xref: /linux/drivers/net/wireless/ath/ath10k/htt_rx.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 "core.h"
19 #include "htc.h"
20 #include "htt.h"
21 #include "txrx.h"
22 #include "debug.h"
23 #include "trace.h"
24 #include "mac.h"
25 
26 #include <linux/log2.h>
27 
28 #define HTT_RX_RING_SIZE HTT_RX_RING_SIZE_MAX
29 #define HTT_RX_RING_FILL_LEVEL (((HTT_RX_RING_SIZE) / 2) - 1)
30 
31 /* when under memory pressure rx ring refill may fail and needs a retry */
32 #define HTT_RX_RING_REFILL_RETRY_MS 50
33 
34 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb);
35 static void ath10k_htt_txrx_compl_task(unsigned long ptr);
36 
37 static struct sk_buff *
38 ath10k_htt_rx_find_skb_paddr(struct ath10k *ar, u32 paddr)
39 {
40 	struct ath10k_skb_rxcb *rxcb;
41 
42 	hash_for_each_possible(ar->htt.rx_ring.skb_table, rxcb, hlist, paddr)
43 		if (rxcb->paddr == paddr)
44 			return ATH10K_RXCB_SKB(rxcb);
45 
46 	WARN_ON_ONCE(1);
47 	return NULL;
48 }
49 
50 static void ath10k_htt_rx_ring_free(struct ath10k_htt *htt)
51 {
52 	struct sk_buff *skb;
53 	struct ath10k_skb_rxcb *rxcb;
54 	struct hlist_node *n;
55 	int i;
56 
57 	if (htt->rx_ring.in_ord_rx) {
58 		hash_for_each_safe(htt->rx_ring.skb_table, i, n, rxcb, hlist) {
59 			skb = ATH10K_RXCB_SKB(rxcb);
60 			dma_unmap_single(htt->ar->dev, rxcb->paddr,
61 					 skb->len + skb_tailroom(skb),
62 					 DMA_FROM_DEVICE);
63 			hash_del(&rxcb->hlist);
64 			dev_kfree_skb_any(skb);
65 		}
66 	} else {
67 		for (i = 0; i < htt->rx_ring.size; i++) {
68 			skb = htt->rx_ring.netbufs_ring[i];
69 			if (!skb)
70 				continue;
71 
72 			rxcb = ATH10K_SKB_RXCB(skb);
73 			dma_unmap_single(htt->ar->dev, rxcb->paddr,
74 					 skb->len + skb_tailroom(skb),
75 					 DMA_FROM_DEVICE);
76 			dev_kfree_skb_any(skb);
77 		}
78 	}
79 
80 	htt->rx_ring.fill_cnt = 0;
81 	hash_init(htt->rx_ring.skb_table);
82 	memset(htt->rx_ring.netbufs_ring, 0,
83 	       htt->rx_ring.size * sizeof(htt->rx_ring.netbufs_ring[0]));
84 }
85 
86 static int __ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
87 {
88 	struct htt_rx_desc *rx_desc;
89 	struct ath10k_skb_rxcb *rxcb;
90 	struct sk_buff *skb;
91 	dma_addr_t paddr;
92 	int ret = 0, idx;
93 
94 	/* The Full Rx Reorder firmware has no way of telling the host
95 	 * implicitly when it copied HTT Rx Ring buffers to MAC Rx Ring.
96 	 * To keep things simple make sure ring is always half empty. This
97 	 * guarantees there'll be no replenishment overruns possible.
98 	 */
99 	BUILD_BUG_ON(HTT_RX_RING_FILL_LEVEL >= HTT_RX_RING_SIZE / 2);
100 
101 	idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
102 	while (num > 0) {
103 		skb = dev_alloc_skb(HTT_RX_BUF_SIZE + HTT_RX_DESC_ALIGN);
104 		if (!skb) {
105 			ret = -ENOMEM;
106 			goto fail;
107 		}
108 
109 		if (!IS_ALIGNED((unsigned long)skb->data, HTT_RX_DESC_ALIGN))
110 			skb_pull(skb,
111 				 PTR_ALIGN(skb->data, HTT_RX_DESC_ALIGN) -
112 				 skb->data);
113 
114 		/* Clear rx_desc attention word before posting to Rx ring */
115 		rx_desc = (struct htt_rx_desc *)skb->data;
116 		rx_desc->attention.flags = __cpu_to_le32(0);
117 
118 		paddr = dma_map_single(htt->ar->dev, skb->data,
119 				       skb->len + skb_tailroom(skb),
120 				       DMA_FROM_DEVICE);
121 
122 		if (unlikely(dma_mapping_error(htt->ar->dev, paddr))) {
123 			dev_kfree_skb_any(skb);
124 			ret = -ENOMEM;
125 			goto fail;
126 		}
127 
128 		rxcb = ATH10K_SKB_RXCB(skb);
129 		rxcb->paddr = paddr;
130 		htt->rx_ring.netbufs_ring[idx] = skb;
131 		htt->rx_ring.paddrs_ring[idx] = __cpu_to_le32(paddr);
132 		htt->rx_ring.fill_cnt++;
133 
134 		if (htt->rx_ring.in_ord_rx) {
135 			hash_add(htt->rx_ring.skb_table,
136 				 &ATH10K_SKB_RXCB(skb)->hlist,
137 				 (u32)paddr);
138 		}
139 
140 		num--;
141 		idx++;
142 		idx &= htt->rx_ring.size_mask;
143 	}
144 
145 fail:
146 	/*
147 	 * Make sure the rx buffer is updated before available buffer
148 	 * index to avoid any potential rx ring corruption.
149 	 */
150 	mb();
151 	*htt->rx_ring.alloc_idx.vaddr = __cpu_to_le32(idx);
152 	return ret;
153 }
154 
155 static int ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
156 {
157 	lockdep_assert_held(&htt->rx_ring.lock);
158 	return __ath10k_htt_rx_ring_fill_n(htt, num);
159 }
160 
161 static void ath10k_htt_rx_msdu_buff_replenish(struct ath10k_htt *htt)
162 {
163 	int ret, num_deficit, num_to_fill;
164 
165 	/* Refilling the whole RX ring buffer proves to be a bad idea. The
166 	 * reason is RX may take up significant amount of CPU cycles and starve
167 	 * other tasks, e.g. TX on an ethernet device while acting as a bridge
168 	 * with ath10k wlan interface. This ended up with very poor performance
169 	 * once CPU the host system was overwhelmed with RX on ath10k.
170 	 *
171 	 * By limiting the number of refills the replenishing occurs
172 	 * progressively. This in turns makes use of the fact tasklets are
173 	 * processed in FIFO order. This means actual RX processing can starve
174 	 * out refilling. If there's not enough buffers on RX ring FW will not
175 	 * report RX until it is refilled with enough buffers. This
176 	 * automatically balances load wrt to CPU power.
177 	 *
178 	 * This probably comes at a cost of lower maximum throughput but
179 	 * improves the average and stability. */
180 	spin_lock_bh(&htt->rx_ring.lock);
181 	num_deficit = htt->rx_ring.fill_level - htt->rx_ring.fill_cnt;
182 	num_to_fill = min(ATH10K_HTT_MAX_NUM_REFILL, num_deficit);
183 	num_deficit -= num_to_fill;
184 	ret = ath10k_htt_rx_ring_fill_n(htt, num_to_fill);
185 	if (ret == -ENOMEM) {
186 		/*
187 		 * Failed to fill it to the desired level -
188 		 * we'll start a timer and try again next time.
189 		 * As long as enough buffers are left in the ring for
190 		 * another A-MPDU rx, no special recovery is needed.
191 		 */
192 		mod_timer(&htt->rx_ring.refill_retry_timer, jiffies +
193 			  msecs_to_jiffies(HTT_RX_RING_REFILL_RETRY_MS));
194 	} else if (num_deficit > 0) {
195 		tasklet_schedule(&htt->rx_replenish_task);
196 	}
197 	spin_unlock_bh(&htt->rx_ring.lock);
198 }
199 
200 static void ath10k_htt_rx_ring_refill_retry(unsigned long arg)
201 {
202 	struct ath10k_htt *htt = (struct ath10k_htt *)arg;
203 
204 	ath10k_htt_rx_msdu_buff_replenish(htt);
205 }
206 
207 int ath10k_htt_rx_ring_refill(struct ath10k *ar)
208 {
209 	struct ath10k_htt *htt = &ar->htt;
210 	int ret;
211 
212 	spin_lock_bh(&htt->rx_ring.lock);
213 	ret = ath10k_htt_rx_ring_fill_n(htt, (htt->rx_ring.fill_level -
214 					      htt->rx_ring.fill_cnt));
215 	spin_unlock_bh(&htt->rx_ring.lock);
216 
217 	if (ret)
218 		ath10k_htt_rx_ring_free(htt);
219 
220 	return ret;
221 }
222 
223 void ath10k_htt_rx_free(struct ath10k_htt *htt)
224 {
225 	del_timer_sync(&htt->rx_ring.refill_retry_timer);
226 	tasklet_kill(&htt->rx_replenish_task);
227 	tasklet_kill(&htt->txrx_compl_task);
228 
229 	skb_queue_purge(&htt->tx_compl_q);
230 	skb_queue_purge(&htt->rx_compl_q);
231 	skb_queue_purge(&htt->rx_in_ord_compl_q);
232 
233 	ath10k_htt_rx_ring_free(htt);
234 
235 	dma_free_coherent(htt->ar->dev,
236 			  (htt->rx_ring.size *
237 			   sizeof(htt->rx_ring.paddrs_ring)),
238 			  htt->rx_ring.paddrs_ring,
239 			  htt->rx_ring.base_paddr);
240 
241 	dma_free_coherent(htt->ar->dev,
242 			  sizeof(*htt->rx_ring.alloc_idx.vaddr),
243 			  htt->rx_ring.alloc_idx.vaddr,
244 			  htt->rx_ring.alloc_idx.paddr);
245 
246 	kfree(htt->rx_ring.netbufs_ring);
247 }
248 
249 static inline struct sk_buff *ath10k_htt_rx_netbuf_pop(struct ath10k_htt *htt)
250 {
251 	struct ath10k *ar = htt->ar;
252 	int idx;
253 	struct sk_buff *msdu;
254 
255 	lockdep_assert_held(&htt->rx_ring.lock);
256 
257 	if (htt->rx_ring.fill_cnt == 0) {
258 		ath10k_warn(ar, "tried to pop sk_buff from an empty rx ring\n");
259 		return NULL;
260 	}
261 
262 	idx = htt->rx_ring.sw_rd_idx.msdu_payld;
263 	msdu = htt->rx_ring.netbufs_ring[idx];
264 	htt->rx_ring.netbufs_ring[idx] = NULL;
265 	htt->rx_ring.paddrs_ring[idx] = 0;
266 
267 	idx++;
268 	idx &= htt->rx_ring.size_mask;
269 	htt->rx_ring.sw_rd_idx.msdu_payld = idx;
270 	htt->rx_ring.fill_cnt--;
271 
272 	dma_unmap_single(htt->ar->dev,
273 			 ATH10K_SKB_RXCB(msdu)->paddr,
274 			 msdu->len + skb_tailroom(msdu),
275 			 DMA_FROM_DEVICE);
276 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt rx netbuf pop: ",
277 			msdu->data, msdu->len + skb_tailroom(msdu));
278 
279 	return msdu;
280 }
281 
282 /* return: < 0 fatal error, 0 - non chained msdu, 1 chained msdu */
283 static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
284 				   u8 **fw_desc, int *fw_desc_len,
285 				   struct sk_buff_head *amsdu)
286 {
287 	struct ath10k *ar = htt->ar;
288 	int msdu_len, msdu_chaining = 0;
289 	struct sk_buff *msdu;
290 	struct htt_rx_desc *rx_desc;
291 
292 	lockdep_assert_held(&htt->rx_ring.lock);
293 
294 	for (;;) {
295 		int last_msdu, msdu_len_invalid, msdu_chained;
296 
297 		msdu = ath10k_htt_rx_netbuf_pop(htt);
298 		if (!msdu) {
299 			__skb_queue_purge(amsdu);
300 			return -ENOENT;
301 		}
302 
303 		__skb_queue_tail(amsdu, msdu);
304 
305 		rx_desc = (struct htt_rx_desc *)msdu->data;
306 
307 		/* FIXME: we must report msdu payload since this is what caller
308 		 *        expects now */
309 		skb_put(msdu, offsetof(struct htt_rx_desc, msdu_payload));
310 		skb_pull(msdu, offsetof(struct htt_rx_desc, msdu_payload));
311 
312 		/*
313 		 * Sanity check - confirm the HW is finished filling in the
314 		 * rx data.
315 		 * If the HW and SW are working correctly, then it's guaranteed
316 		 * that the HW's MAC DMA is done before this point in the SW.
317 		 * To prevent the case that we handle a stale Rx descriptor,
318 		 * just assert for now until we have a way to recover.
319 		 */
320 		if (!(__le32_to_cpu(rx_desc->attention.flags)
321 				& RX_ATTENTION_FLAGS_MSDU_DONE)) {
322 			__skb_queue_purge(amsdu);
323 			return -EIO;
324 		}
325 
326 		/*
327 		 * Copy the FW rx descriptor for this MSDU from the rx
328 		 * indication message into the MSDU's netbuf. HL uses the
329 		 * same rx indication message definition as LL, and simply
330 		 * appends new info (fields from the HW rx desc, and the
331 		 * MSDU payload itself). So, the offset into the rx
332 		 * indication message only has to account for the standard
333 		 * offset of the per-MSDU FW rx desc info within the
334 		 * message, and how many bytes of the per-MSDU FW rx desc
335 		 * info have already been consumed. (And the endianness of
336 		 * the host, since for a big-endian host, the rx ind
337 		 * message contents, including the per-MSDU rx desc bytes,
338 		 * were byteswapped during upload.)
339 		 */
340 		if (*fw_desc_len > 0) {
341 			rx_desc->fw_desc.info0 = **fw_desc;
342 			/*
343 			 * The target is expected to only provide the basic
344 			 * per-MSDU rx descriptors. Just to be sure, verify
345 			 * that the target has not attached extension data
346 			 * (e.g. LRO flow ID).
347 			 */
348 
349 			/* or more, if there's extension data */
350 			(*fw_desc)++;
351 			(*fw_desc_len)--;
352 		} else {
353 			/*
354 			 * When an oversized AMSDU happened, FW will lost
355 			 * some of MSDU status - in this case, the FW
356 			 * descriptors provided will be less than the
357 			 * actual MSDUs inside this MPDU. Mark the FW
358 			 * descriptors so that it will still deliver to
359 			 * upper stack, if no CRC error for this MPDU.
360 			 *
361 			 * FIX THIS - the FW descriptors are actually for
362 			 * MSDUs in the end of this A-MSDU instead of the
363 			 * beginning.
364 			 */
365 			rx_desc->fw_desc.info0 = 0;
366 		}
367 
368 		msdu_len_invalid = !!(__le32_to_cpu(rx_desc->attention.flags)
369 					& (RX_ATTENTION_FLAGS_MPDU_LENGTH_ERR |
370 					   RX_ATTENTION_FLAGS_MSDU_LENGTH_ERR));
371 		msdu_len = MS(__le32_to_cpu(rx_desc->msdu_start.common.info0),
372 			      RX_MSDU_START_INFO0_MSDU_LENGTH);
373 		msdu_chained = rx_desc->frag_info.ring2_more_count;
374 
375 		if (msdu_len_invalid)
376 			msdu_len = 0;
377 
378 		skb_trim(msdu, 0);
379 		skb_put(msdu, min(msdu_len, HTT_RX_MSDU_SIZE));
380 		msdu_len -= msdu->len;
381 
382 		/* Note: Chained buffers do not contain rx descriptor */
383 		while (msdu_chained--) {
384 			msdu = ath10k_htt_rx_netbuf_pop(htt);
385 			if (!msdu) {
386 				__skb_queue_purge(amsdu);
387 				return -ENOENT;
388 			}
389 
390 			__skb_queue_tail(amsdu, msdu);
391 			skb_trim(msdu, 0);
392 			skb_put(msdu, min(msdu_len, HTT_RX_BUF_SIZE));
393 			msdu_len -= msdu->len;
394 			msdu_chaining = 1;
395 		}
396 
397 		last_msdu = __le32_to_cpu(rx_desc->msdu_end.common.info0) &
398 				RX_MSDU_END_INFO0_LAST_MSDU;
399 
400 		trace_ath10k_htt_rx_desc(ar, &rx_desc->attention,
401 					 sizeof(*rx_desc) - sizeof(u32));
402 
403 		if (last_msdu)
404 			break;
405 	}
406 
407 	if (skb_queue_empty(amsdu))
408 		msdu_chaining = -1;
409 
410 	/*
411 	 * Don't refill the ring yet.
412 	 *
413 	 * First, the elements popped here are still in use - it is not
414 	 * safe to overwrite them until the matching call to
415 	 * mpdu_desc_list_next. Second, for efficiency it is preferable to
416 	 * refill the rx ring with 1 PPDU's worth of rx buffers (something
417 	 * like 32 x 3 buffers), rather than one MPDU's worth of rx buffers
418 	 * (something like 3 buffers). Consequently, we'll rely on the txrx
419 	 * SW to tell us when it is done pulling all the PPDU's rx buffers
420 	 * out of the rx ring, and then refill it just once.
421 	 */
422 
423 	return msdu_chaining;
424 }
425 
426 static void ath10k_htt_rx_replenish_task(unsigned long ptr)
427 {
428 	struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
429 
430 	ath10k_htt_rx_msdu_buff_replenish(htt);
431 }
432 
433 static struct sk_buff *ath10k_htt_rx_pop_paddr(struct ath10k_htt *htt,
434 					       u32 paddr)
435 {
436 	struct ath10k *ar = htt->ar;
437 	struct ath10k_skb_rxcb *rxcb;
438 	struct sk_buff *msdu;
439 
440 	lockdep_assert_held(&htt->rx_ring.lock);
441 
442 	msdu = ath10k_htt_rx_find_skb_paddr(ar, paddr);
443 	if (!msdu)
444 		return NULL;
445 
446 	rxcb = ATH10K_SKB_RXCB(msdu);
447 	hash_del(&rxcb->hlist);
448 	htt->rx_ring.fill_cnt--;
449 
450 	dma_unmap_single(htt->ar->dev, rxcb->paddr,
451 			 msdu->len + skb_tailroom(msdu),
452 			 DMA_FROM_DEVICE);
453 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt rx netbuf pop: ",
454 			msdu->data, msdu->len + skb_tailroom(msdu));
455 
456 	return msdu;
457 }
458 
459 static int ath10k_htt_rx_pop_paddr_list(struct ath10k_htt *htt,
460 					struct htt_rx_in_ord_ind *ev,
461 					struct sk_buff_head *list)
462 {
463 	struct ath10k *ar = htt->ar;
464 	struct htt_rx_in_ord_msdu_desc *msdu_desc = ev->msdu_descs;
465 	struct htt_rx_desc *rxd;
466 	struct sk_buff *msdu;
467 	int msdu_count;
468 	bool is_offload;
469 	u32 paddr;
470 
471 	lockdep_assert_held(&htt->rx_ring.lock);
472 
473 	msdu_count = __le16_to_cpu(ev->msdu_count);
474 	is_offload = !!(ev->info & HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK);
475 
476 	while (msdu_count--) {
477 		paddr = __le32_to_cpu(msdu_desc->msdu_paddr);
478 
479 		msdu = ath10k_htt_rx_pop_paddr(htt, paddr);
480 		if (!msdu) {
481 			__skb_queue_purge(list);
482 			return -ENOENT;
483 		}
484 
485 		__skb_queue_tail(list, msdu);
486 
487 		if (!is_offload) {
488 			rxd = (void *)msdu->data;
489 
490 			trace_ath10k_htt_rx_desc(ar, rxd, sizeof(*rxd));
491 
492 			skb_put(msdu, sizeof(*rxd));
493 			skb_pull(msdu, sizeof(*rxd));
494 			skb_put(msdu, __le16_to_cpu(msdu_desc->msdu_len));
495 
496 			if (!(__le32_to_cpu(rxd->attention.flags) &
497 			      RX_ATTENTION_FLAGS_MSDU_DONE)) {
498 				ath10k_warn(htt->ar, "tried to pop an incomplete frame, oops!\n");
499 				return -EIO;
500 			}
501 		}
502 
503 		msdu_desc++;
504 	}
505 
506 	return 0;
507 }
508 
509 int ath10k_htt_rx_alloc(struct ath10k_htt *htt)
510 {
511 	struct ath10k *ar = htt->ar;
512 	dma_addr_t paddr;
513 	void *vaddr;
514 	size_t size;
515 	struct timer_list *timer = &htt->rx_ring.refill_retry_timer;
516 
517 	htt->rx_confused = false;
518 
519 	/* XXX: The fill level could be changed during runtime in response to
520 	 * the host processing latency. Is this really worth it?
521 	 */
522 	htt->rx_ring.size = HTT_RX_RING_SIZE;
523 	htt->rx_ring.size_mask = htt->rx_ring.size - 1;
524 	htt->rx_ring.fill_level = HTT_RX_RING_FILL_LEVEL;
525 
526 	if (!is_power_of_2(htt->rx_ring.size)) {
527 		ath10k_warn(ar, "htt rx ring size is not power of 2\n");
528 		return -EINVAL;
529 	}
530 
531 	htt->rx_ring.netbufs_ring =
532 		kzalloc(htt->rx_ring.size * sizeof(struct sk_buff *),
533 			GFP_KERNEL);
534 	if (!htt->rx_ring.netbufs_ring)
535 		goto err_netbuf;
536 
537 	size = htt->rx_ring.size * sizeof(htt->rx_ring.paddrs_ring);
538 
539 	vaddr = dma_alloc_coherent(htt->ar->dev, size, &paddr, GFP_KERNEL);
540 	if (!vaddr)
541 		goto err_dma_ring;
542 
543 	htt->rx_ring.paddrs_ring = vaddr;
544 	htt->rx_ring.base_paddr = paddr;
545 
546 	vaddr = dma_alloc_coherent(htt->ar->dev,
547 				   sizeof(*htt->rx_ring.alloc_idx.vaddr),
548 				   &paddr, GFP_KERNEL);
549 	if (!vaddr)
550 		goto err_dma_idx;
551 
552 	htt->rx_ring.alloc_idx.vaddr = vaddr;
553 	htt->rx_ring.alloc_idx.paddr = paddr;
554 	htt->rx_ring.sw_rd_idx.msdu_payld = htt->rx_ring.size_mask;
555 	*htt->rx_ring.alloc_idx.vaddr = 0;
556 
557 	/* Initialize the Rx refill retry timer */
558 	setup_timer(timer, ath10k_htt_rx_ring_refill_retry, (unsigned long)htt);
559 
560 	spin_lock_init(&htt->rx_ring.lock);
561 
562 	htt->rx_ring.fill_cnt = 0;
563 	htt->rx_ring.sw_rd_idx.msdu_payld = 0;
564 	hash_init(htt->rx_ring.skb_table);
565 
566 	tasklet_init(&htt->rx_replenish_task, ath10k_htt_rx_replenish_task,
567 		     (unsigned long)htt);
568 
569 	skb_queue_head_init(&htt->tx_compl_q);
570 	skb_queue_head_init(&htt->rx_compl_q);
571 	skb_queue_head_init(&htt->rx_in_ord_compl_q);
572 
573 	tasklet_init(&htt->txrx_compl_task, ath10k_htt_txrx_compl_task,
574 		     (unsigned long)htt);
575 
576 	ath10k_dbg(ar, ATH10K_DBG_BOOT, "htt rx ring size %d fill_level %d\n",
577 		   htt->rx_ring.size, htt->rx_ring.fill_level);
578 	return 0;
579 
580 err_dma_idx:
581 	dma_free_coherent(htt->ar->dev,
582 			  (htt->rx_ring.size *
583 			   sizeof(htt->rx_ring.paddrs_ring)),
584 			  htt->rx_ring.paddrs_ring,
585 			  htt->rx_ring.base_paddr);
586 err_dma_ring:
587 	kfree(htt->rx_ring.netbufs_ring);
588 err_netbuf:
589 	return -ENOMEM;
590 }
591 
592 static int ath10k_htt_rx_crypto_param_len(struct ath10k *ar,
593 					  enum htt_rx_mpdu_encrypt_type type)
594 {
595 	switch (type) {
596 	case HTT_RX_MPDU_ENCRYPT_NONE:
597 		return 0;
598 	case HTT_RX_MPDU_ENCRYPT_WEP40:
599 	case HTT_RX_MPDU_ENCRYPT_WEP104:
600 		return IEEE80211_WEP_IV_LEN;
601 	case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
602 	case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
603 		return IEEE80211_TKIP_IV_LEN;
604 	case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
605 		return IEEE80211_CCMP_HDR_LEN;
606 	case HTT_RX_MPDU_ENCRYPT_WEP128:
607 	case HTT_RX_MPDU_ENCRYPT_WAPI:
608 		break;
609 	}
610 
611 	ath10k_warn(ar, "unsupported encryption type %d\n", type);
612 	return 0;
613 }
614 
615 #define MICHAEL_MIC_LEN 8
616 
617 static int ath10k_htt_rx_crypto_tail_len(struct ath10k *ar,
618 					 enum htt_rx_mpdu_encrypt_type type)
619 {
620 	switch (type) {
621 	case HTT_RX_MPDU_ENCRYPT_NONE:
622 		return 0;
623 	case HTT_RX_MPDU_ENCRYPT_WEP40:
624 	case HTT_RX_MPDU_ENCRYPT_WEP104:
625 		return IEEE80211_WEP_ICV_LEN;
626 	case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
627 	case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
628 		return IEEE80211_TKIP_ICV_LEN;
629 	case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
630 		return IEEE80211_CCMP_MIC_LEN;
631 	case HTT_RX_MPDU_ENCRYPT_WEP128:
632 	case HTT_RX_MPDU_ENCRYPT_WAPI:
633 		break;
634 	}
635 
636 	ath10k_warn(ar, "unsupported encryption type %d\n", type);
637 	return 0;
638 }
639 
640 struct amsdu_subframe_hdr {
641 	u8 dst[ETH_ALEN];
642 	u8 src[ETH_ALEN];
643 	__be16 len;
644 } __packed;
645 
646 #define GROUP_ID_IS_SU_MIMO(x) ((x) == 0 || (x) == 63)
647 
648 static void ath10k_htt_rx_h_rates(struct ath10k *ar,
649 				  struct ieee80211_rx_status *status,
650 				  struct htt_rx_desc *rxd)
651 {
652 	struct ieee80211_supported_band *sband;
653 	u8 cck, rate, bw, sgi, mcs, nss;
654 	u8 preamble = 0;
655 	u8 group_id;
656 	u32 info1, info2, info3;
657 
658 	info1 = __le32_to_cpu(rxd->ppdu_start.info1);
659 	info2 = __le32_to_cpu(rxd->ppdu_start.info2);
660 	info3 = __le32_to_cpu(rxd->ppdu_start.info3);
661 
662 	preamble = MS(info1, RX_PPDU_START_INFO1_PREAMBLE_TYPE);
663 
664 	switch (preamble) {
665 	case HTT_RX_LEGACY:
666 		/* To get legacy rate index band is required. Since band can't
667 		 * be undefined check if freq is non-zero.
668 		 */
669 		if (!status->freq)
670 			return;
671 
672 		cck = info1 & RX_PPDU_START_INFO1_L_SIG_RATE_SELECT;
673 		rate = MS(info1, RX_PPDU_START_INFO1_L_SIG_RATE);
674 		rate &= ~RX_PPDU_START_RATE_FLAG;
675 
676 		sband = &ar->mac.sbands[status->band];
677 		status->rate_idx = ath10k_mac_hw_rate_to_idx(sband, rate, cck);
678 		break;
679 	case HTT_RX_HT:
680 	case HTT_RX_HT_WITH_TXBF:
681 		/* HT-SIG - Table 20-11 in info2 and info3 */
682 		mcs = info2 & 0x1F;
683 		nss = mcs >> 3;
684 		bw = (info2 >> 7) & 1;
685 		sgi = (info3 >> 7) & 1;
686 
687 		status->rate_idx = mcs;
688 		status->flag |= RX_FLAG_HT;
689 		if (sgi)
690 			status->flag |= RX_FLAG_SHORT_GI;
691 		if (bw)
692 			status->flag |= RX_FLAG_40MHZ;
693 		break;
694 	case HTT_RX_VHT:
695 	case HTT_RX_VHT_WITH_TXBF:
696 		/* VHT-SIG-A1 in info2, VHT-SIG-A2 in info3
697 		   TODO check this */
698 		bw = info2 & 3;
699 		sgi = info3 & 1;
700 		group_id = (info2 >> 4) & 0x3F;
701 
702 		if (GROUP_ID_IS_SU_MIMO(group_id)) {
703 			mcs = (info3 >> 4) & 0x0F;
704 			nss = ((info2 >> 10) & 0x07) + 1;
705 		} else {
706 			/* Hardware doesn't decode VHT-SIG-B into Rx descriptor
707 			 * so it's impossible to decode MCS. Also since
708 			 * firmware consumes Group Id Management frames host
709 			 * has no knowledge regarding group/user position
710 			 * mapping so it's impossible to pick the correct Nsts
711 			 * from VHT-SIG-A1.
712 			 *
713 			 * Bandwidth and SGI are valid so report the rateinfo
714 			 * on best-effort basis.
715 			 */
716 			mcs = 0;
717 			nss = 1;
718 		}
719 
720 		if (mcs > 0x09) {
721 			ath10k_warn(ar, "invalid MCS received %u\n", mcs);
722 			ath10k_warn(ar, "rxd %08x mpdu start %08x %08x msdu start %08x %08x ppdu start %08x %08x %08x %08x %08x\n",
723 				    __le32_to_cpu(rxd->attention.flags),
724 				    __le32_to_cpu(rxd->mpdu_start.info0),
725 				    __le32_to_cpu(rxd->mpdu_start.info1),
726 				    __le32_to_cpu(rxd->msdu_start.common.info0),
727 				    __le32_to_cpu(rxd->msdu_start.common.info1),
728 				    rxd->ppdu_start.info0,
729 				    __le32_to_cpu(rxd->ppdu_start.info1),
730 				    __le32_to_cpu(rxd->ppdu_start.info2),
731 				    __le32_to_cpu(rxd->ppdu_start.info3),
732 				    __le32_to_cpu(rxd->ppdu_start.info4));
733 
734 			ath10k_warn(ar, "msdu end %08x mpdu end %08x\n",
735 				    __le32_to_cpu(rxd->msdu_end.common.info0),
736 				    __le32_to_cpu(rxd->mpdu_end.info0));
737 
738 			ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL,
739 					"rx desc msdu payload: ",
740 					rxd->msdu_payload, 50);
741 		}
742 
743 		status->rate_idx = mcs;
744 		status->vht_nss = nss;
745 
746 		if (sgi)
747 			status->flag |= RX_FLAG_SHORT_GI;
748 
749 		switch (bw) {
750 		/* 20MHZ */
751 		case 0:
752 			break;
753 		/* 40MHZ */
754 		case 1:
755 			status->flag |= RX_FLAG_40MHZ;
756 			break;
757 		/* 80MHZ */
758 		case 2:
759 			status->vht_flag |= RX_VHT_FLAG_80MHZ;
760 		}
761 
762 		status->flag |= RX_FLAG_VHT;
763 		break;
764 	default:
765 		break;
766 	}
767 }
768 
769 static struct ieee80211_channel *
770 ath10k_htt_rx_h_peer_channel(struct ath10k *ar, struct htt_rx_desc *rxd)
771 {
772 	struct ath10k_peer *peer;
773 	struct ath10k_vif *arvif;
774 	struct cfg80211_chan_def def;
775 	u16 peer_id;
776 
777 	lockdep_assert_held(&ar->data_lock);
778 
779 	if (!rxd)
780 		return NULL;
781 
782 	if (rxd->attention.flags &
783 	    __cpu_to_le32(RX_ATTENTION_FLAGS_PEER_IDX_INVALID))
784 		return NULL;
785 
786 	if (!(rxd->msdu_end.common.info0 &
787 	      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU)))
788 		return NULL;
789 
790 	peer_id = MS(__le32_to_cpu(rxd->mpdu_start.info0),
791 		     RX_MPDU_START_INFO0_PEER_IDX);
792 
793 	peer = ath10k_peer_find_by_id(ar, peer_id);
794 	if (!peer)
795 		return NULL;
796 
797 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
798 	if (WARN_ON_ONCE(!arvif))
799 		return NULL;
800 
801 	if (WARN_ON(ath10k_mac_vif_chan(arvif->vif, &def)))
802 		return NULL;
803 
804 	return def.chan;
805 }
806 
807 static struct ieee80211_channel *
808 ath10k_htt_rx_h_vdev_channel(struct ath10k *ar, u32 vdev_id)
809 {
810 	struct ath10k_vif *arvif;
811 	struct cfg80211_chan_def def;
812 
813 	lockdep_assert_held(&ar->data_lock);
814 
815 	list_for_each_entry(arvif, &ar->arvifs, list) {
816 		if (arvif->vdev_id == vdev_id &&
817 		    ath10k_mac_vif_chan(arvif->vif, &def) == 0)
818 			return def.chan;
819 	}
820 
821 	return NULL;
822 }
823 
824 static void
825 ath10k_htt_rx_h_any_chan_iter(struct ieee80211_hw *hw,
826 			      struct ieee80211_chanctx_conf *conf,
827 			      void *data)
828 {
829 	struct cfg80211_chan_def *def = data;
830 
831 	*def = conf->def;
832 }
833 
834 static struct ieee80211_channel *
835 ath10k_htt_rx_h_any_channel(struct ath10k *ar)
836 {
837 	struct cfg80211_chan_def def = {};
838 
839 	ieee80211_iter_chan_contexts_atomic(ar->hw,
840 					    ath10k_htt_rx_h_any_chan_iter,
841 					    &def);
842 
843 	return def.chan;
844 }
845 
846 static bool ath10k_htt_rx_h_channel(struct ath10k *ar,
847 				    struct ieee80211_rx_status *status,
848 				    struct htt_rx_desc *rxd,
849 				    u32 vdev_id)
850 {
851 	struct ieee80211_channel *ch;
852 
853 	spin_lock_bh(&ar->data_lock);
854 	ch = ar->scan_channel;
855 	if (!ch)
856 		ch = ar->rx_channel;
857 	if (!ch)
858 		ch = ath10k_htt_rx_h_peer_channel(ar, rxd);
859 	if (!ch)
860 		ch = ath10k_htt_rx_h_vdev_channel(ar, vdev_id);
861 	if (!ch)
862 		ch = ath10k_htt_rx_h_any_channel(ar);
863 	spin_unlock_bh(&ar->data_lock);
864 
865 	if (!ch)
866 		return false;
867 
868 	status->band = ch->band;
869 	status->freq = ch->center_freq;
870 
871 	return true;
872 }
873 
874 static void ath10k_htt_rx_h_signal(struct ath10k *ar,
875 				   struct ieee80211_rx_status *status,
876 				   struct htt_rx_desc *rxd)
877 {
878 	/* FIXME: Get real NF */
879 	status->signal = ATH10K_DEFAULT_NOISE_FLOOR +
880 			 rxd->ppdu_start.rssi_comb;
881 	status->flag &= ~RX_FLAG_NO_SIGNAL_VAL;
882 }
883 
884 static void ath10k_htt_rx_h_mactime(struct ath10k *ar,
885 				    struct ieee80211_rx_status *status,
886 				    struct htt_rx_desc *rxd)
887 {
888 	/* FIXME: TSF is known only at the end of PPDU, in the last MPDU. This
889 	 * means all prior MSDUs in a PPDU are reported to mac80211 without the
890 	 * TSF. Is it worth holding frames until end of PPDU is known?
891 	 *
892 	 * FIXME: Can we get/compute 64bit TSF?
893 	 */
894 	status->mactime = __le32_to_cpu(rxd->ppdu_end.common.tsf_timestamp);
895 	status->flag |= RX_FLAG_MACTIME_END;
896 }
897 
898 static void ath10k_htt_rx_h_ppdu(struct ath10k *ar,
899 				 struct sk_buff_head *amsdu,
900 				 struct ieee80211_rx_status *status,
901 				 u32 vdev_id)
902 {
903 	struct sk_buff *first;
904 	struct htt_rx_desc *rxd;
905 	bool is_first_ppdu;
906 	bool is_last_ppdu;
907 
908 	if (skb_queue_empty(amsdu))
909 		return;
910 
911 	first = skb_peek(amsdu);
912 	rxd = (void *)first->data - sizeof(*rxd);
913 
914 	is_first_ppdu = !!(rxd->attention.flags &
915 			   __cpu_to_le32(RX_ATTENTION_FLAGS_FIRST_MPDU));
916 	is_last_ppdu = !!(rxd->attention.flags &
917 			  __cpu_to_le32(RX_ATTENTION_FLAGS_LAST_MPDU));
918 
919 	if (is_first_ppdu) {
920 		/* New PPDU starts so clear out the old per-PPDU status. */
921 		status->freq = 0;
922 		status->rate_idx = 0;
923 		status->vht_nss = 0;
924 		status->vht_flag &= ~RX_VHT_FLAG_80MHZ;
925 		status->flag &= ~(RX_FLAG_HT |
926 				  RX_FLAG_VHT |
927 				  RX_FLAG_SHORT_GI |
928 				  RX_FLAG_40MHZ |
929 				  RX_FLAG_MACTIME_END);
930 		status->flag |= RX_FLAG_NO_SIGNAL_VAL;
931 
932 		ath10k_htt_rx_h_signal(ar, status, rxd);
933 		ath10k_htt_rx_h_channel(ar, status, rxd, vdev_id);
934 		ath10k_htt_rx_h_rates(ar, status, rxd);
935 	}
936 
937 	if (is_last_ppdu)
938 		ath10k_htt_rx_h_mactime(ar, status, rxd);
939 }
940 
941 static const char * const tid_to_ac[] = {
942 	"BE",
943 	"BK",
944 	"BK",
945 	"BE",
946 	"VI",
947 	"VI",
948 	"VO",
949 	"VO",
950 };
951 
952 static char *ath10k_get_tid(struct ieee80211_hdr *hdr, char *out, size_t size)
953 {
954 	u8 *qc;
955 	int tid;
956 
957 	if (!ieee80211_is_data_qos(hdr->frame_control))
958 		return "";
959 
960 	qc = ieee80211_get_qos_ctl(hdr);
961 	tid = *qc & IEEE80211_QOS_CTL_TID_MASK;
962 	if (tid < 8)
963 		snprintf(out, size, "tid %d (%s)", tid, tid_to_ac[tid]);
964 	else
965 		snprintf(out, size, "tid %d", tid);
966 
967 	return out;
968 }
969 
970 static void ath10k_process_rx(struct ath10k *ar,
971 			      struct ieee80211_rx_status *rx_status,
972 			      struct sk_buff *skb)
973 {
974 	struct ieee80211_rx_status *status;
975 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
976 	char tid[32];
977 
978 	status = IEEE80211_SKB_RXCB(skb);
979 	*status = *rx_status;
980 
981 	ath10k_dbg(ar, ATH10K_DBG_DATA,
982 		   "rx skb %p len %u peer %pM %s %s sn %u %s%s%s%s%s %srate_idx %u vht_nss %u freq %u band %u flag 0x%x fcs-err %i mic-err %i amsdu-more %i\n",
983 		   skb,
984 		   skb->len,
985 		   ieee80211_get_SA(hdr),
986 		   ath10k_get_tid(hdr, tid, sizeof(tid)),
987 		   is_multicast_ether_addr(ieee80211_get_DA(hdr)) ?
988 							"mcast" : "ucast",
989 		   (__le16_to_cpu(hdr->seq_ctrl) & IEEE80211_SCTL_SEQ) >> 4,
990 		   status->flag == 0 ? "legacy" : "",
991 		   status->flag & RX_FLAG_HT ? "ht" : "",
992 		   status->flag & RX_FLAG_VHT ? "vht" : "",
993 		   status->flag & RX_FLAG_40MHZ ? "40" : "",
994 		   status->vht_flag & RX_VHT_FLAG_80MHZ ? "80" : "",
995 		   status->flag & RX_FLAG_SHORT_GI ? "sgi " : "",
996 		   status->rate_idx,
997 		   status->vht_nss,
998 		   status->freq,
999 		   status->band, status->flag,
1000 		   !!(status->flag & RX_FLAG_FAILED_FCS_CRC),
1001 		   !!(status->flag & RX_FLAG_MMIC_ERROR),
1002 		   !!(status->flag & RX_FLAG_AMSDU_MORE));
1003 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "rx skb: ",
1004 			skb->data, skb->len);
1005 	trace_ath10k_rx_hdr(ar, skb->data, skb->len);
1006 	trace_ath10k_rx_payload(ar, skb->data, skb->len);
1007 
1008 	ieee80211_rx(ar->hw, skb);
1009 }
1010 
1011 static int ath10k_htt_rx_nwifi_hdrlen(struct ath10k *ar,
1012 				      struct ieee80211_hdr *hdr)
1013 {
1014 	int len = ieee80211_hdrlen(hdr->frame_control);
1015 
1016 	if (!test_bit(ATH10K_FW_FEATURE_NO_NWIFI_DECAP_4ADDR_PADDING,
1017 		      ar->fw_features))
1018 		len = round_up(len, 4);
1019 
1020 	return len;
1021 }
1022 
1023 static void ath10k_htt_rx_h_undecap_raw(struct ath10k *ar,
1024 					struct sk_buff *msdu,
1025 					struct ieee80211_rx_status *status,
1026 					enum htt_rx_mpdu_encrypt_type enctype,
1027 					bool is_decrypted)
1028 {
1029 	struct ieee80211_hdr *hdr;
1030 	struct htt_rx_desc *rxd;
1031 	size_t hdr_len;
1032 	size_t crypto_len;
1033 	bool is_first;
1034 	bool is_last;
1035 
1036 	rxd = (void *)msdu->data - sizeof(*rxd);
1037 	is_first = !!(rxd->msdu_end.common.info0 &
1038 		      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU));
1039 	is_last = !!(rxd->msdu_end.common.info0 &
1040 		     __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU));
1041 
1042 	/* Delivered decapped frame:
1043 	 * [802.11 header]
1044 	 * [crypto param] <-- can be trimmed if !fcs_err &&
1045 	 *                    !decrypt_err && !peer_idx_invalid
1046 	 * [amsdu header] <-- only if A-MSDU
1047 	 * [rfc1042/llc]
1048 	 * [payload]
1049 	 * [FCS] <-- at end, needs to be trimmed
1050 	 */
1051 
1052 	/* This probably shouldn't happen but warn just in case */
1053 	if (unlikely(WARN_ON_ONCE(!is_first)))
1054 		return;
1055 
1056 	/* This probably shouldn't happen but warn just in case */
1057 	if (unlikely(WARN_ON_ONCE(!(is_first && is_last))))
1058 		return;
1059 
1060 	skb_trim(msdu, msdu->len - FCS_LEN);
1061 
1062 	/* In most cases this will be true for sniffed frames. It makes sense
1063 	 * to deliver them as-is without stripping the crypto param. This is
1064 	 * necessary for software based decryption.
1065 	 *
1066 	 * If there's no error then the frame is decrypted. At least that is
1067 	 * the case for frames that come in via fragmented rx indication.
1068 	 */
1069 	if (!is_decrypted)
1070 		return;
1071 
1072 	/* The payload is decrypted so strip crypto params. Start from tail
1073 	 * since hdr is used to compute some stuff.
1074 	 */
1075 
1076 	hdr = (void *)msdu->data;
1077 
1078 	/* Tail */
1079 	skb_trim(msdu, msdu->len - ath10k_htt_rx_crypto_tail_len(ar, enctype));
1080 
1081 	/* MMIC */
1082 	if (!ieee80211_has_morefrags(hdr->frame_control) &&
1083 	    enctype == HTT_RX_MPDU_ENCRYPT_TKIP_WPA)
1084 		skb_trim(msdu, msdu->len - 8);
1085 
1086 	/* Head */
1087 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1088 	crypto_len = ath10k_htt_rx_crypto_param_len(ar, enctype);
1089 
1090 	memmove((void *)msdu->data + crypto_len,
1091 		(void *)msdu->data, hdr_len);
1092 	skb_pull(msdu, crypto_len);
1093 }
1094 
1095 static void ath10k_htt_rx_h_undecap_nwifi(struct ath10k *ar,
1096 					  struct sk_buff *msdu,
1097 					  struct ieee80211_rx_status *status,
1098 					  const u8 first_hdr[64])
1099 {
1100 	struct ieee80211_hdr *hdr;
1101 	size_t hdr_len;
1102 	u8 da[ETH_ALEN];
1103 	u8 sa[ETH_ALEN];
1104 
1105 	/* Delivered decapped frame:
1106 	 * [nwifi 802.11 header] <-- replaced with 802.11 hdr
1107 	 * [rfc1042/llc]
1108 	 *
1109 	 * Note: The nwifi header doesn't have QoS Control and is
1110 	 * (always?) a 3addr frame.
1111 	 *
1112 	 * Note2: There's no A-MSDU subframe header. Even if it's part
1113 	 * of an A-MSDU.
1114 	 */
1115 
1116 	/* pull decapped header and copy SA & DA */
1117 	if ((ar->hw_params.hw_4addr_pad == ATH10K_HW_4ADDR_PAD_BEFORE) &&
1118 	    ieee80211_has_a4(((struct ieee80211_hdr *)first_hdr)->frame_control)) {
1119 		/* The QCA99X0 4 address mode pad 2 bytes at the
1120 		 * beginning of MSDU
1121 		 */
1122 		hdr = (struct ieee80211_hdr *)(msdu->data + 2);
1123 		/* The skb length need be extended 2 as the 2 bytes at the tail
1124 		 * be excluded due to the padding
1125 		 */
1126 		skb_put(msdu, 2);
1127 	} else {
1128 		hdr = (struct ieee80211_hdr *)(msdu->data);
1129 	}
1130 
1131 	hdr_len = ath10k_htt_rx_nwifi_hdrlen(ar, hdr);
1132 	ether_addr_copy(da, ieee80211_get_DA(hdr));
1133 	ether_addr_copy(sa, ieee80211_get_SA(hdr));
1134 	skb_pull(msdu, hdr_len);
1135 
1136 	/* push original 802.11 header */
1137 	hdr = (struct ieee80211_hdr *)first_hdr;
1138 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1139 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1140 
1141 	/* original 802.11 header has a different DA and in
1142 	 * case of 4addr it may also have different SA
1143 	 */
1144 	hdr = (struct ieee80211_hdr *)msdu->data;
1145 	ether_addr_copy(ieee80211_get_DA(hdr), da);
1146 	ether_addr_copy(ieee80211_get_SA(hdr), sa);
1147 }
1148 
1149 static void *ath10k_htt_rx_h_find_rfc1042(struct ath10k *ar,
1150 					  struct sk_buff *msdu,
1151 					  enum htt_rx_mpdu_encrypt_type enctype)
1152 {
1153 	struct ieee80211_hdr *hdr;
1154 	struct htt_rx_desc *rxd;
1155 	size_t hdr_len, crypto_len;
1156 	void *rfc1042;
1157 	bool is_first, is_last, is_amsdu;
1158 
1159 	rxd = (void *)msdu->data - sizeof(*rxd);
1160 	hdr = (void *)rxd->rx_hdr_status;
1161 
1162 	is_first = !!(rxd->msdu_end.common.info0 &
1163 		      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU));
1164 	is_last = !!(rxd->msdu_end.common.info0 &
1165 		     __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU));
1166 	is_amsdu = !(is_first && is_last);
1167 
1168 	rfc1042 = hdr;
1169 
1170 	if (is_first) {
1171 		hdr_len = ieee80211_hdrlen(hdr->frame_control);
1172 		crypto_len = ath10k_htt_rx_crypto_param_len(ar, enctype);
1173 
1174 		rfc1042 += round_up(hdr_len, 4) +
1175 			   round_up(crypto_len, 4);
1176 	}
1177 
1178 	if (is_amsdu)
1179 		rfc1042 += sizeof(struct amsdu_subframe_hdr);
1180 
1181 	return rfc1042;
1182 }
1183 
1184 static void ath10k_htt_rx_h_undecap_eth(struct ath10k *ar,
1185 					struct sk_buff *msdu,
1186 					struct ieee80211_rx_status *status,
1187 					const u8 first_hdr[64],
1188 					enum htt_rx_mpdu_encrypt_type enctype)
1189 {
1190 	struct ieee80211_hdr *hdr;
1191 	struct ethhdr *eth;
1192 	size_t hdr_len;
1193 	void *rfc1042;
1194 	u8 da[ETH_ALEN];
1195 	u8 sa[ETH_ALEN];
1196 
1197 	/* Delivered decapped frame:
1198 	 * [eth header] <-- replaced with 802.11 hdr & rfc1042/llc
1199 	 * [payload]
1200 	 */
1201 
1202 	rfc1042 = ath10k_htt_rx_h_find_rfc1042(ar, msdu, enctype);
1203 	if (WARN_ON_ONCE(!rfc1042))
1204 		return;
1205 
1206 	/* pull decapped header and copy SA & DA */
1207 	eth = (struct ethhdr *)msdu->data;
1208 	ether_addr_copy(da, eth->h_dest);
1209 	ether_addr_copy(sa, eth->h_source);
1210 	skb_pull(msdu, sizeof(struct ethhdr));
1211 
1212 	/* push rfc1042/llc/snap */
1213 	memcpy(skb_push(msdu, sizeof(struct rfc1042_hdr)), rfc1042,
1214 	       sizeof(struct rfc1042_hdr));
1215 
1216 	/* push original 802.11 header */
1217 	hdr = (struct ieee80211_hdr *)first_hdr;
1218 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1219 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1220 
1221 	/* original 802.11 header has a different DA and in
1222 	 * case of 4addr it may also have different SA
1223 	 */
1224 	hdr = (struct ieee80211_hdr *)msdu->data;
1225 	ether_addr_copy(ieee80211_get_DA(hdr), da);
1226 	ether_addr_copy(ieee80211_get_SA(hdr), sa);
1227 }
1228 
1229 static void ath10k_htt_rx_h_undecap_snap(struct ath10k *ar,
1230 					 struct sk_buff *msdu,
1231 					 struct ieee80211_rx_status *status,
1232 					 const u8 first_hdr[64])
1233 {
1234 	struct ieee80211_hdr *hdr;
1235 	size_t hdr_len;
1236 
1237 	/* Delivered decapped frame:
1238 	 * [amsdu header] <-- replaced with 802.11 hdr
1239 	 * [rfc1042/llc]
1240 	 * [payload]
1241 	 */
1242 
1243 	skb_pull(msdu, sizeof(struct amsdu_subframe_hdr));
1244 
1245 	hdr = (struct ieee80211_hdr *)first_hdr;
1246 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1247 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1248 }
1249 
1250 static void ath10k_htt_rx_h_undecap(struct ath10k *ar,
1251 				    struct sk_buff *msdu,
1252 				    struct ieee80211_rx_status *status,
1253 				    u8 first_hdr[64],
1254 				    enum htt_rx_mpdu_encrypt_type enctype,
1255 				    bool is_decrypted)
1256 {
1257 	struct htt_rx_desc *rxd;
1258 	enum rx_msdu_decap_format decap;
1259 
1260 	/* First msdu's decapped header:
1261 	 * [802.11 header] <-- padded to 4 bytes long
1262 	 * [crypto param] <-- padded to 4 bytes long
1263 	 * [amsdu header] <-- only if A-MSDU
1264 	 * [rfc1042/llc]
1265 	 *
1266 	 * Other (2nd, 3rd, ..) msdu's decapped header:
1267 	 * [amsdu header] <-- only if A-MSDU
1268 	 * [rfc1042/llc]
1269 	 */
1270 
1271 	rxd = (void *)msdu->data - sizeof(*rxd);
1272 	decap = MS(__le32_to_cpu(rxd->msdu_start.common.info1),
1273 		   RX_MSDU_START_INFO1_DECAP_FORMAT);
1274 
1275 	switch (decap) {
1276 	case RX_MSDU_DECAP_RAW:
1277 		ath10k_htt_rx_h_undecap_raw(ar, msdu, status, enctype,
1278 					    is_decrypted);
1279 		break;
1280 	case RX_MSDU_DECAP_NATIVE_WIFI:
1281 		ath10k_htt_rx_h_undecap_nwifi(ar, msdu, status, first_hdr);
1282 		break;
1283 	case RX_MSDU_DECAP_ETHERNET2_DIX:
1284 		ath10k_htt_rx_h_undecap_eth(ar, msdu, status, first_hdr, enctype);
1285 		break;
1286 	case RX_MSDU_DECAP_8023_SNAP_LLC:
1287 		ath10k_htt_rx_h_undecap_snap(ar, msdu, status, first_hdr);
1288 		break;
1289 	}
1290 }
1291 
1292 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb)
1293 {
1294 	struct htt_rx_desc *rxd;
1295 	u32 flags, info;
1296 	bool is_ip4, is_ip6;
1297 	bool is_tcp, is_udp;
1298 	bool ip_csum_ok, tcpudp_csum_ok;
1299 
1300 	rxd = (void *)skb->data - sizeof(*rxd);
1301 	flags = __le32_to_cpu(rxd->attention.flags);
1302 	info = __le32_to_cpu(rxd->msdu_start.common.info1);
1303 
1304 	is_ip4 = !!(info & RX_MSDU_START_INFO1_IPV4_PROTO);
1305 	is_ip6 = !!(info & RX_MSDU_START_INFO1_IPV6_PROTO);
1306 	is_tcp = !!(info & RX_MSDU_START_INFO1_TCP_PROTO);
1307 	is_udp = !!(info & RX_MSDU_START_INFO1_UDP_PROTO);
1308 	ip_csum_ok = !(flags & RX_ATTENTION_FLAGS_IP_CHKSUM_FAIL);
1309 	tcpudp_csum_ok = !(flags & RX_ATTENTION_FLAGS_TCP_UDP_CHKSUM_FAIL);
1310 
1311 	if (!is_ip4 && !is_ip6)
1312 		return CHECKSUM_NONE;
1313 	if (!is_tcp && !is_udp)
1314 		return CHECKSUM_NONE;
1315 	if (!ip_csum_ok)
1316 		return CHECKSUM_NONE;
1317 	if (!tcpudp_csum_ok)
1318 		return CHECKSUM_NONE;
1319 
1320 	return CHECKSUM_UNNECESSARY;
1321 }
1322 
1323 static void ath10k_htt_rx_h_csum_offload(struct sk_buff *msdu)
1324 {
1325 	msdu->ip_summed = ath10k_htt_rx_get_csum_state(msdu);
1326 }
1327 
1328 static void ath10k_htt_rx_h_mpdu(struct ath10k *ar,
1329 				 struct sk_buff_head *amsdu,
1330 				 struct ieee80211_rx_status *status)
1331 {
1332 	struct sk_buff *first;
1333 	struct sk_buff *last;
1334 	struct sk_buff *msdu;
1335 	struct htt_rx_desc *rxd;
1336 	struct ieee80211_hdr *hdr;
1337 	enum htt_rx_mpdu_encrypt_type enctype;
1338 	u8 first_hdr[64];
1339 	u8 *qos;
1340 	size_t hdr_len;
1341 	bool has_fcs_err;
1342 	bool has_crypto_err;
1343 	bool has_tkip_err;
1344 	bool has_peer_idx_invalid;
1345 	bool is_decrypted;
1346 	u32 attention;
1347 
1348 	if (skb_queue_empty(amsdu))
1349 		return;
1350 
1351 	first = skb_peek(amsdu);
1352 	rxd = (void *)first->data - sizeof(*rxd);
1353 
1354 	enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
1355 		     RX_MPDU_START_INFO0_ENCRYPT_TYPE);
1356 
1357 	/* First MSDU's Rx descriptor in an A-MSDU contains full 802.11
1358 	 * decapped header. It'll be used for undecapping of each MSDU.
1359 	 */
1360 	hdr = (void *)rxd->rx_hdr_status;
1361 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1362 	memcpy(first_hdr, hdr, hdr_len);
1363 
1364 	/* Each A-MSDU subframe will use the original header as the base and be
1365 	 * reported as a separate MSDU so strip the A-MSDU bit from QoS Ctl.
1366 	 */
1367 	hdr = (void *)first_hdr;
1368 	qos = ieee80211_get_qos_ctl(hdr);
1369 	qos[0] &= ~IEEE80211_QOS_CTL_A_MSDU_PRESENT;
1370 
1371 	/* Some attention flags are valid only in the last MSDU. */
1372 	last = skb_peek_tail(amsdu);
1373 	rxd = (void *)last->data - sizeof(*rxd);
1374 	attention = __le32_to_cpu(rxd->attention.flags);
1375 
1376 	has_fcs_err = !!(attention & RX_ATTENTION_FLAGS_FCS_ERR);
1377 	has_crypto_err = !!(attention & RX_ATTENTION_FLAGS_DECRYPT_ERR);
1378 	has_tkip_err = !!(attention & RX_ATTENTION_FLAGS_TKIP_MIC_ERR);
1379 	has_peer_idx_invalid = !!(attention & RX_ATTENTION_FLAGS_PEER_IDX_INVALID);
1380 
1381 	/* Note: If hardware captures an encrypted frame that it can't decrypt,
1382 	 * e.g. due to fcs error, missing peer or invalid key data it will
1383 	 * report the frame as raw.
1384 	 */
1385 	is_decrypted = (enctype != HTT_RX_MPDU_ENCRYPT_NONE &&
1386 			!has_fcs_err &&
1387 			!has_crypto_err &&
1388 			!has_peer_idx_invalid);
1389 
1390 	/* Clear per-MPDU flags while leaving per-PPDU flags intact. */
1391 	status->flag &= ~(RX_FLAG_FAILED_FCS_CRC |
1392 			  RX_FLAG_MMIC_ERROR |
1393 			  RX_FLAG_DECRYPTED |
1394 			  RX_FLAG_IV_STRIPPED |
1395 			  RX_FLAG_MMIC_STRIPPED);
1396 
1397 	if (has_fcs_err)
1398 		status->flag |= RX_FLAG_FAILED_FCS_CRC;
1399 
1400 	if (has_tkip_err)
1401 		status->flag |= RX_FLAG_MMIC_ERROR;
1402 
1403 	if (is_decrypted)
1404 		status->flag |= RX_FLAG_DECRYPTED |
1405 				RX_FLAG_IV_STRIPPED |
1406 				RX_FLAG_MMIC_STRIPPED;
1407 
1408 	skb_queue_walk(amsdu, msdu) {
1409 		ath10k_htt_rx_h_csum_offload(msdu);
1410 		ath10k_htt_rx_h_undecap(ar, msdu, status, first_hdr, enctype,
1411 					is_decrypted);
1412 
1413 		/* Undecapping involves copying the original 802.11 header back
1414 		 * to sk_buff. If frame is protected and hardware has decrypted
1415 		 * it then remove the protected bit.
1416 		 */
1417 		if (!is_decrypted)
1418 			continue;
1419 
1420 		hdr = (void *)msdu->data;
1421 		hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
1422 	}
1423 }
1424 
1425 static void ath10k_htt_rx_h_deliver(struct ath10k *ar,
1426 				    struct sk_buff_head *amsdu,
1427 				    struct ieee80211_rx_status *status)
1428 {
1429 	struct sk_buff *msdu;
1430 
1431 	while ((msdu = __skb_dequeue(amsdu))) {
1432 		/* Setup per-MSDU flags */
1433 		if (skb_queue_empty(amsdu))
1434 			status->flag &= ~RX_FLAG_AMSDU_MORE;
1435 		else
1436 			status->flag |= RX_FLAG_AMSDU_MORE;
1437 
1438 		ath10k_process_rx(ar, status, msdu);
1439 	}
1440 }
1441 
1442 static int ath10k_unchain_msdu(struct sk_buff_head *amsdu)
1443 {
1444 	struct sk_buff *skb, *first;
1445 	int space;
1446 	int total_len = 0;
1447 
1448 	/* TODO:  Might could optimize this by using
1449 	 * skb_try_coalesce or similar method to
1450 	 * decrease copying, or maybe get mac80211 to
1451 	 * provide a way to just receive a list of
1452 	 * skb?
1453 	 */
1454 
1455 	first = __skb_dequeue(amsdu);
1456 
1457 	/* Allocate total length all at once. */
1458 	skb_queue_walk(amsdu, skb)
1459 		total_len += skb->len;
1460 
1461 	space = total_len - skb_tailroom(first);
1462 	if ((space > 0) &&
1463 	    (pskb_expand_head(first, 0, space, GFP_ATOMIC) < 0)) {
1464 		/* TODO:  bump some rx-oom error stat */
1465 		/* put it back together so we can free the
1466 		 * whole list at once.
1467 		 */
1468 		__skb_queue_head(amsdu, first);
1469 		return -1;
1470 	}
1471 
1472 	/* Walk list again, copying contents into
1473 	 * msdu_head
1474 	 */
1475 	while ((skb = __skb_dequeue(amsdu))) {
1476 		skb_copy_from_linear_data(skb, skb_put(first, skb->len),
1477 					  skb->len);
1478 		dev_kfree_skb_any(skb);
1479 	}
1480 
1481 	__skb_queue_head(amsdu, first);
1482 	return 0;
1483 }
1484 
1485 static void ath10k_htt_rx_h_unchain(struct ath10k *ar,
1486 				    struct sk_buff_head *amsdu,
1487 				    bool chained)
1488 {
1489 	struct sk_buff *first;
1490 	struct htt_rx_desc *rxd;
1491 	enum rx_msdu_decap_format decap;
1492 
1493 	first = skb_peek(amsdu);
1494 	rxd = (void *)first->data - sizeof(*rxd);
1495 	decap = MS(__le32_to_cpu(rxd->msdu_start.common.info1),
1496 		   RX_MSDU_START_INFO1_DECAP_FORMAT);
1497 
1498 	if (!chained)
1499 		return;
1500 
1501 	/* FIXME: Current unchaining logic can only handle simple case of raw
1502 	 * msdu chaining. If decapping is other than raw the chaining may be
1503 	 * more complex and this isn't handled by the current code. Don't even
1504 	 * try re-constructing such frames - it'll be pretty much garbage.
1505 	 */
1506 	if (decap != RX_MSDU_DECAP_RAW ||
1507 	    skb_queue_len(amsdu) != 1 + rxd->frag_info.ring2_more_count) {
1508 		__skb_queue_purge(amsdu);
1509 		return;
1510 	}
1511 
1512 	ath10k_unchain_msdu(amsdu);
1513 }
1514 
1515 static bool ath10k_htt_rx_amsdu_allowed(struct ath10k *ar,
1516 					struct sk_buff_head *amsdu,
1517 					struct ieee80211_rx_status *rx_status)
1518 {
1519 	struct sk_buff *msdu;
1520 	struct htt_rx_desc *rxd;
1521 	bool is_mgmt;
1522 	bool has_fcs_err;
1523 
1524 	msdu = skb_peek(amsdu);
1525 	rxd = (void *)msdu->data - sizeof(*rxd);
1526 
1527 	/* FIXME: It might be a good idea to do some fuzzy-testing to drop
1528 	 * invalid/dangerous frames.
1529 	 */
1530 
1531 	if (!rx_status->freq) {
1532 		ath10k_warn(ar, "no channel configured; ignoring frame(s)!\n");
1533 		return false;
1534 	}
1535 
1536 	is_mgmt = !!(rxd->attention.flags &
1537 		     __cpu_to_le32(RX_ATTENTION_FLAGS_MGMT_TYPE));
1538 	has_fcs_err = !!(rxd->attention.flags &
1539 			 __cpu_to_le32(RX_ATTENTION_FLAGS_FCS_ERR));
1540 
1541 	/* Management frames are handled via WMI events. The pros of such
1542 	 * approach is that channel is explicitly provided in WMI events
1543 	 * whereas HTT doesn't provide channel information for Rxed frames.
1544 	 *
1545 	 * However some firmware revisions don't report corrupted frames via
1546 	 * WMI so don't drop them.
1547 	 */
1548 	if (is_mgmt && !has_fcs_err) {
1549 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx mgmt ctrl\n");
1550 		return false;
1551 	}
1552 
1553 	if (test_bit(ATH10K_CAC_RUNNING, &ar->dev_flags)) {
1554 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx cac running\n");
1555 		return false;
1556 	}
1557 
1558 	return true;
1559 }
1560 
1561 static void ath10k_htt_rx_h_filter(struct ath10k *ar,
1562 				   struct sk_buff_head *amsdu,
1563 				   struct ieee80211_rx_status *rx_status)
1564 {
1565 	if (skb_queue_empty(amsdu))
1566 		return;
1567 
1568 	if (ath10k_htt_rx_amsdu_allowed(ar, amsdu, rx_status))
1569 		return;
1570 
1571 	__skb_queue_purge(amsdu);
1572 }
1573 
1574 static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
1575 				  struct htt_rx_indication *rx)
1576 {
1577 	struct ath10k *ar = htt->ar;
1578 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1579 	struct htt_rx_indication_mpdu_range *mpdu_ranges;
1580 	struct sk_buff_head amsdu;
1581 	int num_mpdu_ranges;
1582 	int fw_desc_len;
1583 	u8 *fw_desc;
1584 	int i, ret, mpdu_count = 0;
1585 
1586 	lockdep_assert_held(&htt->rx_ring.lock);
1587 
1588 	if (htt->rx_confused)
1589 		return;
1590 
1591 	fw_desc_len = __le16_to_cpu(rx->prefix.fw_rx_desc_bytes);
1592 	fw_desc = (u8 *)&rx->fw_desc;
1593 
1594 	num_mpdu_ranges = MS(__le32_to_cpu(rx->hdr.info1),
1595 			     HTT_RX_INDICATION_INFO1_NUM_MPDU_RANGES);
1596 	mpdu_ranges = htt_rx_ind_get_mpdu_ranges(rx);
1597 
1598 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt rx ind: ",
1599 			rx, sizeof(*rx) +
1600 			(sizeof(struct htt_rx_indication_mpdu_range) *
1601 				num_mpdu_ranges));
1602 
1603 	for (i = 0; i < num_mpdu_ranges; i++)
1604 		mpdu_count += mpdu_ranges[i].mpdu_count;
1605 
1606 	while (mpdu_count--) {
1607 		__skb_queue_head_init(&amsdu);
1608 		ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc,
1609 					      &fw_desc_len, &amsdu);
1610 		if (ret < 0) {
1611 			ath10k_warn(ar, "rx ring became corrupted: %d\n", ret);
1612 			__skb_queue_purge(&amsdu);
1613 			/* FIXME: It's probably a good idea to reboot the
1614 			 * device instead of leaving it inoperable.
1615 			 */
1616 			htt->rx_confused = true;
1617 			break;
1618 		}
1619 
1620 		ath10k_htt_rx_h_ppdu(ar, &amsdu, rx_status, 0xffff);
1621 		ath10k_htt_rx_h_unchain(ar, &amsdu, ret > 0);
1622 		ath10k_htt_rx_h_filter(ar, &amsdu, rx_status);
1623 		ath10k_htt_rx_h_mpdu(ar, &amsdu, rx_status);
1624 		ath10k_htt_rx_h_deliver(ar, &amsdu, rx_status);
1625 	}
1626 
1627 	tasklet_schedule(&htt->rx_replenish_task);
1628 }
1629 
1630 static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
1631 				       struct htt_rx_fragment_indication *frag)
1632 {
1633 	struct ath10k *ar = htt->ar;
1634 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1635 	struct sk_buff_head amsdu;
1636 	int ret;
1637 	u8 *fw_desc;
1638 	int fw_desc_len;
1639 
1640 	fw_desc_len = __le16_to_cpu(frag->fw_rx_desc_bytes);
1641 	fw_desc = (u8 *)frag->fw_msdu_rx_desc;
1642 
1643 	__skb_queue_head_init(&amsdu);
1644 
1645 	spin_lock_bh(&htt->rx_ring.lock);
1646 	ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc, &fw_desc_len,
1647 				      &amsdu);
1648 	spin_unlock_bh(&htt->rx_ring.lock);
1649 
1650 	tasklet_schedule(&htt->rx_replenish_task);
1651 
1652 	ath10k_dbg(ar, ATH10K_DBG_HTT_DUMP, "htt rx frag ahead\n");
1653 
1654 	if (ret) {
1655 		ath10k_warn(ar, "failed to pop amsdu from httr rx ring for fragmented rx %d\n",
1656 			    ret);
1657 		__skb_queue_purge(&amsdu);
1658 		return;
1659 	}
1660 
1661 	if (skb_queue_len(&amsdu) != 1) {
1662 		ath10k_warn(ar, "failed to pop frag amsdu: too many msdus\n");
1663 		__skb_queue_purge(&amsdu);
1664 		return;
1665 	}
1666 
1667 	ath10k_htt_rx_h_ppdu(ar, &amsdu, rx_status, 0xffff);
1668 	ath10k_htt_rx_h_filter(ar, &amsdu, rx_status);
1669 	ath10k_htt_rx_h_mpdu(ar, &amsdu, rx_status);
1670 	ath10k_htt_rx_h_deliver(ar, &amsdu, rx_status);
1671 
1672 	if (fw_desc_len > 0) {
1673 		ath10k_dbg(ar, ATH10K_DBG_HTT,
1674 			   "expecting more fragmented rx in one indication %d\n",
1675 			   fw_desc_len);
1676 	}
1677 }
1678 
1679 static void ath10k_htt_rx_frm_tx_compl(struct ath10k *ar,
1680 				       struct sk_buff *skb)
1681 {
1682 	struct ath10k_htt *htt = &ar->htt;
1683 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1684 	struct htt_tx_done tx_done = {};
1685 	int status = MS(resp->data_tx_completion.flags, HTT_DATA_TX_STATUS);
1686 	__le16 msdu_id;
1687 	int i;
1688 
1689 	switch (status) {
1690 	case HTT_DATA_TX_STATUS_NO_ACK:
1691 		tx_done.no_ack = true;
1692 		break;
1693 	case HTT_DATA_TX_STATUS_OK:
1694 		tx_done.success = true;
1695 		break;
1696 	case HTT_DATA_TX_STATUS_DISCARD:
1697 	case HTT_DATA_TX_STATUS_POSTPONE:
1698 	case HTT_DATA_TX_STATUS_DOWNLOAD_FAIL:
1699 		tx_done.discard = true;
1700 		break;
1701 	default:
1702 		ath10k_warn(ar, "unhandled tx completion status %d\n", status);
1703 		tx_done.discard = true;
1704 		break;
1705 	}
1706 
1707 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx completion num_msdus %d\n",
1708 		   resp->data_tx_completion.num_msdus);
1709 
1710 	for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
1711 		msdu_id = resp->data_tx_completion.msdus[i];
1712 		tx_done.msdu_id = __le16_to_cpu(msdu_id);
1713 		ath10k_txrx_tx_unref(htt, &tx_done);
1714 	}
1715 }
1716 
1717 static void ath10k_htt_rx_addba(struct ath10k *ar, struct htt_resp *resp)
1718 {
1719 	struct htt_rx_addba *ev = &resp->rx_addba;
1720 	struct ath10k_peer *peer;
1721 	struct ath10k_vif *arvif;
1722 	u16 info0, tid, peer_id;
1723 
1724 	info0 = __le16_to_cpu(ev->info0);
1725 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1726 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1727 
1728 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1729 		   "htt rx addba tid %hu peer_id %hu size %hhu\n",
1730 		   tid, peer_id, ev->window_size);
1731 
1732 	spin_lock_bh(&ar->data_lock);
1733 	peer = ath10k_peer_find_by_id(ar, peer_id);
1734 	if (!peer) {
1735 		ath10k_warn(ar, "received addba event for invalid peer_id: %hu\n",
1736 			    peer_id);
1737 		spin_unlock_bh(&ar->data_lock);
1738 		return;
1739 	}
1740 
1741 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1742 	if (!arvif) {
1743 		ath10k_warn(ar, "received addba event for invalid vdev_id: %u\n",
1744 			    peer->vdev_id);
1745 		spin_unlock_bh(&ar->data_lock);
1746 		return;
1747 	}
1748 
1749 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1750 		   "htt rx start rx ba session sta %pM tid %hu size %hhu\n",
1751 		   peer->addr, tid, ev->window_size);
1752 
1753 	ieee80211_start_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1754 	spin_unlock_bh(&ar->data_lock);
1755 }
1756 
1757 static void ath10k_htt_rx_delba(struct ath10k *ar, struct htt_resp *resp)
1758 {
1759 	struct htt_rx_delba *ev = &resp->rx_delba;
1760 	struct ath10k_peer *peer;
1761 	struct ath10k_vif *arvif;
1762 	u16 info0, tid, peer_id;
1763 
1764 	info0 = __le16_to_cpu(ev->info0);
1765 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1766 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1767 
1768 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1769 		   "htt rx delba tid %hu peer_id %hu\n",
1770 		   tid, peer_id);
1771 
1772 	spin_lock_bh(&ar->data_lock);
1773 	peer = ath10k_peer_find_by_id(ar, peer_id);
1774 	if (!peer) {
1775 		ath10k_warn(ar, "received addba event for invalid peer_id: %hu\n",
1776 			    peer_id);
1777 		spin_unlock_bh(&ar->data_lock);
1778 		return;
1779 	}
1780 
1781 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1782 	if (!arvif) {
1783 		ath10k_warn(ar, "received addba event for invalid vdev_id: %u\n",
1784 			    peer->vdev_id);
1785 		spin_unlock_bh(&ar->data_lock);
1786 		return;
1787 	}
1788 
1789 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1790 		   "htt rx stop rx ba session sta %pM tid %hu\n",
1791 		   peer->addr, tid);
1792 
1793 	ieee80211_stop_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1794 	spin_unlock_bh(&ar->data_lock);
1795 }
1796 
1797 static int ath10k_htt_rx_extract_amsdu(struct sk_buff_head *list,
1798 				       struct sk_buff_head *amsdu)
1799 {
1800 	struct sk_buff *msdu;
1801 	struct htt_rx_desc *rxd;
1802 
1803 	if (skb_queue_empty(list))
1804 		return -ENOBUFS;
1805 
1806 	if (WARN_ON(!skb_queue_empty(amsdu)))
1807 		return -EINVAL;
1808 
1809 	while ((msdu = __skb_dequeue(list))) {
1810 		__skb_queue_tail(amsdu, msdu);
1811 
1812 		rxd = (void *)msdu->data - sizeof(*rxd);
1813 		if (rxd->msdu_end.common.info0 &
1814 		    __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))
1815 			break;
1816 	}
1817 
1818 	msdu = skb_peek_tail(amsdu);
1819 	rxd = (void *)msdu->data - sizeof(*rxd);
1820 	if (!(rxd->msdu_end.common.info0 &
1821 	      __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))) {
1822 		skb_queue_splice_init(amsdu, list);
1823 		return -EAGAIN;
1824 	}
1825 
1826 	return 0;
1827 }
1828 
1829 static void ath10k_htt_rx_h_rx_offload_prot(struct ieee80211_rx_status *status,
1830 					    struct sk_buff *skb)
1831 {
1832 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1833 
1834 	if (!ieee80211_has_protected(hdr->frame_control))
1835 		return;
1836 
1837 	/* Offloaded frames are already decrypted but firmware insists they are
1838 	 * protected in the 802.11 header. Strip the flag.  Otherwise mac80211
1839 	 * will drop the frame.
1840 	 */
1841 
1842 	hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
1843 	status->flag |= RX_FLAG_DECRYPTED |
1844 			RX_FLAG_IV_STRIPPED |
1845 			RX_FLAG_MMIC_STRIPPED;
1846 }
1847 
1848 static void ath10k_htt_rx_h_rx_offload(struct ath10k *ar,
1849 				       struct sk_buff_head *list)
1850 {
1851 	struct ath10k_htt *htt = &ar->htt;
1852 	struct ieee80211_rx_status *status = &htt->rx_status;
1853 	struct htt_rx_offload_msdu *rx;
1854 	struct sk_buff *msdu;
1855 	size_t offset;
1856 
1857 	while ((msdu = __skb_dequeue(list))) {
1858 		/* Offloaded frames don't have Rx descriptor. Instead they have
1859 		 * a short meta information header.
1860 		 */
1861 
1862 		rx = (void *)msdu->data;
1863 
1864 		skb_put(msdu, sizeof(*rx));
1865 		skb_pull(msdu, sizeof(*rx));
1866 
1867 		if (skb_tailroom(msdu) < __le16_to_cpu(rx->msdu_len)) {
1868 			ath10k_warn(ar, "dropping frame: offloaded rx msdu is too long!\n");
1869 			dev_kfree_skb_any(msdu);
1870 			continue;
1871 		}
1872 
1873 		skb_put(msdu, __le16_to_cpu(rx->msdu_len));
1874 
1875 		/* Offloaded rx header length isn't multiple of 2 nor 4 so the
1876 		 * actual payload is unaligned. Align the frame.  Otherwise
1877 		 * mac80211 complains.  This shouldn't reduce performance much
1878 		 * because these offloaded frames are rare.
1879 		 */
1880 		offset = 4 - ((unsigned long)msdu->data & 3);
1881 		skb_put(msdu, offset);
1882 		memmove(msdu->data + offset, msdu->data, msdu->len);
1883 		skb_pull(msdu, offset);
1884 
1885 		/* FIXME: The frame is NWifi. Re-construct QoS Control
1886 		 * if possible later.
1887 		 */
1888 
1889 		memset(status, 0, sizeof(*status));
1890 		status->flag |= RX_FLAG_NO_SIGNAL_VAL;
1891 
1892 		ath10k_htt_rx_h_rx_offload_prot(status, msdu);
1893 		ath10k_htt_rx_h_channel(ar, status, NULL, rx->vdev_id);
1894 		ath10k_process_rx(ar, status, msdu);
1895 	}
1896 }
1897 
1898 static void ath10k_htt_rx_in_ord_ind(struct ath10k *ar, struct sk_buff *skb)
1899 {
1900 	struct ath10k_htt *htt = &ar->htt;
1901 	struct htt_resp *resp = (void *)skb->data;
1902 	struct ieee80211_rx_status *status = &htt->rx_status;
1903 	struct sk_buff_head list;
1904 	struct sk_buff_head amsdu;
1905 	u16 peer_id;
1906 	u16 msdu_count;
1907 	u8 vdev_id;
1908 	u8 tid;
1909 	bool offload;
1910 	bool frag;
1911 	int ret;
1912 
1913 	lockdep_assert_held(&htt->rx_ring.lock);
1914 
1915 	if (htt->rx_confused)
1916 		return;
1917 
1918 	skb_pull(skb, sizeof(resp->hdr));
1919 	skb_pull(skb, sizeof(resp->rx_in_ord_ind));
1920 
1921 	peer_id = __le16_to_cpu(resp->rx_in_ord_ind.peer_id);
1922 	msdu_count = __le16_to_cpu(resp->rx_in_ord_ind.msdu_count);
1923 	vdev_id = resp->rx_in_ord_ind.vdev_id;
1924 	tid = SM(resp->rx_in_ord_ind.info, HTT_RX_IN_ORD_IND_INFO_TID);
1925 	offload = !!(resp->rx_in_ord_ind.info &
1926 			HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK);
1927 	frag = !!(resp->rx_in_ord_ind.info & HTT_RX_IN_ORD_IND_INFO_FRAG_MASK);
1928 
1929 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1930 		   "htt rx in ord vdev %i peer %i tid %i offload %i frag %i msdu count %i\n",
1931 		   vdev_id, peer_id, tid, offload, frag, msdu_count);
1932 
1933 	if (skb->len < msdu_count * sizeof(*resp->rx_in_ord_ind.msdu_descs)) {
1934 		ath10k_warn(ar, "dropping invalid in order rx indication\n");
1935 		return;
1936 	}
1937 
1938 	/* The event can deliver more than 1 A-MSDU. Each A-MSDU is later
1939 	 * extracted and processed.
1940 	 */
1941 	__skb_queue_head_init(&list);
1942 	ret = ath10k_htt_rx_pop_paddr_list(htt, &resp->rx_in_ord_ind, &list);
1943 	if (ret < 0) {
1944 		ath10k_warn(ar, "failed to pop paddr list: %d\n", ret);
1945 		htt->rx_confused = true;
1946 		return;
1947 	}
1948 
1949 	/* Offloaded frames are very different and need to be handled
1950 	 * separately.
1951 	 */
1952 	if (offload)
1953 		ath10k_htt_rx_h_rx_offload(ar, &list);
1954 
1955 	while (!skb_queue_empty(&list)) {
1956 		__skb_queue_head_init(&amsdu);
1957 		ret = ath10k_htt_rx_extract_amsdu(&list, &amsdu);
1958 		switch (ret) {
1959 		case 0:
1960 			/* Note: The in-order indication may report interleaved
1961 			 * frames from different PPDUs meaning reported rx rate
1962 			 * to mac80211 isn't accurate/reliable. It's still
1963 			 * better to report something than nothing though. This
1964 			 * should still give an idea about rx rate to the user.
1965 			 */
1966 			ath10k_htt_rx_h_ppdu(ar, &amsdu, status, vdev_id);
1967 			ath10k_htt_rx_h_filter(ar, &amsdu, status);
1968 			ath10k_htt_rx_h_mpdu(ar, &amsdu, status);
1969 			ath10k_htt_rx_h_deliver(ar, &amsdu, status);
1970 			break;
1971 		case -EAGAIN:
1972 			/* fall through */
1973 		default:
1974 			/* Should not happen. */
1975 			ath10k_warn(ar, "failed to extract amsdu: %d\n", ret);
1976 			htt->rx_confused = true;
1977 			__skb_queue_purge(&list);
1978 			return;
1979 		}
1980 	}
1981 
1982 	tasklet_schedule(&htt->rx_replenish_task);
1983 }
1984 
1985 void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
1986 {
1987 	struct ath10k_htt *htt = &ar->htt;
1988 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1989 	enum htt_t2h_msg_type type;
1990 
1991 	/* confirm alignment */
1992 	if (!IS_ALIGNED((unsigned long)skb->data, 4))
1993 		ath10k_warn(ar, "unaligned htt message, expect trouble\n");
1994 
1995 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx, msg_type: 0x%0X\n",
1996 		   resp->hdr.msg_type);
1997 
1998 	if (resp->hdr.msg_type >= ar->htt.t2h_msg_types_max) {
1999 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx, unsupported msg_type: 0x%0X\n max: 0x%0X",
2000 			   resp->hdr.msg_type, ar->htt.t2h_msg_types_max);
2001 		dev_kfree_skb_any(skb);
2002 		return;
2003 	}
2004 	type = ar->htt.t2h_msg_types[resp->hdr.msg_type];
2005 
2006 	switch (type) {
2007 	case HTT_T2H_MSG_TYPE_VERSION_CONF: {
2008 		htt->target_version_major = resp->ver_resp.major;
2009 		htt->target_version_minor = resp->ver_resp.minor;
2010 		complete(&htt->target_version_received);
2011 		break;
2012 	}
2013 	case HTT_T2H_MSG_TYPE_RX_IND:
2014 		spin_lock_bh(&htt->rx_ring.lock);
2015 		__skb_queue_tail(&htt->rx_compl_q, skb);
2016 		spin_unlock_bh(&htt->rx_ring.lock);
2017 		tasklet_schedule(&htt->txrx_compl_task);
2018 		return;
2019 	case HTT_T2H_MSG_TYPE_PEER_MAP: {
2020 		struct htt_peer_map_event ev = {
2021 			.vdev_id = resp->peer_map.vdev_id,
2022 			.peer_id = __le16_to_cpu(resp->peer_map.peer_id),
2023 		};
2024 		memcpy(ev.addr, resp->peer_map.addr, sizeof(ev.addr));
2025 		ath10k_peer_map_event(htt, &ev);
2026 		break;
2027 	}
2028 	case HTT_T2H_MSG_TYPE_PEER_UNMAP: {
2029 		struct htt_peer_unmap_event ev = {
2030 			.peer_id = __le16_to_cpu(resp->peer_unmap.peer_id),
2031 		};
2032 		ath10k_peer_unmap_event(htt, &ev);
2033 		break;
2034 	}
2035 	case HTT_T2H_MSG_TYPE_MGMT_TX_COMPLETION: {
2036 		struct htt_tx_done tx_done = {};
2037 		int status = __le32_to_cpu(resp->mgmt_tx_completion.status);
2038 
2039 		tx_done.msdu_id =
2040 			__le32_to_cpu(resp->mgmt_tx_completion.desc_id);
2041 
2042 		switch (status) {
2043 		case HTT_MGMT_TX_STATUS_OK:
2044 			tx_done.success = true;
2045 			break;
2046 		case HTT_MGMT_TX_STATUS_RETRY:
2047 			tx_done.no_ack = true;
2048 			break;
2049 		case HTT_MGMT_TX_STATUS_DROP:
2050 			tx_done.discard = true;
2051 			break;
2052 		}
2053 
2054 		ath10k_txrx_tx_unref(htt, &tx_done);
2055 		break;
2056 	}
2057 	case HTT_T2H_MSG_TYPE_TX_COMPL_IND:
2058 		skb_queue_tail(&htt->tx_compl_q, skb);
2059 		tasklet_schedule(&htt->txrx_compl_task);
2060 		return;
2061 	case HTT_T2H_MSG_TYPE_SEC_IND: {
2062 		struct ath10k *ar = htt->ar;
2063 		struct htt_security_indication *ev = &resp->security_indication;
2064 
2065 		ath10k_dbg(ar, ATH10K_DBG_HTT,
2066 			   "sec ind peer_id %d unicast %d type %d\n",
2067 			  __le16_to_cpu(ev->peer_id),
2068 			  !!(ev->flags & HTT_SECURITY_IS_UNICAST),
2069 			  MS(ev->flags, HTT_SECURITY_TYPE));
2070 		complete(&ar->install_key_done);
2071 		break;
2072 	}
2073 	case HTT_T2H_MSG_TYPE_RX_FRAG_IND: {
2074 		ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
2075 				skb->data, skb->len);
2076 		ath10k_htt_rx_frag_handler(htt, &resp->rx_frag_ind);
2077 		break;
2078 	}
2079 	case HTT_T2H_MSG_TYPE_TEST:
2080 		break;
2081 	case HTT_T2H_MSG_TYPE_STATS_CONF:
2082 		trace_ath10k_htt_stats(ar, skb->data, skb->len);
2083 		break;
2084 	case HTT_T2H_MSG_TYPE_TX_INSPECT_IND:
2085 		/* Firmware can return tx frames if it's unable to fully
2086 		 * process them and suspects host may be able to fix it. ath10k
2087 		 * sends all tx frames as already inspected so this shouldn't
2088 		 * happen unless fw has a bug.
2089 		 */
2090 		ath10k_warn(ar, "received an unexpected htt tx inspect event\n");
2091 		break;
2092 	case HTT_T2H_MSG_TYPE_RX_ADDBA:
2093 		ath10k_htt_rx_addba(ar, resp);
2094 		break;
2095 	case HTT_T2H_MSG_TYPE_RX_DELBA:
2096 		ath10k_htt_rx_delba(ar, resp);
2097 		break;
2098 	case HTT_T2H_MSG_TYPE_PKTLOG: {
2099 		struct ath10k_pktlog_hdr *hdr =
2100 			(struct ath10k_pktlog_hdr *)resp->pktlog_msg.payload;
2101 
2102 		trace_ath10k_htt_pktlog(ar, resp->pktlog_msg.payload,
2103 					sizeof(*hdr) +
2104 					__le16_to_cpu(hdr->size));
2105 		break;
2106 	}
2107 	case HTT_T2H_MSG_TYPE_RX_FLUSH: {
2108 		/* Ignore this event because mac80211 takes care of Rx
2109 		 * aggregation reordering.
2110 		 */
2111 		break;
2112 	}
2113 	case HTT_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND: {
2114 		spin_lock_bh(&htt->rx_ring.lock);
2115 		__skb_queue_tail(&htt->rx_in_ord_compl_q, skb);
2116 		spin_unlock_bh(&htt->rx_ring.lock);
2117 		tasklet_schedule(&htt->txrx_compl_task);
2118 		return;
2119 	}
2120 	case HTT_T2H_MSG_TYPE_TX_CREDIT_UPDATE_IND:
2121 		break;
2122 	case HTT_T2H_MSG_TYPE_CHAN_CHANGE:
2123 		break;
2124 	case HTT_T2H_MSG_TYPE_AGGR_CONF:
2125 		break;
2126 	case HTT_T2H_MSG_TYPE_EN_STATS:
2127 	case HTT_T2H_MSG_TYPE_TX_FETCH_IND:
2128 	case HTT_T2H_MSG_TYPE_TX_FETCH_CONF:
2129 	case HTT_T2H_MSG_TYPE_TX_LOW_LATENCY_IND:
2130 	default:
2131 		ath10k_warn(ar, "htt event (%d) not handled\n",
2132 			    resp->hdr.msg_type);
2133 		ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
2134 				skb->data, skb->len);
2135 		break;
2136 	};
2137 
2138 	/* Free the indication buffer */
2139 	dev_kfree_skb_any(skb);
2140 }
2141 EXPORT_SYMBOL(ath10k_htt_t2h_msg_handler);
2142 
2143 void ath10k_htt_rx_pktlog_completion_handler(struct ath10k *ar,
2144 					     struct sk_buff *skb)
2145 {
2146 	struct ath10k_pktlog_10_4_hdr *hdr =
2147 		(struct ath10k_pktlog_10_4_hdr *)skb->data;
2148 
2149 	trace_ath10k_htt_pktlog(ar, hdr->payload,
2150 				sizeof(*hdr) + __le16_to_cpu(hdr->size));
2151 	dev_kfree_skb_any(skb);
2152 }
2153 EXPORT_SYMBOL(ath10k_htt_rx_pktlog_completion_handler);
2154 
2155 static void ath10k_htt_txrx_compl_task(unsigned long ptr)
2156 {
2157 	struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
2158 	struct ath10k *ar = htt->ar;
2159 	struct htt_resp *resp;
2160 	struct sk_buff *skb;
2161 
2162 	while ((skb = skb_dequeue(&htt->tx_compl_q))) {
2163 		ath10k_htt_rx_frm_tx_compl(htt->ar, skb);
2164 		dev_kfree_skb_any(skb);
2165 	}
2166 
2167 	spin_lock_bh(&htt->rx_ring.lock);
2168 	while ((skb = __skb_dequeue(&htt->rx_compl_q))) {
2169 		resp = (struct htt_resp *)skb->data;
2170 		ath10k_htt_rx_handler(htt, &resp->rx_ind);
2171 		dev_kfree_skb_any(skb);
2172 	}
2173 
2174 	while ((skb = __skb_dequeue(&htt->rx_in_ord_compl_q))) {
2175 		ath10k_htt_rx_in_ord_ind(ar, skb);
2176 		dev_kfree_skb_any(skb);
2177 	}
2178 	spin_unlock_bh(&htt->rx_ring.lock);
2179 }
2180