2_TID(an, tid);
712 	struct ath_txq *txq = &sc->tx.txq[txtid->ac->qnum];
713 	struct ath_buf *bf;
714 	struct list_head bf_head;
715 	INIT_LIST_HEAD(&bf_head);
716 
717 	if (txtid->state & AGGR_CLEANUP)
718 		return 0;
719 
720 	if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
721 		txtid->state &= ~AGGR_ADDBA_PROGRESS;
722 		txtid->addba_exchangeattempts = 0;
723 		return 0;
724 	}
725 
726 	ath_tx_pause_tid(sc, txtid);
727 
728 	/* drop all software retried frames and mark this TID */
729 	spin_lock_bh(&txq->axq_lock);
730 	while (!list_empty(&txtid->buf_q)) {
731 		bf = list_first_entry(&txtid->buf_q, struct ath_buf, list);
732 		if (!bf_isretried(bf)) {
733 			/*
734 			 * NB: it's based on the assumption that
735 			 * software retried frame will always stay
736 			 * at the head of software queue.
737 			 */
738 			break;
739 		}
740 		list_move_tail(&bf->list, &bf_head);
741 		ath_tx_update_baw(sc, txtid, bf->bf_seqno);
742 		ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
743 	}
744 	spin_unlock_bh(&txq->axq_lock);
745 
746 	if (txtid->baw_head != txtid->baw_tail) {
747 		txtid->state |= AGGR_CLEANUP;
748 	} else {
749 		txtid->state &= ~AGGR_ADDBA_COMPLETE;
750 		txtid->addba_exchangeattempts = 0;
751 		ath_tx_flush_tid(sc, txtid);
752 	}
753 
754 	return 0;
755 }
756 
757 void ath_tx_aggr_resume(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
758 {
759 	struct ath_atx_tid *txtid;
760 	struct ath_node *an;
761 
762 	an = (struct ath_node *)sta->drv_priv;
763 
764 	if (sc->sc_flags & SC_OP_TXAGGR) {
765 		txtid = ATH_AN_2_TID(an, tid);
766 		txtid->baw_size =
767 			IEEE80211_MIN_AMPDU_BUF << sta->ht_cap.ampdu_factor;
768 		txtid->state |= AGGR_ADDBA_COMPLETE;
769 		txtid->state &= ~AGGR_ADDBA_PROGRESS;
770 		ath_tx_resume_tid(sc, txtid);
771 	}
772 }
773 
774 bool ath_tx_aggr_check(struct ath_softc *sc, struct ath_node *an, u8 tidno)
775 {
776 	struct ath_atx_tid *txtid;
777 
778 	if (!(sc->sc_flags & SC_OP_TXAGGR))
779 		return false;
780 
781 	txtid = ATH_AN_2_TID(an, tidno);
782 
783 	if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
784 		if (!(txtid->state & AGGR_ADDBA_PROGRESS) &&
785 		    (txtid->addba_exchangeattempts < ADDBA_EXCHANGE_ATTEMPTS)) {
786 			txtid->addba_exchangeattempts++;
787 			return true;
788 		}
789 	}
790 
791 	return false;
792 }
793 
794 /********************/
795 /* Queue Management */
796 /********************/
797 
798 static void ath_txq_drain_pending_buffers(struct ath_softc *sc,
799 					  struct ath_txq *txq)
800 {
801 	struct ath_atx_ac *ac, *ac_tmp;
802 	struct ath_atx_tid *tid, *tid_tmp;
803 
804 	list_for_each_entry_safe(ac, ac_tmp, &txq->axq_acq, list) {
805 		list_del(&ac->list);
806 		ac->sched = false;
807 		list_for_each_entry_safe(tid, tid_tmp, &ac->tid_q, list) {
808 			list_del(&tid->list);
809 			tid->sched = false;
810 			ath_tid_drain(sc, txq, tid);
811 		}
812 	}
813 }
814 
815 struct ath_txq *ath_txq_setup(struct ath_softc *sc, int qtype, int subtype)
816 {
817 	struct ath_hw *ah = sc->sc_ah;
818 	struct ath9k_tx_queue_info qi;
819 	int qnum;
820 
821 	memset(&qi, 0, sizeof(qi));
822 	qi.tqi_subtype = subtype;
823 	qi.tqi_aifs = ATH9K_TXQ_USEDEFAULT;
824 	qi.tqi_cwmin = ATH9K_TXQ_USEDEFAULT;
825 	qi.tqi_cwmax = ATH9K_TXQ_USEDEFAULT;
826 	qi.tqi_physCompBuf = 0;
827 
828 	/*
829 	 * Enable interrupts only for EOL and DESC conditions.
830 	 * We mark tx descriptors to receive a DESC interrupt
831 	 * when a tx queue gets deep; otherwise waiting for the
832 	 * EOL to reap descriptors.  Note that this is done to
833 	 * reduce interrupt load and this only defers reaping
834 	 * descriptors, never transmitting frames.  Aside from
835 	 * reducing interrupts this also permits more concurrency.
836 	 * The only potential downside is if the tx queue backs
837 	 * up in which case the top half of the kernel may backup
838 	 * due to a lack of tx descriptors.
839 	 *
840 	 * The UAPSD queue is an exception, since we take a desc-
841 	 * based intr on the EOSP frames.
842 	 */
843 	if (qtype == ATH9K_TX_QUEUE_UAPSD)
844 		qi.tqi_qflags = TXQ_FLAG_TXDESCINT_ENABLE;
845 	else
846 		qi.tqi_qflags = TXQ_FLAG_TXEOLINT_ENABLE |
847 			TXQ_FLAG_TXDESCINT_ENABLE;
848 	qnum = ath9k_hw_setuptxqueue(ah, qtype, &qi);
849 	if (qnum == -1) {
850 		/*
851 		 * NB: don't print a message, this happens
852 		 * normally on parts with too few tx queues
853 		 */
854 		return NULL;
855 	}
856 	if (qnum >= ARRAY_SIZE(sc->tx.txq)) {
857 		DPRINTF(sc, ATH_DBG_FATAL,
858 			"qnum %u out of range, max %u!\n",
859 			qnum, (unsigned int)ARRAY_SIZE(sc->tx.txq));
860 		ath9k_hw_releasetxqueue(ah, qnum);
861 		return NULL;
862 	}
863 	if (!ATH_TXQ_SETUP(sc, qnum)) {
864 		struct ath_txq *txq = &sc->tx.txq[qnum];
865 
866 		txq->axq_qnum = qnum;
867 		txq->axq_link = NULL;
868 		INIT_LIST_HEAD(&txq->axq_q);
869 		INIT_LIST_HEAD(&txq->axq_acq);
870 		spin_lock_init(&txq->axq_lock);
871 		txq->axq_depth = 0;
872 		txq->axq_aggr_depth = 0;
873 		txq->axq_totalqueued = 0;
874 		txq->axq_linkbuf = NULL;
875 		sc->tx.txqsetup |= 1<<qnum;
876 	}
877 	return &sc->tx.txq[qnum];
878 }
879 
880 static int ath_tx_get_qnum(struct ath_softc *sc, int qtype, int haltype)
881 {
882 	int qnum;
883 
884 	switch (qtype) {
885 	case ATH9K_TX_QUEUE_DATA:
886 		if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
887 			DPRINTF(sc, ATH_DBG_FATAL,
888 				"HAL AC %u out of range, max %zu!\n",
889 				haltype, ARRAY_SIZE(sc->tx.hwq_map));
890 			return -1;
891 		}
892 		qnum = sc->tx.hwq_map[haltype];
893 		break;
894 	case ATH9K_TX_QUEUE_BEACON:
895 		qnum = sc->beacon.beaconq;
896 		break;
897 	case ATH9K_TX_QUEUE_CAB:
898 		qnum = sc->beacon.cabq->axq_qnum;
899 		break;
900 	default:
901 		qnum = -1;
902 	}
903 	return qnum;
904 }
905 
906 struct ath_txq *ath_test_get_txq(struct ath_softc *sc, struct sk_buff *skb)
907 {
908 	struct ath_txq *txq = NULL;
909 	int qnum;
910 
911 	qnum = ath_get_hal_qnum(skb_get_queue_mapping(skb), sc);
912 	txq = &sc->tx.txq[qnum];
913 
914 	spin_lock_bh(&txq->axq_lock);
915 
916 	if (txq->axq_depth >= (ATH_TXBUF - 20)) {
917 		DPRINTF(sc, ATH_DBG_XMIT,
918 			"TX queue: %d is full, depth: %d\n",
919 			qnum, txq->axq_depth);
920 		ieee80211_stop_queue(sc->hw, skb_get_queue_mapping(skb));
921 		txq->stopped = 1;
922 		spin_unlock_bh(&txq->axq_lock);
923 		return NULL;
924 	}
925 
926 	spin_unlock_bh(&txq->axq_lock);
927 
928 	return txq;
929 }
930 
931 int ath_txq_update(struct ath_softc *sc, int qnum,
932 		   struct ath9k_tx_queue_info *qinfo)
933 {
934 	struct ath_hw *ah = sc->sc_ah;
935 	int error = 0;
936 	struct ath9k_tx_queue_info qi;
937 
938 	if (qnum == sc->beacon.beaconq) {
939 		/*
940 		 * XXX: for beacon queue, we just save the parameter.
941 		 * It will be picked up by ath_beaconq_config when
942 		 * it's necessary.
943 		 */
944 		sc->beacon.beacon_qi = *qinfo;
945 		return 0;
946 	}
947 
948 	ASSERT(sc->tx.txq[qnum].axq_qnum == qnum);
949 
950 	ath9k_hw_get_txq_props(ah, qnum, &qi);
951 	qi.tqi_aifs = qinfo->tqi_aifs;
952 	qi.tqi_cwmin = qinfo->tqi_cwmin;
953 	qi.tqi_cwmax = qinfo->tqi_cwmax;
954 	qi.tqi_burstTime = qinfo->tqi_burstTime;
955 	qi.tqi_readyTime = qinfo->tqi_readyTime;
956 
957 	if (!ath9k_hw_set_txq_props(ah, qnum, &qi)) {
958 		DPRINTF(sc, ATH_DBG_FATAL,
959 			"Unable to update hardware queue %u!\n", qnum);
960 		error = -EIO;
961 	} else {
962 		ath9k_hw_resettxqueue(ah, qnum);
963 	}
964 
965 	return error;
966 }
967 
968 int ath_cabq_update(struct ath_softc *sc)
969 {
970 	struct ath9k_tx_queue_info qi;
971 	int qnum = sc->beacon.cabq->axq_qnum;
972 
973 	ath9k_hw_get_txq_props(sc->sc_ah, qnum, &qi);
974 	/*
975 	 * Ensure the readytime % is within the bounds.
976 	 */
977 	if (sc->config.cabqReadytime < ATH9K_READY_TIME_LO_BOUND)
978 		sc->config.cabqReadytime = ATH9K_READY_TIME_LO_BOUND;
979 	else if (sc->config.cabqReadytime > ATH9K_READY_TIME_HI_BOUND)
980 		sc->config.cabqReadytime = ATH9K_READY_TIME_HI_BOUND;
981 
982 	qi.tqi_readyTime = (sc->beacon_interval *
983 			    sc->config.cabqReadytime) / 100;
984 	ath_txq_update(sc, qnum, &qi);
985 
986 	return 0;
987 }
988 
989 /*
990  * Drain a given TX queue (could be Beacon or Data)
991  *
992  * This assumes output has been stopped and
993  * we do not need to block ath_tx_tasklet.
994  */
995 void ath_draintxq(struct ath_softc *sc, struct ath_txq *txq, bool retry_tx)
996 {
997 	struct ath_buf *bf, *lastbf;
998 	struct list_head bf_head;
999 
1000 	INIT_LIST_HEAD(&bf_head);
1001 
1002 	for (;;) {
1003 		spin_lock_bh(&txq->axq_lock);
1004 
1005 		if (list_empty(&txq->axq_q)) {
1006 			txq->axq_link = NULL;
1007 			txq->axq_linkbuf = NULL;
1008 			spin_unlock_bh(&txq->axq_lock);
1009 			break;
1010 		}
1011 
1012 		bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
1013 
1014 		if (bf->bf_stale) {
1015 			list_del(&bf->list);
1016 			spin_unlock_bh(&txq->axq_lock);
1017 
1018 			spin_lock_bh(&sc->tx.txbuflock);
1019 			list_add_tail(&bf->list, &sc->tx.txbuf);
1020 			spin_unlock_bh(&sc->tx.txbuflock);
1021 			continue;
1022 		}
1023 
1024 		lastbf = bf->bf_lastbf;
1025 		if (!retry_tx)
1026 			lastbf->bf_desc->ds_txstat.ts_flags =
1027 				ATH9K_TX_SW_ABORTED;
1028 
1029 		/* remove ath_buf's of the same mpdu from txq */
1030 		list_cut_position(&bf_head, &txq->axq_q, &lastbf->list);
1031 		txq->axq_depth--;
1032 
1033 		spin_unlock_bh(&txq->axq_lock);
1034 
1035 		if (bf_isampdu(bf))
1036 			ath_tx_complete_aggr(sc, txq, bf, &bf_head, 0);
1037 		else
1038 			ath_tx_complete_buf(sc, bf, &bf_head, 0, 0);
1039 	}
1040 
1041 	/* flush any pending frames if aggregation is enabled */
1042 	if (sc->sc_flags & SC_OP_TXAGGR) {
1043 		if (!retry_tx) {
1044 			spin_lock_bh(&txq->axq_lock);
1045 			ath_txq_drain_pending_buffers(sc, txq);
1046 			spin_unlock_bh(&txq->axq_lock);
1047 		}
1048 	}
1049 }
1050 
1051 void ath_drain_all_txq(struct ath_softc *sc, bool retry_tx)
1052 {
1053 	struct ath_hw *ah = sc->sc_ah;
1054 	struct ath_txq *txq;
1055 	int i, npend = 0;
1056 
1057 	if (sc->sc_flags & SC_OP_INVALID)
1058 		return;
1059 
1060 	/* Stop beacon queue */
1061 	ath9k_hw_stoptxdma(sc->sc_ah, sc->beacon.beaconq);
1062 
1063 	/* Stop data queues */
1064 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
1065 		if (ATH_TXQ_SETUP(sc, i)) {
1066 			txq = &sc->tx.txq[i];
1067 			ath9k_hw_stoptxdma(ah, txq->axq_qnum);
1068 			npend += ath9k_hw_numtxpending(ah, txq->axq_qnum);
1069 		}
1070 	}
1071 
1072 	if (npend) {
1073 		int r;
1074 
1075 		DPRINTF(sc, ATH_DBG_XMIT, "Unable to stop TxDMA. Reset HAL!\n");
1076 
1077 		spin_lock_bh(&sc->sc_resetlock);
1078 		r = ath9k_hw_reset(ah, sc->sc_ah->curchan, true);
1079 		if (r)
1080 			DPRINTF(sc, ATH_DBG_FATAL,
1081 				"Unable to reset hardware; reset status %d\n",
1082 				r);
1083 		spin_unlock_bh(&sc->sc_resetlock);
1084 	}
1085 
1086 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
1087 		if (ATH_TXQ_SETUP(sc, i))
1088 			ath_draintxq(sc, &sc->tx.txq[i], retry_tx);
1089 	}
1090 }
1091 
1092 void ath_tx_cleanupq(struct ath_softc *sc, struct ath_txq *txq)
1093 {
1094 	ath9k_hw_releasetxqueue(sc->sc_ah, txq->axq_qnum);
1095 	sc->tx.txqsetup &= ~(1<<txq->axq_qnum);
1096 }
1097 
1098 void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq)
1099 {
1100 	struct ath_atx_ac *ac;
1101 	struct ath_atx_tid *tid;
1102 
1103 	if (list_empty(&txq->axq_acq))
1104 		return;
1105 
1106 	ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac, list);
1107 	list_del(&ac->list);
1108 	ac->sched = false;
1109 
1110 	do {
1111 		if (list_empty(&ac->tid_q))
1112 			return;
1113 
1114 		tid = list_first_entry(&ac->tid_q, struct ath_atx_tid, list);
1115 		list_del(&tid->list);
1116 		tid->sched = false;
1117 
1118 		if (tid->paused)
1119 			continue;
1120 
1121 		if ((txq->axq_depth % 2) == 0)
1122 			ath_tx_sched_aggr(sc, txq, tid);
1123 
1124 		/*
1125 		 * add tid to round-robin queue if more frames
1126 		 * are pending for the tid
1127 		 */
1128 		if (!list_empty(&tid->buf_q))
1129 			ath_tx_queue_tid(txq, tid);
1130 
1131 		break;
1132 	} while (!list_empty(&ac->tid_q));
1133 
1134 	if (!list_empty(&ac->tid_q)) {
1135 		if (!ac->sched) {
1136 			ac->sched = true;
1137 			list_add_tail(&ac->list, &txq->axq_acq);
1138 		}
1139 	}
1140 }
1141 
1142 int ath_tx_setup(struct ath_softc *sc, int haltype)
1143 {
1144 	struct ath_txq *txq;
1145 
1146 	if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
1147 		DPRINTF(sc, ATH_DBG_FATAL,
1148 			"HAL AC %u out of range, max %zu!\n",
1149 			 haltype, ARRAY_SIZE(sc->tx.hwq_map));
1150 		return 0;
1151 	}
1152 	txq = ath_txq_setup(sc, ATH9K_TX_QUEUE_DATA, haltype);
1153 	if (txq != NULL) {
1154 		sc->tx.hwq_map[haltype] = txq->axq_qnum;
1155 		return 1;
1156 	} else
1157 		return 0;
1158 }
1159 
1160 /***********/
1161 /* TX, DMA */
1162 /***********/
1163 
1164 /*
1165  * Insert a chain of ath_buf (descriptors) on a txq and
1166  * assume the descriptors are already chained together by caller.
1167  */
1168 static void ath_tx_txqaddbuf(struct ath_softc *sc, struct ath_txq *txq,
1169 			     struct list_head *head)
1170 {
1171 	struct ath_hw *ah = sc->sc_ah;
1172 	struct ath_buf *bf;
1173 
1174 	/*
1175 	 * Insert the frame on the outbound list and
1176 	 * pass it on to the hardware.
1177 	 */
1178 
1179 	if (list_empty(head))
1180 		return;
1181 
1182 	bf = list_first_entry(head, struct ath_buf, list);
1183 
1184 	list_splice_tail_init(head, &txq->axq_q);
1185 	txq->axq_depth++;
1186 	txq->axq_totalqueued++;
1187 	txq->axq_linkbuf = list_entry(txq->axq_q.prev, struct ath_buf, list);
1188 
1189 	DPRINTF(sc, ATH_DBG_QUEUE,
1190 		"qnum: %d, txq depth: %d\n", txq->axq_qnum, txq->axq_depth);
1191 
1192 	if (txq->axq_link == NULL) {
1193 		ath9k_hw_puttxbuf(ah, txq->axq_qnum, bf->bf_daddr);
1194 		DPRINTF(sc, ATH_DBG_XMIT,
1195 			"TXDP[%u] = %llx (%p)\n",
1196 			txq->axq_qnum, ito64(bf->bf_daddr), bf->bf_desc);
1197 	} else {
1198 		*txq->axq_link = bf->bf_daddr;
1199 		DPRINTF(sc, ATH_DBG_XMIT, "link[%u] (%p)=%llx (%p)\n",
1200 			txq->axq_qnum, txq->axq_link,
1201 			ito64(bf->bf_daddr), bf->bf_desc);
1202 	}
1203 	txq->axq_link = &(bf->bf_lastbf->bf_desc->ds_link);
1204 	ath9k_hw_txstart(ah, txq->axq_qnum);
1205 }
1206 
1207 static struct ath_buf *ath_tx_get_buffer(struct ath_softc *sc)
1208 {
1209 	struct ath_buf *bf = NULL;
1210 
1211 	spin_lock_bh(&sc->tx.txbuflock);
1212 
1213 	if (unlikely(list_empty(&sc->tx.txbuf))) {
1214 		spin_unlock_bh(&sc->tx.txbuflock);
1215 		return NULL;
1216 	}
1217 
1218 	bf = list_first_entry(&sc->tx.txbuf, struct ath_buf, list);
1219 	list_del(&bf->list);
1220 
1221 	spin_unlock_bh(&sc->tx.txbuflock);
1222 
1223 	return bf;
1224 }
1225 
1226 static void ath_tx_send_ampdu(struct ath_softc *sc, struct ath_atx_tid *tid,
1227 			      struct list_head *bf_head,
1228 			      struct ath_tx_control *txctl)
1229 {
1230 	struct ath_buf *bf;
1231 
1232 	bf = list_first_entry(bf_head, struct ath_buf, list);
1233 	bf->bf_state.bf_type |= BUF_AMPDU;
1234 
1235 	/*
1236 	 * Do not queue to h/w when any of the following conditions is true:
1237 	 * - there are pending frames in software queue
1238 	 * - the TID is currently paused for ADDBA/BAR request
1239 	 * - seqno is not within block-ack window
1240 	 * - h/w queue depth exceeds low water mark
1241 	 */
1242 	if (!list_empty(&tid->buf_q) || tid->paused ||
1243 	    !BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno) ||
1244 	    txctl->txq->axq_depth >= ATH_AGGR_MIN_QDEPTH) {
1245 		/*
1246 		 * Add this frame to software queue for scheduling later
1247 		 * for aggregation.
1248 		 */
1249 		list_move_tail(&bf->list, &tid->buf_q);
1250 		ath_tx_queue_tid(txctl->txq, tid);
1251 		return;
1252 	}
1253 
1254 	/* Add sub-frame to BAW */
1255 	ath_tx_addto_baw(sc, tid, bf);
1256 
1257 	/* Queue to h/w without aggregation */
1258 	bf->bf_nframes = 1;
1259 	bf->bf_lastbf = bf;
1260 	ath_buf_set_rate(sc, bf);
1261 	ath_tx_txqaddbuf(sc, txctl->txq, bf_head);
1262 }
1263 
1264 static void ath_tx_send_ht_normal(struct ath_softc *sc, struct ath_txq *txq,
1265 				  struct ath_atx_tid *tid,
1266 				  struct list_head *bf_head)
1267 {
1268 	struct ath_buf *bf;
1269 
1270 	bf = list_first_entry(bf_head, struct ath_buf, list);
1271 	bf->bf_state.bf_type &= ~BUF_AMPDU;
1272 
1273 	/* update starting sequence number for subsequent ADDBA request */
1274 	INCR(tid->seq_start, IEEE80211_SEQ_MAX);
1275 
1276 	bf->bf_nframes = 1;
1277 	bf->bf_lastbf = bf;
1278 	ath_buf_set_rate(sc, bf);
1279 	ath_tx_txqaddbuf(sc, txq, bf_head);
1280 }
1281 
1282 static void ath_tx_send_normal(struct ath_softc *sc, struct ath_txq *txq,
1283 			       struct list_head *bf_head)
1284 {
1285 	struct ath_buf *bf;
1286 
1287 	bf = list_first_entry(bf_head, struct ath_buf, list);
1288 
1289 	bf->bf_lastbf = bf;
1290 	bf->bf_nframes = 1;
1291 	ath_buf_set_rate(sc, bf);
1292 	ath_tx_txqaddbuf(sc, txq, bf_head);
1293 }
1294 
1295 static enum ath9k_pkt_type get_hw_packet_type(struct sk_buff *skb)
1296 {
1297 	struct ieee80211_hdr *hdr;
1298 	enum ath9k_pkt_type htype;
1299 	__le16 fc;
1300 
1301 	hdr = (struct ieee80211_hdr *)skb->data;
1302 	fc = hdr->frame_control;
1303 
1304 	if (ieee80211_is_beacon(fc))
1305 		htype = ATH9K_PKT_TYPE_BEACON;
1306 	else if (ieee80211_is_probe_resp(fc))
1307 		htype = ATH9K_PKT_TYPE_PROBE_RESP;
1308 	else if (ieee80211_is_atim(fc))
1309 		htype = ATH9K_PKT_TYPE_ATIM;
1310 	else if (ieee80211_is_pspoll(fc))
1311 		htype = ATH9K_PKT_TYPE_PSPOLL;
1312 	else
1313 		htype = ATH9K_PKT_TYPE_NORMAL;
1314 
1315 	return htype;
1316 }
1317 
1318 static bool is_pae(struct sk_buff *skb)
1319 {
1320 	struct ieee80211_hdr *hdr;
1321 	__le16 fc;
1322 
1323 	hdr = (struct ieee80211_hdr *)skb->data;
1324 	fc = hdr->frame_control;
1325 
1326 	if (ieee80211_is_data(fc)) {
1327 		if (ieee80211_is_nullfunc(fc) ||
1328 		    /* Port Access Entity (IEEE 802.1X) */
1329 		    (skb->protocol == cpu_to_be16(ETH_P_PAE))) {
1330 			return true;
1331 		}
1332 	}
1333 
1334 	return false;
1335 }
1336 
1337 static int get_hw_crypto_keytype(struct sk_buff *skb)
1338 {
1339 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1340 
1341 	if (tx_info->control.hw_key) {
1342 		if (tx_info->control.hw_key->alg == ALG_WEP)
1343 			return ATH9K_KEY_TYPE_WEP;
1344 		else if (tx_info->control.hw_key->alg == ALG_TKIP)
1345 			return ATH9K_KEY_TYPE_TKIP;
1346 		else if (tx_info->control.hw_key->alg == ALG_CCMP)
1347 			return ATH9K_KEY_TYPE_AES;
1348 	}
1349 
1350 	return ATH9K_KEY_TYPE_CLEAR;
1351 }
1352 
1353 static void assign_aggr_tid_seqno(struct sk_buff *skb,
1354 				  struct ath_buf *bf)
1355 {
1356 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1357 	struct ieee80211_hdr *hdr;
1358 	struct ath_node *an;
1359 	struct ath_atx_tid *tid;
1360 	__le16 fc;
1361 	u8 *qc;
1362 
1363 	if (!tx_info->control.sta)
1364 		return;
1365 
1366 	an = (struct ath_node *)tx_info->control.sta->drv_priv;
1367 	hdr = (struct ieee80211_hdr *)skb->data;
1368 	fc = hdr->frame_control;
1369 
1370 	if (ieee80211_is_data_qos(fc)) {
1371 		qc = ieee80211_get_qos_ctl(hdr);
1372 		bf->bf_tidno = qc[0] & 0xf;
1373 	}
1374 
1375 	/*
1376 	 * For HT capable stations, we save tidno for later use.
1377 	 * We also override seqno set by upper layer with the one
1378 	 * in tx aggregation state.
1379 	 *
1380 	 * If fragmentation is on, the sequence number is
1381 	 * not overridden, since it has been
1382 	 * incremented by the fragmentation routine.
1383 	 *
1384 	 * FIXME: check if the fragmentation threshold exceeds
1385 	 * IEEE80211 max.
1386 	 */
1387 	tid = ATH_AN_2_TID(an, bf->bf_tidno);
1388 	hdr->seq_ctrl = cpu_to_le16(tid->seq_next <<
1389 			IEEE80211_SEQ_SEQ_SHIFT);
1390 	bf->bf_seqno = tid->seq_next;
1391 	INCR(tid->seq_next, IEEE80211_SEQ_MAX);
1392 }
1393 
1394 static int setup_tx_flags(struct ath_softc *sc, struct sk_buff *skb,
1395 			  struct ath_txq *txq)
1396 {
1397 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1398 	int flags = 0;
1399 
1400 	flags |= ATH9K_TXDESC_CLRDMASK; /* needed for crypto errors */
1401 	flags |= ATH9K_TXDESC_INTREQ;
1402 
1403 	if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK)
1404 		flags |= ATH9K_TXDESC_NOACK;
1405 
1406 	return flags;
1407 }
1408 
1409 /*
1410  * rix - rate index
1411  * pktlen - total bytes (delims + data + fcs + pads + pad delims)
1412  * width  - 0 for 20 MHz, 1 for 40 MHz
1413  * half_gi - to use 4us v/s 3.6 us for symbol time
1414  */
1415 static u32 ath_pkt_duration(struct ath_softc *sc, u8 rix, struct ath_buf *bf,
1416 			    int width, int half_gi, bool shortPreamble)
1417 {
1418 	const struct ath_rate_table *rate_table = sc->cur_rate_table;
1419 	u32 nbits, nsymbits, duration, nsymbols;
1420 	u8 rc;
1421 	int streams, pktlen;
1422 
1423 	pktlen = bf_isaggr(bf) ? bf->bf_al : bf->bf_frmlen;
1424 	rc = rate_table->info[rix].ratecode;
1425 
1426 	/* for legacy rates, use old function to compute packet duration */
1427 	if (!IS_HT_RATE(rc))
1428 		return ath9k_hw_computetxtime(sc->sc_ah, rate_table, pktlen,
1429 					      rix, shortPreamble);
1430 
1431 	/* find number of symbols: PLCP + data */
1432 	nbits = (pktlen << 3) + OFDM_PLCP_BITS;
1433 	nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
1434 	nsymbols = (nbits + nsymbits - 1) / nsymbits;
1435 
1436 	if (!half_gi)
1437 		duration = SYMBOL_TIME(nsymbols);
1438 	else
1439 		duration = SYMBOL_TIME_HALFGI(nsymbols);
1440 
1441 	/* addup duration for legacy/ht training and signal fields */
1442 	streams = HT_RC_2_STREAMS(rc);
1443 	duration += L_STF + L_LTF + L_SIG + HT_SIG + HT_STF + HT_LTF(streams);
1444 
1445 	return duration;
1446 }
1447 
1448 static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf)
1449 {
1450 	const struct ath_rate_table *rt = sc->cur_rate_table;
1451 	struct ath9k_11n_rate_series series[4];
1452 	struct sk_buff *skb;
1453 	struct ieee80211_tx_info *tx_info;
1454 	struct ieee80211_tx_rate *rates;
1455 	struct ieee80211_hdr *hdr;
1456 	int i, flags = 0;
1457 	u8 rix = 0, ctsrate = 0;
1458 	bool is_pspoll;
1459 
1460 	memset(series, 0, sizeof(struct ath9k_11n_rate_series) * 4);
1461 
1462 	skb = bf->bf_mpdu;
1463 	tx_info = IEEE80211_SKB_CB(skb);
1464 	rates = tx_info->control.rates;
1465 	hdr = (struct ieee80211_hdr *)skb->data;
1466 	is_pspoll = ieee80211_is_pspoll(hdr->frame_control);
1467 
1468 	/*
1469 	 * We check if Short Preamble is needed for the CTS rate by
1470 	 * checking the BSS's global flag.
1471 	 * But for the rate series, IEEE80211_TX_RC_USE_SHORT_PREAMBLE is used.
1472 	 */
1473 	if (sc->sc_flags & SC_OP_PREAMBLE_SHORT)
1474 		ctsrate = rt->info[tx_info->control.rts_cts_rate_idx].ratecode |
1475 			rt->info[tx_info->control.rts_cts_rate_idx].short_preamble;
1476 	else
1477 		ctsrate = rt->info[tx_info->control.rts_cts_rate_idx].ratecode;
1478 
1479 	/*
1480 	 * ATH9K_TXDESC_RTSENA and ATH9K_TXDESC_CTSENA are mutually exclusive.
1481 	 * Check the first rate in the series to decide whether RTS/CTS
1482 	 * or CTS-to-self has to be used.
1483 	 */
1484 	if (rates[0].flags & IEEE80211_TX_RC_USE_CTS_PROTECT)
1485 		flags = ATH9K_TXDESC_CTSENA;
1486 	else if (rates[0].flags & IEEE80211_TX_RC_USE_RTS_CTS)
1487 		flags = ATH9K_TXDESC_RTSENA;
1488 
1489 	/* FIXME: Handle aggregation protection */
1490 	if (sc->config.ath_aggr_prot &&
1491 	    (!bf_isaggr(bf) || (bf_isaggr(bf) && bf->bf_al < 8192))) {
1492 		flags = ATH9K_TXDESC_RTSENA;
1493 	}
1494 
1495 	/* For AR5416 - RTS cannot be followed by a frame larger than 8K */
1496 	if (bf_isaggr(bf) && (bf->bf_al > sc->sc_ah->caps.rts_aggr_limit))
1497 		flags &= ~(ATH9K_TXDESC_RTSENA);
1498 
1499 	for (i = 0; i < 4; i++) {
1500 		if (!rates[i].count || (rates[i].idx < 0))
1501 			continue;
1502 
1503 		rix = rates[i].idx;
1504 		series[i].Tries = rates[i].count;
1505 		series[i].ChSel = sc->tx_chainmask;
1506 
1507 		if (rates[i].flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE)
1508 			series[i].Rate = rt->info[rix].ratecode |
1509 				rt->info[rix].short_preamble;
1510 		else
1511 			series[i].Rate = rt->info[rix].ratecode;
1512 
1513 		if (rates[i].flags & IEEE80211_TX_RC_USE_RTS_CTS)
1514 			series[i].RateFlags |= ATH9K_RATESERIES_RTS_CTS;
1515 		if (rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
1516 			series[i].RateFlags |= ATH9K_RATESERIES_2040;
1517 		if (rates[i].flags & IEEE80211_TX_RC_SHORT_GI)
1518 			series[i].RateFlags |= ATH9K_RATESERIES_HALFGI;
1519 
1520 		series[i].PktDuration = ath_pkt_duration(sc, rix, bf,
1521 			 (rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH) != 0,
1522 			 (rates[i].flags & IEEE80211_TX_RC_SHORT_GI),
1523 			 (rates[i].flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE));
1524 	}
1525 
1526 	/* set dur_update_en for l-sig computation except for PS-Poll frames */
1527 	ath9k_hw_set11n_ratescenario(sc->sc_ah, bf->bf_desc,
1528 				     bf->bf_lastbf->bf_desc,
1529 				     !is_pspoll, ctsrate,
1530 				     0, series, 4, flags);
1531 
1532 	if (sc->config.ath_aggr_prot && flags)
1533 		ath9k_hw_set11n_burstduration(sc->sc_ah, bf->bf_desc, 8192);
1534 }
1535 
1536 static int ath_tx_setup_buffer(struct ieee80211_hw *hw, struct ath_buf *bf,
1537 				struct sk_buff *skb,
1538 				struct ath_tx_control *txctl)
1539 {
1540 	struct ath_wiphy *aphy = hw->priv;
1541 	struct ath_softc *sc = aphy->sc;
1542 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1543 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1544 	struct ath_tx_info_priv *tx_info_priv;
1545 	int hdrlen;
1546 	__le16 fc;
1547 
1548 	tx_info_priv = kzalloc(sizeof(*tx_info_priv), GFP_ATOMIC);
1549 	if (unlikely(!tx_info_priv))
1550 		return -ENOMEM;
1551 	tx_info->rate_driver_data[0] = tx_info_priv;
1552 	tx_info_priv->aphy = aphy;
1553 	tx_info_priv->frame_type = txctl->frame_type;
1554 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
1555 	fc = hdr->frame_control;
1556 
1557 	ATH_TXBUF_RESET(bf);
1558 
1559 	bf->bf_frmlen = skb->len + FCS_LEN - (hdrlen & 3);
1560 
1561 	if (conf_is_ht(&sc->hw->conf) && !is_pae(skb))
1562 		bf->bf_state.bf_type |= BUF_HT;
1563 
1564 	bf->bf_flags = setup_tx_flags(sc, skb, txctl->txq);
1565 
1566 	bf->bf_keytype = get_hw_crypto_keytype(skb);
1567 	if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR) {
1568 		bf->bf_frmlen += tx_info->control.hw_key->icv_len;
1569 		bf->bf_keyix = tx_info->control.hw_key->hw_key_idx;
1570 	} else {
1571 		bf->bf_keyix = ATH9K_TXKEYIX_INVALID;
1572 	}
1573 
1574 	if (ieee80211_is_data_qos(fc) && (sc->sc_flags & SC_OP_TXAGGR))
1575 		assign_aggr_tid_seqno(skb, bf);
1576 
1577 	bf->bf_mpdu = skb;
1578 
1579 	bf->bf_dmacontext = dma_map_single(sc->dev, skb->data,
1580 					   skb->len, DMA_TO_DEVICE);
1581 	if (unlikely(dma_mapping_error(sc->dev, bf->bf_dmacontext))) {
1582 		bf->bf_mpdu = NULL;
1583 		kfree(tx_info_priv);
1584 		tx_info->rate_driver_data[0] = NULL;
1585 		DPRINTF(sc, ATH_DBG_FATAL, "dma_mapping_error() on TX\n");
1586 		return -ENOMEM;
1587 	}
1588 
1589 	bf->bf_buf_addr = bf->bf_dmacontext;
1590 	return 0;
1591 }
1592 
1593 /* FIXME: tx power */
1594 static void ath_tx_start_dma(struct ath_softc *sc, struct ath_buf *bf,
1595 			     struct ath_tx_control *txctl)
1596 {
1597 	struct sk_buff *skb = bf->bf_mpdu;
1598 	struct ieee80211_tx_info *tx_info =  IEEE80211_SKB_CB(skb);
1599 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1600 	struct ath_node *an = NULL;
1601 	struct list_head bf_head;
1602 	struct ath_desc *ds;
1603 	struct ath_atx_tid *tid;
1604 	struct ath_hw *ah = sc->sc_ah;
1605 	int frm_type;
1606 	__le16 fc;
1607 
1608 	frm_type = get_hw_packet_type(skb);
1609 	fc = hdr->frame_control;
1610 
1611 	INIT_LIST_HEAD(&bf_head);
1612 	list_add_tail(&bf->list, &bf_head);
1613 
1614 	ds = bf->bf_desc;
1615 	ds->ds_link = 0;
1616 	ds->ds_data = bf->bf_buf_addr;
1617 
1618 	ath9k_hw_set11n_txdesc(ah, ds, bf->bf_frmlen, frm_type, MAX_RATE_POWER,
1619 			       bf->bf_keyix, bf->bf_keytype, bf->bf_flags);
1620 
1621 	ath9k_hw_filltxdesc(ah, ds,
1622 			    skb->len,	/* segment length */
1623 			    true,	/* first segment */
1624 			    true,	/* last segment */
1625 			    ds);	/* first descriptor */
1626 
1627 	spin_lock_bh(&txctl->txq->axq_lock);
1628 
1629 	if (bf_isht(bf) && (sc->sc_flags & SC_OP_TXAGGR) &&
1630 	    tx_info->control.sta) {
1631 		an = (struct ath_node *)tx_info->control.sta->drv_priv;
1632 		tid = ATH_AN_2_TID(an, bf->bf_tidno);
1633 
1634 		if (!ieee80211_is_data_qos(fc)) {
1635 			ath_tx_send_normal(sc, txctl->txq, &bf_head);
1636 			goto tx_done;
1637 		}
1638 
1639 		if (ath_aggr_query(sc, an, bf->bf_tidno)) {
1640 			/*
1641 			 * Try aggregation if it's a unicast data frame
1642 			 * and the destination is HT capable.
1643 			 */
1644 			ath_tx_send_ampdu(sc, tid, &bf_head, txctl);
1645 		} else {
1646 			/*
1647 			 * Send this frame as regular when ADDBA
1648 			 * exchange is neither complete nor pending.
1649 			 */
1650 			ath_tx_send_ht_normal(sc, txctl->txq,
1651 					      tid, &bf_head);
1652 		}
1653 	} else {
1654 		ath_tx_send_normal(sc, txctl->txq, &bf_head);
1655 	}
1656 
1657 tx_done:
1658 	spin_unlock_bh(&txctl->txq->axq_lock);
1659 }
1660 
1661 /* Upon failure caller should free skb */
1662 int ath_tx_start(struct ieee80211_hw *hw, struct sk_buff *skb,
1663 		 struct ath_tx_control *txctl)
1664 {
1665 	struct ath_wiphy *aphy = hw->priv;
1666 	struct ath_softc *sc = aphy->sc;
1667 	struct ath_buf *bf;
1668 	int r;
1669 
1670 	bf = ath_tx_get_buffer(sc);
1671 	if (!bf) {
1672 		DPRINTF(sc, ATH_DBG_XMIT, "TX buffers are full\n");
1673 		return -1;
1674 	}
1675 
1676 	r = ath_tx_setup_buffer(hw, bf, skb, txctl);
1677 	if (unlikely(r)) {
1678 		struct ath_txq *txq = txctl->txq;
1679 
1680 		DPRINTF(sc, ATH_DBG_FATAL, "TX mem alloc failure\n");
1681 
1682 		/* upon ath_tx_processq() this TX queue will be resumed, we
1683 		 * guarantee this will happen by knowing beforehand that
1684 		 * we will at least have to run TX completionon one buffer
1685 		 * on the queue */
1686 		spin_lock_bh(&txq->axq_lock);
1687 		if (sc->tx.txq[txq->axq_qnum].axq_depth > 1) {
1688 			ieee80211_stop_queue(sc->hw,
1689 				skb_get_queue_mapping(skb));
1690 			txq->stopped = 1;
1691 		}
1692 		spin_unlock_bh(&txq->axq_lock);
1693 
1694 		spin_lock_bh(&sc->tx.txbuflock);
1695 		list_add_tail(&bf->list, &sc->tx.txbuf);
1696 		spin_unlock_bh(&sc->tx.txbuflock);
1697 
1698 		return r;
1699 	}
1700 
1701 	ath_tx_start_dma(sc, bf, txctl);
1702 
1703 	return 0;
1704 }
1705 
1706 void ath_tx_cabq(struct ieee80211_hw *hw, struct sk_buff *skb)
1707 {
1708 	struct ath_wiphy *aphy = hw->priv;
1709 	struct ath_softc *sc = aphy->sc;
1710 	int hdrlen, padsize;
1711 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
1712 	struct ath_tx_control txctl;
1713 
1714 	memset(&txctl, 0, sizeof(struct ath_tx_control));
1715 
1716 	/*
1717 	 * As a temporary workaround, assign seq# here; this will likely need
1718 	 * to be cleaned up to work better with Beacon transmission and virtual
1719 	 * BSSes.
1720 	 */
1721 	if (info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ) {
1722 		struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
1723 		if (info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT)
1724 			sc->tx.seq_no += 0x10;
1725 		hdr->seq_ctrl &= cpu_to_le16(IEEE80211_SCTL_FRAG);
1726 		hdr->seq_ctrl |= cpu_to_le16(sc->tx.seq_no);
1727 	}
1728 
1729 	/* Add the padding after the header if this is not already done */
1730 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
1731 	if (hdrlen & 3) {
1732 		padsize = hdrlen % 4;
1733 		if (skb_headroom(skb) < padsize) {
1734 			DPRINTF(sc, ATH_DBG_XMIT, "TX CABQ padding failed\n");
1735 			dev_kfree_skb_any(skb);
1736 			return;
1737 		}
1738 		skb_push(skb, padsize);
1739 		memmove(skb->data, skb->data + padsize, hdrlen);
1740 	}
1741 
1742 	txctl.txq = sc->beacon.cabq;
1743 
1744 	DPRINTF(sc, ATH_DBG_XMIT, "transmitting CABQ packet, skb: %p\n", skb);
1745 
1746 	if (ath_tx_start(hw, skb, &txctl) != 0) {
1747 		DPRINTF(sc, ATH_DBG_XMIT, "CABQ TX failed\n");
1748 		goto exit;
1749 	}
1750 
1751 	return;
1752 exit:
1753 	dev_kfree_skb_any(skb);
1754 }
1755 
1756 /*****************/
1757 /* TX Completion */
1758 /*****************/
1759 
1760 static void ath_tx_complete(struct ath_softc *sc, struct sk_buff *skb,
1761 			    int tx_flags)
1762 {
1763 	struct ieee80211_hw *hw = sc->hw;
1764 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1765 	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
1766 	int hdrlen, padsize;
1767 	int frame_type = ATH9K_NOT_INTERNAL;
1768 
1769 	DPRINTF(sc, ATH_DBG_XMIT, "TX complete: skb: %p\n", skb);
1770 
1771 	if (tx_info_priv) {
1772 		hw = tx_info_priv->aphy->hw;
1773 		frame_type = tx_info_priv->frame_type;
1774 	}
1775 
1776 	if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK ||
1777 	    tx_info->flags & IEEE80211_TX_STAT_TX_FILTERED) {
1778 		kfree(tx_info_priv);
1779 		tx_info->rate_driver_data[0] = NULL;
1780 	}
1781 
1782 	if (tx_flags & ATH_TX_BAR)
1783 		tx_info->flags |= IEEE80211_TX_STAT_AMPDU_NO_BACK;
1784 
1785 	if (!(tx_flags & (ATH_TX_ERROR | ATH_TX_XRETRY))) {
1786 		/* Frame was ACKed */
1787 		tx_info->flags |= IEEE80211_TX_STAT_ACK;
1788 	}
1789 
1790 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
1791 	padsize = hdrlen & 3;
1792 	if (padsize && hdrlen >= 24) {
1793 		/*
1794 		 * Remove MAC header padding before giving the frame back to
1795 		 * mac80211.
1796 		 */
1797 		memmove(skb->data + padsize, skb->data, hdrlen);
1798 		skb_pull(skb, padsize);
1799 	}
1800 
1801 	if (sc->sc_flags & SC_OP_WAIT_FOR_TX_ACK) {
1802 		sc->sc_flags &= ~SC_OP_WAIT_FOR_TX_ACK;
1803 		DPRINTF(sc, ATH_DBG_PS, "Going back to sleep after having "
1804 			"received TX status (0x%x)\n",
1805 			sc->sc_flags & (SC_OP_WAIT_FOR_BEACON |
1806 					SC_OP_WAIT_FOR_CAB |
1807 					SC_OP_WAIT_FOR_PSPOLL_DATA |
1808 					SC_OP_WAIT_FOR_TX_ACK));
1809 	}
1810 
1811 	if (frame_type == ATH9K_NOT_INTERNAL)
1812 		ieee80211_tx_status(hw, skb);
1813 	else
1814 		ath9k_tx_status(hw, skb);
1815 }
1816 
1817 static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
1818 				struct list_head *bf_q,
1819 				int txok, int sendbar)
1820 {
1821 	struct sk_buff *skb = bf->bf_mpdu;
1822 	unsigned long flags;
1823 	int tx_flags = 0;
1824 
1825 
1826 	if (sendbar)
1827 		tx_flags = ATH_TX_BAR;
1828 
1829 	if (!txok) {
1830 		tx_flags |= ATH_TX_ERROR;
1831 
1832 		if (bf_isxretried(bf))
1833 			tx_flags |= ATH_TX_XRETRY;
1834 	}
1835 
1836 	dma_unmap_single(sc->dev, bf->bf_dmacontext, skb->len, DMA_TO_DEVICE);
1837 	ath_tx_complete(sc, skb, tx_flags);
1838 
1839 	/*
1840 	 * Return the list of ath_buf of this mpdu to free queue
1841 	 */
1842 	spin_lock_irqsave(&sc->tx.txbuflock, flags);
1843 	list_splice_tail_init(bf_q, &sc->tx.txbuf);
1844 	spin_unlock_irqrestore(&sc->tx.txbuflock, flags);
1845 }
1846 
1847 static int ath_tx_num_badfrms(struct ath_softc *sc, struct ath_buf *bf,
1848 			      int txok)
1849 {
1850 	struct ath_buf *bf_last = bf->bf_lastbf;
1851 	struct ath_desc *ds = bf_last->bf_desc;
1852 	u16 seq_st = 0;
1853 	u32 ba[WME_BA_BMP_SIZE >> 5];
1854 	int ba_index;
1855 	int nbad = 0;
1856 	int isaggr = 0;
1857 
1858 	if (ds->ds_txstat.ts_flags == ATH9K_TX_SW_ABORTED)
1859 		return 0;
1860 
1861 	isaggr = bf_isaggr(bf);
1862 	if (isaggr) {
1863 		seq_st = ATH_DS_BA_SEQ(ds);
1864 		memcpy(ba, ATH_DS_BA_BITMAP(ds), WME_BA_BMP_SIZE >> 3);
1865 	}
1866 
1867 	while (bf) {
1868 		ba_index = ATH_BA_INDEX(seq_st, bf->bf_seqno);
1869 		if (!txok || (isaggr && !ATH_BA_ISSET(ba, ba_index)))
1870 			nbad++;
1871 
1872 		bf = bf->bf_next;
1873 	}
1874 
1875 	return nbad;
1876 }
1877 
1878 static void ath_tx_rc_status(struct ath_buf *bf, struct ath_desc *ds,
1879 			     int nbad, int txok, bool update_rc)
1880 {
1881 	struct sk_buff *skb = bf->bf_mpdu;
1882 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1883 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1884 	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
1885 	struct ieee80211_hw *hw = tx_info_priv->aphy->hw;
1886 	u8 i, tx_rateindex;
1887 
1888 	if (txok)
1889 		tx_info->status.ack_signal = ds->ds_txstat.ts_rssi;
1890 
1891 	tx_rateindex = ds->ds_txstat.ts_rateindex;
1892 	WARN_ON(tx_rateindex >= hw->max_rates);
1893 
1894 	tx_info_priv->update_rc = update_rc;
1895 	if (ds->ds_txstat.ts_status & ATH9K_TXERR_FILT)
1896 		tx_info->flags |= IEEE80211_TX_STAT_TX_FILTERED;
1897 
1898 	if ((ds->ds_txstat.ts_status & ATH9K_TXERR_FILT) == 0 &&
1899 	    (bf->bf_flags & ATH9K_TXDESC_NOACK) == 0 && update_rc) {
1900 		if (ieee80211_is_data(hdr->frame_control)) {
1901 			memcpy(&tx_info_priv->tx, &ds->ds_txstat,
1902 			       sizeof(tx_info_priv->tx));
1903 			tx_info_priv->n_frames = bf->bf_nframes;
1904 			tx_info_priv->n_bad_frames = nbad;
1905 		}
1906 	}
1907 
1908 	for (i = tx_rateindex + 1; i < hw->max_rates; i++)
1909 		tx_info->status.rates[i].count = 0;
1910 
1911 	tx_info->status.rates[tx_rateindex].count = bf->bf_retries + 1;
1912 }
1913 
1914 static void ath_wake_mac80211_queue(struct ath_softc *sc, struct ath_txq *txq)
1915 {
1916 	int qnum;
1917 
1918 	spin_lock_bh(&txq->axq_lock);
1919 	if (txq->stopped &&
1920 	    sc->tx.txq[txq->axq_qnum].axq_depth <= (ATH_TXBUF - 20)) {
1921 		qnum = ath_get_mac80211_qnum(txq->axq_qnum, sc);
1922 		if (qnum != -1) {
1923 			ieee80211_wake_queue(sc->hw, qnum);
1924 			txq->stopped = 0;
1925 		}
1926 	}
1927 	spin_unlock_bh(&txq->axq_lock);
1928 }
1929 
1930 static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
1931 {
1932 	struct ath_hw *ah = sc->sc_ah;
1933 	struct ath_buf *bf, *lastbf, *bf_held = NULL;
1934 	struct list_head bf_head;
1935 	struct ath_desc *ds;
1936 	int txok;
1937 	int status;
1938 
1939 	DPRINTF(sc, ATH_DBG_QUEUE, "tx queue %d (%x), link %p\n",
1940 		txq->axq_qnum, ath9k_hw_gettxbuf(sc->sc_ah, txq->axq_qnum),
1941 		txq->axq_link);
1942 
1943 	for (;;) {
1944 		spin_lock_bh(&txq->axq_lock);
1945 		if (list_empty(&txq->axq_q)) {
1946 			txq->axq_link = NULL;
1947 			txq->axq_linkbuf = NULL;
1948 			spin_unlock_bh(&txq->axq_lock);
1949 			break;
1950 		}
1951 		bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
1952 
1953 		/*
1954 		 * There is a race condition that a BH gets scheduled
1955 		 * after sw writes TxE and before hw re-load the last
1956 		 * descriptor to get the newly chained one.
1957 		 * Software must keep the last DONE descriptor as a
1958 		 * holding descriptor - software does so by marking
1959 		 * it with the STALE flag.
1960 		 */
1961 		bf_held = NULL;
1962 		if (bf->bf_stale) {
1963 			bf_held = bf;
1964 			if (list_is_last(&bf_held->list, &txq->axq_q)) {
1965 				txq->axq_link = NULL;
1966 				txq->axq_linkbuf = NULL;
1967 				spin_unlock_bh(&txq->axq_lock);
1968 
1969 				/*
1970 				 * The holding descriptor is the last
1971 				 * descriptor in queue. It's safe to remove
1972 				 * the last holding descriptor in BH context.
1973 				 */
1974 				spin_lock_bh(&sc->tx.txbuflock);
1975 				list_move_tail(&bf_held->list, &sc->tx.txbuf);
1976 				spin_unlock_bh(&sc->tx.txbuflock);
1977 
1978 				break;
1979 			} else {
1980 				bf = list_entry(bf_held->list.next,
1981 						struct ath_buf, list);
1982 			}
1983 		}
1984 
1985 		lastbf = bf->bf_lastbf;
1986 		ds = lastbf->bf_desc;
1987 
1988 		status = ath9k_hw_txprocdesc(ah, ds);
1989 		if (status == -EINPROGRESS) {
1990 			spin_unlock_bh(&txq->axq_lock);
1991 			break;
1992 		}
1993 		if (bf->bf_desc == txq->axq_lastdsWithCTS)
1994 			txq->axq_lastdsWithCTS = NULL;
1995 		if (ds == txq->axq_gatingds)
1996 			txq->axq_gatingds = NULL;
1997 
1998 		/*
1999 		 * Remove ath_buf's of the same transmit unit from txq,
2000 		 * however leave the last descriptor back as the holding
2001 		 * descriptor for hw.
2002 		 */
2003 		lastbf->bf_stale = true;
2004 		INIT_LIST_HEAD(&bf_head);
2005 		if (!list_is_singular(&lastbf->list))
2006 			list_cut_position(&bf_head,
2007 				&txq->axq_q, lastbf->list.prev);
2008 
2009 		txq->axq_depth--;
2010 		if (bf_isaggr(bf))
2011 			txq->axq_aggr_depth--;
2012 
2013 		txok = (ds->ds_txstat.ts_status == 0);
2014 		spin_unlock_bh(&txq->axq_lock);
2015 
2016 		if (bf_held) {
2017 			spin_lock_bh(&sc->tx.txbuflock);
2018 			list_move_tail(&bf_held->list, &sc->tx.txbuf);
2019 			spin_unlock_bh(&sc->tx.txbuflock);
2020 		}
2021 
2022 		if (!bf_isampdu(bf)) {
2023 			/*
2024 			 * This frame is sent out as a single frame.
2025 			 * Use hardware retry status for this frame.
2026 			 */
2027 			bf->bf_retries = ds->ds_txstat.ts_longretry;
2028 			if (ds->ds_txstat.ts_status & ATH9K_TXERR_XRETRY)
2029 				bf->bf_state.bf_type |= BUF_XRETRY;
2030 			ath_tx_rc_status(bf, ds, 0, txok, true);
2031 		}
2032 
2033 		if (bf_isampdu(bf))
2034 			ath_tx_complete_aggr(sc, txq, bf, &bf_head, txok);
2035 		else
2036 			ath_tx_complete_buf(sc, bf, &bf_head, txok, 0);
2037 
2038 		ath_wake_mac80211_queue(sc, txq);
2039 
2040 		spin_lock_bh(&txq->axq_lock);
2041 		if (sc->sc_flags & SC_OP_TXAGGR)
2042 			ath_txq_schedule(sc, txq);
2043 		spin_unlock_bh(&txq->axq_lock);
2044 	}
2045 }
2046 
2047 
2048 void ath_tx_tasklet(struct ath_softc *sc)
2049 {
2050 	int i;
2051 	u32 qcumask = ((1 << ATH9K_NUM_TX_QUEUES) - 1);
2052 
2053 	ath9k_hw_gettxintrtxqs(sc->sc_ah, &qcumask);
2054 
2055 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
2056 		if (ATH_TXQ_SETUP(sc, i) && (qcumask & (1 << i)))
2057 			ath_tx_processq(sc, &sc->tx.txq[i]);
2058 	}
2059 }
2060 
2061 /*****************/
2062 /* Init, Cleanup */
2063 /*****************/
2064 
2065 int ath_tx_init(struct ath_softc *sc, int nbufs)
2066 {
2067 	int error = 0;
2068 
2069 	spin_lock_init(&sc->tx.txbuflock);
2070 
2071 	error = ath_descdma_setup(sc, &sc->tx.txdma, &sc->tx.txbuf,
2072 				  "tx", nbufs, 1);
2073 	if (error != 0) {
2074 		DPRINTF(sc, ATH_DBG_FATAL,
2075 			"Failed to allocate tx descriptors: %d\n", error);
2076 		goto err;
2077 	}
2078 
2079 	error = ath_descdma_setup(sc, &sc->beacon.bdma, &sc->beacon.bbuf,
2080 				  "beacon", ATH_BCBUF, 1);
2081 	if (error != 0) {
2082 		DPRINTF(sc, ATH_DBG_FATAL,
2083 			"Failed to allocate beacon descriptors: %d\n", error);
2084 		goto err;
2085 	}
2086 
2087 err:
2088 	if (error != 0)
2089 		ath_tx_cleanup(sc);
2090 
2091 	return error;
2092 }
2093 
2094 void ath_tx_cleanup(struct ath_softc *sc)
2095 {
2096 	if (sc->beacon.bdma.dd_desc_len != 0)
2097 		ath_descdma_cleanup(sc, &sc->beacon.bdma, &sc->beacon.bbuf);
2098 
2099 	if (sc->tx.txdma.dd_desc_len != 0)
2100 		ath_descdma_cleanup(sc, &sc->tx.txdma, &sc->tx.txbuf);
2101 }
2102 
2103 void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
2104 {
2105 	struct ath_atx_tid *tid;
2106 	struct ath_atx_ac *ac;
2107 	int tidno, acno;
2108 
2109 	for (tidno = 0, tid = &an->tid[tidno];
2110 	     tidno < WME_NUM_TID;
2111 	     tidno++, tid++) {
2112 		tid->an        = an;
2113 		tid->tidno     = tidno;
2114 		tid->seq_start = tid->seq_next = 0;
2115 		tid->baw_size  = WME_MAX_BA;
2116 		tid->baw_head  = tid->baw_tail = 0;
2117 		tid->sched     = false;
2118 		tid->paused    = false;
2119 		tid->state &= ~AGGR_CLEANUP;
2120 		INIT_LIST_HEAD(&tid->buf_q);
2121 		acno = TID_TO_WME_AC(tidno);
2122 		tid->ac = &an->ac[acno];
2123 		tid->state &= ~AGGR_ADDBA_COMPLETE;
2124 		tid->state &= ~AGGR_ADDBA_PROGRESS;
2125 		tid->addba_exchangeattempts = 0;
2126 	}
2127 
2128 	for (acno = 0, ac = &an->ac[acno];
2129 	     acno < WME_NUM_AC; acno++, ac++) {
2130 		ac->sched    = false;
2131 		INIT_LIST_HEAD(&ac->tid_q);
2132 
2133 		switch (acno) {
2134 		case WME_AC_BE:
2135 			ac->qnum = ath_tx_get_qnum(sc,
2136 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_BE);
2137 			break;
2138 		case WME_AC_BK:
2139 			ac->qnum = ath_tx_get_qnum(sc,
2140 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_BK);
2141 			break;
2142 		case WME_AC_VI:
2143 			ac->qnum = ath_tx_get_qnum(sc,
2144 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_VI);
2145 			break;
2146 		case WME_AC_VO:
2147 			ac->qnum = ath_tx_get_qnum(sc,
2148 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_VO);
2149 			break;
2150 		}
2151 	}
2152 }
2153 
2154 void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an)
2155 {
2156 	int i;
2157 	struct ath_atx_ac *ac, *ac_tmp;
2158 	struct ath_atx_tid *tid, *tid_tmp;
2159 	struct ath_txq *txq;
2160 
2161 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
2162 		if (ATH_TXQ_SETUP(sc, i)) {
2163 			txq = &sc->tx.txq[i];
2164 
2165 			spin_lock(&txq->axq_lock);
2166 
2167 			list_for_each_entry_safe(ac,
2168 					ac_tmp, &txq->axq_acq, list) {
2169 				tid = list_first_entry(&ac->tid_q,
2170 						struct ath_atx_tid, list);
2171 				if (tid && tid->an != an)
2172 					continue;
2173 				list_del(&ac->list);
2174 				ac->sched = false;
2175 
2176 				list_for_each_entry_safe(tid,
2177 						tid_tmp, &ac->tid_q, list) {
2178 					list_del(&tid->list);
2179 					tid->sched = false;
2180 					ath_tid_drain(sc, txq, tid);
2181 					tid->state &= ~AGGR_ADDBA_COMPLETE;
2182 					tid->addba_exchangeattempts = 0;
2183 					tid->state &= ~AGGR_CLEANUP;
2184 				}
2185 			}
2186 
2187 			spin_unlock(&txq->axq_lock);
2188 		}
2189 	}
2190 }
2191