xref: /linux/drivers/net/wireless/ath/ath10k/htt_tx.c (revision add4b1b02da7e7ec35c34dd04d351ac53f3f0dd8)
1 /*
2  * Copyright (c) 2005-2011 Atheros Communications Inc.
3  * Copyright (c) 2011-2013 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 void __ath10k_htt_tx_dec_pending(struct ath10k_htt *htt)
26 {
27 	htt->num_pending_tx--;
28 	if (htt->num_pending_tx == htt->max_num_pending_tx - 1)
29 		ieee80211_wake_queues(htt->ar->hw);
30 }
31 
32 static void ath10k_htt_tx_dec_pending(struct ath10k_htt *htt)
33 {
34 	spin_lock_bh(&htt->tx_lock);
35 	__ath10k_htt_tx_dec_pending(htt);
36 	spin_unlock_bh(&htt->tx_lock);
37 }
38 
39 static int ath10k_htt_tx_inc_pending(struct ath10k_htt *htt)
40 {
41 	int ret = 0;
42 
43 	spin_lock_bh(&htt->tx_lock);
44 
45 	if (htt->num_pending_tx >= htt->max_num_pending_tx) {
46 		ret = -EBUSY;
47 		goto exit;
48 	}
49 
50 	htt->num_pending_tx++;
51 	if (htt->num_pending_tx == htt->max_num_pending_tx)
52 		ieee80211_stop_queues(htt->ar->hw);
53 
54 exit:
55 	spin_unlock_bh(&htt->tx_lock);
56 	return ret;
57 }
58 
59 int ath10k_htt_tx_alloc_msdu_id(struct ath10k_htt *htt)
60 {
61 	struct ath10k *ar = htt->ar;
62 	int msdu_id;
63 
64 	lockdep_assert_held(&htt->tx_lock);
65 
66 	msdu_id = find_first_zero_bit(htt->used_msdu_ids,
67 				      htt->max_num_pending_tx);
68 	if (msdu_id == htt->max_num_pending_tx)
69 		return -ENOBUFS;
70 
71 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx alloc msdu_id %d\n", msdu_id);
72 	__set_bit(msdu_id, htt->used_msdu_ids);
73 	return msdu_id;
74 }
75 
76 void ath10k_htt_tx_free_msdu_id(struct ath10k_htt *htt, u16 msdu_id)
77 {
78 	struct ath10k *ar = htt->ar;
79 
80 	lockdep_assert_held(&htt->tx_lock);
81 
82 	if (!test_bit(msdu_id, htt->used_msdu_ids))
83 		ath10k_warn(ar, "trying to free unallocated msdu_id %d\n",
84 			    msdu_id);
85 
86 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx free msdu_id %hu\n", msdu_id);
87 	__clear_bit(msdu_id, htt->used_msdu_ids);
88 }
89 
90 int ath10k_htt_tx_alloc(struct ath10k_htt *htt)
91 {
92 	struct ath10k *ar = htt->ar;
93 
94 	spin_lock_init(&htt->tx_lock);
95 
96 	if (test_bit(ATH10K_FW_FEATURE_WMI_10X, htt->ar->fw_features))
97 		htt->max_num_pending_tx = TARGET_10X_NUM_MSDU_DESC;
98 	else
99 		htt->max_num_pending_tx = TARGET_NUM_MSDU_DESC;
100 
101 	ath10k_dbg(ar, ATH10K_DBG_BOOT, "htt tx max num pending tx %d\n",
102 		   htt->max_num_pending_tx);
103 
104 	htt->pending_tx = kzalloc(sizeof(*htt->pending_tx) *
105 				  htt->max_num_pending_tx, GFP_KERNEL);
106 	if (!htt->pending_tx)
107 		return -ENOMEM;
108 
109 	htt->used_msdu_ids = kzalloc(sizeof(unsigned long) *
110 				     BITS_TO_LONGS(htt->max_num_pending_tx),
111 				     GFP_KERNEL);
112 	if (!htt->used_msdu_ids) {
113 		kfree(htt->pending_tx);
114 		return -ENOMEM;
115 	}
116 
117 	htt->tx_pool = dma_pool_create("ath10k htt tx pool", htt->ar->dev,
118 				       sizeof(struct ath10k_htt_txbuf), 4, 0);
119 	if (!htt->tx_pool) {
120 		kfree(htt->used_msdu_ids);
121 		kfree(htt->pending_tx);
122 		return -ENOMEM;
123 	}
124 
125 	return 0;
126 }
127 
128 static void ath10k_htt_tx_free_pending(struct ath10k_htt *htt)
129 {
130 	struct ath10k *ar = htt->ar;
131 	struct htt_tx_done tx_done = {0};
132 	int msdu_id;
133 
134 	spin_lock_bh(&htt->tx_lock);
135 	for (msdu_id = 0; msdu_id < htt->max_num_pending_tx; msdu_id++) {
136 		if (!test_bit(msdu_id, htt->used_msdu_ids))
137 			continue;
138 
139 		ath10k_dbg(ar, ATH10K_DBG_HTT, "force cleanup msdu_id %hu\n",
140 			   msdu_id);
141 
142 		tx_done.discard = 1;
143 		tx_done.msdu_id = msdu_id;
144 
145 		ath10k_txrx_tx_unref(htt, &tx_done);
146 	}
147 	spin_unlock_bh(&htt->tx_lock);
148 }
149 
150 void ath10k_htt_tx_free(struct ath10k_htt *htt)
151 {
152 	ath10k_htt_tx_free_pending(htt);
153 	kfree(htt->pending_tx);
154 	kfree(htt->used_msdu_ids);
155 	dma_pool_destroy(htt->tx_pool);
156 }
157 
158 void ath10k_htt_htc_tx_complete(struct ath10k *ar, struct sk_buff *skb)
159 {
160 	dev_kfree_skb_any(skb);
161 }
162 
163 int ath10k_htt_h2t_ver_req_msg(struct ath10k_htt *htt)
164 {
165 	struct ath10k *ar = htt->ar;
166 	struct sk_buff *skb;
167 	struct htt_cmd *cmd;
168 	int len = 0;
169 	int ret;
170 
171 	len += sizeof(cmd->hdr);
172 	len += sizeof(cmd->ver_req);
173 
174 	skb = ath10k_htc_alloc_skb(ar, len);
175 	if (!skb)
176 		return -ENOMEM;
177 
178 	skb_put(skb, len);
179 	cmd = (struct htt_cmd *)skb->data;
180 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_VERSION_REQ;
181 
182 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
183 	if (ret) {
184 		dev_kfree_skb_any(skb);
185 		return ret;
186 	}
187 
188 	return 0;
189 }
190 
191 int ath10k_htt_h2t_stats_req(struct ath10k_htt *htt, u8 mask, u64 cookie)
192 {
193 	struct ath10k *ar = htt->ar;
194 	struct htt_stats_req *req;
195 	struct sk_buff *skb;
196 	struct htt_cmd *cmd;
197 	int len = 0, ret;
198 
199 	len += sizeof(cmd->hdr);
200 	len += sizeof(cmd->stats_req);
201 
202 	skb = ath10k_htc_alloc_skb(ar, len);
203 	if (!skb)
204 		return -ENOMEM;
205 
206 	skb_put(skb, len);
207 	cmd = (struct htt_cmd *)skb->data;
208 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_STATS_REQ;
209 
210 	req = &cmd->stats_req;
211 
212 	memset(req, 0, sizeof(*req));
213 
214 	/* currently we support only max 8 bit masks so no need to worry
215 	 * about endian support */
216 	req->upload_types[0] = mask;
217 	req->reset_types[0] = mask;
218 	req->stat_type = HTT_STATS_REQ_CFG_STAT_TYPE_INVALID;
219 	req->cookie_lsb = cpu_to_le32(cookie & 0xffffffff);
220 	req->cookie_msb = cpu_to_le32((cookie & 0xffffffff00000000ULL) >> 32);
221 
222 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
223 	if (ret) {
224 		ath10k_warn(ar, "failed to send htt type stats request: %d",
225 			    ret);
226 		dev_kfree_skb_any(skb);
227 		return ret;
228 	}
229 
230 	return 0;
231 }
232 
233 int ath10k_htt_send_rx_ring_cfg_ll(struct ath10k_htt *htt)
234 {
235 	struct ath10k *ar = htt->ar;
236 	struct sk_buff *skb;
237 	struct htt_cmd *cmd;
238 	struct htt_rx_ring_setup_ring *ring;
239 	const int num_rx_ring = 1;
240 	u16 flags;
241 	u32 fw_idx;
242 	int len;
243 	int ret;
244 
245 	/*
246 	 * the HW expects the buffer to be an integral number of 4-byte
247 	 * "words"
248 	 */
249 	BUILD_BUG_ON(!IS_ALIGNED(HTT_RX_BUF_SIZE, 4));
250 	BUILD_BUG_ON((HTT_RX_BUF_SIZE & HTT_MAX_CACHE_LINE_SIZE_MASK) != 0);
251 
252 	len = sizeof(cmd->hdr) + sizeof(cmd->rx_setup.hdr)
253 	    + (sizeof(*ring) * num_rx_ring);
254 	skb = ath10k_htc_alloc_skb(ar, len);
255 	if (!skb)
256 		return -ENOMEM;
257 
258 	skb_put(skb, len);
259 
260 	cmd = (struct htt_cmd *)skb->data;
261 	ring = &cmd->rx_setup.rings[0];
262 
263 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_RX_RING_CFG;
264 	cmd->rx_setup.hdr.num_rings = 1;
265 
266 	/* FIXME: do we need all of this? */
267 	flags = 0;
268 	flags |= HTT_RX_RING_FLAGS_MAC80211_HDR;
269 	flags |= HTT_RX_RING_FLAGS_MSDU_PAYLOAD;
270 	flags |= HTT_RX_RING_FLAGS_PPDU_START;
271 	flags |= HTT_RX_RING_FLAGS_PPDU_END;
272 	flags |= HTT_RX_RING_FLAGS_MPDU_START;
273 	flags |= HTT_RX_RING_FLAGS_MPDU_END;
274 	flags |= HTT_RX_RING_FLAGS_MSDU_START;
275 	flags |= HTT_RX_RING_FLAGS_MSDU_END;
276 	flags |= HTT_RX_RING_FLAGS_RX_ATTENTION;
277 	flags |= HTT_RX_RING_FLAGS_FRAG_INFO;
278 	flags |= HTT_RX_RING_FLAGS_UNICAST_RX;
279 	flags |= HTT_RX_RING_FLAGS_MULTICAST_RX;
280 	flags |= HTT_RX_RING_FLAGS_CTRL_RX;
281 	flags |= HTT_RX_RING_FLAGS_MGMT_RX;
282 	flags |= HTT_RX_RING_FLAGS_NULL_RX;
283 	flags |= HTT_RX_RING_FLAGS_PHY_DATA_RX;
284 
285 	fw_idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
286 
287 	ring->fw_idx_shadow_reg_paddr =
288 		__cpu_to_le32(htt->rx_ring.alloc_idx.paddr);
289 	ring->rx_ring_base_paddr = __cpu_to_le32(htt->rx_ring.base_paddr);
290 	ring->rx_ring_len = __cpu_to_le16(htt->rx_ring.size);
291 	ring->rx_ring_bufsize = __cpu_to_le16(HTT_RX_BUF_SIZE);
292 	ring->flags = __cpu_to_le16(flags);
293 	ring->fw_idx_init_val = __cpu_to_le16(fw_idx);
294 
295 #define desc_offset(x) (offsetof(struct htt_rx_desc, x) / 4)
296 
297 	ring->mac80211_hdr_offset = __cpu_to_le16(desc_offset(rx_hdr_status));
298 	ring->msdu_payload_offset = __cpu_to_le16(desc_offset(msdu_payload));
299 	ring->ppdu_start_offset = __cpu_to_le16(desc_offset(ppdu_start));
300 	ring->ppdu_end_offset = __cpu_to_le16(desc_offset(ppdu_end));
301 	ring->mpdu_start_offset = __cpu_to_le16(desc_offset(mpdu_start));
302 	ring->mpdu_end_offset = __cpu_to_le16(desc_offset(mpdu_end));
303 	ring->msdu_start_offset = __cpu_to_le16(desc_offset(msdu_start));
304 	ring->msdu_end_offset = __cpu_to_le16(desc_offset(msdu_end));
305 	ring->rx_attention_offset = __cpu_to_le16(desc_offset(attention));
306 	ring->frag_info_offset = __cpu_to_le16(desc_offset(frag_info));
307 
308 #undef desc_offset
309 
310 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
311 	if (ret) {
312 		dev_kfree_skb_any(skb);
313 		return ret;
314 	}
315 
316 	return 0;
317 }
318 
319 int ath10k_htt_h2t_aggr_cfg_msg(struct ath10k_htt *htt,
320 				u8 max_subfrms_ampdu,
321 				u8 max_subfrms_amsdu)
322 {
323 	struct ath10k *ar = htt->ar;
324 	struct htt_aggr_conf *aggr_conf;
325 	struct sk_buff *skb;
326 	struct htt_cmd *cmd;
327 	int len;
328 	int ret;
329 
330 	/* Firmware defaults are: amsdu = 3 and ampdu = 64 */
331 
332 	if (max_subfrms_ampdu == 0 || max_subfrms_ampdu > 64)
333 		return -EINVAL;
334 
335 	if (max_subfrms_amsdu == 0 || max_subfrms_amsdu > 31)
336 		return -EINVAL;
337 
338 	len = sizeof(cmd->hdr);
339 	len += sizeof(cmd->aggr_conf);
340 
341 	skb = ath10k_htc_alloc_skb(ar, len);
342 	if (!skb)
343 		return -ENOMEM;
344 
345 	skb_put(skb, len);
346 	cmd = (struct htt_cmd *)skb->data;
347 	cmd->hdr.msg_type = HTT_H2T_MSG_TYPE_AGGR_CFG;
348 
349 	aggr_conf = &cmd->aggr_conf;
350 	aggr_conf->max_num_ampdu_subframes = max_subfrms_ampdu;
351 	aggr_conf->max_num_amsdu_subframes = max_subfrms_amsdu;
352 
353 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt h2t aggr cfg msg amsdu %d ampdu %d",
354 		   aggr_conf->max_num_amsdu_subframes,
355 		   aggr_conf->max_num_ampdu_subframes);
356 
357 	ret = ath10k_htc_send(&htt->ar->htc, htt->eid, skb);
358 	if (ret) {
359 		dev_kfree_skb_any(skb);
360 		return ret;
361 	}
362 
363 	return 0;
364 }
365 
366 int ath10k_htt_mgmt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
367 {
368 	struct ath10k *ar = htt->ar;
369 	struct device *dev = ar->dev;
370 	struct sk_buff *txdesc = NULL;
371 	struct htt_cmd *cmd;
372 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
373 	u8 vdev_id = skb_cb->vdev_id;
374 	int len = 0;
375 	int msdu_id = -1;
376 	int res;
377 
378 	res = ath10k_htt_tx_inc_pending(htt);
379 	if (res)
380 		goto err;
381 
382 	len += sizeof(cmd->hdr);
383 	len += sizeof(cmd->mgmt_tx);
384 
385 	spin_lock_bh(&htt->tx_lock);
386 	res = ath10k_htt_tx_alloc_msdu_id(htt);
387 	if (res < 0) {
388 		spin_unlock_bh(&htt->tx_lock);
389 		goto err_tx_dec;
390 	}
391 	msdu_id = res;
392 	htt->pending_tx[msdu_id] = msdu;
393 	spin_unlock_bh(&htt->tx_lock);
394 
395 	txdesc = ath10k_htc_alloc_skb(ar, len);
396 	if (!txdesc) {
397 		res = -ENOMEM;
398 		goto err_free_msdu_id;
399 	}
400 
401 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
402 				       DMA_TO_DEVICE);
403 	res = dma_mapping_error(dev, skb_cb->paddr);
404 	if (res)
405 		goto err_free_txdesc;
406 
407 	skb_put(txdesc, len);
408 	cmd = (struct htt_cmd *)txdesc->data;
409 	cmd->hdr.msg_type         = HTT_H2T_MSG_TYPE_MGMT_TX;
410 	cmd->mgmt_tx.msdu_paddr = __cpu_to_le32(ATH10K_SKB_CB(msdu)->paddr);
411 	cmd->mgmt_tx.len        = __cpu_to_le32(msdu->len);
412 	cmd->mgmt_tx.desc_id    = __cpu_to_le32(msdu_id);
413 	cmd->mgmt_tx.vdev_id    = __cpu_to_le32(vdev_id);
414 	memcpy(cmd->mgmt_tx.hdr, msdu->data,
415 	       min_t(int, msdu->len, HTT_MGMT_FRM_HDR_DOWNLOAD_LEN));
416 
417 	skb_cb->htt.txbuf = NULL;
418 
419 	res = ath10k_htc_send(&htt->ar->htc, htt->eid, txdesc);
420 	if (res)
421 		goto err_unmap_msdu;
422 
423 	return 0;
424 
425 err_unmap_msdu:
426 	dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
427 err_free_txdesc:
428 	dev_kfree_skb_any(txdesc);
429 err_free_msdu_id:
430 	spin_lock_bh(&htt->tx_lock);
431 	htt->pending_tx[msdu_id] = NULL;
432 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
433 	spin_unlock_bh(&htt->tx_lock);
434 err_tx_dec:
435 	ath10k_htt_tx_dec_pending(htt);
436 err:
437 	return res;
438 }
439 
440 int ath10k_htt_tx(struct ath10k_htt *htt, struct sk_buff *msdu)
441 {
442 	struct ath10k *ar = htt->ar;
443 	struct device *dev = ar->dev;
444 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)msdu->data;
445 	struct ath10k_skb_cb *skb_cb = ATH10K_SKB_CB(msdu);
446 	struct ath10k_hif_sg_item sg_items[2];
447 	struct htt_data_tx_desc_frag *frags;
448 	u8 vdev_id = skb_cb->vdev_id;
449 	u8 tid = skb_cb->htt.tid;
450 	int prefetch_len;
451 	int res;
452 	u8 flags0 = 0;
453 	u16 msdu_id, flags1 = 0;
454 	dma_addr_t paddr;
455 	u32 frags_paddr;
456 	bool use_frags;
457 
458 	res = ath10k_htt_tx_inc_pending(htt);
459 	if (res)
460 		goto err;
461 
462 	spin_lock_bh(&htt->tx_lock);
463 	res = ath10k_htt_tx_alloc_msdu_id(htt);
464 	if (res < 0) {
465 		spin_unlock_bh(&htt->tx_lock);
466 		goto err_tx_dec;
467 	}
468 	msdu_id = res;
469 	htt->pending_tx[msdu_id] = msdu;
470 	spin_unlock_bh(&htt->tx_lock);
471 
472 	prefetch_len = min(htt->prefetch_len, msdu->len);
473 	prefetch_len = roundup(prefetch_len, 4);
474 
475 	/* Since HTT 3.0 there is no separate mgmt tx command. However in case
476 	 * of mgmt tx using TX_FRM there is not tx fragment list. Instead of tx
477 	 * fragment list host driver specifies directly frame pointer. */
478 	use_frags = htt->target_version_major < 3 ||
479 		    !ieee80211_is_mgmt(hdr->frame_control);
480 
481 	skb_cb->htt.txbuf = dma_pool_alloc(htt->tx_pool, GFP_ATOMIC,
482 					   &paddr);
483 	if (!skb_cb->htt.txbuf)
484 		goto err_free_msdu_id;
485 	skb_cb->htt.txbuf_paddr = paddr;
486 
487 	skb_cb->paddr = dma_map_single(dev, msdu->data, msdu->len,
488 				       DMA_TO_DEVICE);
489 	res = dma_mapping_error(dev, skb_cb->paddr);
490 	if (res)
491 		goto err_free_txbuf;
492 
493 	if (likely(use_frags)) {
494 		frags = skb_cb->htt.txbuf->frags;
495 
496 		frags[0].paddr = __cpu_to_le32(skb_cb->paddr);
497 		frags[0].len = __cpu_to_le32(msdu->len);
498 		frags[1].paddr = 0;
499 		frags[1].len = 0;
500 
501 		flags0 |= SM(ATH10K_HW_TXRX_NATIVE_WIFI,
502 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
503 
504 		frags_paddr = skb_cb->htt.txbuf_paddr;
505 	} else {
506 		flags0 |= SM(ATH10K_HW_TXRX_MGMT,
507 			     HTT_DATA_TX_DESC_FLAGS0_PKT_TYPE);
508 
509 		frags_paddr = skb_cb->paddr;
510 	}
511 
512 	/* Normally all commands go through HTC which manages tx credits for
513 	 * each endpoint and notifies when tx is completed.
514 	 *
515 	 * HTT endpoint is creditless so there's no need to care about HTC
516 	 * flags. In that case it is trivial to fill the HTC header here.
517 	 *
518 	 * MSDU transmission is considered completed upon HTT event. This
519 	 * implies no relevant resources can be freed until after the event is
520 	 * received. That's why HTC tx completion handler itself is ignored by
521 	 * setting NULL to transfer_context for all sg items.
522 	 *
523 	 * There is simply no point in pushing HTT TX_FRM through HTC tx path
524 	 * as it's a waste of resources. By bypassing HTC it is possible to
525 	 * avoid extra memory allocations, compress data structures and thus
526 	 * improve performance. */
527 
528 	skb_cb->htt.txbuf->htc_hdr.eid = htt->eid;
529 	skb_cb->htt.txbuf->htc_hdr.len = __cpu_to_le16(
530 			sizeof(skb_cb->htt.txbuf->cmd_hdr) +
531 			sizeof(skb_cb->htt.txbuf->cmd_tx) +
532 			prefetch_len);
533 	skb_cb->htt.txbuf->htc_hdr.flags = 0;
534 
535 	if (!ieee80211_has_protected(hdr->frame_control))
536 		flags0 |= HTT_DATA_TX_DESC_FLAGS0_NO_ENCRYPT;
537 
538 	flags0 |= HTT_DATA_TX_DESC_FLAGS0_MAC_HDR_PRESENT;
539 
540 	flags1 |= SM((u16)vdev_id, HTT_DATA_TX_DESC_FLAGS1_VDEV_ID);
541 	flags1 |= SM((u16)tid, HTT_DATA_TX_DESC_FLAGS1_EXT_TID);
542 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L3_OFFLOAD;
543 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_CKSUM_L4_OFFLOAD;
544 
545 	/* Prevent firmware from sending up tx inspection requests. There's
546 	 * nothing ath10k can do with frames requested for inspection so force
547 	 * it to simply rely a regular tx completion with discard status.
548 	 */
549 	flags1 |= HTT_DATA_TX_DESC_FLAGS1_POSTPONED;
550 
551 	skb_cb->htt.txbuf->cmd_hdr.msg_type = HTT_H2T_MSG_TYPE_TX_FRM;
552 	skb_cb->htt.txbuf->cmd_tx.flags0 = flags0;
553 	skb_cb->htt.txbuf->cmd_tx.flags1 = __cpu_to_le16(flags1);
554 	skb_cb->htt.txbuf->cmd_tx.len = __cpu_to_le16(msdu->len);
555 	skb_cb->htt.txbuf->cmd_tx.id = __cpu_to_le16(msdu_id);
556 	skb_cb->htt.txbuf->cmd_tx.frags_paddr = __cpu_to_le32(frags_paddr);
557 	skb_cb->htt.txbuf->cmd_tx.peerid = __cpu_to_le16(HTT_INVALID_PEERID);
558 	skb_cb->htt.txbuf->cmd_tx.freq = __cpu_to_le16(skb_cb->htt.freq);
559 
560 	trace_ath10k_htt_tx(ar, msdu_id, msdu->len, vdev_id, tid);
561 	ath10k_dbg(ar, ATH10K_DBG_HTT,
562 		   "htt tx flags0 %hhu flags1 %hu len %d id %hu frags_paddr %08x, msdu_paddr %08x vdev %hhu tid %hhu freq %hu\n",
563 		   flags0, flags1, msdu->len, msdu_id, frags_paddr,
564 		   (u32)skb_cb->paddr, vdev_id, tid, skb_cb->htt.freq);
565 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt tx msdu: ",
566 			msdu->data, msdu->len);
567 	trace_ath10k_tx_hdr(ar, msdu->data, msdu->len);
568 	trace_ath10k_tx_payload(ar, msdu->data, msdu->len);
569 
570 	sg_items[0].transfer_id = 0;
571 	sg_items[0].transfer_context = NULL;
572 	sg_items[0].vaddr = &skb_cb->htt.txbuf->htc_hdr;
573 	sg_items[0].paddr = skb_cb->htt.txbuf_paddr +
574 			    sizeof(skb_cb->htt.txbuf->frags);
575 	sg_items[0].len = sizeof(skb_cb->htt.txbuf->htc_hdr) +
576 			  sizeof(skb_cb->htt.txbuf->cmd_hdr) +
577 			  sizeof(skb_cb->htt.txbuf->cmd_tx);
578 
579 	sg_items[1].transfer_id = 0;
580 	sg_items[1].transfer_context = NULL;
581 	sg_items[1].vaddr = msdu->data;
582 	sg_items[1].paddr = skb_cb->paddr;
583 	sg_items[1].len = prefetch_len;
584 
585 	res = ath10k_hif_tx_sg(htt->ar,
586 			       htt->ar->htc.endpoint[htt->eid].ul_pipe_id,
587 			       sg_items, ARRAY_SIZE(sg_items));
588 	if (res)
589 		goto err_unmap_msdu;
590 
591 	return 0;
592 
593 err_unmap_msdu:
594 	dma_unmap_single(dev, skb_cb->paddr, msdu->len, DMA_TO_DEVICE);
595 err_free_txbuf:
596 	dma_pool_free(htt->tx_pool,
597 		      skb_cb->htt.txbuf,
598 		      skb_cb->htt.txbuf_paddr);
599 err_free_msdu_id:
600 	spin_lock_bh(&htt->tx_lock);
601 	htt->pending_tx[msdu_id] = NULL;
602 	ath10k_htt_tx_free_msdu_id(htt, msdu_id);
603 	spin_unlock_bh(&htt->tx_lock);
604 err_tx_dec:
605 	ath10k_htt_tx_dec_pending(htt);
606 err:
607 	return res;
608 }
609