xref: /openbmc/linux/drivers/net/wireless/ath/ath9k/xmit.c (revision fd589a8f)
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 ath_txq *txq,
63 				struct list_head *bf_q,
64 				int txok, int sendbar);
65 static void ath_tx_txqaddbuf(struct ath_softc *sc, struct ath_txq *txq,
66 			     struct list_head *head);
67 static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf);
68 static int ath_tx_num_badfrms(struct ath_softc *sc, struct ath_buf *bf,
69 			      int txok);
70 static void ath_tx_rc_status(struct ath_buf *bf, struct ath_desc *ds,
71 			     int nbad, int txok, bool update_rc);
72 
73 /*********************/
74 /* Aggregation logic */
75 /*********************/
76 
77 static void ath_tx_queue_tid(struct ath_txq *txq, struct ath_atx_tid *tid)
78 {
79 	struct ath_atx_ac *ac = tid->ac;
80 
81 	if (tid->paused)
82 		return;
83 
84 	if (tid->sched)
85 		return;
86 
87 	tid->sched = true;
88 	list_add_tail(&tid->list, &ac->tid_q);
89 
90 	if (ac->sched)
91 		return;
92 
93 	ac->sched = true;
94 	list_add_tail(&ac->list, &txq->axq_acq);
95 }
96 
97 static void ath_tx_pause_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
98 {
99 	struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
100 
101 	spin_lock_bh(&txq->axq_lock);
102 	tid->paused++;
103 	spin_unlock_bh(&txq->axq_lock);
104 }
105 
106 static void ath_tx_resume_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
107 {
108 	struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
109 
110 	ASSERT(tid->paused > 0);
111 	spin_lock_bh(&txq->axq_lock);
112 
113 	tid->paused--;
114 
115 	if (tid->paused > 0)
116 		goto unlock;
117 
118 	if (list_empty(&tid->buf_q))
119 		goto unlock;
120 
121 	ath_tx_queue_tid(txq, tid);
122 	ath_txq_schedule(sc, txq);
123 unlock:
124 	spin_unlock_bh(&txq->axq_lock);
125 }
126 
127 static void ath_tx_flush_tid(struct ath_softc *sc, struct ath_atx_tid *tid)
128 {
129 	struct ath_txq *txq = &sc->tx.txq[tid->ac->qnum];
130 	struct ath_buf *bf;
131 	struct list_head bf_head;
132 	INIT_LIST_HEAD(&bf_head);
133 
134 	ASSERT(tid->paused > 0);
135 	spin_lock_bh(&txq->axq_lock);
136 
137 	tid->paused--;
138 
139 	if (tid->paused > 0) {
140 		spin_unlock_bh(&txq->axq_lock);
141 		return;
142 	}
143 
144 	while (!list_empty(&tid->buf_q)) {
145 		bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
146 		ASSERT(!bf_isretried(bf));
147 		list_move_tail(&bf->list, &bf_head);
148 		ath_tx_send_ht_normal(sc, txq, tid, &bf_head);
149 	}
150 
151 	spin_unlock_bh(&txq->axq_lock);
152 }
153 
154 static void ath_tx_update_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
155 			      int seqno)
156 {
157 	int index, cindex;
158 
159 	index  = ATH_BA_INDEX(tid->seq_start, seqno);
160 	cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
161 
162 	tid->tx_buf[cindex] = NULL;
163 
164 	while (tid->baw_head != tid->baw_tail && !tid->tx_buf[tid->baw_head]) {
165 		INCR(tid->seq_start, IEEE80211_SEQ_MAX);
166 		INCR(tid->baw_head, ATH_TID_MAX_BUFS);
167 	}
168 }
169 
170 static void ath_tx_addto_baw(struct ath_softc *sc, struct ath_atx_tid *tid,
171 			     struct ath_buf *bf)
172 {
173 	int index, cindex;
174 
175 	if (bf_isretried(bf))
176 		return;
177 
178 	index  = ATH_BA_INDEX(tid->seq_start, bf->bf_seqno);
179 	cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
180 
181 	ASSERT(tid->tx_buf[cindex] == NULL);
182 	tid->tx_buf[cindex] = bf;
183 
184 	if (index >= ((tid->baw_tail - tid->baw_head) &
185 		(ATH_TID_MAX_BUFS - 1))) {
186 		tid->baw_tail = cindex;
187 		INCR(tid->baw_tail, ATH_TID_MAX_BUFS);
188 	}
189 }
190 
191 /*
192  * TODO: For frame(s) that are in the retry state, we will reuse the
193  * sequence number(s) without setting the retry bit. The
194  * alternative is to give up on these and BAR the receiver's window
195  * forward.
196  */
197 static void ath_tid_drain(struct ath_softc *sc, struct ath_txq *txq,
198 			  struct ath_atx_tid *tid)
199 
200 {
201 	struct ath_buf *bf;
202 	struct list_head bf_head;
203 	INIT_LIST_HEAD(&bf_head);
204 
205 	for (;;) {
206 		if (list_empty(&tid->buf_q))
207 			break;
208 
209 		bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
210 		list_move_tail(&bf->list, &bf_head);
211 
212 		if (bf_isretried(bf))
213 			ath_tx_update_baw(sc, tid, bf->bf_seqno);
214 
215 		spin_unlock(&txq->axq_lock);
216 		ath_tx_complete_buf(sc, bf, txq, &bf_head, 0, 0);
217 		spin_lock(&txq->axq_lock);
218 	}
219 
220 	tid->seq_next = tid->seq_start;
221 	tid->baw_tail = tid->baw_head;
222 }
223 
224 static void ath_tx_set_retry(struct ath_softc *sc, struct ath_txq *txq,
225 			     struct ath_buf *bf)
226 {
227 	struct sk_buff *skb;
228 	struct ieee80211_hdr *hdr;
229 
230 	bf->bf_state.bf_type |= BUF_RETRY;
231 	bf->bf_retries++;
232 	TX_STAT_INC(txq->axq_qnum, a_retries);
233 
234 	skb = bf->bf_mpdu;
235 	hdr = (struct ieee80211_hdr *)skb->data;
236 	hdr->frame_control |= cpu_to_le16(IEEE80211_FCTL_RETRY);
237 }
238 
239 static struct ath_buf* ath_clone_txbuf(struct ath_softc *sc, struct ath_buf *bf)
240 {
241 	struct ath_buf *tbf;
242 
243 	spin_lock_bh(&sc->tx.txbuflock);
244 	if (WARN_ON(list_empty(&sc->tx.txbuf))) {
245 		spin_unlock_bh(&sc->tx.txbuflock);
246 		return NULL;
247 	}
248 	tbf = list_first_entry(&sc->tx.txbuf, struct ath_buf, list);
249 	list_del(&tbf->list);
250 	spin_unlock_bh(&sc->tx.txbuflock);
251 
252 	ATH_TXBUF_RESET(tbf);
253 
254 	tbf->bf_mpdu = bf->bf_mpdu;
255 	tbf->bf_buf_addr = bf->bf_buf_addr;
256 	*(tbf->bf_desc) = *(bf->bf_desc);
257 	tbf->bf_state = bf->bf_state;
258 	tbf->bf_dmacontext = bf->bf_dmacontext;
259 
260 	return tbf;
261 }
262 
263 static void ath_tx_complete_aggr(struct ath_softc *sc, struct ath_txq *txq,
264 				 struct ath_buf *bf, struct list_head *bf_q,
265 				 int txok)
266 {
267 	struct ath_node *an = NULL;
268 	struct sk_buff *skb;
269 	struct ieee80211_sta *sta;
270 	struct ieee80211_hdr *hdr;
271 	struct ath_atx_tid *tid = NULL;
272 	struct ath_buf *bf_next, *bf_last = bf->bf_lastbf;
273 	struct ath_desc *ds = bf_last->bf_desc;
274 	struct list_head bf_head, bf_pending;
275 	u16 seq_st = 0, acked_cnt = 0, txfail_cnt = 0;
276 	u32 ba[WME_BA_BMP_SIZE >> 5];
277 	int isaggr, txfail, txpending, sendbar = 0, needreset = 0, nbad = 0;
278 	bool rc_update = true;
279 
280 	skb = bf->bf_mpdu;
281 	hdr = (struct ieee80211_hdr *)skb->data;
282 
283 	rcu_read_lock();
284 
285 	sta = ieee80211_find_sta(sc->hw, hdr->addr1);
286 	if (!sta) {
287 		rcu_read_unlock();
288 		return;
289 	}
290 
291 	an = (struct ath_node *)sta->drv_priv;
292 	tid = ATH_AN_2_TID(an, bf->bf_tidno);
293 
294 	isaggr = bf_isaggr(bf);
295 	memset(ba, 0, WME_BA_BMP_SIZE >> 3);
296 
297 	if (isaggr && txok) {
298 		if (ATH_DS_TX_BA(ds)) {
299 			seq_st = ATH_DS_BA_SEQ(ds);
300 			memcpy(ba, ATH_DS_BA_BITMAP(ds),
301 			       WME_BA_BMP_SIZE >> 3);
302 		} else {
303 			/*
304 			 * AR5416 can become deaf/mute when BA
305 			 * issue happens. Chip needs to be reset.
306 			 * But AP code may have sychronization issues
307 			 * when perform internal reset in this routine.
308 			 * Only enable reset in STA mode for now.
309 			 */
310 			if (sc->sc_ah->opmode == NL80211_IFTYPE_STATION)
311 				needreset = 1;
312 		}
313 	}
314 
315 	INIT_LIST_HEAD(&bf_pending);
316 	INIT_LIST_HEAD(&bf_head);
317 
318 	nbad = ath_tx_num_badfrms(sc, bf, txok);
319 	while (bf) {
320 		txfail = txpending = 0;
321 		bf_next = bf->bf_next;
322 
323 		if (ATH_BA_ISSET(ba, ATH_BA_INDEX(seq_st, bf->bf_seqno))) {
324 			/* transmit completion, subframe is
325 			 * acked by block ack */
326 			acked_cnt++;
327 		} else if (!isaggr && txok) {
328 			/* transmit completion */
329 			acked_cnt++;
330 		} else {
331 			if (!(tid->state & AGGR_CLEANUP) &&
332 			    ds->ds_txstat.ts_flags != ATH9K_TX_SW_ABORTED) {
333 				if (bf->bf_retries < ATH_MAX_SW_RETRIES) {
334 					ath_tx_set_retry(sc, txq, bf);
335 					txpending = 1;
336 				} else {
337 					bf->bf_state.bf_type |= BUF_XRETRY;
338 					txfail = 1;
339 					sendbar = 1;
340 					txfail_cnt++;
341 				}
342 			} else {
343 				/*
344 				 * cleanup in progress, just fail
345 				 * the un-acked sub-frames
346 				 */
347 				txfail = 1;
348 			}
349 		}
350 
351 		if (bf_next == NULL) {
352 			/*
353 			 * Make sure the last desc is reclaimed if it
354 			 * not a holding desc.
355 			 */
356 			if (!bf_last->bf_stale)
357 				list_move_tail(&bf->list, &bf_head);
358 			else
359 				INIT_LIST_HEAD(&bf_head);
360 		} else {
361 			ASSERT(!list_empty(bf_q));
362 			list_move_tail(&bf->list, &bf_head);
363 		}
364 
365 		if (!txpending) {
366 			/*
367 			 * complete the acked-ones/xretried ones; update
368 			 * block-ack window
369 			 */
370 			spin_lock_bh(&txq->axq_lock);
371 			ath_tx_update_baw(sc, tid, bf->bf_seqno);
372 			spin_unlock_bh(&txq->axq_lock);
373 
374 			if (rc_update && (acked_cnt == 1 || txfail_cnt == 1)) {
375 				ath_tx_rc_status(bf, ds, nbad, txok, true);
376 				rc_update = false;
377 			} else {
378 				ath_tx_rc_status(bf, ds, nbad, txok, false);
379 			}
380 
381 			ath_tx_complete_buf(sc, bf, txq, &bf_head, !txfail, sendbar);
382 		} else {
383 			/* retry the un-acked ones */
384 			if (bf->bf_next == NULL && bf_last->bf_stale) {
385 				struct ath_buf *tbf;
386 
387 				tbf = ath_clone_txbuf(sc, bf_last);
388 				/*
389 				 * Update tx baw and complete the frame with
390 				 * failed status if we run out of tx buf
391 				 */
392 				if (!tbf) {
393 					spin_lock_bh(&txq->axq_lock);
394 					ath_tx_update_baw(sc, tid,
395 							  bf->bf_seqno);
396 					spin_unlock_bh(&txq->axq_lock);
397 
398 					bf->bf_state.bf_type |= BUF_XRETRY;
399 					ath_tx_rc_status(bf, ds, nbad,
400 							 0, false);
401 					ath_tx_complete_buf(sc, bf, txq,
402 							    &bf_head, 0, 0);
403 					break;
404 				}
405 
406 				ath9k_hw_cleartxdesc(sc->sc_ah, tbf->bf_desc);
407 				list_add_tail(&tbf->list, &bf_head);
408 			} else {
409 				/*
410 				 * Clear descriptor status words for
411 				 * software retry
412 				 */
413 				ath9k_hw_cleartxdesc(sc->sc_ah, bf->bf_desc);
414 			}
415 
416 			/*
417 			 * Put this buffer to the temporary pending
418 			 * queue to retain ordering
419 			 */
420 			list_splice_tail_init(&bf_head, &bf_pending);
421 		}
422 
423 		bf = bf_next;
424 	}
425 
426 	if (tid->state & AGGR_CLEANUP) {
427 		if (tid->baw_head == tid->baw_tail) {
428 			tid->state &= ~AGGR_ADDBA_COMPLETE;
429 			tid->state &= ~AGGR_CLEANUP;
430 
431 			/* send buffered frames as singles */
432 			ath_tx_flush_tid(sc, tid);
433 		}
434 		rcu_read_unlock();
435 		return;
436 	}
437 
438 	/* prepend un-acked frames to the beginning of the pending frame queue */
439 	if (!list_empty(&bf_pending)) {
440 		spin_lock_bh(&txq->axq_lock);
441 		list_splice(&bf_pending, &tid->buf_q);
442 		ath_tx_queue_tid(txq, tid);
443 		spin_unlock_bh(&txq->axq_lock);
444 	}
445 
446 	rcu_read_unlock();
447 
448 	if (needreset)
449 		ath_reset(sc, false);
450 }
451 
452 static u32 ath_lookup_rate(struct ath_softc *sc, struct ath_buf *bf,
453 			   struct ath_atx_tid *tid)
454 {
455 	const struct ath_rate_table *rate_table = sc->cur_rate_table;
456 	struct sk_buff *skb;
457 	struct ieee80211_tx_info *tx_info;
458 	struct ieee80211_tx_rate *rates;
459 	struct ath_tx_info_priv *tx_info_priv;
460 	u32 max_4ms_framelen, frmlen;
461 	u16 aggr_limit, legacy = 0;
462 	int i;
463 
464 	skb = bf->bf_mpdu;
465 	tx_info = IEEE80211_SKB_CB(skb);
466 	rates = tx_info->control.rates;
467 	tx_info_priv = (struct ath_tx_info_priv *)tx_info->rate_driver_data[0];
468 
469 	/*
470 	 * Find the lowest frame length among the rate series that will have a
471 	 * 4ms transmit duration.
472 	 * TODO - TXOP limit needs to be considered.
473 	 */
474 	max_4ms_framelen = ATH_AMPDU_LIMIT_MAX;
475 
476 	for (i = 0; i < 4; i++) {
477 		if (rates[i].count) {
478 			if (!WLAN_RC_PHY_HT(rate_table->info[rates[i].idx].phy)) {
479 				legacy = 1;
480 				break;
481 			}
482 
483 			frmlen = rate_table->info[rates[i].idx].max_4ms_framelen;
484 			max_4ms_framelen = min(max_4ms_framelen, frmlen);
485 		}
486 	}
487 
488 	/*
489 	 * limit aggregate size by the minimum rate if rate selected is
490 	 * not a probe rate, if rate selected is a probe rate then
491 	 * avoid aggregation of this packet.
492 	 */
493 	if (tx_info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE || legacy)
494 		return 0;
495 
496 	if (sc->sc_flags & SC_OP_BT_PRIORITY_DETECTED)
497 		aggr_limit = min((max_4ms_framelen * 3) / 8,
498 				 (u32)ATH_AMPDU_LIMIT_MAX);
499 	else
500 		aggr_limit = min(max_4ms_framelen,
501 				 (u32)ATH_AMPDU_LIMIT_MAX);
502 
503 	/*
504 	 * h/w can accept aggregates upto 16 bit lengths (65535).
505 	 * The IE, however can hold upto 65536, which shows up here
506 	 * as zero. Ignore 65536 since we  are constrained by hw.
507 	 */
508 	if (tid->an->maxampdu)
509 		aggr_limit = min(aggr_limit, tid->an->maxampdu);
510 
511 	return aggr_limit;
512 }
513 
514 /*
515  * Returns the number of delimiters to be added to
516  * meet the minimum required mpdudensity.
517  */
518 static int ath_compute_num_delims(struct ath_softc *sc, struct ath_atx_tid *tid,
519 				  struct ath_buf *bf, u16 frmlen)
520 {
521 	const struct ath_rate_table *rt = sc->cur_rate_table;
522 	struct sk_buff *skb = bf->bf_mpdu;
523 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
524 	u32 nsymbits, nsymbols;
525 	u16 minlen;
526 	u8 rc, flags, rix;
527 	int width, half_gi, ndelim, mindelim;
528 
529 	/* Select standard number of delimiters based on frame length alone */
530 	ndelim = ATH_AGGR_GET_NDELIM(frmlen);
531 
532 	/*
533 	 * If encryption enabled, hardware requires some more padding between
534 	 * subframes.
535 	 * TODO - this could be improved to be dependent on the rate.
536 	 *      The hardware can keep up at lower rates, but not higher rates
537 	 */
538 	if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR)
539 		ndelim += ATH_AGGR_ENCRYPTDELIM;
540 
541 	/*
542 	 * Convert desired mpdu density from microeconds to bytes based
543 	 * on highest rate in rate series (i.e. first rate) to determine
544 	 * required minimum length for subframe. Take into account
545 	 * whether high rate is 20 or 40Mhz and half or full GI.
546 	 *
547 	 * If there is no mpdu density restriction, no further calculation
548 	 * is needed.
549 	 */
550 
551 	if (tid->an->mpdudensity == 0)
552 		return ndelim;
553 
554 	rix = tx_info->control.rates[0].idx;
555 	flags = tx_info->control.rates[0].flags;
556 	rc = rt->info[rix].ratecode;
557 	width = (flags & IEEE80211_TX_RC_40_MHZ_WIDTH) ? 1 : 0;
558 	half_gi = (flags & IEEE80211_TX_RC_SHORT_GI) ? 1 : 0;
559 
560 	if (half_gi)
561 		nsymbols = NUM_SYMBOLS_PER_USEC_HALFGI(tid->an->mpdudensity);
562 	else
563 		nsymbols = NUM_SYMBOLS_PER_USEC(tid->an->mpdudensity);
564 
565 	if (nsymbols == 0)
566 		nsymbols = 1;
567 
568 	nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
569 	minlen = (nsymbols * nsymbits) / BITS_PER_BYTE;
570 
571 	if (frmlen < minlen) {
572 		mindelim = (minlen - frmlen) / ATH_AGGR_DELIM_SZ;
573 		ndelim = max(mindelim, ndelim);
574 	}
575 
576 	return ndelim;
577 }
578 
579 static enum ATH_AGGR_STATUS ath_tx_form_aggr(struct ath_softc *sc,
580 					     struct ath_txq *txq,
581 					     struct ath_atx_tid *tid,
582 					     struct list_head *bf_q)
583 {
584 #define PADBYTES(_len) ((4 - ((_len) % 4)) % 4)
585 	struct ath_buf *bf, *bf_first, *bf_prev = NULL;
586 	int rl = 0, nframes = 0, ndelim, prev_al = 0;
587 	u16 aggr_limit = 0, al = 0, bpad = 0,
588 		al_delta, h_baw = tid->baw_size / 2;
589 	enum ATH_AGGR_STATUS status = ATH_AGGR_DONE;
590 
591 	bf_first = list_first_entry(&tid->buf_q, struct ath_buf, list);
592 
593 	do {
594 		bf = list_first_entry(&tid->buf_q, struct ath_buf, list);
595 
596 		/* do not step over block-ack window */
597 		if (!BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno)) {
598 			status = ATH_AGGR_BAW_CLOSED;
599 			break;
600 		}
601 
602 		if (!rl) {
603 			aggr_limit = ath_lookup_rate(sc, bf, tid);
604 			rl = 1;
605 		}
606 
607 		/* do not exceed aggregation limit */
608 		al_delta = ATH_AGGR_DELIM_SZ + bf->bf_frmlen;
609 
610 		if (nframes &&
611 		    (aggr_limit < (al + bpad + al_delta + prev_al))) {
612 			status = ATH_AGGR_LIMITED;
613 			break;
614 		}
615 
616 		/* do not exceed subframe limit */
617 		if (nframes >= min((int)h_baw, ATH_AMPDU_SUBFRAME_DEFAULT)) {
618 			status = ATH_AGGR_LIMITED;
619 			break;
620 		}
621 		nframes++;
622 
623 		/* add padding for previous frame to aggregation length */
624 		al += bpad + al_delta;
625 
626 		/*
627 		 * Get the delimiters needed to meet the MPDU
628 		 * density for this node.
629 		 */
630 		ndelim = ath_compute_num_delims(sc, tid, bf_first, bf->bf_frmlen);
631 		bpad = PADBYTES(al_delta) + (ndelim << 2);
632 
633 		bf->bf_next = NULL;
634 		bf->bf_desc->ds_link = 0;
635 
636 		/* link buffers of this frame to the aggregate */
637 		ath_tx_addto_baw(sc, tid, bf);
638 		ath9k_hw_set11n_aggr_middle(sc->sc_ah, bf->bf_desc, ndelim);
639 		list_move_tail(&bf->list, bf_q);
640 		if (bf_prev) {
641 			bf_prev->bf_next = bf;
642 			bf_prev->bf_desc->ds_link = bf->bf_daddr;
643 		}
644 		bf_prev = bf;
645 
646 	} while (!list_empty(&tid->buf_q));
647 
648 	bf_first->bf_al = al;
649 	bf_first->bf_nframes = nframes;
650 
651 	return status;
652 #undef PADBYTES
653 }
654 
655 static void ath_tx_sched_aggr(struct ath_softc *sc, struct ath_txq *txq,
656 			      struct ath_atx_tid *tid)
657 {
658 	struct ath_buf *bf;
659 	enum ATH_AGGR_STATUS status;
660 	struct list_head bf_q;
661 
662 	do {
663 		if (list_empty(&tid->buf_q))
664 			return;
665 
666 		INIT_LIST_HEAD(&bf_q);
667 
668 		status = ath_tx_form_aggr(sc, txq, tid, &bf_q);
669 
670 		/*
671 		 * no frames picked up to be aggregated;
672 		 * block-ack window is not open.
673 		 */
674 		if (list_empty(&bf_q))
675 			break;
676 
677 		bf = list_first_entry(&bf_q, struct ath_buf, list);
678 		bf->bf_lastbf = list_entry(bf_q.prev, struct ath_buf, list);
679 
680 		/* if only one frame, send as non-aggregate */
681 		if (bf->bf_nframes == 1) {
682 			bf->bf_state.bf_type &= ~BUF_AGGR;
683 			ath9k_hw_clr11n_aggr(sc->sc_ah, bf->bf_desc);
684 			ath_buf_set_rate(sc, bf);
685 			ath_tx_txqaddbuf(sc, txq, &bf_q);
686 			continue;
687 		}
688 
689 		/* setup first desc of aggregate */
690 		bf->bf_state.bf_type |= BUF_AGGR;
691 		ath_buf_set_rate(sc, bf);
692 		ath9k_hw_set11n_aggr_first(sc->sc_ah, bf->bf_desc, bf->bf_al);
693 
694 		/* anchor last desc of aggregate */
695 		ath9k_hw_set11n_aggr_last(sc->sc_ah, bf->bf_lastbf->bf_desc);
696 
697 		txq->axq_aggr_depth++;
698 		ath_tx_txqaddbuf(sc, txq, &bf_q);
699 		TX_STAT_INC(txq->axq_qnum, a_aggr);
700 
701 	} while (txq->axq_depth < ATH_AGGR_MIN_QDEPTH &&
702 		 status != ATH_AGGR_BAW_CLOSED);
703 }
704 
705 void ath_tx_aggr_start(struct ath_softc *sc, struct ieee80211_sta *sta,
706 		       u16 tid, u16 *ssn)
707 {
708 	struct ath_atx_tid *txtid;
709 	struct ath_node *an;
710 
711 	an = (struct ath_node *)sta->drv_priv;
712 	txtid = ATH_AN_2_TID(an, tid);
713 	txtid->state |= AGGR_ADDBA_PROGRESS;
714 	ath_tx_pause_tid(sc, txtid);
715 	*ssn = txtid->seq_start;
716 }
717 
718 void ath_tx_aggr_stop(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
719 {
720 	struct ath_node *an = (struct ath_node *)sta->drv_priv;
721 	struct ath_atx_tid *txtid = ATH_AN_2_TID(an, tid);
722 	struct ath_txq *txq = &sc->tx.txq[txtid->ac->qnum];
723 	struct ath_buf *bf;
724 	struct list_head bf_head;
725 	INIT_LIST_HEAD(&bf_head);
726 
727 	if (txtid->state & AGGR_CLEANUP)
728 		return;
729 
730 	if (!(txtid->state & AGGR_ADDBA_COMPLETE)) {
731 		txtid->state &= ~AGGR_ADDBA_PROGRESS;
732 		return;
733 	}
734 
735 	ath_tx_pause_tid(sc, txtid);
736 
737 	/* drop all software retried frames and mark this TID */
738 	spin_lock_bh(&txq->axq_lock);
739 	while (!list_empty(&txtid->buf_q)) {
740 		bf = list_first_entry(&txtid->buf_q, struct ath_buf, list);
741 		if (!bf_isretried(bf)) {
742 			/*
743 			 * NB: it's based on the assumption that
744 			 * software retried frame will always stay
745 			 * at the head of software queue.
746 			 */
747 			break;
748 		}
749 		list_move_tail(&bf->list, &bf_head);
750 		ath_tx_update_baw(sc, txtid, bf->bf_seqno);
751 		ath_tx_complete_buf(sc, bf, txq, &bf_head, 0, 0);
752 	}
753 	spin_unlock_bh(&txq->axq_lock);
754 
755 	if (txtid->baw_head != txtid->baw_tail) {
756 		txtid->state |= AGGR_CLEANUP;
757 	} else {
758 		txtid->state &= ~AGGR_ADDBA_COMPLETE;
759 		ath_tx_flush_tid(sc, txtid);
760 	}
761 }
762 
763 void ath_tx_aggr_resume(struct ath_softc *sc, struct ieee80211_sta *sta, u16 tid)
764 {
765 	struct ath_atx_tid *txtid;
766 	struct ath_node *an;
767 
768 	an = (struct ath_node *)sta->drv_priv;
769 
770 	if (sc->sc_flags & SC_OP_TXAGGR) {
771 		txtid = ATH_AN_2_TID(an, tid);
772 		txtid->baw_size =
773 			IEEE80211_MIN_AMPDU_BUF << sta->ht_cap.ampdu_factor;
774 		txtid->state |= AGGR_ADDBA_COMPLETE;
775 		txtid->state &= ~AGGR_ADDBA_PROGRESS;
776 		ath_tx_resume_tid(sc, txtid);
777 	}
778 }
779 
780 bool ath_tx_aggr_check(struct ath_softc *sc, struct ath_node *an, u8 tidno)
781 {
782 	struct ath_atx_tid *txtid;
783 
784 	if (!(sc->sc_flags & SC_OP_TXAGGR))
785 		return false;
786 
787 	txtid = ATH_AN_2_TID(an, tidno);
788 
789 	if (!(txtid->state & (AGGR_ADDBA_COMPLETE | AGGR_ADDBA_PROGRESS)))
790 			return true;
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_linkbuf = NULL;
874 		txq->axq_tx_inprogress = false;
875 		sc->tx.txqsetup |= 1<<qnum;
876 	}
877 	return &sc->tx.txq[qnum];
878 }
879 
880 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, txq, &bf_head, 0, 0);
1039 	}
1040 
1041 	spin_lock_bh(&txq->axq_lock);
1042 	txq->axq_tx_inprogress = false;
1043 	spin_unlock_bh(&txq->axq_lock);
1044 
1045 	/* flush any pending frames if aggregation is enabled */
1046 	if (sc->sc_flags & SC_OP_TXAGGR) {
1047 		if (!retry_tx) {
1048 			spin_lock_bh(&txq->axq_lock);
1049 			ath_txq_drain_pending_buffers(sc, txq);
1050 			spin_unlock_bh(&txq->axq_lock);
1051 		}
1052 	}
1053 }
1054 
1055 void ath_drain_all_txq(struct ath_softc *sc, bool retry_tx)
1056 {
1057 	struct ath_hw *ah = sc->sc_ah;
1058 	struct ath_txq *txq;
1059 	int i, npend = 0;
1060 
1061 	if (sc->sc_flags & SC_OP_INVALID)
1062 		return;
1063 
1064 	/* Stop beacon queue */
1065 	ath9k_hw_stoptxdma(sc->sc_ah, sc->beacon.beaconq);
1066 
1067 	/* Stop data queues */
1068 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
1069 		if (ATH_TXQ_SETUP(sc, i)) {
1070 			txq = &sc->tx.txq[i];
1071 			ath9k_hw_stoptxdma(ah, txq->axq_qnum);
1072 			npend += ath9k_hw_numtxpending(ah, txq->axq_qnum);
1073 		}
1074 	}
1075 
1076 	if (npend) {
1077 		int r;
1078 
1079 		DPRINTF(sc, ATH_DBG_XMIT, "Unable to stop TxDMA. Reset HAL!\n");
1080 
1081 		spin_lock_bh(&sc->sc_resetlock);
1082 		r = ath9k_hw_reset(ah, sc->sc_ah->curchan, true);
1083 		if (r)
1084 			DPRINTF(sc, ATH_DBG_FATAL,
1085 				"Unable to reset hardware; reset status %d\n",
1086 				r);
1087 		spin_unlock_bh(&sc->sc_resetlock);
1088 	}
1089 
1090 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
1091 		if (ATH_TXQ_SETUP(sc, i))
1092 			ath_draintxq(sc, &sc->tx.txq[i], retry_tx);
1093 	}
1094 }
1095 
1096 void ath_tx_cleanupq(struct ath_softc *sc, struct ath_txq *txq)
1097 {
1098 	ath9k_hw_releasetxqueue(sc->sc_ah, txq->axq_qnum);
1099 	sc->tx.txqsetup &= ~(1<<txq->axq_qnum);
1100 }
1101 
1102 void ath_txq_schedule(struct ath_softc *sc, struct ath_txq *txq)
1103 {
1104 	struct ath_atx_ac *ac;
1105 	struct ath_atx_tid *tid;
1106 
1107 	if (list_empty(&txq->axq_acq))
1108 		return;
1109 
1110 	ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac, list);
1111 	list_del(&ac->list);
1112 	ac->sched = false;
1113 
1114 	do {
1115 		if (list_empty(&ac->tid_q))
1116 			return;
1117 
1118 		tid = list_first_entry(&ac->tid_q, struct ath_atx_tid, list);
1119 		list_del(&tid->list);
1120 		tid->sched = false;
1121 
1122 		if (tid->paused)
1123 			continue;
1124 
1125 		ath_tx_sched_aggr(sc, txq, tid);
1126 
1127 		/*
1128 		 * add tid to round-robin queue if more frames
1129 		 * are pending for the tid
1130 		 */
1131 		if (!list_empty(&tid->buf_q))
1132 			ath_tx_queue_tid(txq, tid);
1133 
1134 		break;
1135 	} while (!list_empty(&ac->tid_q));
1136 
1137 	if (!list_empty(&ac->tid_q)) {
1138 		if (!ac->sched) {
1139 			ac->sched = true;
1140 			list_add_tail(&ac->list, &txq->axq_acq);
1141 		}
1142 	}
1143 }
1144 
1145 int ath_tx_setup(struct ath_softc *sc, int haltype)
1146 {
1147 	struct ath_txq *txq;
1148 
1149 	if (haltype >= ARRAY_SIZE(sc->tx.hwq_map)) {
1150 		DPRINTF(sc, ATH_DBG_FATAL,
1151 			"HAL AC %u out of range, max %zu!\n",
1152 			 haltype, ARRAY_SIZE(sc->tx.hwq_map));
1153 		return 0;
1154 	}
1155 	txq = ath_txq_setup(sc, ATH9K_TX_QUEUE_DATA, haltype);
1156 	if (txq != NULL) {
1157 		sc->tx.hwq_map[haltype] = txq->axq_qnum;
1158 		return 1;
1159 	} else
1160 		return 0;
1161 }
1162 
1163 /***********/
1164 /* TX, DMA */
1165 /***********/
1166 
1167 /*
1168  * Insert a chain of ath_buf (descriptors) on a txq and
1169  * assume the descriptors are already chained together by caller.
1170  */
1171 static void ath_tx_txqaddbuf(struct ath_softc *sc, struct ath_txq *txq,
1172 			     struct list_head *head)
1173 {
1174 	struct ath_hw *ah = sc->sc_ah;
1175 	struct ath_buf *bf;
1176 
1177 	/*
1178 	 * Insert the frame on the outbound list and
1179 	 * pass it on to the hardware.
1180 	 */
1181 
1182 	if (list_empty(head))
1183 		return;
1184 
1185 	bf = list_first_entry(head, struct ath_buf, list);
1186 
1187 	list_splice_tail_init(head, &txq->axq_q);
1188 	txq->axq_depth++;
1189 	txq->axq_linkbuf = list_entry(txq->axq_q.prev, struct ath_buf, list);
1190 
1191 	DPRINTF(sc, ATH_DBG_QUEUE,
1192 		"qnum: %d, txq depth: %d\n", txq->axq_qnum, txq->axq_depth);
1193 
1194 	if (txq->axq_link == NULL) {
1195 		ath9k_hw_puttxbuf(ah, txq->axq_qnum, bf->bf_daddr);
1196 		DPRINTF(sc, ATH_DBG_XMIT,
1197 			"TXDP[%u] = %llx (%p)\n",
1198 			txq->axq_qnum, ito64(bf->bf_daddr), bf->bf_desc);
1199 	} else {
1200 		*txq->axq_link = bf->bf_daddr;
1201 		DPRINTF(sc, ATH_DBG_XMIT, "link[%u] (%p)=%llx (%p)\n",
1202 			txq->axq_qnum, txq->axq_link,
1203 			ito64(bf->bf_daddr), bf->bf_desc);
1204 	}
1205 	txq->axq_link = &(bf->bf_lastbf->bf_desc->ds_link);
1206 	ath9k_hw_txstart(ah, txq->axq_qnum);
1207 }
1208 
1209 static struct ath_buf *ath_tx_get_buffer(struct ath_softc *sc)
1210 {
1211 	struct ath_buf *bf = NULL;
1212 
1213 	spin_lock_bh(&sc->tx.txbuflock);
1214 
1215 	if (unlikely(list_empty(&sc->tx.txbuf))) {
1216 		spin_unlock_bh(&sc->tx.txbuflock);
1217 		return NULL;
1218 	}
1219 
1220 	bf = list_first_entry(&sc->tx.txbuf, struct ath_buf, list);
1221 	list_del(&bf->list);
1222 
1223 	spin_unlock_bh(&sc->tx.txbuflock);
1224 
1225 	return bf;
1226 }
1227 
1228 static void ath_tx_send_ampdu(struct ath_softc *sc, struct ath_atx_tid *tid,
1229 			      struct list_head *bf_head,
1230 			      struct ath_tx_control *txctl)
1231 {
1232 	struct ath_buf *bf;
1233 
1234 	bf = list_first_entry(bf_head, struct ath_buf, list);
1235 	bf->bf_state.bf_type |= BUF_AMPDU;
1236 	TX_STAT_INC(txctl->txq->axq_qnum, a_queued);
1237 
1238 	/*
1239 	 * Do not queue to h/w when any of the following conditions is true:
1240 	 * - there are pending frames in software queue
1241 	 * - the TID is currently paused for ADDBA/BAR request
1242 	 * - seqno is not within block-ack window
1243 	 * - h/w queue depth exceeds low water mark
1244 	 */
1245 	if (!list_empty(&tid->buf_q) || tid->paused ||
1246 	    !BAW_WITHIN(tid->seq_start, tid->baw_size, bf->bf_seqno) ||
1247 	    txctl->txq->axq_depth >= ATH_AGGR_MIN_QDEPTH) {
1248 		/*
1249 		 * Add this frame to software queue for scheduling later
1250 		 * for aggregation.
1251 		 */
1252 		list_move_tail(&bf->list, &tid->buf_q);
1253 		ath_tx_queue_tid(txctl->txq, tid);
1254 		return;
1255 	}
1256 
1257 	/* Add sub-frame to BAW */
1258 	ath_tx_addto_baw(sc, tid, bf);
1259 
1260 	/* Queue to h/w without aggregation */
1261 	bf->bf_nframes = 1;
1262 	bf->bf_lastbf = bf;
1263 	ath_buf_set_rate(sc, bf);
1264 	ath_tx_txqaddbuf(sc, txctl->txq, bf_head);
1265 }
1266 
1267 static void ath_tx_send_ht_normal(struct ath_softc *sc, struct ath_txq *txq,
1268 				  struct ath_atx_tid *tid,
1269 				  struct list_head *bf_head)
1270 {
1271 	struct ath_buf *bf;
1272 
1273 	bf = list_first_entry(bf_head, struct ath_buf, list);
1274 	bf->bf_state.bf_type &= ~BUF_AMPDU;
1275 
1276 	/* update starting sequence number for subsequent ADDBA request */
1277 	INCR(tid->seq_start, IEEE80211_SEQ_MAX);
1278 
1279 	bf->bf_nframes = 1;
1280 	bf->bf_lastbf = bf;
1281 	ath_buf_set_rate(sc, bf);
1282 	ath_tx_txqaddbuf(sc, txq, bf_head);
1283 	TX_STAT_INC(txq->axq_qnum, queued);
1284 }
1285 
1286 static void ath_tx_send_normal(struct ath_softc *sc, struct ath_txq *txq,
1287 			       struct list_head *bf_head)
1288 {
1289 	struct ath_buf *bf;
1290 
1291 	bf = list_first_entry(bf_head, struct ath_buf, list);
1292 
1293 	bf->bf_lastbf = bf;
1294 	bf->bf_nframes = 1;
1295 	ath_buf_set_rate(sc, bf);
1296 	ath_tx_txqaddbuf(sc, txq, bf_head);
1297 	TX_STAT_INC(txq->axq_qnum, queued);
1298 }
1299 
1300 static enum ath9k_pkt_type get_hw_packet_type(struct sk_buff *skb)
1301 {
1302 	struct ieee80211_hdr *hdr;
1303 	enum ath9k_pkt_type htype;
1304 	__le16 fc;
1305 
1306 	hdr = (struct ieee80211_hdr *)skb->data;
1307 	fc = hdr->frame_control;
1308 
1309 	if (ieee80211_is_beacon(fc))
1310 		htype = ATH9K_PKT_TYPE_BEACON;
1311 	else if (ieee80211_is_probe_resp(fc))
1312 		htype = ATH9K_PKT_TYPE_PROBE_RESP;
1313 	else if (ieee80211_is_atim(fc))
1314 		htype = ATH9K_PKT_TYPE_ATIM;
1315 	else if (ieee80211_is_pspoll(fc))
1316 		htype = ATH9K_PKT_TYPE_PSPOLL;
1317 	else
1318 		htype = ATH9K_PKT_TYPE_NORMAL;
1319 
1320 	return htype;
1321 }
1322 
1323 static bool is_pae(struct sk_buff *skb)
1324 {
1325 	struct ieee80211_hdr *hdr;
1326 	__le16 fc;
1327 
1328 	hdr = (struct ieee80211_hdr *)skb->data;
1329 	fc = hdr->frame_control;
1330 
1331 	if (ieee80211_is_data(fc)) {
1332 		if (ieee80211_is_nullfunc(fc) ||
1333 		    /* Port Access Entity (IEEE 802.1X) */
1334 		    (skb->protocol == cpu_to_be16(ETH_P_PAE))) {
1335 			return true;
1336 		}
1337 	}
1338 
1339 	return false;
1340 }
1341 
1342 static int get_hw_crypto_keytype(struct sk_buff *skb)
1343 {
1344 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1345 
1346 	if (tx_info->control.hw_key) {
1347 		if (tx_info->control.hw_key->alg == ALG_WEP)
1348 			return ATH9K_KEY_TYPE_WEP;
1349 		else if (tx_info->control.hw_key->alg == ALG_TKIP)
1350 			return ATH9K_KEY_TYPE_TKIP;
1351 		else if (tx_info->control.hw_key->alg == ALG_CCMP)
1352 			return ATH9K_KEY_TYPE_AES;
1353 	}
1354 
1355 	return ATH9K_KEY_TYPE_CLEAR;
1356 }
1357 
1358 static void assign_aggr_tid_seqno(struct sk_buff *skb,
1359 				  struct ath_buf *bf)
1360 {
1361 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1362 	struct ieee80211_hdr *hdr;
1363 	struct ath_node *an;
1364 	struct ath_atx_tid *tid;
1365 	__le16 fc;
1366 	u8 *qc;
1367 
1368 	if (!tx_info->control.sta)
1369 		return;
1370 
1371 	an = (struct ath_node *)tx_info->control.sta->drv_priv;
1372 	hdr = (struct ieee80211_hdr *)skb->data;
1373 	fc = hdr->frame_control;
1374 
1375 	if (ieee80211_is_data_qos(fc)) {
1376 		qc = ieee80211_get_qos_ctl(hdr);
1377 		bf->bf_tidno = qc[0] & 0xf;
1378 	}
1379 
1380 	/*
1381 	 * For HT capable stations, we save tidno for later use.
1382 	 * We also override seqno set by upper layer with the one
1383 	 * in tx aggregation state.
1384 	 *
1385 	 * If fragmentation is on, the sequence number is
1386 	 * not overridden, since it has been
1387 	 * incremented by the fragmentation routine.
1388 	 *
1389 	 * FIXME: check if the fragmentation threshold exceeds
1390 	 * IEEE80211 max.
1391 	 */
1392 	tid = ATH_AN_2_TID(an, bf->bf_tidno);
1393 	hdr->seq_ctrl = cpu_to_le16(tid->seq_next <<
1394 			IEEE80211_SEQ_SEQ_SHIFT);
1395 	bf->bf_seqno = tid->seq_next;
1396 	INCR(tid->seq_next, IEEE80211_SEQ_MAX);
1397 }
1398 
1399 static int setup_tx_flags(struct ath_softc *sc, struct sk_buff *skb,
1400 			  struct ath_txq *txq)
1401 {
1402 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1403 	int flags = 0;
1404 
1405 	flags |= ATH9K_TXDESC_CLRDMASK; /* needed for crypto errors */
1406 	flags |= ATH9K_TXDESC_INTREQ;
1407 
1408 	if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK)
1409 		flags |= ATH9K_TXDESC_NOACK;
1410 
1411 	return flags;
1412 }
1413 
1414 /*
1415  * rix - rate index
1416  * pktlen - total bytes (delims + data + fcs + pads + pad delims)
1417  * width  - 0 for 20 MHz, 1 for 40 MHz
1418  * half_gi - to use 4us v/s 3.6 us for symbol time
1419  */
1420 static u32 ath_pkt_duration(struct ath_softc *sc, u8 rix, struct ath_buf *bf,
1421 			    int width, int half_gi, bool shortPreamble)
1422 {
1423 	const struct ath_rate_table *rate_table = sc->cur_rate_table;
1424 	u32 nbits, nsymbits, duration, nsymbols;
1425 	u8 rc;
1426 	int streams, pktlen;
1427 
1428 	pktlen = bf_isaggr(bf) ? bf->bf_al : bf->bf_frmlen;
1429 	rc = rate_table->info[rix].ratecode;
1430 
1431 	/* for legacy rates, use old function to compute packet duration */
1432 	if (!IS_HT_RATE(rc))
1433 		return ath9k_hw_computetxtime(sc->sc_ah, rate_table, pktlen,
1434 					      rix, shortPreamble);
1435 
1436 	/* find number of symbols: PLCP + data */
1437 	nbits = (pktlen << 3) + OFDM_PLCP_BITS;
1438 	nsymbits = bits_per_symbol[HT_RC_2_MCS(rc)][width];
1439 	nsymbols = (nbits + nsymbits - 1) / nsymbits;
1440 
1441 	if (!half_gi)
1442 		duration = SYMBOL_TIME(nsymbols);
1443 	else
1444 		duration = SYMBOL_TIME_HALFGI(nsymbols);
1445 
1446 	/* addup duration for legacy/ht training and signal fields */
1447 	streams = HT_RC_2_STREAMS(rc);
1448 	duration += L_STF + L_LTF + L_SIG + HT_SIG + HT_STF + HT_LTF(streams);
1449 
1450 	return duration;
1451 }
1452 
1453 static void ath_buf_set_rate(struct ath_softc *sc, struct ath_buf *bf)
1454 {
1455 	const struct ath_rate_table *rt = sc->cur_rate_table;
1456 	struct ath9k_11n_rate_series series[4];
1457 	struct sk_buff *skb;
1458 	struct ieee80211_tx_info *tx_info;
1459 	struct ieee80211_tx_rate *rates;
1460 	struct ieee80211_hdr *hdr;
1461 	int i, flags = 0;
1462 	u8 rix = 0, ctsrate = 0;
1463 	bool is_pspoll;
1464 
1465 	memset(series, 0, sizeof(struct ath9k_11n_rate_series) * 4);
1466 
1467 	skb = bf->bf_mpdu;
1468 	tx_info = IEEE80211_SKB_CB(skb);
1469 	rates = tx_info->control.rates;
1470 	hdr = (struct ieee80211_hdr *)skb->data;
1471 	is_pspoll = ieee80211_is_pspoll(hdr->frame_control);
1472 
1473 	/*
1474 	 * We check if Short Preamble is needed for the CTS rate by
1475 	 * checking the BSS's global flag.
1476 	 * But for the rate series, IEEE80211_TX_RC_USE_SHORT_PREAMBLE is used.
1477 	 */
1478 	if (sc->sc_flags & SC_OP_PREAMBLE_SHORT)
1479 		ctsrate = rt->info[tx_info->control.rts_cts_rate_idx].ratecode |
1480 			rt->info[tx_info->control.rts_cts_rate_idx].short_preamble;
1481 	else
1482 		ctsrate = rt->info[tx_info->control.rts_cts_rate_idx].ratecode;
1483 
1484 	/*
1485 	 * ATH9K_TXDESC_RTSENA and ATH9K_TXDESC_CTSENA are mutually exclusive.
1486 	 * Check the first rate in the series to decide whether RTS/CTS
1487 	 * or CTS-to-self has to be used.
1488 	 */
1489 	if (rates[0].flags & IEEE80211_TX_RC_USE_CTS_PROTECT)
1490 		flags = ATH9K_TXDESC_CTSENA;
1491 	else if (rates[0].flags & IEEE80211_TX_RC_USE_RTS_CTS)
1492 		flags = ATH9K_TXDESC_RTSENA;
1493 
1494 	/* FIXME: Handle aggregation protection */
1495 	if (sc->config.ath_aggr_prot &&
1496 	    (!bf_isaggr(bf) || (bf_isaggr(bf) && bf->bf_al < 8192))) {
1497 		flags = ATH9K_TXDESC_RTSENA;
1498 	}
1499 
1500 	/* For AR5416 - RTS cannot be followed by a frame larger than 8K */
1501 	if (bf_isaggr(bf) && (bf->bf_al > sc->sc_ah->caps.rts_aggr_limit))
1502 		flags &= ~(ATH9K_TXDESC_RTSENA);
1503 
1504 	for (i = 0; i < 4; i++) {
1505 		if (!rates[i].count || (rates[i].idx < 0))
1506 			continue;
1507 
1508 		rix = rates[i].idx;
1509 		series[i].Tries = rates[i].count;
1510 		series[i].ChSel = sc->tx_chainmask;
1511 
1512 		if (rates[i].flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE)
1513 			series[i].Rate = rt->info[rix].ratecode |
1514 				rt->info[rix].short_preamble;
1515 		else
1516 			series[i].Rate = rt->info[rix].ratecode;
1517 
1518 		if (rates[i].flags & IEEE80211_TX_RC_USE_RTS_CTS)
1519 			series[i].RateFlags |= ATH9K_RATESERIES_RTS_CTS;
1520 		if (rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH)
1521 			series[i].RateFlags |= ATH9K_RATESERIES_2040;
1522 		if (rates[i].flags & IEEE80211_TX_RC_SHORT_GI)
1523 			series[i].RateFlags |= ATH9K_RATESERIES_HALFGI;
1524 
1525 		series[i].PktDuration = ath_pkt_duration(sc, rix, bf,
1526 			 (rates[i].flags & IEEE80211_TX_RC_40_MHZ_WIDTH) != 0,
1527 			 (rates[i].flags & IEEE80211_TX_RC_SHORT_GI),
1528 			 (rates[i].flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE));
1529 	}
1530 
1531 	/* set dur_update_en for l-sig computation except for PS-Poll frames */
1532 	ath9k_hw_set11n_ratescenario(sc->sc_ah, bf->bf_desc,
1533 				     bf->bf_lastbf->bf_desc,
1534 				     !is_pspoll, ctsrate,
1535 				     0, series, 4, flags);
1536 
1537 	if (sc->config.ath_aggr_prot && flags)
1538 		ath9k_hw_set11n_burstduration(sc->sc_ah, bf->bf_desc, 8192);
1539 }
1540 
1541 static int ath_tx_setup_buffer(struct ieee80211_hw *hw, struct ath_buf *bf,
1542 				struct sk_buff *skb,
1543 				struct ath_tx_control *txctl)
1544 {
1545 	struct ath_wiphy *aphy = hw->priv;
1546 	struct ath_softc *sc = aphy->sc;
1547 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1548 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1549 	struct ath_tx_info_priv *tx_info_priv;
1550 	int hdrlen;
1551 	__le16 fc;
1552 
1553 	tx_info_priv = kzalloc(sizeof(*tx_info_priv), GFP_ATOMIC);
1554 	if (unlikely(!tx_info_priv))
1555 		return -ENOMEM;
1556 	tx_info->rate_driver_data[0] = tx_info_priv;
1557 	tx_info_priv->aphy = aphy;
1558 	tx_info_priv->frame_type = txctl->frame_type;
1559 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
1560 	fc = hdr->frame_control;
1561 
1562 	ATH_TXBUF_RESET(bf);
1563 
1564 	bf->bf_frmlen = skb->len + FCS_LEN - (hdrlen & 3);
1565 
1566 	if (conf_is_ht(&sc->hw->conf) && !is_pae(skb))
1567 		bf->bf_state.bf_type |= BUF_HT;
1568 
1569 	bf->bf_flags = setup_tx_flags(sc, skb, txctl->txq);
1570 
1571 	bf->bf_keytype = get_hw_crypto_keytype(skb);
1572 	if (bf->bf_keytype != ATH9K_KEY_TYPE_CLEAR) {
1573 		bf->bf_frmlen += tx_info->control.hw_key->icv_len;
1574 		bf->bf_keyix = tx_info->control.hw_key->hw_key_idx;
1575 	} else {
1576 		bf->bf_keyix = ATH9K_TXKEYIX_INVALID;
1577 	}
1578 
1579 	if (ieee80211_is_data_qos(fc) && (sc->sc_flags & SC_OP_TXAGGR))
1580 		assign_aggr_tid_seqno(skb, bf);
1581 
1582 	bf->bf_mpdu = skb;
1583 
1584 	bf->bf_dmacontext = dma_map_single(sc->dev, skb->data,
1585 					   skb->len, DMA_TO_DEVICE);
1586 	if (unlikely(dma_mapping_error(sc->dev, bf->bf_dmacontext))) {
1587 		bf->bf_mpdu = NULL;
1588 		kfree(tx_info_priv);
1589 		tx_info->rate_driver_data[0] = NULL;
1590 		DPRINTF(sc, ATH_DBG_FATAL, "dma_mapping_error() on TX\n");
1591 		return -ENOMEM;
1592 	}
1593 
1594 	bf->bf_buf_addr = bf->bf_dmacontext;
1595 	return 0;
1596 }
1597 
1598 /* FIXME: tx power */
1599 static void ath_tx_start_dma(struct ath_softc *sc, struct ath_buf *bf,
1600 			     struct ath_tx_control *txctl)
1601 {
1602 	struct sk_buff *skb = bf->bf_mpdu;
1603 	struct ieee80211_tx_info *tx_info =  IEEE80211_SKB_CB(skb);
1604 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1605 	struct ath_node *an = NULL;
1606 	struct list_head bf_head;
1607 	struct ath_desc *ds;
1608 	struct ath_atx_tid *tid;
1609 	struct ath_hw *ah = sc->sc_ah;
1610 	int frm_type;
1611 	__le16 fc;
1612 
1613 	frm_type = get_hw_packet_type(skb);
1614 	fc = hdr->frame_control;
1615 
1616 	INIT_LIST_HEAD(&bf_head);
1617 	list_add_tail(&bf->list, &bf_head);
1618 
1619 	ds = bf->bf_desc;
1620 	ds->ds_link = 0;
1621 	ds->ds_data = bf->bf_buf_addr;
1622 
1623 	ath9k_hw_set11n_txdesc(ah, ds, bf->bf_frmlen, frm_type, MAX_RATE_POWER,
1624 			       bf->bf_keyix, bf->bf_keytype, bf->bf_flags);
1625 
1626 	ath9k_hw_filltxdesc(ah, ds,
1627 			    skb->len,	/* segment length */
1628 			    true,	/* first segment */
1629 			    true,	/* last segment */
1630 			    ds);	/* first descriptor */
1631 
1632 	spin_lock_bh(&txctl->txq->axq_lock);
1633 
1634 	if (bf_isht(bf) && (sc->sc_flags & SC_OP_TXAGGR) &&
1635 	    tx_info->control.sta) {
1636 		an = (struct ath_node *)tx_info->control.sta->drv_priv;
1637 		tid = ATH_AN_2_TID(an, bf->bf_tidno);
1638 
1639 		if (!ieee80211_is_data_qos(fc)) {
1640 			ath_tx_send_normal(sc, txctl->txq, &bf_head);
1641 			goto tx_done;
1642 		}
1643 
1644 		if (tx_info->flags & IEEE80211_TX_CTL_AMPDU) {
1645 			/*
1646 			 * Try aggregation if it's a unicast data frame
1647 			 * and the destination is HT capable.
1648 			 */
1649 			ath_tx_send_ampdu(sc, tid, &bf_head, txctl);
1650 		} else {
1651 			/*
1652 			 * Send this frame as regular when ADDBA
1653 			 * exchange is neither complete nor pending.
1654 			 */
1655 			ath_tx_send_ht_normal(sc, txctl->txq,
1656 					      tid, &bf_head);
1657 		}
1658 	} else {
1659 		ath_tx_send_normal(sc, txctl->txq, &bf_head);
1660 	}
1661 
1662 tx_done:
1663 	spin_unlock_bh(&txctl->txq->axq_lock);
1664 }
1665 
1666 /* Upon failure caller should free skb */
1667 int ath_tx_start(struct ieee80211_hw *hw, struct sk_buff *skb,
1668 		 struct ath_tx_control *txctl)
1669 {
1670 	struct ath_wiphy *aphy = hw->priv;
1671 	struct ath_softc *sc = aphy->sc;
1672 	struct ath_buf *bf;
1673 	int r;
1674 
1675 	bf = ath_tx_get_buffer(sc);
1676 	if (!bf) {
1677 		DPRINTF(sc, ATH_DBG_XMIT, "TX buffers are full\n");
1678 		return -1;
1679 	}
1680 
1681 	r = ath_tx_setup_buffer(hw, bf, skb, txctl);
1682 	if (unlikely(r)) {
1683 		struct ath_txq *txq = txctl->txq;
1684 
1685 		DPRINTF(sc, ATH_DBG_FATAL, "TX mem alloc failure\n");
1686 
1687 		/* upon ath_tx_processq() this TX queue will be resumed, we
1688 		 * guarantee this will happen by knowing beforehand that
1689 		 * we will at least have to run TX completionon one buffer
1690 		 * on the queue */
1691 		spin_lock_bh(&txq->axq_lock);
1692 		if (sc->tx.txq[txq->axq_qnum].axq_depth > 1) {
1693 			ieee80211_stop_queue(sc->hw,
1694 				skb_get_queue_mapping(skb));
1695 			txq->stopped = 1;
1696 		}
1697 		spin_unlock_bh(&txq->axq_lock);
1698 
1699 		spin_lock_bh(&sc->tx.txbuflock);
1700 		list_add_tail(&bf->list, &sc->tx.txbuf);
1701 		spin_unlock_bh(&sc->tx.txbuflock);
1702 
1703 		return r;
1704 	}
1705 
1706 	ath_tx_start_dma(sc, bf, txctl);
1707 
1708 	return 0;
1709 }
1710 
1711 void ath_tx_cabq(struct ieee80211_hw *hw, struct sk_buff *skb)
1712 {
1713 	struct ath_wiphy *aphy = hw->priv;
1714 	struct ath_softc *sc = aphy->sc;
1715 	int hdrlen, padsize;
1716 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
1717 	struct ath_tx_control txctl;
1718 
1719 	memset(&txctl, 0, sizeof(struct ath_tx_control));
1720 
1721 	/*
1722 	 * As a temporary workaround, assign seq# here; this will likely need
1723 	 * to be cleaned up to work better with Beacon transmission and virtual
1724 	 * BSSes.
1725 	 */
1726 	if (info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ) {
1727 		struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
1728 		if (info->flags & IEEE80211_TX_CTL_FIRST_FRAGMENT)
1729 			sc->tx.seq_no += 0x10;
1730 		hdr->seq_ctrl &= cpu_to_le16(IEEE80211_SCTL_FRAG);
1731 		hdr->seq_ctrl |= cpu_to_le16(sc->tx.seq_no);
1732 	}
1733 
1734 	/* Add the padding after the header if this is not already done */
1735 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
1736 	if (hdrlen & 3) {
1737 		padsize = hdrlen % 4;
1738 		if (skb_headroom(skb) < padsize) {
1739 			DPRINTF(sc, ATH_DBG_XMIT, "TX CABQ padding failed\n");
1740 			dev_kfree_skb_any(skb);
1741 			return;
1742 		}
1743 		skb_push(skb, padsize);
1744 		memmove(skb->data, skb->data + padsize, hdrlen);
1745 	}
1746 
1747 	txctl.txq = sc->beacon.cabq;
1748 
1749 	DPRINTF(sc, ATH_DBG_XMIT, "transmitting CABQ packet, skb: %p\n", skb);
1750 
1751 	if (ath_tx_start(hw, skb, &txctl) != 0) {
1752 		DPRINTF(sc, ATH_DBG_XMIT, "CABQ TX failed\n");
1753 		goto exit;
1754 	}
1755 
1756 	return;
1757 exit:
1758 	dev_kfree_skb_any(skb);
1759 }
1760 
1761 /*****************/
1762 /* TX Completion */
1763 /*****************/
1764 
1765 static void ath_tx_complete(struct ath_softc *sc, struct sk_buff *skb,
1766 			    int tx_flags)
1767 {
1768 	struct ieee80211_hw *hw = sc->hw;
1769 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1770 	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
1771 	int hdrlen, padsize;
1772 	int frame_type = ATH9K_NOT_INTERNAL;
1773 
1774 	DPRINTF(sc, ATH_DBG_XMIT, "TX complete: skb: %p\n", skb);
1775 
1776 	if (tx_info_priv) {
1777 		hw = tx_info_priv->aphy->hw;
1778 		frame_type = tx_info_priv->frame_type;
1779 	}
1780 
1781 	if (tx_info->flags & IEEE80211_TX_CTL_NO_ACK ||
1782 	    tx_info->flags & IEEE80211_TX_STAT_TX_FILTERED) {
1783 		kfree(tx_info_priv);
1784 		tx_info->rate_driver_data[0] = NULL;
1785 	}
1786 
1787 	if (tx_flags & ATH_TX_BAR)
1788 		tx_info->flags |= IEEE80211_TX_STAT_AMPDU_NO_BACK;
1789 
1790 	if (!(tx_flags & (ATH_TX_ERROR | ATH_TX_XRETRY))) {
1791 		/* Frame was ACKed */
1792 		tx_info->flags |= IEEE80211_TX_STAT_ACK;
1793 	}
1794 
1795 	hdrlen = ieee80211_get_hdrlen_from_skb(skb);
1796 	padsize = hdrlen & 3;
1797 	if (padsize && hdrlen >= 24) {
1798 		/*
1799 		 * Remove MAC header padding before giving the frame back to
1800 		 * mac80211.
1801 		 */
1802 		memmove(skb->data + padsize, skb->data, hdrlen);
1803 		skb_pull(skb, padsize);
1804 	}
1805 
1806 	if (sc->sc_flags & SC_OP_WAIT_FOR_TX_ACK) {
1807 		sc->sc_flags &= ~SC_OP_WAIT_FOR_TX_ACK;
1808 		DPRINTF(sc, ATH_DBG_PS, "Going back to sleep after having "
1809 			"received TX status (0x%x)\n",
1810 			sc->sc_flags & (SC_OP_WAIT_FOR_BEACON |
1811 					SC_OP_WAIT_FOR_CAB |
1812 					SC_OP_WAIT_FOR_PSPOLL_DATA |
1813 					SC_OP_WAIT_FOR_TX_ACK));
1814 	}
1815 
1816 	if (frame_type == ATH9K_NOT_INTERNAL)
1817 		ieee80211_tx_status(hw, skb);
1818 	else
1819 		ath9k_tx_status(hw, skb);
1820 }
1821 
1822 static void ath_tx_complete_buf(struct ath_softc *sc, struct ath_buf *bf,
1823 				struct ath_txq *txq,
1824 				struct list_head *bf_q,
1825 				int txok, int sendbar)
1826 {
1827 	struct sk_buff *skb = bf->bf_mpdu;
1828 	unsigned long flags;
1829 	int tx_flags = 0;
1830 
1831 	if (sendbar)
1832 		tx_flags = ATH_TX_BAR;
1833 
1834 	if (!txok) {
1835 		tx_flags |= ATH_TX_ERROR;
1836 
1837 		if (bf_isxretried(bf))
1838 			tx_flags |= ATH_TX_XRETRY;
1839 	}
1840 
1841 	dma_unmap_single(sc->dev, bf->bf_dmacontext, skb->len, DMA_TO_DEVICE);
1842 	ath_tx_complete(sc, skb, tx_flags);
1843 	ath_debug_stat_tx(sc, txq, bf);
1844 
1845 	/*
1846 	 * Return the list of ath_buf of this mpdu to free queue
1847 	 */
1848 	spin_lock_irqsave(&sc->tx.txbuflock, flags);
1849 	list_splice_tail_init(bf_q, &sc->tx.txbuf);
1850 	spin_unlock_irqrestore(&sc->tx.txbuflock, flags);
1851 }
1852 
1853 static int ath_tx_num_badfrms(struct ath_softc *sc, struct ath_buf *bf,
1854 			      int txok)
1855 {
1856 	struct ath_buf *bf_last = bf->bf_lastbf;
1857 	struct ath_desc *ds = bf_last->bf_desc;
1858 	u16 seq_st = 0;
1859 	u32 ba[WME_BA_BMP_SIZE >> 5];
1860 	int ba_index;
1861 	int nbad = 0;
1862 	int isaggr = 0;
1863 
1864 	if (ds->ds_txstat.ts_flags == ATH9K_TX_SW_ABORTED)
1865 		return 0;
1866 
1867 	isaggr = bf_isaggr(bf);
1868 	if (isaggr) {
1869 		seq_st = ATH_DS_BA_SEQ(ds);
1870 		memcpy(ba, ATH_DS_BA_BITMAP(ds), WME_BA_BMP_SIZE >> 3);
1871 	}
1872 
1873 	while (bf) {
1874 		ba_index = ATH_BA_INDEX(seq_st, bf->bf_seqno);
1875 		if (!txok || (isaggr && !ATH_BA_ISSET(ba, ba_index)))
1876 			nbad++;
1877 
1878 		bf = bf->bf_next;
1879 	}
1880 
1881 	return nbad;
1882 }
1883 
1884 static void ath_tx_rc_status(struct ath_buf *bf, struct ath_desc *ds,
1885 			     int nbad, int txok, bool update_rc)
1886 {
1887 	struct sk_buff *skb = bf->bf_mpdu;
1888 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1889 	struct ieee80211_tx_info *tx_info = IEEE80211_SKB_CB(skb);
1890 	struct ath_tx_info_priv *tx_info_priv = ATH_TX_INFO_PRIV(tx_info);
1891 	struct ieee80211_hw *hw = tx_info_priv->aphy->hw;
1892 	u8 i, tx_rateindex;
1893 
1894 	if (txok)
1895 		tx_info->status.ack_signal = ds->ds_txstat.ts_rssi;
1896 
1897 	tx_rateindex = ds->ds_txstat.ts_rateindex;
1898 	WARN_ON(tx_rateindex >= hw->max_rates);
1899 
1900 	tx_info_priv->update_rc = update_rc;
1901 	if (ds->ds_txstat.ts_status & ATH9K_TXERR_FILT)
1902 		tx_info->flags |= IEEE80211_TX_STAT_TX_FILTERED;
1903 
1904 	if ((ds->ds_txstat.ts_status & ATH9K_TXERR_FILT) == 0 &&
1905 	    (bf->bf_flags & ATH9K_TXDESC_NOACK) == 0 && update_rc) {
1906 		if (ieee80211_is_data(hdr->frame_control)) {
1907 			memcpy(&tx_info_priv->tx, &ds->ds_txstat,
1908 			       sizeof(tx_info_priv->tx));
1909 			tx_info_priv->n_frames = bf->bf_nframes;
1910 			tx_info_priv->n_bad_frames = nbad;
1911 		}
1912 	}
1913 
1914 	for (i = tx_rateindex + 1; i < hw->max_rates; i++)
1915 		tx_info->status.rates[i].count = 0;
1916 
1917 	tx_info->status.rates[tx_rateindex].count = bf->bf_retries + 1;
1918 }
1919 
1920 static void ath_wake_mac80211_queue(struct ath_softc *sc, struct ath_txq *txq)
1921 {
1922 	int qnum;
1923 
1924 	spin_lock_bh(&txq->axq_lock);
1925 	if (txq->stopped &&
1926 	    sc->tx.txq[txq->axq_qnum].axq_depth <= (ATH_TXBUF - 20)) {
1927 		qnum = ath_get_mac80211_qnum(txq->axq_qnum, sc);
1928 		if (qnum != -1) {
1929 			ieee80211_wake_queue(sc->hw, qnum);
1930 			txq->stopped = 0;
1931 		}
1932 	}
1933 	spin_unlock_bh(&txq->axq_lock);
1934 }
1935 
1936 static void ath_tx_processq(struct ath_softc *sc, struct ath_txq *txq)
1937 {
1938 	struct ath_hw *ah = sc->sc_ah;
1939 	struct ath_buf *bf, *lastbf, *bf_held = NULL;
1940 	struct list_head bf_head;
1941 	struct ath_desc *ds;
1942 	int txok;
1943 	int status;
1944 
1945 	DPRINTF(sc, ATH_DBG_QUEUE, "tx queue %d (%x), link %p\n",
1946 		txq->axq_qnum, ath9k_hw_gettxbuf(sc->sc_ah, txq->axq_qnum),
1947 		txq->axq_link);
1948 
1949 	for (;;) {
1950 		spin_lock_bh(&txq->axq_lock);
1951 		if (list_empty(&txq->axq_q)) {
1952 			txq->axq_link = NULL;
1953 			txq->axq_linkbuf = NULL;
1954 			spin_unlock_bh(&txq->axq_lock);
1955 			break;
1956 		}
1957 		bf = list_first_entry(&txq->axq_q, struct ath_buf, list);
1958 
1959 		/*
1960 		 * There is a race condition that a BH gets scheduled
1961 		 * after sw writes TxE and before hw re-load the last
1962 		 * descriptor to get the newly chained one.
1963 		 * Software must keep the last DONE descriptor as a
1964 		 * holding descriptor - software does so by marking
1965 		 * it with the STALE flag.
1966 		 */
1967 		bf_held = NULL;
1968 		if (bf->bf_stale) {
1969 			bf_held = bf;
1970 			if (list_is_last(&bf_held->list, &txq->axq_q)) {
1971 				spin_unlock_bh(&txq->axq_lock);
1972 				break;
1973 			} else {
1974 				bf = list_entry(bf_held->list.next,
1975 						struct ath_buf, list);
1976 			}
1977 		}
1978 
1979 		lastbf = bf->bf_lastbf;
1980 		ds = lastbf->bf_desc;
1981 
1982 		status = ath9k_hw_txprocdesc(ah, ds);
1983 		if (status == -EINPROGRESS) {
1984 			spin_unlock_bh(&txq->axq_lock);
1985 			break;
1986 		}
1987 		if (bf->bf_desc == txq->axq_lastdsWithCTS)
1988 			txq->axq_lastdsWithCTS = NULL;
1989 		if (ds == txq->axq_gatingds)
1990 			txq->axq_gatingds = NULL;
1991 
1992 		/*
1993 		 * Remove ath_buf's of the same transmit unit from txq,
1994 		 * however leave the last descriptor back as the holding
1995 		 * descriptor for hw.
1996 		 */
1997 		lastbf->bf_stale = true;
1998 		INIT_LIST_HEAD(&bf_head);
1999 		if (!list_is_singular(&lastbf->list))
2000 			list_cut_position(&bf_head,
2001 				&txq->axq_q, lastbf->list.prev);
2002 
2003 		txq->axq_depth--;
2004 		if (bf_isaggr(bf))
2005 			txq->axq_aggr_depth--;
2006 
2007 		txok = (ds->ds_txstat.ts_status == 0);
2008 		txq->axq_tx_inprogress = false;
2009 		spin_unlock_bh(&txq->axq_lock);
2010 
2011 		if (bf_held) {
2012 			spin_lock_bh(&sc->tx.txbuflock);
2013 			list_move_tail(&bf_held->list, &sc->tx.txbuf);
2014 			spin_unlock_bh(&sc->tx.txbuflock);
2015 		}
2016 
2017 		if (!bf_isampdu(bf)) {
2018 			/*
2019 			 * This frame is sent out as a single frame.
2020 			 * Use hardware retry status for this frame.
2021 			 */
2022 			bf->bf_retries = ds->ds_txstat.ts_longretry;
2023 			if (ds->ds_txstat.ts_status & ATH9K_TXERR_XRETRY)
2024 				bf->bf_state.bf_type |= BUF_XRETRY;
2025 			ath_tx_rc_status(bf, ds, 0, txok, true);
2026 		}
2027 
2028 		if (bf_isampdu(bf))
2029 			ath_tx_complete_aggr(sc, txq, bf, &bf_head, txok);
2030 		else
2031 			ath_tx_complete_buf(sc, bf, txq, &bf_head, txok, 0);
2032 
2033 		ath_wake_mac80211_queue(sc, txq);
2034 
2035 		spin_lock_bh(&txq->axq_lock);
2036 		if (sc->sc_flags & SC_OP_TXAGGR)
2037 			ath_txq_schedule(sc, txq);
2038 		spin_unlock_bh(&txq->axq_lock);
2039 	}
2040 }
2041 
2042 static void ath_tx_complete_poll_work(struct work_struct *work)
2043 {
2044 	struct ath_softc *sc = container_of(work, struct ath_softc,
2045 			tx_complete_work.work);
2046 	struct ath_txq *txq;
2047 	int i;
2048 	bool needreset = false;
2049 
2050 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++)
2051 		if (ATH_TXQ_SETUP(sc, i)) {
2052 			txq = &sc->tx.txq[i];
2053 			spin_lock_bh(&txq->axq_lock);
2054 			if (txq->axq_depth) {
2055 				if (txq->axq_tx_inprogress) {
2056 					needreset = true;
2057 					spin_unlock_bh(&txq->axq_lock);
2058 					break;
2059 				} else {
2060 					txq->axq_tx_inprogress = true;
2061 				}
2062 			}
2063 			spin_unlock_bh(&txq->axq_lock);
2064 		}
2065 
2066 	if (needreset) {
2067 		DPRINTF(sc, ATH_DBG_RESET, "tx hung, resetting the chip\n");
2068 		ath_reset(sc, false);
2069 	}
2070 
2071 	ieee80211_queue_delayed_work(sc->hw, &sc->tx_complete_work,
2072 			msecs_to_jiffies(ATH_TX_COMPLETE_POLL_INT));
2073 }
2074 
2075 
2076 
2077 void ath_tx_tasklet(struct ath_softc *sc)
2078 {
2079 	int i;
2080 	u32 qcumask = ((1 << ATH9K_NUM_TX_QUEUES) - 1);
2081 
2082 	ath9k_hw_gettxintrtxqs(sc->sc_ah, &qcumask);
2083 
2084 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
2085 		if (ATH_TXQ_SETUP(sc, i) && (qcumask & (1 << i)))
2086 			ath_tx_processq(sc, &sc->tx.txq[i]);
2087 	}
2088 }
2089 
2090 /*****************/
2091 /* Init, Cleanup */
2092 /*****************/
2093 
2094 int ath_tx_init(struct ath_softc *sc, int nbufs)
2095 {
2096 	int error = 0;
2097 
2098 	spin_lock_init(&sc->tx.txbuflock);
2099 
2100 	error = ath_descdma_setup(sc, &sc->tx.txdma, &sc->tx.txbuf,
2101 				  "tx", nbufs, 1);
2102 	if (error != 0) {
2103 		DPRINTF(sc, ATH_DBG_FATAL,
2104 			"Failed to allocate tx descriptors: %d\n", error);
2105 		goto err;
2106 	}
2107 
2108 	error = ath_descdma_setup(sc, &sc->beacon.bdma, &sc->beacon.bbuf,
2109 				  "beacon", ATH_BCBUF, 1);
2110 	if (error != 0) {
2111 		DPRINTF(sc, ATH_DBG_FATAL,
2112 			"Failed to allocate beacon descriptors: %d\n", error);
2113 		goto err;
2114 	}
2115 
2116 	INIT_DELAYED_WORK(&sc->tx_complete_work, ath_tx_complete_poll_work);
2117 
2118 err:
2119 	if (error != 0)
2120 		ath_tx_cleanup(sc);
2121 
2122 	return error;
2123 }
2124 
2125 void ath_tx_cleanup(struct ath_softc *sc)
2126 {
2127 	if (sc->beacon.bdma.dd_desc_len != 0)
2128 		ath_descdma_cleanup(sc, &sc->beacon.bdma, &sc->beacon.bbuf);
2129 
2130 	if (sc->tx.txdma.dd_desc_len != 0)
2131 		ath_descdma_cleanup(sc, &sc->tx.txdma, &sc->tx.txbuf);
2132 }
2133 
2134 void ath_tx_node_init(struct ath_softc *sc, struct ath_node *an)
2135 {
2136 	struct ath_atx_tid *tid;
2137 	struct ath_atx_ac *ac;
2138 	int tidno, acno;
2139 
2140 	for (tidno = 0, tid = &an->tid[tidno];
2141 	     tidno < WME_NUM_TID;
2142 	     tidno++, tid++) {
2143 		tid->an        = an;
2144 		tid->tidno     = tidno;
2145 		tid->seq_start = tid->seq_next = 0;
2146 		tid->baw_size  = WME_MAX_BA;
2147 		tid->baw_head  = tid->baw_tail = 0;
2148 		tid->sched     = false;
2149 		tid->paused    = false;
2150 		tid->state &= ~AGGR_CLEANUP;
2151 		INIT_LIST_HEAD(&tid->buf_q);
2152 		acno = TID_TO_WME_AC(tidno);
2153 		tid->ac = &an->ac[acno];
2154 		tid->state &= ~AGGR_ADDBA_COMPLETE;
2155 		tid->state &= ~AGGR_ADDBA_PROGRESS;
2156 	}
2157 
2158 	for (acno = 0, ac = &an->ac[acno];
2159 	     acno < WME_NUM_AC; acno++, ac++) {
2160 		ac->sched    = false;
2161 		INIT_LIST_HEAD(&ac->tid_q);
2162 
2163 		switch (acno) {
2164 		case WME_AC_BE:
2165 			ac->qnum = ath_tx_get_qnum(sc,
2166 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_BE);
2167 			break;
2168 		case WME_AC_BK:
2169 			ac->qnum = ath_tx_get_qnum(sc,
2170 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_BK);
2171 			break;
2172 		case WME_AC_VI:
2173 			ac->qnum = ath_tx_get_qnum(sc,
2174 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_VI);
2175 			break;
2176 		case WME_AC_VO:
2177 			ac->qnum = ath_tx_get_qnum(sc,
2178 				   ATH9K_TX_QUEUE_DATA, ATH9K_WME_AC_VO);
2179 			break;
2180 		}
2181 	}
2182 }
2183 
2184 void ath_tx_node_cleanup(struct ath_softc *sc, struct ath_node *an)
2185 {
2186 	int i;
2187 	struct ath_atx_ac *ac, *ac_tmp;
2188 	struct ath_atx_tid *tid, *tid_tmp;
2189 	struct ath_txq *txq;
2190 
2191 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++) {
2192 		if (ATH_TXQ_SETUP(sc, i)) {
2193 			txq = &sc->tx.txq[i];
2194 
2195 			spin_lock(&txq->axq_lock);
2196 
2197 			list_for_each_entry_safe(ac,
2198 					ac_tmp, &txq->axq_acq, list) {
2199 				tid = list_first_entry(&ac->tid_q,
2200 						struct ath_atx_tid, list);
2201 				if (tid && tid->an != an)
2202 					continue;
2203 				list_del(&ac->list);
2204 				ac->sched = false;
2205 
2206 				list_for_each_entry_safe(tid,
2207 						tid_tmp, &ac->tid_q, list) {
2208 					list_del(&tid->list);
2209 					tid->sched = false;
2210 					ath_tid_drain(sc, txq, tid);
2211 					tid->state &= ~AGGR_ADDBA_COMPLETE;
2212 					tid->state &= ~AGGR_CLEANUP;
2213 				}
2214 			}
2215 
2216 			spin_unlock(&txq->axq_lock);
2217 		}
2218 	}
2219 }
2220