1 /*
2  * Copyright (c) 2005-2011 Atheros Communications Inc.
3  * Copyright (c) 2011-2017 Qualcomm Atheros, Inc.
4  *
5  * Permission to use, copy, modify, and/or distribute this software for any
6  * purpose with or without fee is hereby granted, provided that the above
7  * copyright notice and this permission notice appear in all copies.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
16  */
17 
18 #include <linux/etherdevice.h>
19 #include "htt.h"
20 #include "mac.h"
21 #include "hif.h"
22 #include "txrx.h"
23 #include "debug.h"
24 
25 static u8 ath10k_htt_tx_txq_calc_size(size_t count)
26 {
27 	int exp;
28 	int factor;
29 
30 	exp = 0;
31 	factor = count >> 7;
32 
33 	while (factor >= 64 && exp < 4) {
34 		factor >>= 3;
35 		exp++;
36 	}
37 
38 	if (exp == 4)
39 		return 0xff;
40 
41 	if (count > 0)
42 		factor = max(1, factor);
43 
44 	return SM(exp, HTT_TX_Q_STATE_ENTRY_EXP) |
45 	       SM(factor, HTT_TX_Q_STATE_ENTRY_FACTOR);
46 }
47 
48 static void __ath10k_htt_tx_txq_recalc(struct ieee80211_hw *hw,
49 				       struct ieee80211_txq *txq)
50 {
51 	struct ath10k *ar = hw->priv;
52 	struct ath10k_sta *arsta;
53 	struct ath10k_vif *arvif = (void *)txq->vif->drv_priv;
54 	unsigned long frame_cnt;
55 	unsigned long byte_cnt;
56 	int idx;
57 	u32 bit;
58 	u16 peer_id;
59 	u8 tid;
60 	u8 count;
61 
62 	lockdep_assert_held(&ar->htt.tx_lock);
63 
64 	if (!ar->htt.tx_q_state.enabled)
65 		return;
66 
67 	if (ar->htt.tx_q_state.mode != HTT_TX_MODE_SWITCH_PUSH_PULL)
68 		return;
69 
70 	if (txq->sta) {
71 		arsta = (void *)txq->sta->drv_priv;
72 		peer_id = arsta->peer_id;
73 	} else {
74 		peer_id = arvif->peer_id;
75 	}
76 
77 	tid = txq->tid;
78 	bit = BIT(peer_id % 32);
79 	idx = peer_id / 32;
80 
81 	ieee80211_txq_get_depth(txq, &frame_cnt, &byte_cnt);
82 	count = ath10k_htt_tx_txq_calc_size(byte_cnt);
83 
84 	if (unlikely(peer_id >= ar->htt.tx_q_state.num_peers) ||
85 	    unlikely(tid >= ar->htt.tx_q_state.num_tids)) {
86 		ath10k_warn(ar, "refusing to update txq for peer_id %hu tid %hhu due to out of bounds\n",
87 			    peer_id, tid);
88 		return;
89 	}
90 
91 	ar->htt.tx_q_state.vaddr->count[tid][peer_id] = count;
92 	ar->htt.tx_q_state.vaddr->map[tid][idx] &= ~bit;
93 	ar->htt.tx_q_state.vaddr->map[tid][idx] |= count ? bit : 0;
94 
95 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx txq state update peer_id %hu tid %hhu count %hhu\n",
96 		   peer_id, tid, count);
97 }
98 
99 static void __ath10k_htt_tx_txq_sync(struct ath10k *ar)
100 {
101 	u32 seq;
102 	size_t size;
103 
104 	lockdep_assert_held(&ar->htt.tx_lock);
105 
106 	if (!ar->htt.tx_q_state.enabled)
107 		return;
108 
109 	if (ar->htt.tx_q_state.mode != HTT_TX_MODE_SWITCH_PUSH_PULL)
110 		return;
111 
112 	seq = le32_to_cpu(ar->htt.tx_q_state.vaddr->seq);
113 	seq++;
114 	ar->htt.tx_q_state.vaddr->seq = cpu_to_le32(seq);
115 
116 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx txq state update commit seq %u\n",
117 		   seq);
118 
119 	size = sizeof(*ar->htt.tx_q_state.vaddr);
120 	dma_sync_single_for_device(ar->dev,
121 				   ar->htt.tx_q_state.paddr,
122 				   size,
123 				   DMA_TO_DEVICE);
124 }
125 
126 void ath10k_htt_tx_txq_recalc(struct ieee80211_hw *hw,
127 			      struct ieee80211_txq *txq)
128 {
129 	struct ath10k *ar = hw->priv;
130 
131 	spin_lock_bh(&ar->htt.tx_lock);
132 	__ath10k_htt_tx_txq_recalc(hw, txq);
133 	spin_unlock_bh(&ar->htt.tx_lock);
134 }
135 
136 void ath10k_htt_tx_txq_sync(struct ath10k *ar)
137 {
138 	spin_lock_bh(&ar->htt.tx_lock);
139 	__ath10k_htt_tx_txq_sync(ar);
140 	spin_unlock_bh(&ar->htt.tx_lock);
141 }
142 
143 void ath10k_htt_tx_txq_update(struct ieee80211_hw *hw,
144 			      struct ieee80211_txq *txq)
145 {
146 	struct ath10k *ar = hw->priv;
147 
148 	spin_lock_bh(&ar->htt.tx_lock);
149 	__ath10k_htt_tx_txq_recalc(hw, txq);
150 	__ath10k_htt_tx_txq_sync(ar);
151 	spin_unlock_bh(&ar->htt.tx_lock);
152 }
153 
154 void ath10k_htt_tx_dec_pending(struct ath10k_htt *htt)
155 {
156 	lockdep_assert_held(&htt->tx_lock);
157 
158 	htt->num_pending_tx--;
159 	if (htt->num_pending_tx == htt->max_num_pending_tx - 1)
160 		ath10k_mac_tx_unlock(htt->ar, ATH10K_TX_PAUSE_Q_FULL);
161 }
162 
163 int ath10k_htt_tx_inc_pending(struct ath10k_htt *htt)
164 {
165 	lockdep_assert_held(&htt->tx_lock);
166 
167 	if (htt->num_pending_tx >= htt->max_num_pending_tx)
168 		return -EBUSY;
169 
170 	htt->num_pending_tx++;
171 	if (htt->num_pending_tx == htt->max_num_pending_tx)
172 		ath10k_mac_tx_lock(htt->ar, ATH10K_TX_PAUSE_Q_FULL);
173 
174 	return 0;
175 }
176 
177 int ath10k_htt_tx_mgmt_inc_pending(struct ath10k_htt *htt, bool is_mgmt,
178 				   bool is_presp)
179 {
180 	struct ath10k *ar = htt->ar;
181 
182 	lockdep_assert_held(&htt->tx_lock);
183 
184 	if (!is_mgmt || !ar->hw_params.max_probe_resp_desc_thres)
185 		return 0;
186 
187 	if (is_presp &&
188 	    ar->hw_params.max_probe_resp_desc_thres < htt->num_pending_mgmt_tx)
189 		return -EBUSY;
190 
191 	htt->num_pending_mgmt_tx++;
192 
193 	return 0;
194 }
195 
196 void ath10k_htt_tx_mgmt_dec_pending(struct ath10k_htt *htt)
197 {
198 	lockdep_assert_held(&htt->tx_lock);
199 
200 	if (!htt->ar->hw_params.max_probe_resp_desc_thres)
201 		return;
202 
203 	htt->num_pending_mgmt_tx--;
204 }
205 
206 int ath10k_htt_tx_alloc_msdu_id(struct ath10k_htt *htt, struct sk_buff *skb)
207 {
208 	struct ath10k *ar = htt->ar;
209 	int ret;
210 
211 	spin_lock_bh(&htt->tx_lock);
212 	ret = idr_alloc(&htt->pending_tx, skb, 0,
213 			htt->max_num_pending_tx, GFP_ATOMIC);
214 	spin_unlock_bh(&htt->tx_lock);
215 
216 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx alloc msdu_id %d\n", ret);
217 
218 	return ret;
219 }
220 
221 void ath10k_htt_tx_free_msdu_id(struct ath10k_htt *htt, u16 msdu_id)
222 {
223 	struct ath10k *ar = htt->ar;
224 
225 	lockdep_assert_held(&htt->tx_lock);
226 
227 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx free msdu_id %hu\n", msdu_id);
228 
229 	idr_remove(&htt->pending_tx, msdu_id);
230 }
231 
232 static void ath10k_htt_tx_free_cont_txbuf_32(struct ath10k_htt *htt)
233 {
234 	struct ath10k *ar = htt->ar;
235 	size_t size;
236 
237 	if (!htt->txbuf.vaddr_txbuff_32)
238 		return;
239 
240 	size = htt->txbuf.size;
241 	dma_free_coherent(ar->dev, size, htt->txbuf.vaddr_txbuff_32,
242 			  htt->txbuf.paddr);
243 	htt->txbuf.vaddr_txbuff_32 = NULL;
244 }
245 
246 static int ath10k_htt_tx_alloc_cont_txbuf_32(struct ath10k_htt *htt)
247 {
248 	struct ath10k *ar = htt->ar;
249 	size_t size;
250 
251 	size = htt->max_num_pending_tx *
252 			sizeof(struct ath10k_htt_txbuf_32);
253 
254 	htt->txbuf.vaddr_txbuff_32 = dma_alloc_coherent(ar->dev, size,
255 							&htt->txbuf.paddr,
256 							GFP_KERNEL);
257 	if (!htt->txbuf.vaddr_txbuff_32)
258 		return -ENOMEM;
259 
260 	htt->txbuf.size = size;
261 
262 	return 0;
263 }
264 
265 static void ath10k_htt_tx_free_cont_txbuf_64(struct ath10k_htt *htt)
266 {
267 	struct ath10k *ar = htt->ar;
268 	size_t size;
269 
270 	if (!htt->txbuf.vaddr_txbuff_64)
271 		return;
272 
273 	size = htt->txbuf.size;
274 	dma_free_coherent(ar->dev, size, htt->txbuf.vaddr_txbuff_64,
275 			  htt->txbuf.paddr);
276 	htt->txbuf.vaddr_txbuff_64 = NULL;
277 }
278 
279 static int ath10k_htt_tx_alloc_cont_txbuf_64(struct ath10k_htt *htt)
280 {
281 	struct ath10k *ar = htt->ar;
282 	size_t size;
283 
284 	size = htt->max_num_pending_tx *
285 			sizeof(struct ath10k_htt_txbuf_64);
286 
287 	htt->txbuf.vaddr_txbuff_64 = dma_alloc_coherent(ar->dev, size,
288 							&htt->txbuf.paddr,
289 							GFP_KERNEL);
290 	if (!htt->txbuf.vaddr_txbuff_64)
291 		return -ENOMEM;
292 
293 	htt->txbuf.size = size;
294 
295 	return 0;
296 }
297 
298 static void ath10k_htt_tx_free_cont_frag_desc_32(struct ath10k_htt *htt)
299 {
300 	size_t size;
301 
302 	if (!htt->frag_desc.vaddr_desc_32)
303 		return;
304 
305 	size = htt->max_num_pending_tx *
306 			sizeof(struct htt_msdu_ext_desc);
307 
308 	dma_free_coherent(htt->ar->dev,
309 			  size,
310 			  htt->frag_desc.vaddr_desc_32,
311 			  htt->frag_desc.paddr);
312 
313 	htt->frag_desc.vaddr_desc_32 = NULL;
314 }
315 
316 static int ath10k_htt_tx_alloc_cont_frag_desc_32(struct ath10k_htt *htt)
317 {
318 	struct ath10k *ar = htt->ar;
319 	size_t size;
320 
321 	if (!ar->hw_params.continuous_frag_desc)
322 		return 0;
323 
324 	size = htt->max_num_pending_tx *
325 			sizeof(struct htt_msdu_ext_desc);
326 	htt->frag_desc.vaddr_desc_32 = dma_alloc_coherent(ar->dev, size,
327 							  &htt->frag_desc.paddr,
328 							  GFP_KERNEL);
329 	if (!htt->frag_desc.vaddr_desc_32) {
330 		ath10k_err(ar, "failed to alloc fragment desc memory\n");
331 		return -ENOMEM;
332 	}
333 	htt->frag_desc.size = size;
334 
335 	return 0;
336 }
337 
338 static void ath10k_htt_tx_free_cont_frag_desc_64(struct ath10k_htt *htt)
339 {
340 	size_t size;
341 
342 	if (!htt->frag_desc.vaddr_desc_64)
343 		return;
344 
345 	size = htt->max_num_pending_tx *
346 			sizeof(struct htt_msdu_ext_desc_64);
347 
348 	dma_free_coherent(htt->ar->dev,
349 			  size,
350 			  htt->frag_desc.vaddr_desc_64,
351 			  htt->frag_desc.paddr);
352 
353 	htt->frag_desc.vaddr_desc_64 = NULL;
354 }
355 
356 static int ath10k_htt_tx_alloc_cont_frag_desc_64(struct ath10k_htt *htt)
357 {
358 	struct ath10k *ar = htt->ar;
359 	size_t size;
360 
361 	if (!ar->hw_params.continuous_frag_desc)
362 		return 0;
363 
364 	size = htt->max_num_pending_tx *
365 			sizeof(struct htt_msdu_ext_desc_64);
366 
367 	htt->frag_desc.vaddr_desc_64 = dma_alloc_coherent(ar->dev, size,
368 							  &htt->frag_desc.paddr,
369 							  GFP_KERNEL);
370 	if (!htt->frag_desc.vaddr_desc_64) {
371 		ath10k_err(ar, "failed to alloc fragment desc memory\n");
372 		return -ENOMEM;
373 	}
374 	htt->frag_desc.size = size;
375 
376 	return 0;
377 }
378 
379 static void ath10k_htt_tx_free_txq(struct ath10k_htt *htt)
380 {
381 	struct ath10k *ar = htt->ar;
382 	size_t size;
383 
384 	if (!test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
385 		      ar->running_fw->fw_file.fw_features))
386 		return;
387 
388 	size = sizeof(*htt->tx_q_state.vaddr);
389 
390 	dma_unmap_single(ar->dev, htt->tx_q_state.paddr, size, DMA_TO_DEVICE);
391 	kfree(htt->tx_q_state.vaddr);
392 }
393 
394 static int ath10k_htt_tx_alloc_txq(struct ath10k_htt *htt)
395 {
396 	struct ath10k *ar = htt->ar;
397 	size_t size;
398 	int ret;
399 
400 	if (!test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
401 		      ar->running_fw->fw_file.fw_features))
402 		return 0;
403 
404 	htt->tx_q_state.num_peers = HTT_TX_Q_STATE_NUM_PEERS;
405 	htt->tx_q_state.num_tids = HTT_TX_Q_STATE_NUM_TIDS;
406 	htt->tx_q_state.type = HTT_Q_DEPTH_TYPE_BYTES;
407 
408 	size = sizeof(*htt->tx_q_state.vaddr);
409 	htt->tx_q_state.vaddr = kzalloc(size, GFP_KERNEL);
410 	if (!htt->tx_q_state.vaddr)
411 		return -ENOMEM;
412 
413 	htt->tx_q_state.paddr = dma_map_single(ar->dev, htt->tx_q_state.vaddr,
414 					       size, DMA_TO_DEVICE);
415 	ret = dma_mapping_error(ar->dev, htt->tx_q_state.paddr);
416 	if (ret) {
417 		ath10k_warn(ar, "failed to dma map tx_q_state: %d\n", ret);
418 		kfree(htt->tx_q_state.vaddr);
419 		return -EIO;
420 	}
421 
422 	return 0;
423 }
424 
425 static void ath10k_htt_tx_free_txdone_fifo(struct ath10k_htt *htt)
426 {
427 	WARN_ON(!kfifo_is_empty(&htt->txdone_fifo));
428 	kfifo_free(&htt->txdone_fifo);
429 }
430 
431 static int ath10k_htt_tx_alloc_txdone_fifo(struct ath10k_htt *htt)
432 {
433 	int ret;
434 	size_t size;
435 
436 	size = roundup_pow_of_two(htt->max_num_pending_tx);
437 	ret = kfifo_alloc(&htt->txdone_fifo, size, GFP_KERNEL);
438 	return ret;
439 }
440 
441 static int ath10k_htt_tx_alloc_buf(struct ath10k_htt *htt)
442 {
443 	struct ath10k *ar = htt->ar;
444 	int ret;
445 
446 	ret = ath10k_htt_alloc_txbuff(htt);
447 	if (ret) {
448 		ath10k_err(ar, "failed to alloc cont tx buffer: %d\n", ret);
449 		return ret;
450 	}
451 
452 	ret = ath10k_htt_alloc_frag_desc(htt);
453 	if (ret) {
454 		ath10k_err(ar, "failed to alloc cont frag desc: %d\n", ret);
455 		goto free_txbuf;
456 	}
457 
458 	ret = ath10k_htt_tx_alloc_txq(htt);
459 	if (ret) {
460 		ath10k_err(ar, "failed to alloc txq: %d\n", ret);
461 		goto free_frag_desc;
462 	}
463 
464 	ret = ath10k_htt_tx_alloc_txdone_fifo(htt);
465 	if (ret) {
466 		ath10k_err(ar, "failed to alloc txdone fifo: %d\n", ret);
467 		goto free_txq;
468 	}
469 
470 	return 0;
471 
472 free_txq:
473 	ath10k_htt_tx_free_txq(htt);
474 
475 free_frag_desc:
476 	ath10k_htt_free_frag_desc(htt);
477 
478 free_txbuf:
479 	ath10k_htt_free_txbuff(htt);
480 
481 	return ret;
482 }
483 
484 int ath10k_htt_tx_start(struct ath10k_htt *htt)
485 {
486 	struct ath10k *ar = htt->ar;
487 	int ret;
488 
489 	ath10k_dbg(ar, ATH10K_DBG_BOOT, "htt tx max num pending tx %d\n",
490 		   htt->max_num_pending_tx);
491 
492 	spin_lock_init(&htt->tx_lock);
493 	idr_init(&htt->pending_tx);
494 
495 	if (htt->tx_mem_allocated)
496 		return 0;
497 
498 	if (ar->dev_type == ATH10K_DEV_TYPE_HL)
499 		return 0;
500 
501 	ret = ath10k_htt_tx_alloc_buf(htt);
502 	if (ret)
503 		goto free_idr_pending_tx;
504 
505 	htt->tx_mem_allocated = true;
506 
507 	return 0;
508 
509 free_idr_pending_tx:
510 	idr_destroy(&htt->pending_tx);
511 
512 	return ret;
513 }
514 
515 static int ath10k_htt_tx_clean_up_pending(int msdu_id, void *skb, void *ctx)
516 {
517 	struct ath10k *ar = ctx;
518 	struct ath10k_htt *htt = &ar->htt;
519 	struct htt_tx_done tx_done = {0};
520 
521 	ath10k_dbg(ar, ATH10K_DBG_HTT, "force cleanup msdu_id %hu\n", msdu_id);
522 
523 	tx_done.msdu_id = msdu_id;
524 	tx_done.status = HTT_TX_COMPL_STATE_DISCARD;
525 
526 	ath10k_txrx_tx_unref(htt, &tx_done);
527 
528 	return 0;
529 }
530 
531 void ath10k_htt_tx_destroy(struct ath10k_htt *htt)
532 {
533 	if (!htt->tx_mem_allocated)
534 		return;
535 
536 	ath10k_htt_free_txbuff(htt);
537 	ath10k_htt_tx_free_txq(htt);
538 	ath10k_htt_free_frag_desc(htt);
539 	ath10k_htt_tx_free_txdone_fifo(htt);
540 	htt->tx_mem_allocated = false;
541 }
542 
543 void ath10k_htt_tx_stop(struct ath10k_htt *htt)
544 {
545 	idr_for_each(&htt->pending_tx, ath10k_htt_tx_clean_up_pending, htt->ar);
546 	idr_destroy(&htt->pending_tx);
547 }
548 
549 void ath10k_htt_tx_free(struct ath10k_htt *htt)
550 {
551 	ath10k_htt_tx_stop(htt);
552 	ath10k_htt_tx_destroy(htt);
553 }
554 
555 void ath10k_htt_htc_tx_complete(struct ath10k *ar, struct sk_buff *skb)
556 {
557 	dev_kfree_skb_any(skb);
558 }
559 
560 void ath10k_htt_hif_tx_complete(struct ath10k *ar, struct sk_buff *skb)
561 {
562 	dev_kfree_skb_any(skb);
563 }
564 EXPORT_SYMBOL(ath10k_htt_hif_tx_complete);
565 
566 int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
567 {
568 	struct ath10k *ar = htt->ar;
569 	struct sk_buff *skb;
570 	struct htt_cmd *cmd;
571 	int len = 0;
572 	int ret;
573 
574 	len += sizeof(cmd->hdr);
575 	len += sizeof(cmd->ver_req);
576 
577 	skb = ath10k_htc_alloc_skb(ar, len);
578 	if (!skb)
579 		return -ENOMEM;
580 
581 	skb_put(skb, len);
582 	cmd = (struct htt_cmd *)skb->data;
583 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_VERSION_REQ;
584 
585 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
586 	if (ret) {
587 		dev_kfree_skb_any(skb);
588 		return ret;
589 	}
590 
591 	return 0;
592 }
593 
594 int ath10k_htt_h2t_stats_req(struct ath10k_htt *htt, u8 mask, u64 cookie)
595 {
596 	struct ath10k *ar = htt->ar;
597 	struct htt_stats_req *req;
598 	struct sk_buff *skb;
599 	struct htt_cmd *cmd;
600 	int len = 0, ret;
601 
602 	len += sizeof(cmd->hdr);
603 	len += sizeof(cmd->stats_req);
604 
605 	skb = ath10k_htc_alloc_skb(ar, len);
606 	if (!skb)
607 		return -ENOMEM;
608 
609 	skb_put(skb, len);
610 	cmd = (struct htt_cmd *)skb->data;
611 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_STATS_REQ;
612 
613 	req = &cmd->stats_req;
614 
615 	memset(req, 0, sizeof(*req));
616 
617 	/* currently we support only max 8 bit masks so no need to worry
618 	 * about endian support
619 	 */
620 	req->upload_types[0] = mask;
621 	req->reset_types[0] = mask;
622 	req->stat_type = HTT_STATS_REQ_CFG_STAT_TYPE_INVALID;
623 	req->cookie_lsb = cpu_to_le32(cookie & 0xffffffff);
624 	req->cookie_msb = cpu_to_le32((cookie & 0xffffffff00000000ULL) >> 32);
625 
626 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
627 	if (ret) {
628 		ath10k_warn(ar, "failed to send htt type stats request: %d",
629 			    ret);
630 		dev_kfree_skb_any(skb);
631 		return ret;
632 	}
633 
634 	return 0;
635 }
636 
637 static int ath10k_htt_send_frag_desc_bank_cfg_32(struct ath10k_htt *htt)
638 {
639 	struct ath10k *ar = htt->ar;
640 	struct sk_buff *skb;
641 	struct htt_cmd *cmd;
642 	struct htt_frag_desc_bank_cfg32 *cfg;
643 	int ret, size;
644 	u8 info;
645 
646 	if (!ar->hw_params.continuous_frag_desc)
647 		return 0;
648 
649 	if (!htt->frag_desc.paddr) {
650 		ath10k_warn(ar, "invalid frag desc memory\n");
651 		return -EINVAL;
652 	}
653 
654 	size = sizeof(cmd->hdr) + sizeof(cmd->frag_desc_bank_cfg32);
655 	skb = ath10k_htc_alloc_skb(ar, size);
656 	if (!skb)
657 		return -ENOMEM;
658 
659 	skb_put(skb, size);
660 	cmd = (struct htt_cmd *)skb->data;
661 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_FRAG_DESC_BANK_CFG;
662 
663 	info = 0;
664 	info |= SM(htt->tx_q_state.type,
665 		   HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_DEPTH_TYPE);
666 
667 	if (test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
668 		     ar->running_fw->fw_file.fw_features))
669 		info |= HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_VALID;
670 
671 	cfg = &cmd->frag_desc_bank_cfg32;
672 	cfg->info = info;
673 	cfg->num_banks = 1;
674 	cfg->desc_size = sizeof(struct htt_msdu_ext_desc);
675 	cfg->bank_base_addrs[0] = __cpu_to_le32(htt->frag_desc.paddr);
676 	cfg->bank_id[0].bank_min_id = 0;
677 	cfg->bank_id[0].bank_max_id = __cpu_to_le16(htt->max_num_pending_tx -
678 						    1);
679 
680 	cfg->q_state.paddr = cpu_to_le32(htt->tx_q_state.paddr);
681 	cfg->q_state.num_peers = cpu_to_le16(htt->tx_q_state.num_peers);
682 	cfg->q_state.num_tids = cpu_to_le16(htt->tx_q_state.num_tids);
683 	cfg->q_state.record_size = HTT_TX_Q_STATE_ENTRY_SIZE;
684 	cfg->q_state.record_multiplier = HTT_TX_Q_STATE_ENTRY_MULTIPLIER;
685 
686 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt frag desc bank cmd\n");
687 
688 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
689 	if (ret) {
690 		ath10k_warn(ar, "failed to send frag desc bank cfg request: %d\n",
691 			    ret);
692 		dev_kfree_skb_any(skb);
693 		return ret;
694 	}
695 
696 	return 0;
697 }
698 
699 static int ath10k_htt_send_frag_desc_bank_cfg_64(struct ath10k_htt *htt)
700 {
701 	struct ath10k *ar = htt->ar;
702 	struct sk_buff *skb;
703 	struct htt_cmd *cmd;
704 	struct htt_frag_desc_bank_cfg64 *cfg;
705 	int ret, size;
706 	u8 info;
707 
708 	if (!ar->hw_params.continuous_frag_desc)
709 		return 0;
710 
711 	if (!htt->frag_desc.paddr) {
712 		ath10k_warn(ar, "invalid frag desc memory\n");
713 		return -EINVAL;
714 	}
715 
716 	size = sizeof(cmd->hdr) + sizeof(cmd->frag_desc_bank_cfg64);
717 	skb = ath10k_htc_alloc_skb(ar, size);
718 	if (!skb)
719 		return -ENOMEM;
720 
721 	skb_put(skb, size);
722 	cmd = (struct htt_cmd *)skb->data;
723 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_FRAG_DESC_BANK_CFG;
724 
725 	info = 0;
726 	info |= SM(htt->tx_q_state.type,
727 		   HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_DEPTH_TYPE);
728 
729 	if (test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
730 		     ar->running_fw->fw_file.fw_features))
731 		info |= HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_VALID;
732 
733 	cfg = &cmd->frag_desc_bank_cfg64;
734 	cfg->info = info;
735 	cfg->num_banks = 1;
736 	cfg->desc_size = sizeof(struct htt_msdu_ext_desc_64);
737 	cfg->bank_base_addrs[0] =  __cpu_to_le64(htt->frag_desc.paddr);
738 	cfg->bank_id[0].bank_min_id = 0;
739 	cfg->bank_id[0].bank_max_id = __cpu_to_le16(htt->max_num_pending_tx -
740 						    1);
741 
742 	cfg->q_state.paddr = cpu_to_le32(htt->tx_q_state.paddr);
743 	cfg->q_state.num_peers = cpu_to_le16(htt->tx_q_state.num_peers);
744 	cfg->q_state.num_tids = cpu_to_le16(htt->tx_q_state.num_tids);
745 	cfg->q_state.record_size = HTT_TX_Q_STATE_ENTRY_SIZE;
746 	cfg->q_state.record_multiplier = HTT_TX_Q_STATE_ENTRY_MULTIPLIER;
747 
748 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt frag desc bank cmd\n");
749 
750 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
751 	if (ret) {
752 		ath10k_warn(ar, "failed to send frag desc bank cfg request: %d\n",
753 			    ret);
754 		dev_kfree_skb_any(skb);
755 		return ret;
756 	}
757 
758 	return 0;
759 }
760 
761 static void ath10k_htt_fill_rx_desc_offset_32(void *rx_ring)
762 {
763 	struct htt_rx_ring_setup_ring32 *ring =
764 			(struct htt_rx_ring_setup_ring32 *)rx_ring;
765 
766 #define desc_offset(x) (offsetof(struct htt_rx_desc, x) / 4)
767 	ring->mac80211_hdr_offset = __cpu_to_le16(desc_offset(rx_hdr_status));
768 	ring->msdu_payload_offset = __cpu_to_le16(desc_offset(msdu_payload));
769 	ring->ppdu_start_offset = __cpu_to_le16(desc_offset(ppdu_start));
770 	ring->ppdu_end_offset = __cpu_to_le16(desc_offset(ppdu_end));
771 	ring->mpdu_start_offset = __cpu_to_le16(desc_offset(mpdu_start));
772 	ring->mpdu_end_offset = __cpu_to_le16(desc_offset(mpdu_end));
773 	ring->msdu_start_offset = __cpu_to_le16(desc_offset(msdu_start));
774 	ring->msdu_end_offset = __cpu_to_le16(desc_offset(msdu_end));
775 	ring->rx_attention_offset = __cpu_to_le16(desc_offset(attention));
776 	ring->frag_info_offset = __cpu_to_le16(desc_offset(frag_info));
777 #undef desc_offset
778 }
779 
780 static void ath10k_htt_fill_rx_desc_offset_64(void *rx_ring)
781 {
782 	struct htt_rx_ring_setup_ring64 *ring =
783 			(struct htt_rx_ring_setup_ring64 *)rx_ring;
784 
785 #define desc_offset(x) (offsetof(struct htt_rx_desc, x) / 4)
786 	ring->mac80211_hdr_offset = __cpu_to_le16(desc_offset(rx_hdr_status));
787 	ring->msdu_payload_offset = __cpu_to_le16(desc_offset(msdu_payload));
788 	ring->ppdu_start_offset = __cpu_to_le16(desc_offset(ppdu_start));
789 	ring->ppdu_end_offset = __cpu_to_le16(desc_offset(ppdu_end));
790 	ring->mpdu_start_offset = __cpu_to_le16(desc_offset(mpdu_start));
791 	ring->mpdu_end_offset = __cpu_to_le16(desc_offset(mpdu_end));
792 	ring->msdu_start_offset = __cpu_to_le16(desc_offset(msdu_start));
793 	ring->msdu_end_offset = __cpu_to_le16(desc_offset(msdu_end));
794 	ring->rx_attention_offset = __cpu_to_le16(desc_offset(attention));
795 	ring->frag_info_offset = __cpu_to_le16(desc_offset(frag_info));
796 #undef desc_offset
797 }
798 
799 static int ath10k_htt_send_rx_ring_cfg_32(struct ath10k_htt *htt)
800 {
801 	struct ath10k *ar = htt->ar;
802 	struct sk_buff *skb;
803 	struct htt_cmd *cmd;
804 	struct htt_rx_ring_setup_ring32 *ring;
805 	const int num_rx_ring = 1;
806 	u16 flags;
807 	u32 fw_idx;
808 	int len;
809 	int ret;
810 
811 	/*
812 	 * the HW expects the buffer to be an integral number of 4-byte
813 	 * "words"
814 	 */
815 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
816 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
817 
818 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup_32.hdr)
819 	    + (sizeof(*ring) * num_rx_ring);
820 	skb = ath10k_htc_alloc_skb(ar, len);
821 	if (!skb)
822 		return -ENOMEM;
823 
824 	skb_put(skb, len);
825 
826 	cmd = (struct htt_cmd *)skb->data;
827 	ring = &cmd->rx_setup_32.rings[0];
828 
829 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
830 	cmd->rx_setup_32.hdr.num_rings = 1;
831 
832 	/* FIXME: do we need all of this? */
833 	flags = 0;
834 	flags |= HTT_RX_RING_FLAGS_MAC80211_HDR;
835 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
836 	flags |= HTT_RX_RING_FLAGS_PPDU_START;
837 	flags |= HTT_RX_RING_FLAGS_PPDU_END;
838 	flags |= HTT_RX_RING_FLAGS_MPDU_START;
839 	flags |= HTT_RX_RING_FLAGS_MPDU_END;
840 	flags |= HTT_RX_RING_FLAGS_MSDU_START;
841 	flags |= HTT_RX_RING_FLAGS_MSDU_END;
842 	flags |= HTT_RX_RING_FLAGS_RX_ATTENTION;
843 	flags |= HTT_RX_RING_FLAGS_FRAG_INFO;
844 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
845 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
846 	flags |= HTT_RX_RING_FLAGS_CTRL_RX;
847 	flags |= HTT_RX_RING_FLAGS_MGMT_RX;
848 	flags |= HTT_RX_RING_FLAGS_NULL_RX;
849 	flags |= HTT_RX_RING_FLAGS_PHY_DATA_RX;
850 
851 	fw_idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
852 
853 	ring->fw_idx_shadow_reg_paddr =
854 		__cpu_to_le32(htt->rx_ring.alloc_idx.paddr);
855 	ring->rx_ring_base_paddr = __cpu_to_le32(htt->rx_ring.base_paddr);
856 	ring->rx_ring_len = __cpu_to_le16(htt->rx_ring.size);
857 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
858 	ring->flags = __cpu_to_le16(flags);
859 	ring->fw_idx_init_val = __cpu_to_le16(fw_idx);
860 
861 	ath10k_htt_fill_rx_desc_offset_32(ring);
862 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
863 	if (ret) {
864 		dev_kfree_skb_any(skb);
865 		return ret;
866 	}
867 
868 	return 0;
869 }
870 
871 static int ath10k_htt_send_rx_ring_cfg_64(struct ath10k_htt *htt)
872 {
873 	struct ath10k *ar = htt->ar;
874 	struct sk_buff *skb;
875 	struct htt_cmd *cmd;
876 	struct htt_rx_ring_setup_ring64 *ring;
877 	const int num_rx_ring = 1;
878 	u16 flags;
879 	u32 fw_idx;
880 	int len;
881 	int ret;
882 
883 	/* HW expects the buffer to be an integral number of 4-byte
884 	 * "words"
885 	 */
886 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
887 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
888 
889 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup_64.hdr)
890 	    + (sizeof(*ring) * num_rx_ring);
891 	skb = ath10k_htc_alloc_skb(ar, len);
892 	if (!skb)
893 		return -ENOMEM;
894 
895 	skb_put(skb, len);
896 
897 	cmd = (struct htt_cmd *)skb->data;
898 	ring = &cmd->rx_setup_64.rings[0];
899 
900 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
901 	cmd->rx_setup_64.hdr.num_rings = 1;
902 
903 	flags = 0;
904 	flags |= HTT_RX_RING_FLAGS_MAC80211_HDR;
905 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
906 	flags |= HTT_RX_RING_FLAGS_PPDU_START;
907 	flags |= HTT_RX_RING_FLAGS_PPDU_END;
908 	flags |= HTT_RX_RING_FLAGS_MPDU_START;
909 	flags |= HTT_RX_RING_FLAGS_MPDU_END;
910 	flags |= HTT_RX_RING_FLAGS_MSDU_START;
911 	flags |= HTT_RX_RING_FLAGS_MSDU_END;
912 	flags |= HTT_RX_RING_FLAGS_RX_ATTENTION;
913 	flags |= HTT_RX_RING_FLAGS_FRAG_INFO;
914 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
915 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
916 	flags |= HTT_RX_RING_FLAGS_CTRL_RX;
917 	flags |= HTT_RX_RING_FLAGS_MGMT_RX;
918 	flags |= HTT_RX_RING_FLAGS_NULL_RX;
919 	flags |= HTT_RX_RING_FLAGS_PHY_DATA_RX;
920 
921 	fw_idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
922 
923 	ring->fw_idx_shadow_reg_paddr = __cpu_to_le64(htt->rx_ring.alloc_idx.paddr);
924 	ring->rx_ring_base_paddr = __cpu_to_le64(htt->rx_ring.base_paddr);
925 	ring->rx_ring_len = __cpu_to_le16(htt->rx_ring.size);
926 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
927 	ring->flags = __cpu_to_le16(flags);
928 	ring->fw_idx_init_val = __cpu_to_le16(fw_idx);
929 
930 	ath10k_htt_fill_rx_desc_offset_64(ring);
931 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
932 	if (ret) {
933 		dev_kfree_skb_any(skb);
934 		return ret;
935 	}
936 
937 	return 0;
938 }
939 
940 static int ath10k_htt_send_rx_ring_cfg_hl(struct ath10k_htt *htt)
941 {
942 	struct ath10k *ar = htt->ar;
943 	struct sk_buff *skb;
944 	struct htt_cmd *cmd;
945 	struct htt_rx_ring_setup_ring32 *ring;
946 	const int num_rx_ring = 1;
947 	u16 flags;
948 	int len;
949 	int ret;
950 
951 	/*
952 	 * the HW expects the buffer to be an integral number of 4-byte
953 	 * "words"
954 	 */
955 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
956 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
957 
958 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup_32.hdr)
959 	    + (sizeof(*ring) * num_rx_ring);
960 	skb = ath10k_htc_alloc_skb(ar, len);
961 	if (!skb)
962 		return -ENOMEM;
963 
964 	skb_put(skb, len);
965 
966 	cmd = (struct htt_cmd *)skb->data;
967 	ring = &cmd->rx_setup_32.rings[0];
968 
969 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
970 	cmd->rx_setup_32.hdr.num_rings = 1;
971 
972 	flags = 0;
973 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
974 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
975 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
976 
977 	memset(ring, 0, sizeof(*ring));
978 	ring->rx_ring_len = __cpu_to_le16(HTT_RX_RING_SIZE_MIN);
979 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
980 	ring->flags = __cpu_to_le16(flags);
981 
982 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
983 	if (ret) {
984 		dev_kfree_skb_any(skb);
985 		return ret;
986 	}
987 
988 	return 0;
989 }
990 
991 int ath10k_htt_h2t_aggr_cfg_msg(struct ath10k_htt *htt,
992 				u8 max_subfrms_ampdu,
993 				u8 max_subfrms_amsdu)
994 {
995 	struct ath10k *ar = htt->ar;
996 	struct htt_aggr_conf *aggr_conf;
997 	struct sk_buff *skb;
998 	struct htt_cmd *cmd;
999 	int len;
1000 	int ret;
1001 
1002 	/* Firmware defaults are: amsdu = 3 and ampdu = 64 */
1003 
1004 	if (max_subfrms_ampdu == 0 || max_subfrms_ampdu > 64)
1005 		return -EINVAL;
1006 
1007 	if (max_subfrms_amsdu == 0 || max_subfrms_amsdu > 31)
1008 		return -EINVAL;
1009 
1010 	len = sizeof(cmd->hdr);
1011 	len += sizeof(cmd->aggr_conf);
1012 
1013 	skb = ath10k_htc_alloc_skb(ar, len);
1014 	if (!skb)
1015 		return -ENOMEM;
1016 
1017 	skb_put(skb, len);
1018 	cmd = (struct htt_cmd *)skb->data;
1019 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_AGGR_CFG;
1020 
1021 	aggr_conf = &cmd->aggr_conf;
1022 	aggr_conf->max_num_ampdu_subframes = max_subfrms_ampdu;
1023 	aggr_conf->max_num_amsdu_subframes = max_subfrms_amsdu;
1024 
1025 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt h2t aggr cfg msg amsdu %d ampdu %d",
1026 		   aggr_conf->max_num_amsdu_subframes,
1027 		   aggr_conf->max_num_ampdu_subframes);
1028 
1029 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
1030 	if (ret) {
1031 		dev_kfree_skb_any(skb);
1032 		return ret;
1033 	}
1034 
1035 	return 0;
1036 }
1037 
1038 int ath10k_htt_tx_fetch_resp(struct ath10k *ar,
1039 			     __le32 token,
1040 			     __le16 fetch_seq_num,
1041 			     struct htt_tx_fetch_record *records,
1042 			     size_t num_records)
1043 {
1044 	struct sk_buff *skb;
1045 	struct htt_cmd *cmd;
1046 	const u16 resp_id = 0;
1047 	int len = 0;
1048 	int ret;
1049 
1050 	/* Response IDs are echo-ed back only for host driver convienence
1051 	 * purposes. They aren't used for anything in the driver yet so use 0.
1052 	 */
1053 
1054 	len += sizeof(cmd->hdr);
1055 	len += sizeof(cmd->tx_fetch_resp);
1056 	len += sizeof(cmd->tx_fetch_resp.records[0]) * num_records;
1057 
1058 	skb = ath10k_htc_alloc_skb(ar, len);
1059 	if (!skb)
1060 		return -ENOMEM;
1061 
1062 	skb_put(skb, len);
1063 	cmd = (struct htt_cmd *)skb->data;
1064 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FETCH_RESP;
1065 	cmd->tx_fetch_resp.resp_id = cpu_to_le16(resp_id);
1066 	cmd->tx_fetch_resp.fetch_seq_num = fetch_seq_num;
1067 	cmd->tx_fetch_resp.num_records = cpu_to_le16(num_records);
1068 	cmd->tx_fetch_resp.token = token;
1069 
1070 	memcpy(cmd->tx_fetch_resp.records, records,
1071 	       sizeof(records[0]) * num_records);
1072 
1073 	ret = ath10k_htc_send(&ar->htc, ar->htt.eid, skb);
1074 	if (ret) {
1075 		ath10k_warn(ar, "failed to submit htc command: %d\n", ret);
1076 		goto err_free_skb;
1077 	}
1078 
1079 	return 0;
1080 
1081 err_free_skb:
1082 	dev_kfree_skb_any(skb);
1083 
1084 	return ret;
1085 }
1086 
1087 static u8 ath10k_htt_tx_get_vdev_id(struct ath10k *ar, struct sk_buff *skb)
1088 {
1089 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
1090 	struct ath10k_skb_cb *cb = ATH10K_SKB_CB(skb);
1091 	struct ath10k_vif *arvif;
1092 
1093 	if (info->flags & IEEE80211_TX_CTL_TX_OFFCHAN) {
1094 		return ar->scan.vdev_id;
1095 	} else if (cb->vif) {
1096 		arvif = (void *)cb->vif->drv_priv;
1097 		return arvif->vdev_id;
1098 	} else if (ar->monitor_started) {
1099 		return ar->monitor_vdev_id;
1100 	} else {
1101 		return 0;
1102 	}
1103 }
1104 
1105 static u8 ath10k_htt_tx_get_tid(struct sk_buff *skb, bool is_eth)
1106 {
1107 	struct ieee80211_hdr *hdr = (void *)skb->data;
1108 	struct ath10k_skb_cb *cb = ATH10K_SKB_CB(skb);
1109 
1110 	if (!is_eth && ieee80211_is_mgmt(hdr->frame_control))
1111 		return HTT_DATA_TX_EXT_TID_MGMT;
1112 	else if (cb->flags & ATH10K_SKB_F_QOS)
1113 		return skb->priority & IEEE80211_QOS_CTL_TID_MASK;
1114 	else
1115 		return HTT_DATA_TX_EXT_TID_NON_QOS_MCAST_BCAST;
1116 }
1117 
1118 int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
1119 {
1120 	struct ath10k *ar = htt->ar;
1121 	struct device *dev = ar->dev;
1122 	struct sk_buff *txdesc = NULL;
1123 	struct htt_cmd *cmd;
1124 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1125 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1126 	int len = 0;
1127 	int msdu_id = -1;
1128 	int res;
1129 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1130 
1131 	len += sizeof(cmd->hdr);
1132 	len += sizeof(cmd->mgmt_tx);
1133 
1134 	res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1135 	if (res < 0)
1136 		goto err;
1137 
1138 	msdu_id = res;
1139 
1140 	if ((ieee80211_is_action(hdr->frame_control) ||
1141 	     ieee80211_is_deauth(hdr->frame_control) ||
1142 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1143 	     ieee80211_has_protected(hdr->frame_control)) {
1144 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1145 	}
1146 
1147 	txdesc = ath10k_htc_alloc_skb(ar, len);
1148 	if (!txdesc) {
1149 		res = -ENOMEM;
1150 		goto err_free_msdu_id;
1151 	}
1152 
1153 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
1154 				       DMA_TO_DEVICE);
1155 	res = dma_mapping_error(dev, skb_cb->paddr);
1156 	if (res) {
1157 		res = -EIO;
1158 		goto err_free_txdesc;
1159 	}
1160 
1161 	skb_put(txdesc, len);
1162 	cmd = (struct htt_cmd *)txdesc->data;
1163 	memset(cmd, 0, len);
1164 
1165 	cmd->hdr.msg_type         = HTT_H2T_MSG_TYPE_MGMT_TX;
1166 	cmd->mgmt_tx.msdu_paddr = __cpu_to_le32(ATH10K_SKB_CB(msdu)->paddr);
1167 	cmd->mgmt_tx.len        = __cpu_to_le32(msdu->len);
1168 	cmd->mgmt_tx.desc_id    = __cpu_to_le32(msdu_id);
1169 	cmd->mgmt_tx.vdev_id    = __cpu_to_le32(vdev_id);
1170 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
1171 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
1172 
1173 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
1174 	if (res)
1175 		goto err_unmap_msdu;
1176 
1177 	return 0;
1178 
1179 err_unmap_msdu:
1180 	if (ar->dev_type != ATH10K_DEV_TYPE_HL)
1181 		dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
1182 err_free_txdesc:
1183 	dev_kfree_skb_any(txdesc);
1184 err_free_msdu_id:
1185 	spin_lock_bh(&htt->tx_lock);
1186 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
1187 	spin_unlock_bh(&htt->tx_lock);
1188 err:
1189 	return res;
1190 }
1191 
1192 #define HTT_TX_HL_NEEDED_HEADROOM \
1193 	(unsigned int)(sizeof(struct htt_cmd_hdr) + \
1194 	sizeof(struct htt_data_tx_desc) + \
1195 	sizeof(struct ath10k_htc_hdr))
1196 
1197 static int ath10k_htt_tx_hl(struct ath10k_htt *htt, enum ath10k_hw_txrx_mode txmode,
1198 			    struct sk_buff *msdu)
1199 {
1200 	struct ath10k *ar = htt->ar;
1201 	int res, data_len;
1202 	struct htt_cmd_hdr *cmd_hdr;
1203 	struct htt_data_tx_desc *tx_desc;
1204 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1205 	struct sk_buff *tmp_skb;
1206 	bool is_eth = (txmode == ATH10K_HW_TXRX_ETHERNET);
1207 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1208 	u8 tid = ath10k_htt_tx_get_tid(msdu, is_eth);
1209 	u8 flags0 = 0;
1210 	u16 flags1 = 0;
1211 
1212 	data_len = msdu->len;
1213 
1214 	switch (txmode) {
1215 	case ATH10K_HW_TXRX_RAW:
1216 	case ATH10K_HW_TXRX_NATIVE_WIFI:
1217 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1218 		/* fall through */
1219 	case ATH10K_HW_TXRX_ETHERNET:
1220 		flags0 |= SM(txmode, HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1221 		break;
1222 	case ATH10K_HW_TXRX_MGMT:
1223 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
1224 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1225 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1226 		break;
1227 	}
1228 
1229 	if (skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT)
1230 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
1231 
1232 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
1233 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
1234 	if (msdu->ip_summed == CHECKSUM_PARTIAL &&
1235 	    !test_bit(ATH10K_FLAG_RAW_MODE, &ar->dev_flags)) {
1236 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
1237 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
1238 	}
1239 
1240 	/* Prepend the HTT header and TX desc struct to the data message
1241 	 * and realloc the skb if it does not have enough headroom.
1242 	 */
1243 	if (skb_headroom(msdu) < HTT_TX_HL_NEEDED_HEADROOM) {
1244 		tmp_skb = msdu;
1245 
1246 		ath10k_dbg(htt->ar, ATH10K_DBG_HTT,
1247 			   "Not enough headroom in skb. Current headroom: %u, needed: %u. Reallocating...\n",
1248 			   skb_headroom(msdu), HTT_TX_HL_NEEDED_HEADROOM);
1249 		msdu = skb_realloc_headroom(msdu, HTT_TX_HL_NEEDED_HEADROOM);
1250 		kfree_skb(tmp_skb);
1251 		if (!msdu) {
1252 			ath10k_warn(htt->ar, "htt hl tx: Unable to realloc skb!\n");
1253 			res = -ENOMEM;
1254 			goto out;
1255 		}
1256 	}
1257 
1258 	skb_push(msdu, sizeof(*cmd_hdr));
1259 	skb_push(msdu, sizeof(*tx_desc));
1260 	cmd_hdr = (struct htt_cmd_hdr *)msdu->data;
1261 	tx_desc = (struct htt_data_tx_desc *)(msdu->data + sizeof(*cmd_hdr));
1262 
1263 	cmd_hdr->msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
1264 	tx_desc->flags0 = flags0;
1265 	tx_desc->flags1 = __cpu_to_le16(flags1);
1266 	tx_desc->len = __cpu_to_le16(data_len);
1267 	tx_desc->id = 0;
1268 	tx_desc->frags_paddr = 0; /* always zero */
1269 	/* Initialize peer_id to INVALID_PEER because this is NOT
1270 	 * Reinjection path
1271 	 */
1272 	tx_desc->peerid = __cpu_to_le32(HTT_INVALID_PEERID);
1273 
1274 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, msdu);
1275 
1276 out:
1277 	return res;
1278 }
1279 
1280 static int ath10k_htt_tx_32(struct ath10k_htt *htt,
1281 			    enum ath10k_hw_txrx_mode txmode,
1282 			    struct sk_buff *msdu)
1283 {
1284 	struct ath10k *ar = htt->ar;
1285 	struct device *dev = ar->dev;
1286 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1287 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(msdu);
1288 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1289 	struct ath10k_hif_sg_item sg_items[2];
1290 	struct ath10k_htt_txbuf_32 *txbuf;
1291 	struct htt_data_tx_desc_frag *frags;
1292 	bool is_eth = (txmode == ATH10K_HW_TXRX_ETHERNET);
1293 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1294 	u8 tid = ath10k_htt_tx_get_tid(msdu, is_eth);
1295 	int prefetch_len;
1296 	int res;
1297 	u8 flags0 = 0;
1298 	u16 msdu_id, flags1 = 0;
1299 	u16 freq = 0;
1300 	u32 frags_paddr = 0;
1301 	u32 txbuf_paddr;
1302 	struct htt_msdu_ext_desc *ext_desc = NULL;
1303 	struct htt_msdu_ext_desc *ext_desc_t = NULL;
1304 
1305 	res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1306 	if (res < 0)
1307 		goto err;
1308 
1309 	msdu_id = res;
1310 
1311 	prefetch_len = min(htt->prefetch_len, msdu->len);
1312 	prefetch_len = roundup(prefetch_len, 4);
1313 
1314 	txbuf = htt->txbuf.vaddr_txbuff_32 + msdu_id;
1315 	txbuf_paddr = htt->txbuf.paddr +
1316 		      (sizeof(struct ath10k_htt_txbuf_32) * msdu_id);
1317 
1318 	if ((ieee80211_is_action(hdr->frame_control) ||
1319 	     ieee80211_is_deauth(hdr->frame_control) ||
1320 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1321 	     ieee80211_has_protected(hdr->frame_control)) {
1322 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1323 	} else if (!(skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT) &&
1324 		   txmode == ATH10K_HW_TXRX_RAW &&
1325 		   ieee80211_has_protected(hdr->frame_control)) {
1326 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1327 	}
1328 
1329 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
1330 				       DMA_TO_DEVICE);
1331 	res = dma_mapping_error(dev, skb_cb->paddr);
1332 	if (res) {
1333 		res = -EIO;
1334 		goto err_free_msdu_id;
1335 	}
1336 
1337 	if (unlikely(info->flags & IEEE80211_TX_CTL_TX_OFFCHAN))
1338 		freq = ar->scan.roc_freq;
1339 
1340 	switch (txmode) {
1341 	case ATH10K_HW_TXRX_RAW:
1342 	case ATH10K_HW_TXRX_NATIVE_WIFI:
1343 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1344 		/* fall through */
1345 	case ATH10K_HW_TXRX_ETHERNET:
1346 		if (ar->hw_params.continuous_frag_desc) {
1347 			ext_desc_t = htt->frag_desc.vaddr_desc_32;
1348 			memset(&ext_desc_t[msdu_id], 0,
1349 			       sizeof(struct htt_msdu_ext_desc));
1350 			frags = (struct htt_data_tx_desc_frag *)
1351 				&ext_desc_t[msdu_id].frags;
1352 			ext_desc = &ext_desc_t[msdu_id];
1353 			frags[0].tword_addr.paddr_lo =
1354 				__cpu_to_le32(skb_cb->paddr);
1355 			frags[0].tword_addr.paddr_hi = 0;
1356 			frags[0].tword_addr.len_16 = __cpu_to_le16(msdu->len);
1357 
1358 			frags_paddr =  htt->frag_desc.paddr +
1359 				(sizeof(struct htt_msdu_ext_desc) * msdu_id);
1360 		} else {
1361 			frags = txbuf->frags;
1362 			frags[0].dword_addr.paddr =
1363 				__cpu_to_le32(skb_cb->paddr);
1364 			frags[0].dword_addr.len = __cpu_to_le32(msdu->len);
1365 			frags[1].dword_addr.paddr = 0;
1366 			frags[1].dword_addr.len = 0;
1367 
1368 			frags_paddr = txbuf_paddr;
1369 		}
1370 		flags0 |= SM(txmode, HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1371 		break;
1372 	case ATH10K_HW_TXRX_MGMT:
1373 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
1374 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1375 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1376 
1377 		frags_paddr = skb_cb->paddr;
1378 		break;
1379 	}
1380 
1381 	/* Normally all commands go through HTC which manages tx credits for
1382 	 * each endpoint and notifies when tx is completed.
1383 	 *
1384 	 * HTT endpoint is creditless so there's no need to care about HTC
1385 	 * flags. In that case it is trivial to fill the HTC header here.
1386 	 *
1387 	 * MSDU transmission is considered completed upon HTT event. This
1388 	 * implies no relevant resources can be freed until after the event is
1389 	 * received. That's why HTC tx completion handler itself is ignored by
1390 	 * setting NULL to transfer_context for all sg items.
1391 	 *
1392 	 * There is simply no point in pushing HTT TX_FRM through HTC tx path
1393 	 * as it's a waste of resources. By bypassing HTC it is possible to
1394 	 * avoid extra memory allocations, compress data structures and thus
1395 	 * improve performance.
1396 	 */
1397 
1398 	txbuf->htc_hdr.eid = htt->eid;
1399 	txbuf->htc_hdr.len = __cpu_to_le16(sizeof(txbuf->cmd_hdr) +
1400 					   sizeof(txbuf->cmd_tx) +
1401 					   prefetch_len);
1402 	txbuf->htc_hdr.flags = 0;
1403 
1404 	if (skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT)
1405 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
1406 
1407 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
1408 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
1409 	if (msdu->ip_summed == CHECKSUM_PARTIAL &&
1410 	    !test_bit(ATH10K_FLAG_RAW_MODE, &ar->dev_flags)) {
1411 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
1412 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
1413 		if (ar->hw_params.continuous_frag_desc)
1414 			ext_desc->flags |= HTT_MSDU_CHECKSUM_ENABLE;
1415 	}
1416 
1417 	/* Prevent firmware from sending up tx inspection requests. There's
1418 	 * nothing ath10k can do with frames requested for inspection so force
1419 	 * it to simply rely a regular tx completion with discard status.
1420 	 */
1421 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_POSTPONED;
1422 
1423 	txbuf->cmd_hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
1424 	txbuf->cmd_tx.flags0 = flags0;
1425 	txbuf->cmd_tx.flags1 = __cpu_to_le16(flags1);
1426 	txbuf->cmd_tx.len = __cpu_to_le16(msdu->len);
1427 	txbuf->cmd_tx.id = __cpu_to_le16(msdu_id);
1428 	txbuf->cmd_tx.frags_paddr = __cpu_to_le32(frags_paddr);
1429 	if (ath10k_mac_tx_frm_has_freq(ar)) {
1430 		txbuf->cmd_tx.offchan_tx.peerid =
1431 				__cpu_to_le16(HTT_INVALID_PEERID);
1432 		txbuf->cmd_tx.offchan_tx.freq =
1433 				__cpu_to_le16(freq);
1434 	} else {
1435 		txbuf->cmd_tx.peerid =
1436 				__cpu_to_le32(HTT_INVALID_PEERID);
1437 	}
1438 
1439 	trace_ath10k_htt_tx(ar, msdu_id, msdu->len, vdev_id, tid);
1440 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1441 		   "htt tx flags0 %hhu flags1 %hu len %d id %hu frags_paddr %pad, msdu_paddr %pad vdev %hhu tid %hhu freq %hu\n",
1442 		   flags0, flags1, msdu->len, msdu_id, &frags_paddr,
1443 		   &skb_cb->paddr, vdev_id, tid, freq);
1444 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt tx msdu: ",
1445 			msdu->data, msdu->len);
1446 	trace_ath10k_tx_hdr(ar, msdu->data, msdu->len);
1447 	trace_ath10k_tx_payload(ar, msdu->data, msdu->len);
1448 
1449 	sg_items[0].transfer_id = 0;
1450 	sg_items[0].transfer_context = NULL;
1451 	sg_items[0].vaddr = &txbuf->htc_hdr;
1452 	sg_items[0].paddr = txbuf_paddr +
1453 			    sizeof(txbuf->frags);
1454 	sg_items[0].len = sizeof(txbuf->htc_hdr) +
1455 			  sizeof(txbuf->cmd_hdr) +
1456 			  sizeof(txbuf->cmd_tx);
1457 
1458 	sg_items[1].transfer_id = 0;
1459 	sg_items[1].transfer_context = NULL;
1460 	sg_items[1].vaddr = msdu->data;
1461 	sg_items[1].paddr = skb_cb->paddr;
1462 	sg_items[1].len = prefetch_len;
1463 
1464 	res = ath10k_hif_tx_sg(htt->ar,
1465 			       htt->ar->htc.endpoint[htt->eid].ul_pipe_id,
1466 			       sg_items, ARRAY_SIZE(sg_items));
1467 	if (res)
1468 		goto err_unmap_msdu;
1469 
1470 	return 0;
1471 
1472 err_unmap_msdu:
1473 	dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
1474 err_free_msdu_id:
1475 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
1476 err:
1477 	return res;
1478 }
1479 
1480 static int ath10k_htt_tx_64(struct ath10k_htt *htt,
1481 			    enum ath10k_hw_txrx_mode txmode,
1482 			    struct sk_buff *msdu)
1483 {
1484 	struct ath10k *ar = htt->ar;
1485 	struct device *dev = ar->dev;
1486 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1487 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(msdu);
1488 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1489 	struct ath10k_hif_sg_item sg_items[2];
1490 	struct ath10k_htt_txbuf_64 *txbuf;
1491 	struct htt_data_tx_desc_frag *frags;
1492 	bool is_eth = (txmode == ATH10K_HW_TXRX_ETHERNET);
1493 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1494 	u8 tid = ath10k_htt_tx_get_tid(msdu, is_eth);
1495 	int prefetch_len;
1496 	int res;
1497 	u8 flags0 = 0;
1498 	u16 msdu_id, flags1 = 0;
1499 	u16 freq = 0;
1500 	dma_addr_t frags_paddr = 0;
1501 	u32 txbuf_paddr;
1502 	struct htt_msdu_ext_desc_64 *ext_desc = NULL;
1503 	struct htt_msdu_ext_desc_64 *ext_desc_t = NULL;
1504 
1505 	res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1506 	if (res < 0)
1507 		goto err;
1508 
1509 	msdu_id = res;
1510 
1511 	prefetch_len = min(htt->prefetch_len, msdu->len);
1512 	prefetch_len = roundup(prefetch_len, 4);
1513 
1514 	txbuf = htt->txbuf.vaddr_txbuff_64 + msdu_id;
1515 	txbuf_paddr = htt->txbuf.paddr +
1516 		      (sizeof(struct ath10k_htt_txbuf_64) * msdu_id);
1517 
1518 	if ((ieee80211_is_action(hdr->frame_control) ||
1519 	     ieee80211_is_deauth(hdr->frame_control) ||
1520 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1521 	     ieee80211_has_protected(hdr->frame_control)) {
1522 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1523 	} else if (!(skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT) &&
1524 		   txmode == ATH10K_HW_TXRX_RAW &&
1525 		   ieee80211_has_protected(hdr->frame_control)) {
1526 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1527 	}
1528 
1529 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
1530 				       DMA_TO_DEVICE);
1531 	res = dma_mapping_error(dev, skb_cb->paddr);
1532 	if (res) {
1533 		res = -EIO;
1534 		goto err_free_msdu_id;
1535 	}
1536 
1537 	if (unlikely(info->flags & IEEE80211_TX_CTL_TX_OFFCHAN))
1538 		freq = ar->scan.roc_freq;
1539 
1540 	switch (txmode) {
1541 	case ATH10K_HW_TXRX_RAW:
1542 	case ATH10K_HW_TXRX_NATIVE_WIFI:
1543 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1544 		/* fall through */
1545 	case ATH10K_HW_TXRX_ETHERNET:
1546 		if (ar->hw_params.continuous_frag_desc) {
1547 			ext_desc_t = htt->frag_desc.vaddr_desc_64;
1548 			memset(&ext_desc_t[msdu_id], 0,
1549 			       sizeof(struct htt_msdu_ext_desc_64));
1550 			frags = (struct htt_data_tx_desc_frag *)
1551 				&ext_desc_t[msdu_id].frags;
1552 			ext_desc = &ext_desc_t[msdu_id];
1553 			frags[0].tword_addr.paddr_lo =
1554 				__cpu_to_le32(skb_cb->paddr);
1555 			frags[0].tword_addr.paddr_hi =
1556 				__cpu_to_le16(upper_32_bits(skb_cb->paddr));
1557 			frags[0].tword_addr.len_16 = __cpu_to_le16(msdu->len);
1558 
1559 			frags_paddr =  htt->frag_desc.paddr +
1560 			   (sizeof(struct htt_msdu_ext_desc_64) * msdu_id);
1561 		} else {
1562 			frags = txbuf->frags;
1563 			frags[0].tword_addr.paddr_lo =
1564 						__cpu_to_le32(skb_cb->paddr);
1565 			frags[0].tword_addr.paddr_hi =
1566 				__cpu_to_le16(upper_32_bits(skb_cb->paddr));
1567 			frags[0].tword_addr.len_16 = __cpu_to_le16(msdu->len);
1568 			frags[1].tword_addr.paddr_lo = 0;
1569 			frags[1].tword_addr.paddr_hi = 0;
1570 			frags[1].tword_addr.len_16 = 0;
1571 		}
1572 		flags0 |= SM(txmode, HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1573 		break;
1574 	case ATH10K_HW_TXRX_MGMT:
1575 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
1576 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1577 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1578 
1579 		frags_paddr = skb_cb->paddr;
1580 		break;
1581 	}
1582 
1583 	/* Normally all commands go through HTC which manages tx credits for
1584 	 * each endpoint and notifies when tx is completed.
1585 	 *
1586 	 * HTT endpoint is creditless so there's no need to care about HTC
1587 	 * flags. In that case it is trivial to fill the HTC header here.
1588 	 *
1589 	 * MSDU transmission is considered completed upon HTT event. This
1590 	 * implies no relevant resources can be freed until after the event is
1591 	 * received. That's why HTC tx completion handler itself is ignored by
1592 	 * setting NULL to transfer_context for all sg items.
1593 	 *
1594 	 * There is simply no point in pushing HTT TX_FRM through HTC tx path
1595 	 * as it's a waste of resources. By bypassing HTC it is possible to
1596 	 * avoid extra memory allocations, compress data structures and thus
1597 	 * improve performance.
1598 	 */
1599 
1600 	txbuf->htc_hdr.eid = htt->eid;
1601 	txbuf->htc_hdr.len = __cpu_to_le16(sizeof(txbuf->cmd_hdr) +
1602 					   sizeof(txbuf->cmd_tx) +
1603 					   prefetch_len);
1604 	txbuf->htc_hdr.flags = 0;
1605 
1606 	if (skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT)
1607 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
1608 
1609 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
1610 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
1611 	if (msdu->ip_summed == CHECKSUM_PARTIAL &&
1612 	    !test_bit(ATH10K_FLAG_RAW_MODE, &ar->dev_flags)) {
1613 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
1614 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
1615 		if (ar->hw_params.continuous_frag_desc) {
1616 			memset(ext_desc->tso_flag, 0, sizeof(ext_desc->tso_flag));
1617 			ext_desc->tso_flag[3] |=
1618 				__cpu_to_le32(HTT_MSDU_CHECKSUM_ENABLE_64);
1619 		}
1620 	}
1621 
1622 	/* Prevent firmware from sending up tx inspection requests. There's
1623 	 * nothing ath10k can do with frames requested for inspection so force
1624 	 * it to simply rely a regular tx completion with discard status.
1625 	 */
1626 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_POSTPONED;
1627 
1628 	txbuf->cmd_hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
1629 	txbuf->cmd_tx.flags0 = flags0;
1630 	txbuf->cmd_tx.flags1 = __cpu_to_le16(flags1);
1631 	txbuf->cmd_tx.len = __cpu_to_le16(msdu->len);
1632 	txbuf->cmd_tx.id = __cpu_to_le16(msdu_id);
1633 
1634 	/* fill fragment descriptor */
1635 	txbuf->cmd_tx.frags_paddr = __cpu_to_le64(frags_paddr);
1636 	if (ath10k_mac_tx_frm_has_freq(ar)) {
1637 		txbuf->cmd_tx.offchan_tx.peerid =
1638 				__cpu_to_le16(HTT_INVALID_PEERID);
1639 		txbuf->cmd_tx.offchan_tx.freq =
1640 				__cpu_to_le16(freq);
1641 	} else {
1642 		txbuf->cmd_tx.peerid =
1643 				__cpu_to_le32(HTT_INVALID_PEERID);
1644 	}
1645 
1646 	trace_ath10k_htt_tx(ar, msdu_id, msdu->len, vdev_id, tid);
1647 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1648 		   "htt tx flags0 %hhu flags1 %hu len %d id %hu frags_paddr %pad, msdu_paddr %pad vdev %hhu tid %hhu freq %hu\n",
1649 		   flags0, flags1, msdu->len, msdu_id, &frags_paddr,
1650 		   &skb_cb->paddr, vdev_id, tid, freq);
1651 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt tx msdu: ",
1652 			msdu->data, msdu->len);
1653 	trace_ath10k_tx_hdr(ar, msdu->data, msdu->len);
1654 	trace_ath10k_tx_payload(ar, msdu->data, msdu->len);
1655 
1656 	sg_items[0].transfer_id = 0;
1657 	sg_items[0].transfer_context = NULL;
1658 	sg_items[0].vaddr = &txbuf->htc_hdr;
1659 	sg_items[0].paddr = txbuf_paddr +
1660 			    sizeof(txbuf->frags);
1661 	sg_items[0].len = sizeof(txbuf->htc_hdr) +
1662 			  sizeof(txbuf->cmd_hdr) +
1663 			  sizeof(txbuf->cmd_tx);
1664 
1665 	sg_items[1].transfer_id = 0;
1666 	sg_items[1].transfer_context = NULL;
1667 	sg_items[1].vaddr = msdu->data;
1668 	sg_items[1].paddr = skb_cb->paddr;
1669 	sg_items[1].len = prefetch_len;
1670 
1671 	res = ath10k_hif_tx_sg(htt->ar,
1672 			       htt->ar->htc.endpoint[htt->eid].ul_pipe_id,
1673 			       sg_items, ARRAY_SIZE(sg_items));
1674 	if (res)
1675 		goto err_unmap_msdu;
1676 
1677 	return 0;
1678 
1679 err_unmap_msdu:
1680 	dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
1681 err_free_msdu_id:
1682 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
1683 err:
1684 	return res;
1685 }
1686 
1687 static const struct ath10k_htt_tx_ops htt_tx_ops_32 = {
1688 	.htt_send_rx_ring_cfg = ath10k_htt_send_rx_ring_cfg_32,
1689 	.htt_send_frag_desc_bank_cfg = ath10k_htt_send_frag_desc_bank_cfg_32,
1690 	.htt_alloc_frag_desc = ath10k_htt_tx_alloc_cont_frag_desc_32,
1691 	.htt_free_frag_desc = ath10k_htt_tx_free_cont_frag_desc_32,
1692 	.htt_tx = ath10k_htt_tx_32,
1693 	.htt_alloc_txbuff = ath10k_htt_tx_alloc_cont_txbuf_32,
1694 	.htt_free_txbuff = ath10k_htt_tx_free_cont_txbuf_32,
1695 };
1696 
1697 static const struct ath10k_htt_tx_ops htt_tx_ops_64 = {
1698 	.htt_send_rx_ring_cfg = ath10k_htt_send_rx_ring_cfg_64,
1699 	.htt_send_frag_desc_bank_cfg = ath10k_htt_send_frag_desc_bank_cfg_64,
1700 	.htt_alloc_frag_desc = ath10k_htt_tx_alloc_cont_frag_desc_64,
1701 	.htt_free_frag_desc = ath10k_htt_tx_free_cont_frag_desc_64,
1702 	.htt_tx = ath10k_htt_tx_64,
1703 	.htt_alloc_txbuff = ath10k_htt_tx_alloc_cont_txbuf_64,
1704 	.htt_free_txbuff = ath10k_htt_tx_free_cont_txbuf_64,
1705 };
1706 
1707 static const struct ath10k_htt_tx_ops htt_tx_ops_hl = {
1708 	.htt_send_rx_ring_cfg = ath10k_htt_send_rx_ring_cfg_hl,
1709 	.htt_send_frag_desc_bank_cfg = ath10k_htt_send_frag_desc_bank_cfg_32,
1710 	.htt_tx = ath10k_htt_tx_hl,
1711 };
1712 
1713 void ath10k_htt_set_tx_ops(struct ath10k_htt *htt)
1714 {
1715 	struct ath10k *ar = htt->ar;
1716 
1717 	if (ar->dev_type == ATH10K_DEV_TYPE_HL)
1718 		htt->tx_ops = &htt_tx_ops_hl;
1719 	else if (ar->hw_params.target_64bit)
1720 		htt->tx_ops = &htt_tx_ops_64;
1721 	else
1722 		htt->tx_ops = &htt_tx_ops_32;
1723 }
1724