xref: /linux/drivers/net/wireless/ath/ath10k/htt_tx.c (revision dec1c62e91ba268ab2a6e339d4d7a59287d5eba1)
1 // SPDX-License-Identifier: ISC
2 /*
3  * Copyright (c) 2005-2011 Atheros Communications Inc.
4  * Copyright (c) 2011-2017 Qualcomm Atheros, Inc.
5  */
6 
7 #include <linux/etherdevice.h>
8 #include "htt.h"
9 #include "mac.h"
10 #include "hif.h"
11 #include "txrx.h"
12 #include "debug.h"
13 
14 static u8 ath10k_htt_tx_txq_calc_size(size_t count)
15 {
16 	int exp;
17 	int factor;
18 
19 	exp = 0;
20 	factor = count >> 7;
21 
22 	while (factor >= 64 && exp < 4) {
23 		factor >>= 3;
24 		exp++;
25 	}
26 
27 	if (exp == 4)
28 		return 0xff;
29 
30 	if (count > 0)
31 		factor = max(1, factor);
32 
33 	return SM(exp, HTT_TX_Q_STATE_ENTRY_EXP) |
34 	       SM(factor, HTT_TX_Q_STATE_ENTRY_FACTOR);
35 }
36 
37 static void __ath10k_htt_tx_txq_recalc(struct ieee80211_hw *hw,
38 				       struct ieee80211_txq *txq)
39 {
40 	struct ath10k *ar = hw->priv;
41 	struct ath10k_sta *arsta;
42 	struct ath10k_vif *arvif = (void *)txq->vif->drv_priv;
43 	unsigned long frame_cnt;
44 	unsigned long byte_cnt;
45 	int idx;
46 	u32 bit;
47 	u16 peer_id;
48 	u8 tid;
49 	u8 count;
50 
51 	lockdep_assert_held(&ar->htt.tx_lock);
52 
53 	if (!ar->htt.tx_q_state.enabled)
54 		return;
55 
56 	if (ar->htt.tx_q_state.mode != HTT_TX_MODE_SWITCH_PUSH_PULL)
57 		return;
58 
59 	if (txq->sta) {
60 		arsta = (void *)txq->sta->drv_priv;
61 		peer_id = arsta->peer_id;
62 	} else {
63 		peer_id = arvif->peer_id;
64 	}
65 
66 	tid = txq->tid;
67 	bit = BIT(peer_id % 32);
68 	idx = peer_id / 32;
69 
70 	ieee80211_txq_get_depth(txq, &frame_cnt, &byte_cnt);
71 	count = ath10k_htt_tx_txq_calc_size(byte_cnt);
72 
73 	if (unlikely(peer_id >= ar->htt.tx_q_state.num_peers) ||
74 	    unlikely(tid >= ar->htt.tx_q_state.num_tids)) {
75 		ath10k_warn(ar, "refusing to update txq for peer_id %u tid %u due to out of bounds\n",
76 			    peer_id, tid);
77 		return;
78 	}
79 
80 	ar->htt.tx_q_state.vaddr->count[tid][peer_id] = count;
81 	ar->htt.tx_q_state.vaddr->map[tid][idx] &= ~bit;
82 	ar->htt.tx_q_state.vaddr->map[tid][idx] |= count ? bit : 0;
83 
84 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx txq state update peer_id %u tid %u count %u\n",
85 		   peer_id, tid, count);
86 }
87 
88 static void __ath10k_htt_tx_txq_sync(struct ath10k *ar)
89 {
90 	u32 seq;
91 	size_t size;
92 
93 	lockdep_assert_held(&ar->htt.tx_lock);
94 
95 	if (!ar->htt.tx_q_state.enabled)
96 		return;
97 
98 	if (ar->htt.tx_q_state.mode != HTT_TX_MODE_SWITCH_PUSH_PULL)
99 		return;
100 
101 	seq = le32_to_cpu(ar->htt.tx_q_state.vaddr->seq);
102 	seq++;
103 	ar->htt.tx_q_state.vaddr->seq = cpu_to_le32(seq);
104 
105 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx txq state update commit seq %u\n",
106 		   seq);
107 
108 	size = sizeof(*ar->htt.tx_q_state.vaddr);
109 	dma_sync_single_for_device(ar->dev,
110 				   ar->htt.tx_q_state.paddr,
111 				   size,
112 				   DMA_TO_DEVICE);
113 }
114 
115 void ath10k_htt_tx_txq_recalc(struct ieee80211_hw *hw,
116 			      struct ieee80211_txq *txq)
117 {
118 	struct ath10k *ar = hw->priv;
119 
120 	spin_lock_bh(&ar->htt.tx_lock);
121 	__ath10k_htt_tx_txq_recalc(hw, txq);
122 	spin_unlock_bh(&ar->htt.tx_lock);
123 }
124 
125 void ath10k_htt_tx_txq_sync(struct ath10k *ar)
126 {
127 	spin_lock_bh(&ar->htt.tx_lock);
128 	__ath10k_htt_tx_txq_sync(ar);
129 	spin_unlock_bh(&ar->htt.tx_lock);
130 }
131 
132 void ath10k_htt_tx_txq_update(struct ieee80211_hw *hw,
133 			      struct ieee80211_txq *txq)
134 {
135 	struct ath10k *ar = hw->priv;
136 
137 	spin_lock_bh(&ar->htt.tx_lock);
138 	__ath10k_htt_tx_txq_recalc(hw, txq);
139 	__ath10k_htt_tx_txq_sync(ar);
140 	spin_unlock_bh(&ar->htt.tx_lock);
141 }
142 
143 void ath10k_htt_tx_dec_pending(struct ath10k_htt *htt)
144 {
145 	lockdep_assert_held(&htt->tx_lock);
146 
147 	htt->num_pending_tx--;
148 	if (htt->num_pending_tx == htt->max_num_pending_tx - 1)
149 		ath10k_mac_tx_unlock(htt->ar, ATH10K_TX_PAUSE_Q_FULL);
150 
151 	if (htt->num_pending_tx == 0)
152 		wake_up(&htt->empty_tx_wq);
153 }
154 
155 int ath10k_htt_tx_inc_pending(struct ath10k_htt *htt)
156 {
157 	lockdep_assert_held(&htt->tx_lock);
158 
159 	if (htt->num_pending_tx >= htt->max_num_pending_tx)
160 		return -EBUSY;
161 
162 	htt->num_pending_tx++;
163 	if (htt->num_pending_tx == htt->max_num_pending_tx)
164 		ath10k_mac_tx_lock(htt->ar, ATH10K_TX_PAUSE_Q_FULL);
165 
166 	return 0;
167 }
168 
169 int ath10k_htt_tx_mgmt_inc_pending(struct ath10k_htt *htt, bool is_mgmt,
170 				   bool is_presp)
171 {
172 	struct ath10k *ar = htt->ar;
173 
174 	lockdep_assert_held(&htt->tx_lock);
175 
176 	if (!is_mgmt || !ar->hw_params.max_probe_resp_desc_thres)
177 		return 0;
178 
179 	if (is_presp &&
180 	    ar->hw_params.max_probe_resp_desc_thres < htt->num_pending_mgmt_tx)
181 		return -EBUSY;
182 
183 	htt->num_pending_mgmt_tx++;
184 
185 	return 0;
186 }
187 
188 void ath10k_htt_tx_mgmt_dec_pending(struct ath10k_htt *htt)
189 {
190 	lockdep_assert_held(&htt->tx_lock);
191 
192 	if (!htt->ar->hw_params.max_probe_resp_desc_thres)
193 		return;
194 
195 	htt->num_pending_mgmt_tx--;
196 }
197 
198 int ath10k_htt_tx_alloc_msdu_id(struct ath10k_htt *htt, struct sk_buff *skb)
199 {
200 	struct ath10k *ar = htt->ar;
201 	int ret;
202 
203 	spin_lock_bh(&htt->tx_lock);
204 	ret = idr_alloc(&htt->pending_tx, skb, 0,
205 			htt->max_num_pending_tx, GFP_ATOMIC);
206 	spin_unlock_bh(&htt->tx_lock);
207 
208 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx alloc msdu_id %d\n", ret);
209 
210 	return ret;
211 }
212 
213 void ath10k_htt_tx_free_msdu_id(struct ath10k_htt *htt, u16 msdu_id)
214 {
215 	struct ath10k *ar = htt->ar;
216 
217 	lockdep_assert_held(&htt->tx_lock);
218 
219 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx free msdu_id %u\n", msdu_id);
220 
221 	idr_remove(&htt->pending_tx, msdu_id);
222 }
223 
224 static void ath10k_htt_tx_free_cont_txbuf_32(struct ath10k_htt *htt)
225 {
226 	struct ath10k *ar = htt->ar;
227 	size_t size;
228 
229 	if (!htt->txbuf.vaddr_txbuff_32)
230 		return;
231 
232 	size = htt->txbuf.size;
233 	dma_free_coherent(ar->dev, size, htt->txbuf.vaddr_txbuff_32,
234 			  htt->txbuf.paddr);
235 	htt->txbuf.vaddr_txbuff_32 = NULL;
236 }
237 
238 static int ath10k_htt_tx_alloc_cont_txbuf_32(struct ath10k_htt *htt)
239 {
240 	struct ath10k *ar = htt->ar;
241 	size_t size;
242 
243 	size = htt->max_num_pending_tx *
244 			sizeof(struct ath10k_htt_txbuf_32);
245 
246 	htt->txbuf.vaddr_txbuff_32 = dma_alloc_coherent(ar->dev, size,
247 							&htt->txbuf.paddr,
248 							GFP_KERNEL);
249 	if (!htt->txbuf.vaddr_txbuff_32)
250 		return -ENOMEM;
251 
252 	htt->txbuf.size = size;
253 
254 	return 0;
255 }
256 
257 static void ath10k_htt_tx_free_cont_txbuf_64(struct ath10k_htt *htt)
258 {
259 	struct ath10k *ar = htt->ar;
260 	size_t size;
261 
262 	if (!htt->txbuf.vaddr_txbuff_64)
263 		return;
264 
265 	size = htt->txbuf.size;
266 	dma_free_coherent(ar->dev, size, htt->txbuf.vaddr_txbuff_64,
267 			  htt->txbuf.paddr);
268 	htt->txbuf.vaddr_txbuff_64 = NULL;
269 }
270 
271 static int ath10k_htt_tx_alloc_cont_txbuf_64(struct ath10k_htt *htt)
272 {
273 	struct ath10k *ar = htt->ar;
274 	size_t size;
275 
276 	size = htt->max_num_pending_tx *
277 			sizeof(struct ath10k_htt_txbuf_64);
278 
279 	htt->txbuf.vaddr_txbuff_64 = dma_alloc_coherent(ar->dev, size,
280 							&htt->txbuf.paddr,
281 							GFP_KERNEL);
282 	if (!htt->txbuf.vaddr_txbuff_64)
283 		return -ENOMEM;
284 
285 	htt->txbuf.size = size;
286 
287 	return 0;
288 }
289 
290 static void ath10k_htt_tx_free_cont_frag_desc_32(struct ath10k_htt *htt)
291 {
292 	size_t size;
293 
294 	if (!htt->frag_desc.vaddr_desc_32)
295 		return;
296 
297 	size = htt->max_num_pending_tx *
298 			sizeof(struct htt_msdu_ext_desc);
299 
300 	dma_free_coherent(htt->ar->dev,
301 			  size,
302 			  htt->frag_desc.vaddr_desc_32,
303 			  htt->frag_desc.paddr);
304 
305 	htt->frag_desc.vaddr_desc_32 = NULL;
306 }
307 
308 static int ath10k_htt_tx_alloc_cont_frag_desc_32(struct ath10k_htt *htt)
309 {
310 	struct ath10k *ar = htt->ar;
311 	size_t size;
312 
313 	if (!ar->hw_params.continuous_frag_desc)
314 		return 0;
315 
316 	size = htt->max_num_pending_tx *
317 			sizeof(struct htt_msdu_ext_desc);
318 	htt->frag_desc.vaddr_desc_32 = dma_alloc_coherent(ar->dev, size,
319 							  &htt->frag_desc.paddr,
320 							  GFP_KERNEL);
321 	if (!htt->frag_desc.vaddr_desc_32) {
322 		ath10k_err(ar, "failed to alloc fragment desc memory\n");
323 		return -ENOMEM;
324 	}
325 	htt->frag_desc.size = size;
326 
327 	return 0;
328 }
329 
330 static void ath10k_htt_tx_free_cont_frag_desc_64(struct ath10k_htt *htt)
331 {
332 	size_t size;
333 
334 	if (!htt->frag_desc.vaddr_desc_64)
335 		return;
336 
337 	size = htt->max_num_pending_tx *
338 			sizeof(struct htt_msdu_ext_desc_64);
339 
340 	dma_free_coherent(htt->ar->dev,
341 			  size,
342 			  htt->frag_desc.vaddr_desc_64,
343 			  htt->frag_desc.paddr);
344 
345 	htt->frag_desc.vaddr_desc_64 = NULL;
346 }
347 
348 static int ath10k_htt_tx_alloc_cont_frag_desc_64(struct ath10k_htt *htt)
349 {
350 	struct ath10k *ar = htt->ar;
351 	size_t size;
352 
353 	if (!ar->hw_params.continuous_frag_desc)
354 		return 0;
355 
356 	size = htt->max_num_pending_tx *
357 			sizeof(struct htt_msdu_ext_desc_64);
358 
359 	htt->frag_desc.vaddr_desc_64 = dma_alloc_coherent(ar->dev, size,
360 							  &htt->frag_desc.paddr,
361 							  GFP_KERNEL);
362 	if (!htt->frag_desc.vaddr_desc_64) {
363 		ath10k_err(ar, "failed to alloc fragment desc memory\n");
364 		return -ENOMEM;
365 	}
366 	htt->frag_desc.size = size;
367 
368 	return 0;
369 }
370 
371 static void ath10k_htt_tx_free_txq(struct ath10k_htt *htt)
372 {
373 	struct ath10k *ar = htt->ar;
374 	size_t size;
375 
376 	if (!test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
377 		      ar->running_fw->fw_file.fw_features))
378 		return;
379 
380 	size = sizeof(*htt->tx_q_state.vaddr);
381 
382 	dma_unmap_single(ar->dev, htt->tx_q_state.paddr, size, DMA_TO_DEVICE);
383 	kfree(htt->tx_q_state.vaddr);
384 }
385 
386 static int ath10k_htt_tx_alloc_txq(struct ath10k_htt *htt)
387 {
388 	struct ath10k *ar = htt->ar;
389 	size_t size;
390 	int ret;
391 
392 	if (!test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
393 		      ar->running_fw->fw_file.fw_features))
394 		return 0;
395 
396 	htt->tx_q_state.num_peers = HTT_TX_Q_STATE_NUM_PEERS;
397 	htt->tx_q_state.num_tids = HTT_TX_Q_STATE_NUM_TIDS;
398 	htt->tx_q_state.type = HTT_Q_DEPTH_TYPE_BYTES;
399 
400 	size = sizeof(*htt->tx_q_state.vaddr);
401 	htt->tx_q_state.vaddr = kzalloc(size, GFP_KERNEL);
402 	if (!htt->tx_q_state.vaddr)
403 		return -ENOMEM;
404 
405 	htt->tx_q_state.paddr = dma_map_single(ar->dev, htt->tx_q_state.vaddr,
406 					       size, DMA_TO_DEVICE);
407 	ret = dma_mapping_error(ar->dev, htt->tx_q_state.paddr);
408 	if (ret) {
409 		ath10k_warn(ar, "failed to dma map tx_q_state: %d\n", ret);
410 		kfree(htt->tx_q_state.vaddr);
411 		return -EIO;
412 	}
413 
414 	return 0;
415 }
416 
417 static void ath10k_htt_tx_free_txdone_fifo(struct ath10k_htt *htt)
418 {
419 	WARN_ON(!kfifo_is_empty(&htt->txdone_fifo));
420 	kfifo_free(&htt->txdone_fifo);
421 }
422 
423 static int ath10k_htt_tx_alloc_txdone_fifo(struct ath10k_htt *htt)
424 {
425 	int ret;
426 	size_t size;
427 
428 	size = roundup_pow_of_two(htt->max_num_pending_tx);
429 	ret = kfifo_alloc(&htt->txdone_fifo, size, GFP_KERNEL);
430 	return ret;
431 }
432 
433 static int ath10k_htt_tx_alloc_buf(struct ath10k_htt *htt)
434 {
435 	struct ath10k *ar = htt->ar;
436 	int ret;
437 
438 	ret = ath10k_htt_alloc_txbuff(htt);
439 	if (ret) {
440 		ath10k_err(ar, "failed to alloc cont tx buffer: %d\n", ret);
441 		return ret;
442 	}
443 
444 	ret = ath10k_htt_alloc_frag_desc(htt);
445 	if (ret) {
446 		ath10k_err(ar, "failed to alloc cont frag desc: %d\n", ret);
447 		goto free_txbuf;
448 	}
449 
450 	ret = ath10k_htt_tx_alloc_txq(htt);
451 	if (ret) {
452 		ath10k_err(ar, "failed to alloc txq: %d\n", ret);
453 		goto free_frag_desc;
454 	}
455 
456 	ret = ath10k_htt_tx_alloc_txdone_fifo(htt);
457 	if (ret) {
458 		ath10k_err(ar, "failed to alloc txdone fifo: %d\n", ret);
459 		goto free_txq;
460 	}
461 
462 	return 0;
463 
464 free_txq:
465 	ath10k_htt_tx_free_txq(htt);
466 
467 free_frag_desc:
468 	ath10k_htt_free_frag_desc(htt);
469 
470 free_txbuf:
471 	ath10k_htt_free_txbuff(htt);
472 
473 	return ret;
474 }
475 
476 int ath10k_htt_tx_start(struct ath10k_htt *htt)
477 {
478 	struct ath10k *ar = htt->ar;
479 	int ret;
480 
481 	ath10k_dbg(ar, ATH10K_DBG_BOOT, "htt tx max num pending tx %d\n",
482 		   htt->max_num_pending_tx);
483 
484 	spin_lock_init(&htt->tx_lock);
485 	idr_init(&htt->pending_tx);
486 
487 	if (htt->tx_mem_allocated)
488 		return 0;
489 
490 	if (ar->bus_param.dev_type == ATH10K_DEV_TYPE_HL)
491 		return 0;
492 
493 	ret = ath10k_htt_tx_alloc_buf(htt);
494 	if (ret)
495 		goto free_idr_pending_tx;
496 
497 	htt->tx_mem_allocated = true;
498 
499 	return 0;
500 
501 free_idr_pending_tx:
502 	idr_destroy(&htt->pending_tx);
503 
504 	return ret;
505 }
506 
507 static int ath10k_htt_tx_clean_up_pending(int msdu_id, void *skb, void *ctx)
508 {
509 	struct ath10k *ar = ctx;
510 	struct ath10k_htt *htt = &ar->htt;
511 	struct htt_tx_done tx_done = {0};
512 
513 	ath10k_dbg(ar, ATH10K_DBG_HTT, "force cleanup msdu_id %u\n", msdu_id);
514 
515 	tx_done.msdu_id = msdu_id;
516 	tx_done.status = HTT_TX_COMPL_STATE_DISCARD;
517 
518 	ath10k_txrx_tx_unref(htt, &tx_done);
519 
520 	return 0;
521 }
522 
523 void ath10k_htt_tx_destroy(struct ath10k_htt *htt)
524 {
525 	if (!htt->tx_mem_allocated)
526 		return;
527 
528 	ath10k_htt_free_txbuff(htt);
529 	ath10k_htt_tx_free_txq(htt);
530 	ath10k_htt_free_frag_desc(htt);
531 	ath10k_htt_tx_free_txdone_fifo(htt);
532 	htt->tx_mem_allocated = false;
533 }
534 
535 static void ath10k_htt_flush_tx_queue(struct ath10k_htt *htt)
536 {
537 	ath10k_htc_stop_hl(htt->ar);
538 	idr_for_each(&htt->pending_tx, ath10k_htt_tx_clean_up_pending, htt->ar);
539 }
540 
541 void ath10k_htt_tx_stop(struct ath10k_htt *htt)
542 {
543 	ath10k_htt_flush_tx_queue(htt);
544 	idr_destroy(&htt->pending_tx);
545 }
546 
547 void ath10k_htt_tx_free(struct ath10k_htt *htt)
548 {
549 	ath10k_htt_tx_stop(htt);
550 	ath10k_htt_tx_destroy(htt);
551 }
552 
553 void ath10k_htt_op_ep_tx_credits(struct ath10k *ar)
554 {
555 	queue_work(ar->workqueue, &ar->bundle_tx_work);
556 }
557 
558 void ath10k_htt_htc_tx_complete(struct ath10k *ar, struct sk_buff *skb)
559 {
560 	struct ath10k_htt *htt = &ar->htt;
561 	struct htt_tx_done tx_done = {0};
562 	struct htt_cmd_hdr *htt_hdr;
563 	struct htt_data_tx_desc *desc_hdr = NULL;
564 	u16 flags1 = 0;
565 	u8 msg_type = 0;
566 
567 	if (htt->disable_tx_comp) {
568 		htt_hdr = (struct htt_cmd_hdr *)skb->data;
569 		msg_type = htt_hdr->msg_type;
570 
571 		if (msg_type == HTT_H2T_MSG_TYPE_TX_FRM) {
572 			desc_hdr = (struct htt_data_tx_desc *)
573 				(skb->data + sizeof(*htt_hdr));
574 			flags1 = __le16_to_cpu(desc_hdr->flags1);
575 			skb_pull(skb, sizeof(struct htt_cmd_hdr));
576 			skb_pull(skb, sizeof(struct htt_data_tx_desc));
577 		}
578 	}
579 
580 	dev_kfree_skb_any(skb);
581 
582 	if ((!htt->disable_tx_comp) || (msg_type != HTT_H2T_MSG_TYPE_TX_FRM))
583 		return;
584 
585 	ath10k_dbg(ar, ATH10K_DBG_HTT,
586 		   "htt tx complete msdu id:%u ,flags1:%x\n",
587 		   __le16_to_cpu(desc_hdr->id), flags1);
588 
589 	if (flags1 & HTT_DATA_TX_DESC_FLAGS1_TX_COMPLETE)
590 		return;
591 
592 	tx_done.status = HTT_TX_COMPL_STATE_ACK;
593 	tx_done.msdu_id = __le16_to_cpu(desc_hdr->id);
594 	ath10k_txrx_tx_unref(&ar->htt, &tx_done);
595 }
596 
597 void ath10k_htt_hif_tx_complete(struct ath10k *ar, struct sk_buff *skb)
598 {
599 	dev_kfree_skb_any(skb);
600 }
601 EXPORT_SYMBOL(ath10k_htt_hif_tx_complete);
602 
603 int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
604 {
605 	struct ath10k *ar = htt->ar;
606 	struct sk_buff *skb;
607 	struct htt_cmd *cmd;
608 	int len = 0;
609 	int ret;
610 
611 	len += sizeof(cmd->hdr);
612 	len += sizeof(cmd->ver_req);
613 
614 	skb = ath10k_htc_alloc_skb(ar, len);
615 	if (!skb)
616 		return -ENOMEM;
617 
618 	skb_put(skb, len);
619 	cmd = (struct htt_cmd *)skb->data;
620 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_VERSION_REQ;
621 
622 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
623 	if (ret) {
624 		dev_kfree_skb_any(skb);
625 		return ret;
626 	}
627 
628 	return 0;
629 }
630 
631 int ath10k_htt_h2t_stats_req(struct ath10k_htt *htt, u32 mask, u32 reset_mask,
632 			     u64 cookie)
633 {
634 	struct ath10k *ar = htt->ar;
635 	struct htt_stats_req *req;
636 	struct sk_buff *skb;
637 	struct htt_cmd *cmd;
638 	int len = 0, ret;
639 
640 	len += sizeof(cmd->hdr);
641 	len += sizeof(cmd->stats_req);
642 
643 	skb = ath10k_htc_alloc_skb(ar, len);
644 	if (!skb)
645 		return -ENOMEM;
646 
647 	skb_put(skb, len);
648 	cmd = (struct htt_cmd *)skb->data;
649 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_STATS_REQ;
650 
651 	req = &cmd->stats_req;
652 
653 	memset(req, 0, sizeof(*req));
654 
655 	/* currently we support only max 24 bit masks so no need to worry
656 	 * about endian support
657 	 */
658 	memcpy(req->upload_types, &mask, 3);
659 	memcpy(req->reset_types, &reset_mask, 3);
660 	req->stat_type = HTT_STATS_REQ_CFG_STAT_TYPE_INVALID;
661 	req->cookie_lsb = cpu_to_le32(cookie & 0xffffffff);
662 	req->cookie_msb = cpu_to_le32((cookie & 0xffffffff00000000ULL) >> 32);
663 
664 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
665 	if (ret) {
666 		ath10k_warn(ar, "failed to send htt type stats request: %d",
667 			    ret);
668 		dev_kfree_skb_any(skb);
669 		return ret;
670 	}
671 
672 	return 0;
673 }
674 
675 static int ath10k_htt_send_frag_desc_bank_cfg_32(struct ath10k_htt *htt)
676 {
677 	struct ath10k *ar = htt->ar;
678 	struct sk_buff *skb;
679 	struct htt_cmd *cmd;
680 	struct htt_frag_desc_bank_cfg32 *cfg;
681 	int ret, size;
682 	u8 info;
683 
684 	if (!ar->hw_params.continuous_frag_desc)
685 		return 0;
686 
687 	if (!htt->frag_desc.paddr) {
688 		ath10k_warn(ar, "invalid frag desc memory\n");
689 		return -EINVAL;
690 	}
691 
692 	size = sizeof(cmd->hdr) + sizeof(cmd->frag_desc_bank_cfg32);
693 	skb = ath10k_htc_alloc_skb(ar, size);
694 	if (!skb)
695 		return -ENOMEM;
696 
697 	skb_put(skb, size);
698 	cmd = (struct htt_cmd *)skb->data;
699 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_FRAG_DESC_BANK_CFG;
700 
701 	info = 0;
702 	info |= SM(htt->tx_q_state.type,
703 		   HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_DEPTH_TYPE);
704 
705 	if (test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
706 		     ar->running_fw->fw_file.fw_features))
707 		info |= HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_VALID;
708 
709 	cfg = &cmd->frag_desc_bank_cfg32;
710 	cfg->info = info;
711 	cfg->num_banks = 1;
712 	cfg->desc_size = sizeof(struct htt_msdu_ext_desc);
713 	cfg->bank_base_addrs[0] = __cpu_to_le32(htt->frag_desc.paddr);
714 	cfg->bank_id[0].bank_min_id = 0;
715 	cfg->bank_id[0].bank_max_id = __cpu_to_le16(htt->max_num_pending_tx -
716 						    1);
717 
718 	cfg->q_state.paddr = cpu_to_le32(htt->tx_q_state.paddr);
719 	cfg->q_state.num_peers = cpu_to_le16(htt->tx_q_state.num_peers);
720 	cfg->q_state.num_tids = cpu_to_le16(htt->tx_q_state.num_tids);
721 	cfg->q_state.record_size = HTT_TX_Q_STATE_ENTRY_SIZE;
722 	cfg->q_state.record_multiplier = HTT_TX_Q_STATE_ENTRY_MULTIPLIER;
723 
724 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt frag desc bank cmd\n");
725 
726 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
727 	if (ret) {
728 		ath10k_warn(ar, "failed to send frag desc bank cfg request: %d\n",
729 			    ret);
730 		dev_kfree_skb_any(skb);
731 		return ret;
732 	}
733 
734 	return 0;
735 }
736 
737 static int ath10k_htt_send_frag_desc_bank_cfg_64(struct ath10k_htt *htt)
738 {
739 	struct ath10k *ar = htt->ar;
740 	struct sk_buff *skb;
741 	struct htt_cmd *cmd;
742 	struct htt_frag_desc_bank_cfg64 *cfg;
743 	int ret, size;
744 	u8 info;
745 
746 	if (!ar->hw_params.continuous_frag_desc)
747 		return 0;
748 
749 	if (!htt->frag_desc.paddr) {
750 		ath10k_warn(ar, "invalid frag desc memory\n");
751 		return -EINVAL;
752 	}
753 
754 	size = sizeof(cmd->hdr) + sizeof(cmd->frag_desc_bank_cfg64);
755 	skb = ath10k_htc_alloc_skb(ar, size);
756 	if (!skb)
757 		return -ENOMEM;
758 
759 	skb_put(skb, size);
760 	cmd = (struct htt_cmd *)skb->data;
761 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_FRAG_DESC_BANK_CFG;
762 
763 	info = 0;
764 	info |= SM(htt->tx_q_state.type,
765 		   HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_DEPTH_TYPE);
766 
767 	if (test_bit(ATH10K_FW_FEATURE_PEER_FLOW_CONTROL,
768 		     ar->running_fw->fw_file.fw_features))
769 		info |= HTT_FRAG_DESC_BANK_CFG_INFO_Q_STATE_VALID;
770 
771 	cfg = &cmd->frag_desc_bank_cfg64;
772 	cfg->info = info;
773 	cfg->num_banks = 1;
774 	cfg->desc_size = sizeof(struct htt_msdu_ext_desc_64);
775 	cfg->bank_base_addrs[0] =  __cpu_to_le64(htt->frag_desc.paddr);
776 	cfg->bank_id[0].bank_min_id = 0;
777 	cfg->bank_id[0].bank_max_id = __cpu_to_le16(htt->max_num_pending_tx -
778 						    1);
779 
780 	cfg->q_state.paddr = cpu_to_le32(htt->tx_q_state.paddr);
781 	cfg->q_state.num_peers = cpu_to_le16(htt->tx_q_state.num_peers);
782 	cfg->q_state.num_tids = cpu_to_le16(htt->tx_q_state.num_tids);
783 	cfg->q_state.record_size = HTT_TX_Q_STATE_ENTRY_SIZE;
784 	cfg->q_state.record_multiplier = HTT_TX_Q_STATE_ENTRY_MULTIPLIER;
785 
786 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt frag desc bank cmd\n");
787 
788 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
789 	if (ret) {
790 		ath10k_warn(ar, "failed to send frag desc bank cfg request: %d\n",
791 			    ret);
792 		dev_kfree_skb_any(skb);
793 		return ret;
794 	}
795 
796 	return 0;
797 }
798 
799 static void ath10k_htt_fill_rx_desc_offset_32(struct ath10k_hw_params *hw, void *rx_ring)
800 {
801 	struct htt_rx_ring_setup_ring32 *ring =
802 			(struct htt_rx_ring_setup_ring32 *)rx_ring;
803 
804 	ath10k_htt_rx_desc_get_offsets(hw, &ring->offsets);
805 }
806 
807 static void ath10k_htt_fill_rx_desc_offset_64(struct ath10k_hw_params *hw, void *rx_ring)
808 {
809 	struct htt_rx_ring_setup_ring64 *ring =
810 			(struct htt_rx_ring_setup_ring64 *)rx_ring;
811 
812 	ath10k_htt_rx_desc_get_offsets(hw, &ring->offsets);
813 }
814 
815 static int ath10k_htt_send_rx_ring_cfg_32(struct ath10k_htt *htt)
816 {
817 	struct ath10k *ar = htt->ar;
818 	struct ath10k_hw_params *hw = &ar->hw_params;
819 	struct sk_buff *skb;
820 	struct htt_cmd *cmd;
821 	struct htt_rx_ring_setup_ring32 *ring;
822 	const int num_rx_ring = 1;
823 	u16 flags;
824 	u32 fw_idx;
825 	int len;
826 	int ret;
827 
828 	/*
829 	 * the HW expects the buffer to be an integral number of 4-byte
830 	 * "words"
831 	 */
832 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
833 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
834 
835 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup_32.hdr)
836 	    + (sizeof(*ring) * num_rx_ring);
837 	skb = ath10k_htc_alloc_skb(ar, len);
838 	if (!skb)
839 		return -ENOMEM;
840 
841 	skb_put(skb, len);
842 
843 	cmd = (struct htt_cmd *)skb->data;
844 	ring = &cmd->rx_setup_32.rings[0];
845 
846 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
847 	cmd->rx_setup_32.hdr.num_rings = 1;
848 
849 	/* FIXME: do we need all of this? */
850 	flags = 0;
851 	flags |= HTT_RX_RING_FLAGS_MAC80211_HDR;
852 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
853 	flags |= HTT_RX_RING_FLAGS_PPDU_START;
854 	flags |= HTT_RX_RING_FLAGS_PPDU_END;
855 	flags |= HTT_RX_RING_FLAGS_MPDU_START;
856 	flags |= HTT_RX_RING_FLAGS_MPDU_END;
857 	flags |= HTT_RX_RING_FLAGS_MSDU_START;
858 	flags |= HTT_RX_RING_FLAGS_MSDU_END;
859 	flags |= HTT_RX_RING_FLAGS_RX_ATTENTION;
860 	flags |= HTT_RX_RING_FLAGS_FRAG_INFO;
861 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
862 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
863 	flags |= HTT_RX_RING_FLAGS_CTRL_RX;
864 	flags |= HTT_RX_RING_FLAGS_MGMT_RX;
865 	flags |= HTT_RX_RING_FLAGS_NULL_RX;
866 	flags |= HTT_RX_RING_FLAGS_PHY_DATA_RX;
867 
868 	fw_idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
869 
870 	ring->fw_idx_shadow_reg_paddr =
871 		__cpu_to_le32(htt->rx_ring.alloc_idx.paddr);
872 	ring->rx_ring_base_paddr = __cpu_to_le32(htt->rx_ring.base_paddr);
873 	ring->rx_ring_len = __cpu_to_le16(htt->rx_ring.size);
874 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
875 	ring->flags = __cpu_to_le16(flags);
876 	ring->fw_idx_init_val = __cpu_to_le16(fw_idx);
877 
878 	ath10k_htt_fill_rx_desc_offset_32(hw, ring);
879 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
880 	if (ret) {
881 		dev_kfree_skb_any(skb);
882 		return ret;
883 	}
884 
885 	return 0;
886 }
887 
888 static int ath10k_htt_send_rx_ring_cfg_64(struct ath10k_htt *htt)
889 {
890 	struct ath10k *ar = htt->ar;
891 	struct ath10k_hw_params *hw = &ar->hw_params;
892 	struct sk_buff *skb;
893 	struct htt_cmd *cmd;
894 	struct htt_rx_ring_setup_ring64 *ring;
895 	const int num_rx_ring = 1;
896 	u16 flags;
897 	u32 fw_idx;
898 	int len;
899 	int ret;
900 
901 	/* HW expects the buffer to be an integral number of 4-byte
902 	 * "words"
903 	 */
904 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
905 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
906 
907 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup_64.hdr)
908 	    + (sizeof(*ring) * num_rx_ring);
909 	skb = ath10k_htc_alloc_skb(ar, len);
910 	if (!skb)
911 		return -ENOMEM;
912 
913 	skb_put(skb, len);
914 
915 	cmd = (struct htt_cmd *)skb->data;
916 	ring = &cmd->rx_setup_64.rings[0];
917 
918 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
919 	cmd->rx_setup_64.hdr.num_rings = 1;
920 
921 	flags = 0;
922 	flags |= HTT_RX_RING_FLAGS_MAC80211_HDR;
923 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
924 	flags |= HTT_RX_RING_FLAGS_PPDU_START;
925 	flags |= HTT_RX_RING_FLAGS_PPDU_END;
926 	flags |= HTT_RX_RING_FLAGS_MPDU_START;
927 	flags |= HTT_RX_RING_FLAGS_MPDU_END;
928 	flags |= HTT_RX_RING_FLAGS_MSDU_START;
929 	flags |= HTT_RX_RING_FLAGS_MSDU_END;
930 	flags |= HTT_RX_RING_FLAGS_RX_ATTENTION;
931 	flags |= HTT_RX_RING_FLAGS_FRAG_INFO;
932 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
933 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
934 	flags |= HTT_RX_RING_FLAGS_CTRL_RX;
935 	flags |= HTT_RX_RING_FLAGS_MGMT_RX;
936 	flags |= HTT_RX_RING_FLAGS_NULL_RX;
937 	flags |= HTT_RX_RING_FLAGS_PHY_DATA_RX;
938 
939 	fw_idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
940 
941 	ring->fw_idx_shadow_reg_paddr = __cpu_to_le64(htt->rx_ring.alloc_idx.paddr);
942 	ring->rx_ring_base_paddr = __cpu_to_le64(htt->rx_ring.base_paddr);
943 	ring->rx_ring_len = __cpu_to_le16(htt->rx_ring.size);
944 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
945 	ring->flags = __cpu_to_le16(flags);
946 	ring->fw_idx_init_val = __cpu_to_le16(fw_idx);
947 
948 	ath10k_htt_fill_rx_desc_offset_64(hw, ring);
949 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
950 	if (ret) {
951 		dev_kfree_skb_any(skb);
952 		return ret;
953 	}
954 
955 	return 0;
956 }
957 
958 static int ath10k_htt_send_rx_ring_cfg_hl(struct ath10k_htt *htt)
959 {
960 	struct ath10k *ar = htt->ar;
961 	struct sk_buff *skb;
962 	struct htt_cmd *cmd;
963 	struct htt_rx_ring_setup_ring32 *ring;
964 	const int num_rx_ring = 1;
965 	u16 flags;
966 	int len;
967 	int ret;
968 
969 	/*
970 	 * the HW expects the buffer to be an integral number of 4-byte
971 	 * "words"
972 	 */
973 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
974 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
975 
976 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup_32.hdr)
977 	    + (sizeof(*ring) * num_rx_ring);
978 	skb = ath10k_htc_alloc_skb(ar, len);
979 	if (!skb)
980 		return -ENOMEM;
981 
982 	skb_put(skb, len);
983 
984 	cmd = (struct htt_cmd *)skb->data;
985 	ring = &cmd->rx_setup_32.rings[0];
986 
987 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
988 	cmd->rx_setup_32.hdr.num_rings = 1;
989 
990 	flags = 0;
991 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
992 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
993 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
994 
995 	memset(ring, 0, sizeof(*ring));
996 	ring->rx_ring_len = __cpu_to_le16(HTT_RX_RING_SIZE_MIN);
997 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
998 	ring->flags = __cpu_to_le16(flags);
999 
1000 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
1001 	if (ret) {
1002 		dev_kfree_skb_any(skb);
1003 		return ret;
1004 	}
1005 
1006 	return 0;
1007 }
1008 
1009 static int ath10k_htt_h2t_aggr_cfg_msg_32(struct ath10k_htt *htt,
1010 					  u8 max_subfrms_ampdu,
1011 					  u8 max_subfrms_amsdu)
1012 {
1013 	struct ath10k *ar = htt->ar;
1014 	struct htt_aggr_conf *aggr_conf;
1015 	struct sk_buff *skb;
1016 	struct htt_cmd *cmd;
1017 	int len;
1018 	int ret;
1019 
1020 	/* Firmware defaults are: amsdu = 3 and ampdu = 64 */
1021 
1022 	if (max_subfrms_ampdu == 0 || max_subfrms_ampdu > 64)
1023 		return -EINVAL;
1024 
1025 	if (max_subfrms_amsdu == 0 || max_subfrms_amsdu > 31)
1026 		return -EINVAL;
1027 
1028 	len = sizeof(cmd->hdr);
1029 	len += sizeof(cmd->aggr_conf);
1030 
1031 	skb = ath10k_htc_alloc_skb(ar, len);
1032 	if (!skb)
1033 		return -ENOMEM;
1034 
1035 	skb_put(skb, len);
1036 	cmd = (struct htt_cmd *)skb->data;
1037 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_AGGR_CFG;
1038 
1039 	aggr_conf = &cmd->aggr_conf;
1040 	aggr_conf->max_num_ampdu_subframes = max_subfrms_ampdu;
1041 	aggr_conf->max_num_amsdu_subframes = max_subfrms_amsdu;
1042 
1043 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt h2t aggr cfg msg amsdu %d ampdu %d",
1044 		   aggr_conf->max_num_amsdu_subframes,
1045 		   aggr_conf->max_num_ampdu_subframes);
1046 
1047 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
1048 	if (ret) {
1049 		dev_kfree_skb_any(skb);
1050 		return ret;
1051 	}
1052 
1053 	return 0;
1054 }
1055 
1056 static int ath10k_htt_h2t_aggr_cfg_msg_v2(struct ath10k_htt *htt,
1057 					  u8 max_subfrms_ampdu,
1058 					  u8 max_subfrms_amsdu)
1059 {
1060 	struct ath10k *ar = htt->ar;
1061 	struct htt_aggr_conf_v2 *aggr_conf;
1062 	struct sk_buff *skb;
1063 	struct htt_cmd *cmd;
1064 	int len;
1065 	int ret;
1066 
1067 	/* Firmware defaults are: amsdu = 3 and ampdu = 64 */
1068 
1069 	if (max_subfrms_ampdu == 0 || max_subfrms_ampdu > 64)
1070 		return -EINVAL;
1071 
1072 	if (max_subfrms_amsdu == 0 || max_subfrms_amsdu > 31)
1073 		return -EINVAL;
1074 
1075 	len = sizeof(cmd->hdr);
1076 	len += sizeof(cmd->aggr_conf_v2);
1077 
1078 	skb = ath10k_htc_alloc_skb(ar, len);
1079 	if (!skb)
1080 		return -ENOMEM;
1081 
1082 	skb_put(skb, len);
1083 	cmd = (struct htt_cmd *)skb->data;
1084 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_AGGR_CFG;
1085 
1086 	aggr_conf = &cmd->aggr_conf_v2;
1087 	aggr_conf->max_num_ampdu_subframes = max_subfrms_ampdu;
1088 	aggr_conf->max_num_amsdu_subframes = max_subfrms_amsdu;
1089 
1090 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt h2t aggr cfg msg amsdu %d ampdu %d",
1091 		   aggr_conf->max_num_amsdu_subframes,
1092 		   aggr_conf->max_num_ampdu_subframes);
1093 
1094 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
1095 	if (ret) {
1096 		dev_kfree_skb_any(skb);
1097 		return ret;
1098 	}
1099 
1100 	return 0;
1101 }
1102 
1103 int ath10k_htt_tx_fetch_resp(struct ath10k *ar,
1104 			     __le32 token,
1105 			     __le16 fetch_seq_num,
1106 			     struct htt_tx_fetch_record *records,
1107 			     size_t num_records)
1108 {
1109 	struct sk_buff *skb;
1110 	struct htt_cmd *cmd;
1111 	const u16 resp_id = 0;
1112 	int len = 0;
1113 	int ret;
1114 
1115 	/* Response IDs are echo-ed back only for host driver convienence
1116 	 * purposes. They aren't used for anything in the driver yet so use 0.
1117 	 */
1118 
1119 	len += sizeof(cmd->hdr);
1120 	len += sizeof(cmd->tx_fetch_resp);
1121 	len += sizeof(cmd->tx_fetch_resp.records[0]) * num_records;
1122 
1123 	skb = ath10k_htc_alloc_skb(ar, len);
1124 	if (!skb)
1125 		return -ENOMEM;
1126 
1127 	skb_put(skb, len);
1128 	cmd = (struct htt_cmd *)skb->data;
1129 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FETCH_RESP;
1130 	cmd->tx_fetch_resp.resp_id = cpu_to_le16(resp_id);
1131 	cmd->tx_fetch_resp.fetch_seq_num = fetch_seq_num;
1132 	cmd->tx_fetch_resp.num_records = cpu_to_le16(num_records);
1133 	cmd->tx_fetch_resp.token = token;
1134 
1135 	memcpy(cmd->tx_fetch_resp.records, records,
1136 	       sizeof(records[0]) * num_records);
1137 
1138 	ret = ath10k_htc_send(&ar->htc, ar->htt.eid, skb);
1139 	if (ret) {
1140 		ath10k_warn(ar, "failed to submit htc command: %d\n", ret);
1141 		goto err_free_skb;
1142 	}
1143 
1144 	return 0;
1145 
1146 err_free_skb:
1147 	dev_kfree_skb_any(skb);
1148 
1149 	return ret;
1150 }
1151 
1152 static u8 ath10k_htt_tx_get_vdev_id(struct ath10k *ar, struct sk_buff *skb)
1153 {
1154 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
1155 	struct ath10k_skb_cb *cb = ATH10K_SKB_CB(skb);
1156 	struct ath10k_vif *arvif;
1157 
1158 	if (info->flags & IEEE80211_TX_CTL_TX_OFFCHAN) {
1159 		return ar->scan.vdev_id;
1160 	} else if (cb->vif) {
1161 		arvif = (void *)cb->vif->drv_priv;
1162 		return arvif->vdev_id;
1163 	} else if (ar->monitor_started) {
1164 		return ar->monitor_vdev_id;
1165 	} else {
1166 		return 0;
1167 	}
1168 }
1169 
1170 static u8 ath10k_htt_tx_get_tid(struct sk_buff *skb, bool is_eth)
1171 {
1172 	struct ieee80211_hdr *hdr = (void *)skb->data;
1173 	struct ath10k_skb_cb *cb = ATH10K_SKB_CB(skb);
1174 
1175 	if (!is_eth && ieee80211_is_mgmt(hdr->frame_control))
1176 		return HTT_DATA_TX_EXT_TID_MGMT;
1177 	else if (cb->flags & ATH10K_SKB_F_QOS)
1178 		return skb->priority & IEEE80211_QOS_CTL_TID_MASK;
1179 	else
1180 		return HTT_DATA_TX_EXT_TID_NON_QOS_MCAST_BCAST;
1181 }
1182 
1183 int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
1184 {
1185 	struct ath10k *ar = htt->ar;
1186 	struct device *dev = ar->dev;
1187 	struct sk_buff *txdesc = NULL;
1188 	struct htt_cmd *cmd;
1189 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1190 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1191 	int len = 0;
1192 	int msdu_id = -1;
1193 	int res;
1194 	const u8 *peer_addr;
1195 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1196 
1197 	len += sizeof(cmd->hdr);
1198 	len += sizeof(cmd->mgmt_tx);
1199 
1200 	res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1201 	if (res < 0)
1202 		goto err;
1203 
1204 	msdu_id = res;
1205 
1206 	if ((ieee80211_is_action(hdr->frame_control) ||
1207 	     ieee80211_is_deauth(hdr->frame_control) ||
1208 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1209 	     ieee80211_has_protected(hdr->frame_control)) {
1210 		peer_addr = hdr->addr1;
1211 		if (is_multicast_ether_addr(peer_addr)) {
1212 			skb_put(msdu, sizeof(struct ieee80211_mmie_16));
1213 		} else {
1214 			if (skb_cb->ucast_cipher == WLAN_CIPHER_SUITE_GCMP ||
1215 			    skb_cb->ucast_cipher == WLAN_CIPHER_SUITE_GCMP_256)
1216 				skb_put(msdu, IEEE80211_GCMP_MIC_LEN);
1217 			else
1218 				skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1219 		}
1220 	}
1221 
1222 	txdesc = ath10k_htc_alloc_skb(ar, len);
1223 	if (!txdesc) {
1224 		res = -ENOMEM;
1225 		goto err_free_msdu_id;
1226 	}
1227 
1228 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
1229 				       DMA_TO_DEVICE);
1230 	res = dma_mapping_error(dev, skb_cb->paddr);
1231 	if (res) {
1232 		res = -EIO;
1233 		goto err_free_txdesc;
1234 	}
1235 
1236 	skb_put(txdesc, len);
1237 	cmd = (struct htt_cmd *)txdesc->data;
1238 	memset(cmd, 0, len);
1239 
1240 	cmd->hdr.msg_type         = HTT_H2T_MSG_TYPE_MGMT_TX;
1241 	cmd->mgmt_tx.msdu_paddr = __cpu_to_le32(ATH10K_SKB_CB(msdu)->paddr);
1242 	cmd->mgmt_tx.len        = __cpu_to_le32(msdu->len);
1243 	cmd->mgmt_tx.desc_id    = __cpu_to_le32(msdu_id);
1244 	cmd->mgmt_tx.vdev_id    = __cpu_to_le32(vdev_id);
1245 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
1246 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
1247 
1248 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
1249 	if (res)
1250 		goto err_unmap_msdu;
1251 
1252 	return 0;
1253 
1254 err_unmap_msdu:
1255 	if (ar->bus_param.dev_type != ATH10K_DEV_TYPE_HL)
1256 		dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
1257 err_free_txdesc:
1258 	dev_kfree_skb_any(txdesc);
1259 err_free_msdu_id:
1260 	spin_lock_bh(&htt->tx_lock);
1261 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
1262 	spin_unlock_bh(&htt->tx_lock);
1263 err:
1264 	return res;
1265 }
1266 
1267 #define HTT_TX_HL_NEEDED_HEADROOM \
1268 	(unsigned int)(sizeof(struct htt_cmd_hdr) + \
1269 	sizeof(struct htt_data_tx_desc) + \
1270 	sizeof(struct ath10k_htc_hdr))
1271 
1272 static int ath10k_htt_tx_hl(struct ath10k_htt *htt, enum ath10k_hw_txrx_mode txmode,
1273 			    struct sk_buff *msdu)
1274 {
1275 	struct ath10k *ar = htt->ar;
1276 	int res, data_len;
1277 	struct htt_cmd_hdr *cmd_hdr;
1278 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1279 	struct htt_data_tx_desc *tx_desc;
1280 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1281 	struct sk_buff *tmp_skb;
1282 	bool is_eth = (txmode == ATH10K_HW_TXRX_ETHERNET);
1283 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1284 	u8 tid = ath10k_htt_tx_get_tid(msdu, is_eth);
1285 	u8 flags0 = 0;
1286 	u16 flags1 = 0;
1287 	u16 msdu_id = 0;
1288 
1289 	if ((ieee80211_is_action(hdr->frame_control) ||
1290 	     ieee80211_is_deauth(hdr->frame_control) ||
1291 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1292 	     ieee80211_has_protected(hdr->frame_control)) {
1293 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1294 	}
1295 
1296 	data_len = msdu->len;
1297 
1298 	switch (txmode) {
1299 	case ATH10K_HW_TXRX_RAW:
1300 	case ATH10K_HW_TXRX_NATIVE_WIFI:
1301 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1302 		fallthrough;
1303 	case ATH10K_HW_TXRX_ETHERNET:
1304 		flags0 |= SM(txmode, HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1305 		break;
1306 	case ATH10K_HW_TXRX_MGMT:
1307 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
1308 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1309 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1310 
1311 		if (htt->disable_tx_comp)
1312 			flags1 |= HTT_DATA_TX_DESC_FLAGS1_TX_COMPLETE;
1313 		break;
1314 	}
1315 
1316 	if (skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT)
1317 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
1318 
1319 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
1320 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
1321 	if (msdu->ip_summed == CHECKSUM_PARTIAL &&
1322 	    !test_bit(ATH10K_FLAG_RAW_MODE, &ar->dev_flags)) {
1323 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
1324 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
1325 	}
1326 
1327 	/* Prepend the HTT header and TX desc struct to the data message
1328 	 * and realloc the skb if it does not have enough headroom.
1329 	 */
1330 	if (skb_headroom(msdu) < HTT_TX_HL_NEEDED_HEADROOM) {
1331 		tmp_skb = msdu;
1332 
1333 		ath10k_dbg(htt->ar, ATH10K_DBG_HTT,
1334 			   "Not enough headroom in skb. Current headroom: %u, needed: %u. Reallocating...\n",
1335 			   skb_headroom(msdu), HTT_TX_HL_NEEDED_HEADROOM);
1336 		msdu = skb_realloc_headroom(msdu, HTT_TX_HL_NEEDED_HEADROOM);
1337 		kfree_skb(tmp_skb);
1338 		if (!msdu) {
1339 			ath10k_warn(htt->ar, "htt hl tx: Unable to realloc skb!\n");
1340 			res = -ENOMEM;
1341 			goto out;
1342 		}
1343 	}
1344 
1345 	if (ar->bus_param.hl_msdu_ids) {
1346 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_POSTPONED;
1347 		res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1348 		if (res < 0) {
1349 			ath10k_err(ar, "msdu_id allocation failed %d\n", res);
1350 			goto out;
1351 		}
1352 		msdu_id = res;
1353 	}
1354 
1355 	/* As msdu is freed by mac80211 (in ieee80211_tx_status()) and by
1356 	 * ath10k (in ath10k_htt_htc_tx_complete()) we have to increase
1357 	 * reference by one to avoid a use-after-free case and a double
1358 	 * free.
1359 	 */
1360 	skb_get(msdu);
1361 
1362 	skb_push(msdu, sizeof(*cmd_hdr));
1363 	skb_push(msdu, sizeof(*tx_desc));
1364 	cmd_hdr = (struct htt_cmd_hdr *)msdu->data;
1365 	tx_desc = (struct htt_data_tx_desc *)(msdu->data + sizeof(*cmd_hdr));
1366 
1367 	cmd_hdr->msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
1368 	tx_desc->flags0 = flags0;
1369 	tx_desc->flags1 = __cpu_to_le16(flags1);
1370 	tx_desc->len = __cpu_to_le16(data_len);
1371 	tx_desc->id = __cpu_to_le16(msdu_id);
1372 	tx_desc->frags_paddr = 0; /* always zero */
1373 	/* Initialize peer_id to INVALID_PEER because this is NOT
1374 	 * Reinjection path
1375 	 */
1376 	tx_desc->peerid = __cpu_to_le32(HTT_INVALID_PEERID);
1377 
1378 	res = ath10k_htc_send_hl(&htt->ar->htc, htt->eid, msdu);
1379 
1380 out:
1381 	return res;
1382 }
1383 
1384 static int ath10k_htt_tx_32(struct ath10k_htt *htt,
1385 			    enum ath10k_hw_txrx_mode txmode,
1386 			    struct sk_buff *msdu)
1387 {
1388 	struct ath10k *ar = htt->ar;
1389 	struct device *dev = ar->dev;
1390 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1391 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(msdu);
1392 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1393 	struct ath10k_hif_sg_item sg_items[2];
1394 	struct ath10k_htt_txbuf_32 *txbuf;
1395 	struct htt_data_tx_desc_frag *frags;
1396 	bool is_eth = (txmode == ATH10K_HW_TXRX_ETHERNET);
1397 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1398 	u8 tid = ath10k_htt_tx_get_tid(msdu, is_eth);
1399 	int prefetch_len;
1400 	int res;
1401 	u8 flags0 = 0;
1402 	u16 msdu_id, flags1 = 0;
1403 	u16 freq = 0;
1404 	u32 frags_paddr = 0;
1405 	u32 txbuf_paddr;
1406 	struct htt_msdu_ext_desc *ext_desc = NULL;
1407 	struct htt_msdu_ext_desc *ext_desc_t = NULL;
1408 
1409 	res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1410 	if (res < 0)
1411 		goto err;
1412 
1413 	msdu_id = res;
1414 
1415 	prefetch_len = min(htt->prefetch_len, msdu->len);
1416 	prefetch_len = roundup(prefetch_len, 4);
1417 
1418 	txbuf = htt->txbuf.vaddr_txbuff_32 + msdu_id;
1419 	txbuf_paddr = htt->txbuf.paddr +
1420 		      (sizeof(struct ath10k_htt_txbuf_32) * msdu_id);
1421 
1422 	if ((ieee80211_is_action(hdr->frame_control) ||
1423 	     ieee80211_is_deauth(hdr->frame_control) ||
1424 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1425 	     ieee80211_has_protected(hdr->frame_control)) {
1426 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1427 	} else if (!(skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT) &&
1428 		   txmode == ATH10K_HW_TXRX_RAW &&
1429 		   ieee80211_has_protected(hdr->frame_control)) {
1430 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1431 	}
1432 
1433 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
1434 				       DMA_TO_DEVICE);
1435 	res = dma_mapping_error(dev, skb_cb->paddr);
1436 	if (res) {
1437 		res = -EIO;
1438 		goto err_free_msdu_id;
1439 	}
1440 
1441 	if (unlikely(info->flags & IEEE80211_TX_CTL_TX_OFFCHAN))
1442 		freq = ar->scan.roc_freq;
1443 
1444 	switch (txmode) {
1445 	case ATH10K_HW_TXRX_RAW:
1446 	case ATH10K_HW_TXRX_NATIVE_WIFI:
1447 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1448 		fallthrough;
1449 	case ATH10K_HW_TXRX_ETHERNET:
1450 		if (ar->hw_params.continuous_frag_desc) {
1451 			ext_desc_t = htt->frag_desc.vaddr_desc_32;
1452 			memset(&ext_desc_t[msdu_id], 0,
1453 			       sizeof(struct htt_msdu_ext_desc));
1454 			frags = (struct htt_data_tx_desc_frag *)
1455 				&ext_desc_t[msdu_id].frags;
1456 			ext_desc = &ext_desc_t[msdu_id];
1457 			frags[0].tword_addr.paddr_lo =
1458 				__cpu_to_le32(skb_cb->paddr);
1459 			frags[0].tword_addr.paddr_hi = 0;
1460 			frags[0].tword_addr.len_16 = __cpu_to_le16(msdu->len);
1461 
1462 			frags_paddr =  htt->frag_desc.paddr +
1463 				(sizeof(struct htt_msdu_ext_desc) * msdu_id);
1464 		} else {
1465 			frags = txbuf->frags;
1466 			frags[0].dword_addr.paddr =
1467 				__cpu_to_le32(skb_cb->paddr);
1468 			frags[0].dword_addr.len = __cpu_to_le32(msdu->len);
1469 			frags[1].dword_addr.paddr = 0;
1470 			frags[1].dword_addr.len = 0;
1471 
1472 			frags_paddr = txbuf_paddr;
1473 		}
1474 		flags0 |= SM(txmode, HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1475 		break;
1476 	case ATH10K_HW_TXRX_MGMT:
1477 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
1478 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1479 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1480 
1481 		frags_paddr = skb_cb->paddr;
1482 		break;
1483 	}
1484 
1485 	/* Normally all commands go through HTC which manages tx credits for
1486 	 * each endpoint and notifies when tx is completed.
1487 	 *
1488 	 * HTT endpoint is creditless so there's no need to care about HTC
1489 	 * flags. In that case it is trivial to fill the HTC header here.
1490 	 *
1491 	 * MSDU transmission is considered completed upon HTT event. This
1492 	 * implies no relevant resources can be freed until after the event is
1493 	 * received. That's why HTC tx completion handler itself is ignored by
1494 	 * setting NULL to transfer_context for all sg items.
1495 	 *
1496 	 * There is simply no point in pushing HTT TX_FRM through HTC tx path
1497 	 * as it's a waste of resources. By bypassing HTC it is possible to
1498 	 * avoid extra memory allocations, compress data structures and thus
1499 	 * improve performance.
1500 	 */
1501 
1502 	txbuf->htc_hdr.eid = htt->eid;
1503 	txbuf->htc_hdr.len = __cpu_to_le16(sizeof(txbuf->cmd_hdr) +
1504 					   sizeof(txbuf->cmd_tx) +
1505 					   prefetch_len);
1506 	txbuf->htc_hdr.flags = 0;
1507 
1508 	if (skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT)
1509 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
1510 
1511 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
1512 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
1513 	if (msdu->ip_summed == CHECKSUM_PARTIAL &&
1514 	    !test_bit(ATH10K_FLAG_RAW_MODE, &ar->dev_flags)) {
1515 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
1516 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
1517 		if (ar->hw_params.continuous_frag_desc)
1518 			ext_desc->flags |= HTT_MSDU_CHECKSUM_ENABLE;
1519 	}
1520 
1521 	/* Prevent firmware from sending up tx inspection requests. There's
1522 	 * nothing ath10k can do with frames requested for inspection so force
1523 	 * it to simply rely a regular tx completion with discard status.
1524 	 */
1525 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_POSTPONED;
1526 
1527 	txbuf->cmd_hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
1528 	txbuf->cmd_tx.flags0 = flags0;
1529 	txbuf->cmd_tx.flags1 = __cpu_to_le16(flags1);
1530 	txbuf->cmd_tx.len = __cpu_to_le16(msdu->len);
1531 	txbuf->cmd_tx.id = __cpu_to_le16(msdu_id);
1532 	txbuf->cmd_tx.frags_paddr = __cpu_to_le32(frags_paddr);
1533 	if (ath10k_mac_tx_frm_has_freq(ar)) {
1534 		txbuf->cmd_tx.offchan_tx.peerid =
1535 				__cpu_to_le16(HTT_INVALID_PEERID);
1536 		txbuf->cmd_tx.offchan_tx.freq =
1537 				__cpu_to_le16(freq);
1538 	} else {
1539 		txbuf->cmd_tx.peerid =
1540 				__cpu_to_le32(HTT_INVALID_PEERID);
1541 	}
1542 
1543 	trace_ath10k_htt_tx(ar, msdu_id, msdu->len, vdev_id, tid);
1544 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1545 		   "htt tx flags0 %u flags1 %u len %d id %u frags_paddr %pad, msdu_paddr %pad vdev %u tid %u freq %u\n",
1546 		   flags0, flags1, msdu->len, msdu_id, &frags_paddr,
1547 		   &skb_cb->paddr, vdev_id, tid, freq);
1548 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt tx msdu: ",
1549 			msdu->data, msdu->len);
1550 	trace_ath10k_tx_hdr(ar, msdu->data, msdu->len);
1551 	trace_ath10k_tx_payload(ar, msdu->data, msdu->len);
1552 
1553 	sg_items[0].transfer_id = 0;
1554 	sg_items[0].transfer_context = NULL;
1555 	sg_items[0].vaddr = &txbuf->htc_hdr;
1556 	sg_items[0].paddr = txbuf_paddr +
1557 			    sizeof(txbuf->frags);
1558 	sg_items[0].len = sizeof(txbuf->htc_hdr) +
1559 			  sizeof(txbuf->cmd_hdr) +
1560 			  sizeof(txbuf->cmd_tx);
1561 
1562 	sg_items[1].transfer_id = 0;
1563 	sg_items[1].transfer_context = NULL;
1564 	sg_items[1].vaddr = msdu->data;
1565 	sg_items[1].paddr = skb_cb->paddr;
1566 	sg_items[1].len = prefetch_len;
1567 
1568 	res = ath10k_hif_tx_sg(htt->ar,
1569 			       htt->ar->htc.endpoint[htt->eid].ul_pipe_id,
1570 			       sg_items, ARRAY_SIZE(sg_items));
1571 	if (res)
1572 		goto err_unmap_msdu;
1573 
1574 	return 0;
1575 
1576 err_unmap_msdu:
1577 	dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
1578 err_free_msdu_id:
1579 	spin_lock_bh(&htt->tx_lock);
1580 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
1581 	spin_unlock_bh(&htt->tx_lock);
1582 err:
1583 	return res;
1584 }
1585 
1586 static int ath10k_htt_tx_64(struct ath10k_htt *htt,
1587 			    enum ath10k_hw_txrx_mode txmode,
1588 			    struct sk_buff *msdu)
1589 {
1590 	struct ath10k *ar = htt->ar;
1591 	struct device *dev = ar->dev;
1592 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
1593 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(msdu);
1594 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
1595 	struct ath10k_hif_sg_item sg_items[2];
1596 	struct ath10k_htt_txbuf_64 *txbuf;
1597 	struct htt_data_tx_desc_frag *frags;
1598 	bool is_eth = (txmode == ATH10K_HW_TXRX_ETHERNET);
1599 	u8 vdev_id = ath10k_htt_tx_get_vdev_id(ar, msdu);
1600 	u8 tid = ath10k_htt_tx_get_tid(msdu, is_eth);
1601 	int prefetch_len;
1602 	int res;
1603 	u8 flags0 = 0;
1604 	u16 msdu_id, flags1 = 0;
1605 	u16 freq = 0;
1606 	dma_addr_t frags_paddr = 0;
1607 	dma_addr_t txbuf_paddr;
1608 	struct htt_msdu_ext_desc_64 *ext_desc = NULL;
1609 	struct htt_msdu_ext_desc_64 *ext_desc_t = NULL;
1610 
1611 	res = ath10k_htt_tx_alloc_msdu_id(htt, msdu);
1612 	if (res < 0)
1613 		goto err;
1614 
1615 	msdu_id = res;
1616 
1617 	prefetch_len = min(htt->prefetch_len, msdu->len);
1618 	prefetch_len = roundup(prefetch_len, 4);
1619 
1620 	txbuf = htt->txbuf.vaddr_txbuff_64 + msdu_id;
1621 	txbuf_paddr = htt->txbuf.paddr +
1622 		      (sizeof(struct ath10k_htt_txbuf_64) * msdu_id);
1623 
1624 	if ((ieee80211_is_action(hdr->frame_control) ||
1625 	     ieee80211_is_deauth(hdr->frame_control) ||
1626 	     ieee80211_is_disassoc(hdr->frame_control)) &&
1627 	     ieee80211_has_protected(hdr->frame_control)) {
1628 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1629 	} else if (!(skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT) &&
1630 		   txmode == ATH10K_HW_TXRX_RAW &&
1631 		   ieee80211_has_protected(hdr->frame_control)) {
1632 		skb_put(msdu, IEEE80211_CCMP_MIC_LEN);
1633 	}
1634 
1635 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
1636 				       DMA_TO_DEVICE);
1637 	res = dma_mapping_error(dev, skb_cb->paddr);
1638 	if (res) {
1639 		res = -EIO;
1640 		goto err_free_msdu_id;
1641 	}
1642 
1643 	if (unlikely(info->flags & IEEE80211_TX_CTL_TX_OFFCHAN))
1644 		freq = ar->scan.roc_freq;
1645 
1646 	switch (txmode) {
1647 	case ATH10K_HW_TXRX_RAW:
1648 	case ATH10K_HW_TXRX_NATIVE_WIFI:
1649 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1650 		fallthrough;
1651 	case ATH10K_HW_TXRX_ETHERNET:
1652 		if (ar->hw_params.continuous_frag_desc) {
1653 			ext_desc_t = htt->frag_desc.vaddr_desc_64;
1654 			memset(&ext_desc_t[msdu_id], 0,
1655 			       sizeof(struct htt_msdu_ext_desc_64));
1656 			frags = (struct htt_data_tx_desc_frag *)
1657 				&ext_desc_t[msdu_id].frags;
1658 			ext_desc = &ext_desc_t[msdu_id];
1659 			frags[0].tword_addr.paddr_lo =
1660 				__cpu_to_le32(skb_cb->paddr);
1661 			frags[0].tword_addr.paddr_hi =
1662 				__cpu_to_le16(upper_32_bits(skb_cb->paddr));
1663 			frags[0].tword_addr.len_16 = __cpu_to_le16(msdu->len);
1664 
1665 			frags_paddr =  htt->frag_desc.paddr +
1666 			   (sizeof(struct htt_msdu_ext_desc_64) * msdu_id);
1667 		} else {
1668 			frags = txbuf->frags;
1669 			frags[0].tword_addr.paddr_lo =
1670 						__cpu_to_le32(skb_cb->paddr);
1671 			frags[0].tword_addr.paddr_hi =
1672 				__cpu_to_le16(upper_32_bits(skb_cb->paddr));
1673 			frags[0].tword_addr.len_16 = __cpu_to_le16(msdu->len);
1674 			frags[1].tword_addr.paddr_lo = 0;
1675 			frags[1].tword_addr.paddr_hi = 0;
1676 			frags[1].tword_addr.len_16 = 0;
1677 		}
1678 		flags0 |= SM(txmode, HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1679 		break;
1680 	case ATH10K_HW_TXRX_MGMT:
1681 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
1682 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
1683 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
1684 
1685 		frags_paddr = skb_cb->paddr;
1686 		break;
1687 	}
1688 
1689 	/* Normally all commands go through HTC which manages tx credits for
1690 	 * each endpoint and notifies when tx is completed.
1691 	 *
1692 	 * HTT endpoint is creditless so there's no need to care about HTC
1693 	 * flags. In that case it is trivial to fill the HTC header here.
1694 	 *
1695 	 * MSDU transmission is considered completed upon HTT event. This
1696 	 * implies no relevant resources can be freed until after the event is
1697 	 * received. That's why HTC tx completion handler itself is ignored by
1698 	 * setting NULL to transfer_context for all sg items.
1699 	 *
1700 	 * There is simply no point in pushing HTT TX_FRM through HTC tx path
1701 	 * as it's a waste of resources. By bypassing HTC it is possible to
1702 	 * avoid extra memory allocations, compress data structures and thus
1703 	 * improve performance.
1704 	 */
1705 
1706 	txbuf->htc_hdr.eid = htt->eid;
1707 	txbuf->htc_hdr.len = __cpu_to_le16(sizeof(txbuf->cmd_hdr) +
1708 					   sizeof(txbuf->cmd_tx) +
1709 					   prefetch_len);
1710 	txbuf->htc_hdr.flags = 0;
1711 
1712 	if (skb_cb->flags & ATH10K_SKB_F_NO_HWCRYPT)
1713 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
1714 
1715 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
1716 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
1717 	if (msdu->ip_summed == CHECKSUM_PARTIAL &&
1718 	    !test_bit(ATH10K_FLAG_RAW_MODE, &ar->dev_flags)) {
1719 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
1720 		flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
1721 		if (ar->hw_params.continuous_frag_desc) {
1722 			memset(ext_desc->tso_flag, 0, sizeof(ext_desc->tso_flag));
1723 			ext_desc->tso_flag[3] |=
1724 				__cpu_to_le32(HTT_MSDU_CHECKSUM_ENABLE_64);
1725 		}
1726 	}
1727 
1728 	/* Prevent firmware from sending up tx inspection requests. There's
1729 	 * nothing ath10k can do with frames requested for inspection so force
1730 	 * it to simply rely a regular tx completion with discard status.
1731 	 */
1732 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_POSTPONED;
1733 
1734 	txbuf->cmd_hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
1735 	txbuf->cmd_tx.flags0 = flags0;
1736 	txbuf->cmd_tx.flags1 = __cpu_to_le16(flags1);
1737 	txbuf->cmd_tx.len = __cpu_to_le16(msdu->len);
1738 	txbuf->cmd_tx.id = __cpu_to_le16(msdu_id);
1739 
1740 	/* fill fragment descriptor */
1741 	txbuf->cmd_tx.frags_paddr = __cpu_to_le64(frags_paddr);
1742 	if (ath10k_mac_tx_frm_has_freq(ar)) {
1743 		txbuf->cmd_tx.offchan_tx.peerid =
1744 				__cpu_to_le16(HTT_INVALID_PEERID);
1745 		txbuf->cmd_tx.offchan_tx.freq =
1746 				__cpu_to_le16(freq);
1747 	} else {
1748 		txbuf->cmd_tx.peerid =
1749 				__cpu_to_le32(HTT_INVALID_PEERID);
1750 	}
1751 
1752 	trace_ath10k_htt_tx(ar, msdu_id, msdu->len, vdev_id, tid);
1753 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1754 		   "htt tx flags0 %u flags1 %u len %d id %u frags_paddr %pad, msdu_paddr %pad vdev %u tid %u freq %u\n",
1755 		   flags0, flags1, msdu->len, msdu_id, &frags_paddr,
1756 		   &skb_cb->paddr, vdev_id, tid, freq);
1757 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt tx msdu: ",
1758 			msdu->data, msdu->len);
1759 	trace_ath10k_tx_hdr(ar, msdu->data, msdu->len);
1760 	trace_ath10k_tx_payload(ar, msdu->data, msdu->len);
1761 
1762 	sg_items[0].transfer_id = 0;
1763 	sg_items[0].transfer_context = NULL;
1764 	sg_items[0].vaddr = &txbuf->htc_hdr;
1765 	sg_items[0].paddr = txbuf_paddr +
1766 			    sizeof(txbuf->frags);
1767 	sg_items[0].len = sizeof(txbuf->htc_hdr) +
1768 			  sizeof(txbuf->cmd_hdr) +
1769 			  sizeof(txbuf->cmd_tx);
1770 
1771 	sg_items[1].transfer_id = 0;
1772 	sg_items[1].transfer_context = NULL;
1773 	sg_items[1].vaddr = msdu->data;
1774 	sg_items[1].paddr = skb_cb->paddr;
1775 	sg_items[1].len = prefetch_len;
1776 
1777 	res = ath10k_hif_tx_sg(htt->ar,
1778 			       htt->ar->htc.endpoint[htt->eid].ul_pipe_id,
1779 			       sg_items, ARRAY_SIZE(sg_items));
1780 	if (res)
1781 		goto err_unmap_msdu;
1782 
1783 	return 0;
1784 
1785 err_unmap_msdu:
1786 	dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
1787 err_free_msdu_id:
1788 	spin_lock_bh(&htt->tx_lock);
1789 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
1790 	spin_unlock_bh(&htt->tx_lock);
1791 err:
1792 	return res;
1793 }
1794 
1795 static const struct ath10k_htt_tx_ops htt_tx_ops_32 = {
1796 	.htt_send_rx_ring_cfg = ath10k_htt_send_rx_ring_cfg_32,
1797 	.htt_send_frag_desc_bank_cfg = ath10k_htt_send_frag_desc_bank_cfg_32,
1798 	.htt_alloc_frag_desc = ath10k_htt_tx_alloc_cont_frag_desc_32,
1799 	.htt_free_frag_desc = ath10k_htt_tx_free_cont_frag_desc_32,
1800 	.htt_tx = ath10k_htt_tx_32,
1801 	.htt_alloc_txbuff = ath10k_htt_tx_alloc_cont_txbuf_32,
1802 	.htt_free_txbuff = ath10k_htt_tx_free_cont_txbuf_32,
1803 	.htt_h2t_aggr_cfg_msg = ath10k_htt_h2t_aggr_cfg_msg_32,
1804 };
1805 
1806 static const struct ath10k_htt_tx_ops htt_tx_ops_64 = {
1807 	.htt_send_rx_ring_cfg = ath10k_htt_send_rx_ring_cfg_64,
1808 	.htt_send_frag_desc_bank_cfg = ath10k_htt_send_frag_desc_bank_cfg_64,
1809 	.htt_alloc_frag_desc = ath10k_htt_tx_alloc_cont_frag_desc_64,
1810 	.htt_free_frag_desc = ath10k_htt_tx_free_cont_frag_desc_64,
1811 	.htt_tx = ath10k_htt_tx_64,
1812 	.htt_alloc_txbuff = ath10k_htt_tx_alloc_cont_txbuf_64,
1813 	.htt_free_txbuff = ath10k_htt_tx_free_cont_txbuf_64,
1814 	.htt_h2t_aggr_cfg_msg = ath10k_htt_h2t_aggr_cfg_msg_v2,
1815 };
1816 
1817 static const struct ath10k_htt_tx_ops htt_tx_ops_hl = {
1818 	.htt_send_rx_ring_cfg = ath10k_htt_send_rx_ring_cfg_hl,
1819 	.htt_send_frag_desc_bank_cfg = ath10k_htt_send_frag_desc_bank_cfg_32,
1820 	.htt_tx = ath10k_htt_tx_hl,
1821 	.htt_h2t_aggr_cfg_msg = ath10k_htt_h2t_aggr_cfg_msg_32,
1822 	.htt_flush_tx = ath10k_htt_flush_tx_queue,
1823 };
1824 
1825 void ath10k_htt_set_tx_ops(struct ath10k_htt *htt)
1826 {
1827 	struct ath10k *ar = htt->ar;
1828 
1829 	if (ar->bus_param.dev_type == ATH10K_DEV_TYPE_HL)
1830 		htt->tx_ops = &htt_tx_ops_hl;
1831 	else if (ar->hw_params.target_64bit)
1832 		htt->tx_ops = &htt_tx_ops_64;
1833 	else
1834 		htt->tx_ops = &htt_tx_ops_32;
1835 }
1836