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.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.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_DMA);
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_DMA);
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 static void ath10k_htt_rx_h_rates(struct ath10k *ar,
647 				  struct ieee80211_rx_status *status,
648 				  struct htt_rx_desc *rxd)
649 {
650 	struct ieee80211_supported_band *sband;
651 	u8 cck, rate, bw, sgi, mcs, nss;
652 	u8 preamble = 0;
653 	u32 info1, info2, info3;
654 
655 	info1 = __le32_to_cpu(rxd->ppdu_start.info1);
656 	info2 = __le32_to_cpu(rxd->ppdu_start.info2);
657 	info3 = __le32_to_cpu(rxd->ppdu_start.info3);
658 
659 	preamble = MS(info1, RX_PPDU_START_INFO1_PREAMBLE_TYPE);
660 
661 	switch (preamble) {
662 	case HTT_RX_LEGACY:
663 		/* To get legacy rate index band is required. Since band can't
664 		 * be undefined check if freq is non-zero.
665 		 */
666 		if (!status->freq)
667 			return;
668 
669 		cck = info1 & RX_PPDU_START_INFO1_L_SIG_RATE_SELECT;
670 		rate = MS(info1, RX_PPDU_START_INFO1_L_SIG_RATE);
671 		rate &= ~RX_PPDU_START_RATE_FLAG;
672 
673 		sband = &ar->mac.sbands[status->band];
674 		status->rate_idx = ath10k_mac_hw_rate_to_idx(sband, rate);
675 		break;
676 	case HTT_RX_HT:
677 	case HTT_RX_HT_WITH_TXBF:
678 		/* HT-SIG - Table 20-11 in info2 and info3 */
679 		mcs = info2 & 0x1F;
680 		nss = mcs >> 3;
681 		bw = (info2 >> 7) & 1;
682 		sgi = (info3 >> 7) & 1;
683 
684 		status->rate_idx = mcs;
685 		status->flag |= RX_FLAG_HT;
686 		if (sgi)
687 			status->flag |= RX_FLAG_SHORT_GI;
688 		if (bw)
689 			status->flag |= RX_FLAG_40MHZ;
690 		break;
691 	case HTT_RX_VHT:
692 	case HTT_RX_VHT_WITH_TXBF:
693 		/* VHT-SIG-A1 in info2, VHT-SIG-A2 in info3
694 		   TODO check this */
695 		mcs = (info3 >> 4) & 0x0F;
696 		nss = ((info2 >> 10) & 0x07) + 1;
697 		bw = info2 & 3;
698 		sgi = info3 & 1;
699 
700 		status->rate_idx = mcs;
701 		status->vht_nss = nss;
702 
703 		if (sgi)
704 			status->flag |= RX_FLAG_SHORT_GI;
705 
706 		switch (bw) {
707 		/* 20MHZ */
708 		case 0:
709 			break;
710 		/* 40MHZ */
711 		case 1:
712 			status->flag |= RX_FLAG_40MHZ;
713 			break;
714 		/* 80MHZ */
715 		case 2:
716 			status->vht_flag |= RX_VHT_FLAG_80MHZ;
717 		}
718 
719 		status->flag |= RX_FLAG_VHT;
720 		break;
721 	default:
722 		break;
723 	}
724 }
725 
726 static struct ieee80211_channel *
727 ath10k_htt_rx_h_peer_channel(struct ath10k *ar, struct htt_rx_desc *rxd)
728 {
729 	struct ath10k_peer *peer;
730 	struct ath10k_vif *arvif;
731 	struct cfg80211_chan_def def;
732 	u16 peer_id;
733 
734 	lockdep_assert_held(&ar->data_lock);
735 
736 	if (!rxd)
737 		return NULL;
738 
739 	if (rxd->attention.flags &
740 	    __cpu_to_le32(RX_ATTENTION_FLAGS_PEER_IDX_INVALID))
741 		return NULL;
742 
743 	if (!(rxd->msdu_end.info0 &
744 	      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU)))
745 		return NULL;
746 
747 	peer_id = MS(__le32_to_cpu(rxd->mpdu_start.info0),
748 		     RX_MPDU_START_INFO0_PEER_IDX);
749 
750 	peer = ath10k_peer_find_by_id(ar, peer_id);
751 	if (!peer)
752 		return NULL;
753 
754 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
755 	if (WARN_ON_ONCE(!arvif))
756 		return NULL;
757 
758 	if (WARN_ON(ath10k_mac_vif_chan(arvif->vif, &def)))
759 		return NULL;
760 
761 	return def.chan;
762 }
763 
764 static struct ieee80211_channel *
765 ath10k_htt_rx_h_vdev_channel(struct ath10k *ar, u32 vdev_id)
766 {
767 	struct ath10k_vif *arvif;
768 	struct cfg80211_chan_def def;
769 
770 	lockdep_assert_held(&ar->data_lock);
771 
772 	list_for_each_entry(arvif, &ar->arvifs, list) {
773 		if (arvif->vdev_id == vdev_id &&
774 		    ath10k_mac_vif_chan(arvif->vif, &def) == 0)
775 			return def.chan;
776 	}
777 
778 	return NULL;
779 }
780 
781 static void
782 ath10k_htt_rx_h_any_chan_iter(struct ieee80211_hw *hw,
783 			      struct ieee80211_chanctx_conf *conf,
784 			      void *data)
785 {
786 	struct cfg80211_chan_def *def = data;
787 
788 	*def = conf->def;
789 }
790 
791 static struct ieee80211_channel *
792 ath10k_htt_rx_h_any_channel(struct ath10k *ar)
793 {
794 	struct cfg80211_chan_def def = {};
795 
796 	ieee80211_iter_chan_contexts_atomic(ar->hw,
797 					    ath10k_htt_rx_h_any_chan_iter,
798 					    &def);
799 
800 	return def.chan;
801 }
802 
803 static bool ath10k_htt_rx_h_channel(struct ath10k *ar,
804 				    struct ieee80211_rx_status *status,
805 				    struct htt_rx_desc *rxd,
806 				    u32 vdev_id)
807 {
808 	struct ieee80211_channel *ch;
809 
810 	spin_lock_bh(&ar->data_lock);
811 	ch = ar->scan_channel;
812 	if (!ch)
813 		ch = ar->rx_channel;
814 	if (!ch)
815 		ch = ath10k_htt_rx_h_peer_channel(ar, rxd);
816 	if (!ch)
817 		ch = ath10k_htt_rx_h_vdev_channel(ar, vdev_id);
818 	if (!ch)
819 		ch = ath10k_htt_rx_h_any_channel(ar);
820 	spin_unlock_bh(&ar->data_lock);
821 
822 	if (!ch)
823 		return false;
824 
825 	status->band = ch->band;
826 	status->freq = ch->center_freq;
827 
828 	return true;
829 }
830 
831 static void ath10k_htt_rx_h_signal(struct ath10k *ar,
832 				   struct ieee80211_rx_status *status,
833 				   struct htt_rx_desc *rxd)
834 {
835 	/* FIXME: Get real NF */
836 	status->signal = ATH10K_DEFAULT_NOISE_FLOOR +
837 			 rxd->ppdu_start.rssi_comb;
838 	status->flag &= ~RX_FLAG_NO_SIGNAL_VAL;
839 }
840 
841 static void ath10k_htt_rx_h_mactime(struct ath10k *ar,
842 				    struct ieee80211_rx_status *status,
843 				    struct htt_rx_desc *rxd)
844 {
845 	/* FIXME: TSF is known only at the end of PPDU, in the last MPDU. This
846 	 * means all prior MSDUs in a PPDU are reported to mac80211 without the
847 	 * TSF. Is it worth holding frames until end of PPDU is known?
848 	 *
849 	 * FIXME: Can we get/compute 64bit TSF?
850 	 */
851 	status->mactime = __le32_to_cpu(rxd->ppdu_end.common.tsf_timestamp);
852 	status->flag |= RX_FLAG_MACTIME_END;
853 }
854 
855 static void ath10k_htt_rx_h_ppdu(struct ath10k *ar,
856 				 struct sk_buff_head *amsdu,
857 				 struct ieee80211_rx_status *status,
858 				 u32 vdev_id)
859 {
860 	struct sk_buff *first;
861 	struct htt_rx_desc *rxd;
862 	bool is_first_ppdu;
863 	bool is_last_ppdu;
864 
865 	if (skb_queue_empty(amsdu))
866 		return;
867 
868 	first = skb_peek(amsdu);
869 	rxd = (void *)first->data - sizeof(*rxd);
870 
871 	is_first_ppdu = !!(rxd->attention.flags &
872 			   __cpu_to_le32(RX_ATTENTION_FLAGS_FIRST_MPDU));
873 	is_last_ppdu = !!(rxd->attention.flags &
874 			  __cpu_to_le32(RX_ATTENTION_FLAGS_LAST_MPDU));
875 
876 	if (is_first_ppdu) {
877 		/* New PPDU starts so clear out the old per-PPDU status. */
878 		status->freq = 0;
879 		status->rate_idx = 0;
880 		status->vht_nss = 0;
881 		status->vht_flag &= ~RX_VHT_FLAG_80MHZ;
882 		status->flag &= ~(RX_FLAG_HT |
883 				  RX_FLAG_VHT |
884 				  RX_FLAG_SHORT_GI |
885 				  RX_FLAG_40MHZ |
886 				  RX_FLAG_MACTIME_END);
887 		status->flag |= RX_FLAG_NO_SIGNAL_VAL;
888 
889 		ath10k_htt_rx_h_signal(ar, status, rxd);
890 		ath10k_htt_rx_h_channel(ar, status, rxd, vdev_id);
891 		ath10k_htt_rx_h_rates(ar, status, rxd);
892 	}
893 
894 	if (is_last_ppdu)
895 		ath10k_htt_rx_h_mactime(ar, status, rxd);
896 }
897 
898 static const char * const tid_to_ac[] = {
899 	"BE",
900 	"BK",
901 	"BK",
902 	"BE",
903 	"VI",
904 	"VI",
905 	"VO",
906 	"VO",
907 };
908 
909 static char *ath10k_get_tid(struct ieee80211_hdr *hdr, char *out, size_t size)
910 {
911 	u8 *qc;
912 	int tid;
913 
914 	if (!ieee80211_is_data_qos(hdr->frame_control))
915 		return "";
916 
917 	qc = ieee80211_get_qos_ctl(hdr);
918 	tid = *qc & IEEE80211_QOS_CTL_TID_MASK;
919 	if (tid < 8)
920 		snprintf(out, size, "tid %d (%s)", tid, tid_to_ac[tid]);
921 	else
922 		snprintf(out, size, "tid %d", tid);
923 
924 	return out;
925 }
926 
927 static void ath10k_process_rx(struct ath10k *ar,
928 			      struct ieee80211_rx_status *rx_status,
929 			      struct sk_buff *skb)
930 {
931 	struct ieee80211_rx_status *status;
932 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
933 	char tid[32];
934 
935 	status = IEEE80211_SKB_RXCB(skb);
936 	*status = *rx_status;
937 
938 	ath10k_dbg(ar, ATH10K_DBG_DATA,
939 		   "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",
940 		   skb,
941 		   skb->len,
942 		   ieee80211_get_SA(hdr),
943 		   ath10k_get_tid(hdr, tid, sizeof(tid)),
944 		   is_multicast_ether_addr(ieee80211_get_DA(hdr)) ?
945 							"mcast" : "ucast",
946 		   (__le16_to_cpu(hdr->seq_ctrl) & IEEE80211_SCTL_SEQ) >> 4,
947 		   status->flag == 0 ? "legacy" : "",
948 		   status->flag & RX_FLAG_HT ? "ht" : "",
949 		   status->flag & RX_FLAG_VHT ? "vht" : "",
950 		   status->flag & RX_FLAG_40MHZ ? "40" : "",
951 		   status->vht_flag & RX_VHT_FLAG_80MHZ ? "80" : "",
952 		   status->flag & RX_FLAG_SHORT_GI ? "sgi " : "",
953 		   status->rate_idx,
954 		   status->vht_nss,
955 		   status->freq,
956 		   status->band, status->flag,
957 		   !!(status->flag & RX_FLAG_FAILED_FCS_CRC),
958 		   !!(status->flag & RX_FLAG_MMIC_ERROR),
959 		   !!(status->flag & RX_FLAG_AMSDU_MORE));
960 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "rx skb: ",
961 			skb->data, skb->len);
962 	trace_ath10k_rx_hdr(ar, skb->data, skb->len);
963 	trace_ath10k_rx_payload(ar, skb->data, skb->len);
964 
965 	ieee80211_rx(ar->hw, skb);
966 }
967 
968 static int ath10k_htt_rx_nwifi_hdrlen(struct ath10k *ar,
969 				      struct ieee80211_hdr *hdr)
970 {
971 	int len = ieee80211_hdrlen(hdr->frame_control);
972 
973 	if (!test_bit(ATH10K_FW_FEATURE_NO_NWIFI_DECAP_4ADDR_PADDING,
974 		      ar->fw_features))
975 		len = round_up(len, 4);
976 
977 	return len;
978 }
979 
980 static void ath10k_htt_rx_h_undecap_raw(struct ath10k *ar,
981 					struct sk_buff *msdu,
982 					struct ieee80211_rx_status *status,
983 					enum htt_rx_mpdu_encrypt_type enctype,
984 					bool is_decrypted)
985 {
986 	struct ieee80211_hdr *hdr;
987 	struct htt_rx_desc *rxd;
988 	size_t hdr_len;
989 	size_t crypto_len;
990 	bool is_first;
991 	bool is_last;
992 
993 	rxd = (void *)msdu->data - sizeof(*rxd);
994 	is_first = !!(rxd->msdu_end.info0 &
995 		      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU));
996 	is_last = !!(rxd->msdu_end.info0 &
997 		     __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU));
998 
999 	/* Delivered decapped frame:
1000 	 * [802.11 header]
1001 	 * [crypto param] <-- can be trimmed if !fcs_err &&
1002 	 *                    !decrypt_err && !peer_idx_invalid
1003 	 * [amsdu header] <-- only if A-MSDU
1004 	 * [rfc1042/llc]
1005 	 * [payload]
1006 	 * [FCS] <-- at end, needs to be trimmed
1007 	 */
1008 
1009 	/* This probably shouldn't happen but warn just in case */
1010 	if (unlikely(WARN_ON_ONCE(!is_first)))
1011 		return;
1012 
1013 	/* This probably shouldn't happen but warn just in case */
1014 	if (unlikely(WARN_ON_ONCE(!(is_first && is_last))))
1015 		return;
1016 
1017 	skb_trim(msdu, msdu->len - FCS_LEN);
1018 
1019 	/* In most cases this will be true for sniffed frames. It makes sense
1020 	 * to deliver them as-is without stripping the crypto param. This would
1021 	 * also make sense for software based decryption (which is not
1022 	 * implemented in ath10k).
1023 	 *
1024 	 * If there's no error then the frame is decrypted. At least that is
1025 	 * the case for frames that come in via fragmented rx indication.
1026 	 */
1027 	if (!is_decrypted)
1028 		return;
1029 
1030 	/* The payload is decrypted so strip crypto params. Start from tail
1031 	 * since hdr is used to compute some stuff.
1032 	 */
1033 
1034 	hdr = (void *)msdu->data;
1035 
1036 	/* Tail */
1037 	skb_trim(msdu, msdu->len - ath10k_htt_rx_crypto_tail_len(ar, enctype));
1038 
1039 	/* MMIC */
1040 	if (!ieee80211_has_morefrags(hdr->frame_control) &&
1041 	    enctype == HTT_RX_MPDU_ENCRYPT_TKIP_WPA)
1042 		skb_trim(msdu, msdu->len - 8);
1043 
1044 	/* Head */
1045 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1046 	crypto_len = ath10k_htt_rx_crypto_param_len(ar, enctype);
1047 
1048 	memmove((void *)msdu->data + crypto_len,
1049 		(void *)msdu->data, hdr_len);
1050 	skb_pull(msdu, crypto_len);
1051 }
1052 
1053 static void ath10k_htt_rx_h_undecap_nwifi(struct ath10k *ar,
1054 					  struct sk_buff *msdu,
1055 					  struct ieee80211_rx_status *status,
1056 					  const u8 first_hdr[64])
1057 {
1058 	struct ieee80211_hdr *hdr;
1059 	size_t hdr_len;
1060 	u8 da[ETH_ALEN];
1061 	u8 sa[ETH_ALEN];
1062 
1063 	/* Delivered decapped frame:
1064 	 * [nwifi 802.11 header] <-- replaced with 802.11 hdr
1065 	 * [rfc1042/llc]
1066 	 *
1067 	 * Note: The nwifi header doesn't have QoS Control and is
1068 	 * (always?) a 3addr frame.
1069 	 *
1070 	 * Note2: There's no A-MSDU subframe header. Even if it's part
1071 	 * of an A-MSDU.
1072 	 */
1073 
1074 	/* pull decapped header and copy SA & DA */
1075 	hdr = (struct ieee80211_hdr *)msdu->data;
1076 	hdr_len = ath10k_htt_rx_nwifi_hdrlen(ar, hdr);
1077 	ether_addr_copy(da, ieee80211_get_DA(hdr));
1078 	ether_addr_copy(sa, ieee80211_get_SA(hdr));
1079 	skb_pull(msdu, hdr_len);
1080 
1081 	/* push original 802.11 header */
1082 	hdr = (struct ieee80211_hdr *)first_hdr;
1083 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1084 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1085 
1086 	/* original 802.11 header has a different DA and in
1087 	 * case of 4addr it may also have different SA
1088 	 */
1089 	hdr = (struct ieee80211_hdr *)msdu->data;
1090 	ether_addr_copy(ieee80211_get_DA(hdr), da);
1091 	ether_addr_copy(ieee80211_get_SA(hdr), sa);
1092 }
1093 
1094 static void *ath10k_htt_rx_h_find_rfc1042(struct ath10k *ar,
1095 					  struct sk_buff *msdu,
1096 					  enum htt_rx_mpdu_encrypt_type enctype)
1097 {
1098 	struct ieee80211_hdr *hdr;
1099 	struct htt_rx_desc *rxd;
1100 	size_t hdr_len, crypto_len;
1101 	void *rfc1042;
1102 	bool is_first, is_last, is_amsdu;
1103 
1104 	rxd = (void *)msdu->data - sizeof(*rxd);
1105 	hdr = (void *)rxd->rx_hdr_status;
1106 
1107 	is_first = !!(rxd->msdu_end.info0 &
1108 		      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU));
1109 	is_last = !!(rxd->msdu_end.info0 &
1110 		     __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU));
1111 	is_amsdu = !(is_first && is_last);
1112 
1113 	rfc1042 = hdr;
1114 
1115 	if (is_first) {
1116 		hdr_len = ieee80211_hdrlen(hdr->frame_control);
1117 		crypto_len = ath10k_htt_rx_crypto_param_len(ar, enctype);
1118 
1119 		rfc1042 += round_up(hdr_len, 4) +
1120 			   round_up(crypto_len, 4);
1121 	}
1122 
1123 	if (is_amsdu)
1124 		rfc1042 += sizeof(struct amsdu_subframe_hdr);
1125 
1126 	return rfc1042;
1127 }
1128 
1129 static void ath10k_htt_rx_h_undecap_eth(struct ath10k *ar,
1130 					struct sk_buff *msdu,
1131 					struct ieee80211_rx_status *status,
1132 					const u8 first_hdr[64],
1133 					enum htt_rx_mpdu_encrypt_type enctype)
1134 {
1135 	struct ieee80211_hdr *hdr;
1136 	struct ethhdr *eth;
1137 	size_t hdr_len;
1138 	void *rfc1042;
1139 	u8 da[ETH_ALEN];
1140 	u8 sa[ETH_ALEN];
1141 
1142 	/* Delivered decapped frame:
1143 	 * [eth header] <-- replaced with 802.11 hdr & rfc1042/llc
1144 	 * [payload]
1145 	 */
1146 
1147 	rfc1042 = ath10k_htt_rx_h_find_rfc1042(ar, msdu, enctype);
1148 	if (WARN_ON_ONCE(!rfc1042))
1149 		return;
1150 
1151 	/* pull decapped header and copy SA & DA */
1152 	eth = (struct ethhdr *)msdu->data;
1153 	ether_addr_copy(da, eth->h_dest);
1154 	ether_addr_copy(sa, eth->h_source);
1155 	skb_pull(msdu, sizeof(struct ethhdr));
1156 
1157 	/* push rfc1042/llc/snap */
1158 	memcpy(skb_push(msdu, sizeof(struct rfc1042_hdr)), rfc1042,
1159 	       sizeof(struct rfc1042_hdr));
1160 
1161 	/* push original 802.11 header */
1162 	hdr = (struct ieee80211_hdr *)first_hdr;
1163 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1164 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1165 
1166 	/* original 802.11 header has a different DA and in
1167 	 * case of 4addr it may also have different SA
1168 	 */
1169 	hdr = (struct ieee80211_hdr *)msdu->data;
1170 	ether_addr_copy(ieee80211_get_DA(hdr), da);
1171 	ether_addr_copy(ieee80211_get_SA(hdr), sa);
1172 }
1173 
1174 static void ath10k_htt_rx_h_undecap_snap(struct ath10k *ar,
1175 					 struct sk_buff *msdu,
1176 					 struct ieee80211_rx_status *status,
1177 					 const u8 first_hdr[64])
1178 {
1179 	struct ieee80211_hdr *hdr;
1180 	size_t hdr_len;
1181 
1182 	/* Delivered decapped frame:
1183 	 * [amsdu header] <-- replaced with 802.11 hdr
1184 	 * [rfc1042/llc]
1185 	 * [payload]
1186 	 */
1187 
1188 	skb_pull(msdu, sizeof(struct amsdu_subframe_hdr));
1189 
1190 	hdr = (struct ieee80211_hdr *)first_hdr;
1191 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1192 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1193 }
1194 
1195 static void ath10k_htt_rx_h_undecap(struct ath10k *ar,
1196 				    struct sk_buff *msdu,
1197 				    struct ieee80211_rx_status *status,
1198 				    u8 first_hdr[64],
1199 				    enum htt_rx_mpdu_encrypt_type enctype,
1200 				    bool is_decrypted)
1201 {
1202 	struct htt_rx_desc *rxd;
1203 	enum rx_msdu_decap_format decap;
1204 	struct ieee80211_hdr *hdr;
1205 
1206 	/* First msdu's decapped header:
1207 	 * [802.11 header] <-- padded to 4 bytes long
1208 	 * [crypto param] <-- padded to 4 bytes long
1209 	 * [amsdu header] <-- only if A-MSDU
1210 	 * [rfc1042/llc]
1211 	 *
1212 	 * Other (2nd, 3rd, ..) msdu's decapped header:
1213 	 * [amsdu header] <-- only if A-MSDU
1214 	 * [rfc1042/llc]
1215 	 */
1216 
1217 	rxd = (void *)msdu->data - sizeof(*rxd);
1218 	hdr = (void *)rxd->rx_hdr_status;
1219 	decap = MS(__le32_to_cpu(rxd->msdu_start.info1),
1220 		   RX_MSDU_START_INFO1_DECAP_FORMAT);
1221 
1222 	switch (decap) {
1223 	case RX_MSDU_DECAP_RAW:
1224 		ath10k_htt_rx_h_undecap_raw(ar, msdu, status, enctype,
1225 					    is_decrypted);
1226 		break;
1227 	case RX_MSDU_DECAP_NATIVE_WIFI:
1228 		ath10k_htt_rx_h_undecap_nwifi(ar, msdu, status, first_hdr);
1229 		break;
1230 	case RX_MSDU_DECAP_ETHERNET2_DIX:
1231 		ath10k_htt_rx_h_undecap_eth(ar, msdu, status, first_hdr, enctype);
1232 		break;
1233 	case RX_MSDU_DECAP_8023_SNAP_LLC:
1234 		ath10k_htt_rx_h_undecap_snap(ar, msdu, status, first_hdr);
1235 		break;
1236 	}
1237 }
1238 
1239 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb)
1240 {
1241 	struct htt_rx_desc *rxd;
1242 	u32 flags, info;
1243 	bool is_ip4, is_ip6;
1244 	bool is_tcp, is_udp;
1245 	bool ip_csum_ok, tcpudp_csum_ok;
1246 
1247 	rxd = (void *)skb->data - sizeof(*rxd);
1248 	flags = __le32_to_cpu(rxd->attention.flags);
1249 	info = __le32_to_cpu(rxd->msdu_start.info1);
1250 
1251 	is_ip4 = !!(info & RX_MSDU_START_INFO1_IPV4_PROTO);
1252 	is_ip6 = !!(info & RX_MSDU_START_INFO1_IPV6_PROTO);
1253 	is_tcp = !!(info & RX_MSDU_START_INFO1_TCP_PROTO);
1254 	is_udp = !!(info & RX_MSDU_START_INFO1_UDP_PROTO);
1255 	ip_csum_ok = !(flags & RX_ATTENTION_FLAGS_IP_CHKSUM_FAIL);
1256 	tcpudp_csum_ok = !(flags & RX_ATTENTION_FLAGS_TCP_UDP_CHKSUM_FAIL);
1257 
1258 	if (!is_ip4 && !is_ip6)
1259 		return CHECKSUM_NONE;
1260 	if (!is_tcp && !is_udp)
1261 		return CHECKSUM_NONE;
1262 	if (!ip_csum_ok)
1263 		return CHECKSUM_NONE;
1264 	if (!tcpudp_csum_ok)
1265 		return CHECKSUM_NONE;
1266 
1267 	return CHECKSUM_UNNECESSARY;
1268 }
1269 
1270 static void ath10k_htt_rx_h_csum_offload(struct sk_buff *msdu)
1271 {
1272 	msdu->ip_summed = ath10k_htt_rx_get_csum_state(msdu);
1273 }
1274 
1275 static void ath10k_htt_rx_h_mpdu(struct ath10k *ar,
1276 				 struct sk_buff_head *amsdu,
1277 				 struct ieee80211_rx_status *status)
1278 {
1279 	struct sk_buff *first;
1280 	struct sk_buff *last;
1281 	struct sk_buff *msdu;
1282 	struct htt_rx_desc *rxd;
1283 	struct ieee80211_hdr *hdr;
1284 	enum htt_rx_mpdu_encrypt_type enctype;
1285 	u8 first_hdr[64];
1286 	u8 *qos;
1287 	size_t hdr_len;
1288 	bool has_fcs_err;
1289 	bool has_crypto_err;
1290 	bool has_tkip_err;
1291 	bool has_peer_idx_invalid;
1292 	bool is_decrypted;
1293 	u32 attention;
1294 
1295 	if (skb_queue_empty(amsdu))
1296 		return;
1297 
1298 	first = skb_peek(amsdu);
1299 	rxd = (void *)first->data - sizeof(*rxd);
1300 
1301 	enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
1302 		     RX_MPDU_START_INFO0_ENCRYPT_TYPE);
1303 
1304 	/* First MSDU's Rx descriptor in an A-MSDU contains full 802.11
1305 	 * decapped header. It'll be used for undecapping of each MSDU.
1306 	 */
1307 	hdr = (void *)rxd->rx_hdr_status;
1308 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1309 	memcpy(first_hdr, hdr, hdr_len);
1310 
1311 	/* Each A-MSDU subframe will use the original header as the base and be
1312 	 * reported as a separate MSDU so strip the A-MSDU bit from QoS Ctl.
1313 	 */
1314 	hdr = (void *)first_hdr;
1315 	qos = ieee80211_get_qos_ctl(hdr);
1316 	qos[0] &= ~IEEE80211_QOS_CTL_A_MSDU_PRESENT;
1317 
1318 	/* Some attention flags are valid only in the last MSDU. */
1319 	last = skb_peek_tail(amsdu);
1320 	rxd = (void *)last->data - sizeof(*rxd);
1321 	attention = __le32_to_cpu(rxd->attention.flags);
1322 
1323 	has_fcs_err = !!(attention & RX_ATTENTION_FLAGS_FCS_ERR);
1324 	has_crypto_err = !!(attention & RX_ATTENTION_FLAGS_DECRYPT_ERR);
1325 	has_tkip_err = !!(attention & RX_ATTENTION_FLAGS_TKIP_MIC_ERR);
1326 	has_peer_idx_invalid = !!(attention & RX_ATTENTION_FLAGS_PEER_IDX_INVALID);
1327 
1328 	/* Note: If hardware captures an encrypted frame that it can't decrypt,
1329 	 * e.g. due to fcs error, missing peer or invalid key data it will
1330 	 * report the frame as raw.
1331 	 */
1332 	is_decrypted = (enctype != HTT_RX_MPDU_ENCRYPT_NONE &&
1333 			!has_fcs_err &&
1334 			!has_crypto_err &&
1335 			!has_peer_idx_invalid);
1336 
1337 	/* Clear per-MPDU flags while leaving per-PPDU flags intact. */
1338 	status->flag &= ~(RX_FLAG_FAILED_FCS_CRC |
1339 			  RX_FLAG_MMIC_ERROR |
1340 			  RX_FLAG_DECRYPTED |
1341 			  RX_FLAG_IV_STRIPPED |
1342 			  RX_FLAG_MMIC_STRIPPED);
1343 
1344 	if (has_fcs_err)
1345 		status->flag |= RX_FLAG_FAILED_FCS_CRC;
1346 
1347 	if (has_tkip_err)
1348 		status->flag |= RX_FLAG_MMIC_ERROR;
1349 
1350 	if (is_decrypted)
1351 		status->flag |= RX_FLAG_DECRYPTED |
1352 				RX_FLAG_IV_STRIPPED |
1353 				RX_FLAG_MMIC_STRIPPED;
1354 
1355 	skb_queue_walk(amsdu, msdu) {
1356 		ath10k_htt_rx_h_csum_offload(msdu);
1357 		ath10k_htt_rx_h_undecap(ar, msdu, status, first_hdr, enctype,
1358 					is_decrypted);
1359 
1360 		/* Undecapping involves copying the original 802.11 header back
1361 		 * to sk_buff. If frame is protected and hardware has decrypted
1362 		 * it then remove the protected bit.
1363 		 */
1364 		if (!is_decrypted)
1365 			continue;
1366 
1367 		hdr = (void *)msdu->data;
1368 		hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
1369 	}
1370 }
1371 
1372 static void ath10k_htt_rx_h_deliver(struct ath10k *ar,
1373 				    struct sk_buff_head *amsdu,
1374 				    struct ieee80211_rx_status *status)
1375 {
1376 	struct sk_buff *msdu;
1377 
1378 	while ((msdu = __skb_dequeue(amsdu))) {
1379 		/* Setup per-MSDU flags */
1380 		if (skb_queue_empty(amsdu))
1381 			status->flag &= ~RX_FLAG_AMSDU_MORE;
1382 		else
1383 			status->flag |= RX_FLAG_AMSDU_MORE;
1384 
1385 		ath10k_process_rx(ar, status, msdu);
1386 	}
1387 }
1388 
1389 static int ath10k_unchain_msdu(struct sk_buff_head *amsdu)
1390 {
1391 	struct sk_buff *skb, *first;
1392 	int space;
1393 	int total_len = 0;
1394 
1395 	/* TODO:  Might could optimize this by using
1396 	 * skb_try_coalesce or similar method to
1397 	 * decrease copying, or maybe get mac80211 to
1398 	 * provide a way to just receive a list of
1399 	 * skb?
1400 	 */
1401 
1402 	first = __skb_dequeue(amsdu);
1403 
1404 	/* Allocate total length all at once. */
1405 	skb_queue_walk(amsdu, skb)
1406 		total_len += skb->len;
1407 
1408 	space = total_len - skb_tailroom(first);
1409 	if ((space > 0) &&
1410 	    (pskb_expand_head(first, 0, space, GFP_ATOMIC) < 0)) {
1411 		/* TODO:  bump some rx-oom error stat */
1412 		/* put it back together so we can free the
1413 		 * whole list at once.
1414 		 */
1415 		__skb_queue_head(amsdu, first);
1416 		return -1;
1417 	}
1418 
1419 	/* Walk list again, copying contents into
1420 	 * msdu_head
1421 	 */
1422 	while ((skb = __skb_dequeue(amsdu))) {
1423 		skb_copy_from_linear_data(skb, skb_put(first, skb->len),
1424 					  skb->len);
1425 		dev_kfree_skb_any(skb);
1426 	}
1427 
1428 	__skb_queue_head(amsdu, first);
1429 	return 0;
1430 }
1431 
1432 static void ath10k_htt_rx_h_unchain(struct ath10k *ar,
1433 				    struct sk_buff_head *amsdu,
1434 				    bool chained)
1435 {
1436 	struct sk_buff *first;
1437 	struct htt_rx_desc *rxd;
1438 	enum rx_msdu_decap_format decap;
1439 
1440 	first = skb_peek(amsdu);
1441 	rxd = (void *)first->data - sizeof(*rxd);
1442 	decap = MS(__le32_to_cpu(rxd->msdu_start.info1),
1443 		   RX_MSDU_START_INFO1_DECAP_FORMAT);
1444 
1445 	if (!chained)
1446 		return;
1447 
1448 	/* FIXME: Current unchaining logic can only handle simple case of raw
1449 	 * msdu chaining. If decapping is other than raw the chaining may be
1450 	 * more complex and this isn't handled by the current code. Don't even
1451 	 * try re-constructing such frames - it'll be pretty much garbage.
1452 	 */
1453 	if (decap != RX_MSDU_DECAP_RAW ||
1454 	    skb_queue_len(amsdu) != 1 + rxd->frag_info.ring2_more_count) {
1455 		__skb_queue_purge(amsdu);
1456 		return;
1457 	}
1458 
1459 	ath10k_unchain_msdu(amsdu);
1460 }
1461 
1462 static bool ath10k_htt_rx_amsdu_allowed(struct ath10k *ar,
1463 					struct sk_buff_head *amsdu,
1464 					struct ieee80211_rx_status *rx_status)
1465 {
1466 	struct sk_buff *msdu;
1467 	struct htt_rx_desc *rxd;
1468 	bool is_mgmt;
1469 	bool has_fcs_err;
1470 
1471 	msdu = skb_peek(amsdu);
1472 	rxd = (void *)msdu->data - sizeof(*rxd);
1473 
1474 	/* FIXME: It might be a good idea to do some fuzzy-testing to drop
1475 	 * invalid/dangerous frames.
1476 	 */
1477 
1478 	if (!rx_status->freq) {
1479 		ath10k_warn(ar, "no channel configured; ignoring frame(s)!\n");
1480 		return false;
1481 	}
1482 
1483 	is_mgmt = !!(rxd->attention.flags &
1484 		     __cpu_to_le32(RX_ATTENTION_FLAGS_MGMT_TYPE));
1485 	has_fcs_err = !!(rxd->attention.flags &
1486 			 __cpu_to_le32(RX_ATTENTION_FLAGS_FCS_ERR));
1487 
1488 	/* Management frames are handled via WMI events. The pros of such
1489 	 * approach is that channel is explicitly provided in WMI events
1490 	 * whereas HTT doesn't provide channel information for Rxed frames.
1491 	 *
1492 	 * However some firmware revisions don't report corrupted frames via
1493 	 * WMI so don't drop them.
1494 	 */
1495 	if (is_mgmt && !has_fcs_err) {
1496 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx mgmt ctrl\n");
1497 		return false;
1498 	}
1499 
1500 	if (test_bit(ATH10K_CAC_RUNNING, &ar->dev_flags)) {
1501 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx cac running\n");
1502 		return false;
1503 	}
1504 
1505 	return true;
1506 }
1507 
1508 static void ath10k_htt_rx_h_filter(struct ath10k *ar,
1509 				   struct sk_buff_head *amsdu,
1510 				   struct ieee80211_rx_status *rx_status)
1511 {
1512 	if (skb_queue_empty(amsdu))
1513 		return;
1514 
1515 	if (ath10k_htt_rx_amsdu_allowed(ar, amsdu, rx_status))
1516 		return;
1517 
1518 	__skb_queue_purge(amsdu);
1519 }
1520 
1521 static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
1522 				  struct htt_rx_indication *rx)
1523 {
1524 	struct ath10k *ar = htt->ar;
1525 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1526 	struct htt_rx_indication_mpdu_range *mpdu_ranges;
1527 	struct sk_buff_head amsdu;
1528 	int num_mpdu_ranges;
1529 	int fw_desc_len;
1530 	u8 *fw_desc;
1531 	int i, ret, mpdu_count = 0;
1532 
1533 	lockdep_assert_held(&htt->rx_ring.lock);
1534 
1535 	if (htt->rx_confused)
1536 		return;
1537 
1538 	fw_desc_len = __le16_to_cpu(rx->prefix.fw_rx_desc_bytes);
1539 	fw_desc = (u8 *)&rx->fw_desc;
1540 
1541 	num_mpdu_ranges = MS(__le32_to_cpu(rx->hdr.info1),
1542 			     HTT_RX_INDICATION_INFO1_NUM_MPDU_RANGES);
1543 	mpdu_ranges = htt_rx_ind_get_mpdu_ranges(rx);
1544 
1545 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt rx ind: ",
1546 			rx, sizeof(*rx) +
1547 			(sizeof(struct htt_rx_indication_mpdu_range) *
1548 				num_mpdu_ranges));
1549 
1550 	for (i = 0; i < num_mpdu_ranges; i++)
1551 		mpdu_count += mpdu_ranges[i].mpdu_count;
1552 
1553 	while (mpdu_count--) {
1554 		__skb_queue_head_init(&amsdu);
1555 		ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc,
1556 					      &fw_desc_len, &amsdu);
1557 		if (ret < 0) {
1558 			ath10k_warn(ar, "rx ring became corrupted: %d\n", ret);
1559 			__skb_queue_purge(&amsdu);
1560 			/* FIXME: It's probably a good idea to reboot the
1561 			 * device instead of leaving it inoperable.
1562 			 */
1563 			htt->rx_confused = true;
1564 			break;
1565 		}
1566 
1567 		ath10k_htt_rx_h_ppdu(ar, &amsdu, rx_status, 0xffff);
1568 		ath10k_htt_rx_h_unchain(ar, &amsdu, ret > 0);
1569 		ath10k_htt_rx_h_filter(ar, &amsdu, rx_status);
1570 		ath10k_htt_rx_h_mpdu(ar, &amsdu, rx_status);
1571 		ath10k_htt_rx_h_deliver(ar, &amsdu, rx_status);
1572 	}
1573 
1574 	tasklet_schedule(&htt->rx_replenish_task);
1575 }
1576 
1577 static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
1578 				       struct htt_rx_fragment_indication *frag)
1579 {
1580 	struct ath10k *ar = htt->ar;
1581 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1582 	struct sk_buff_head amsdu;
1583 	int ret;
1584 	u8 *fw_desc;
1585 	int fw_desc_len;
1586 
1587 	fw_desc_len = __le16_to_cpu(frag->fw_rx_desc_bytes);
1588 	fw_desc = (u8 *)frag->fw_msdu_rx_desc;
1589 
1590 	__skb_queue_head_init(&amsdu);
1591 
1592 	spin_lock_bh(&htt->rx_ring.lock);
1593 	ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc, &fw_desc_len,
1594 				      &amsdu);
1595 	spin_unlock_bh(&htt->rx_ring.lock);
1596 
1597 	tasklet_schedule(&htt->rx_replenish_task);
1598 
1599 	ath10k_dbg(ar, ATH10K_DBG_HTT_DUMP, "htt rx frag ahead\n");
1600 
1601 	if (ret) {
1602 		ath10k_warn(ar, "failed to pop amsdu from httr rx ring for fragmented rx %d\n",
1603 			    ret);
1604 		__skb_queue_purge(&amsdu);
1605 		return;
1606 	}
1607 
1608 	if (skb_queue_len(&amsdu) != 1) {
1609 		ath10k_warn(ar, "failed to pop frag amsdu: too many msdus\n");
1610 		__skb_queue_purge(&amsdu);
1611 		return;
1612 	}
1613 
1614 	ath10k_htt_rx_h_ppdu(ar, &amsdu, rx_status, 0xffff);
1615 	ath10k_htt_rx_h_filter(ar, &amsdu, rx_status);
1616 	ath10k_htt_rx_h_mpdu(ar, &amsdu, rx_status);
1617 	ath10k_htt_rx_h_deliver(ar, &amsdu, rx_status);
1618 
1619 	if (fw_desc_len > 0) {
1620 		ath10k_dbg(ar, ATH10K_DBG_HTT,
1621 			   "expecting more fragmented rx in one indication %d\n",
1622 			   fw_desc_len);
1623 	}
1624 }
1625 
1626 static void ath10k_htt_rx_frm_tx_compl(struct ath10k *ar,
1627 				       struct sk_buff *skb)
1628 {
1629 	struct ath10k_htt *htt = &ar->htt;
1630 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1631 	struct htt_tx_done tx_done = {};
1632 	int status = MS(resp->data_tx_completion.flags, HTT_DATA_TX_STATUS);
1633 	__le16 msdu_id;
1634 	int i;
1635 
1636 	lockdep_assert_held(&htt->tx_lock);
1637 
1638 	switch (status) {
1639 	case HTT_DATA_TX_STATUS_NO_ACK:
1640 		tx_done.no_ack = true;
1641 		break;
1642 	case HTT_DATA_TX_STATUS_OK:
1643 		tx_done.success = true;
1644 		break;
1645 	case HTT_DATA_TX_STATUS_DISCARD:
1646 	case HTT_DATA_TX_STATUS_POSTPONE:
1647 	case HTT_DATA_TX_STATUS_DOWNLOAD_FAIL:
1648 		tx_done.discard = true;
1649 		break;
1650 	default:
1651 		ath10k_warn(ar, "unhandled tx completion status %d\n", status);
1652 		tx_done.discard = true;
1653 		break;
1654 	}
1655 
1656 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx completion num_msdus %d\n",
1657 		   resp->data_tx_completion.num_msdus);
1658 
1659 	for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
1660 		msdu_id = resp->data_tx_completion.msdus[i];
1661 		tx_done.msdu_id = __le16_to_cpu(msdu_id);
1662 		ath10k_txrx_tx_unref(htt, &tx_done);
1663 	}
1664 }
1665 
1666 static void ath10k_htt_rx_addba(struct ath10k *ar, struct htt_resp *resp)
1667 {
1668 	struct htt_rx_addba *ev = &resp->rx_addba;
1669 	struct ath10k_peer *peer;
1670 	struct ath10k_vif *arvif;
1671 	u16 info0, tid, peer_id;
1672 
1673 	info0 = __le16_to_cpu(ev->info0);
1674 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1675 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1676 
1677 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1678 		   "htt rx addba tid %hu peer_id %hu size %hhu\n",
1679 		   tid, peer_id, ev->window_size);
1680 
1681 	spin_lock_bh(&ar->data_lock);
1682 	peer = ath10k_peer_find_by_id(ar, peer_id);
1683 	if (!peer) {
1684 		ath10k_warn(ar, "received addba event for invalid peer_id: %hu\n",
1685 			    peer_id);
1686 		spin_unlock_bh(&ar->data_lock);
1687 		return;
1688 	}
1689 
1690 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1691 	if (!arvif) {
1692 		ath10k_warn(ar, "received addba event for invalid vdev_id: %u\n",
1693 			    peer->vdev_id);
1694 		spin_unlock_bh(&ar->data_lock);
1695 		return;
1696 	}
1697 
1698 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1699 		   "htt rx start rx ba session sta %pM tid %hu size %hhu\n",
1700 		   peer->addr, tid, ev->window_size);
1701 
1702 	ieee80211_start_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1703 	spin_unlock_bh(&ar->data_lock);
1704 }
1705 
1706 static void ath10k_htt_rx_delba(struct ath10k *ar, struct htt_resp *resp)
1707 {
1708 	struct htt_rx_delba *ev = &resp->rx_delba;
1709 	struct ath10k_peer *peer;
1710 	struct ath10k_vif *arvif;
1711 	u16 info0, tid, peer_id;
1712 
1713 	info0 = __le16_to_cpu(ev->info0);
1714 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1715 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1716 
1717 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1718 		   "htt rx delba tid %hu peer_id %hu\n",
1719 		   tid, peer_id);
1720 
1721 	spin_lock_bh(&ar->data_lock);
1722 	peer = ath10k_peer_find_by_id(ar, peer_id);
1723 	if (!peer) {
1724 		ath10k_warn(ar, "received addba event for invalid peer_id: %hu\n",
1725 			    peer_id);
1726 		spin_unlock_bh(&ar->data_lock);
1727 		return;
1728 	}
1729 
1730 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1731 	if (!arvif) {
1732 		ath10k_warn(ar, "received addba event for invalid vdev_id: %u\n",
1733 			    peer->vdev_id);
1734 		spin_unlock_bh(&ar->data_lock);
1735 		return;
1736 	}
1737 
1738 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1739 		   "htt rx stop rx ba session sta %pM tid %hu\n",
1740 		   peer->addr, tid);
1741 
1742 	ieee80211_stop_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1743 	spin_unlock_bh(&ar->data_lock);
1744 }
1745 
1746 static int ath10k_htt_rx_extract_amsdu(struct sk_buff_head *list,
1747 				       struct sk_buff_head *amsdu)
1748 {
1749 	struct sk_buff *msdu;
1750 	struct htt_rx_desc *rxd;
1751 
1752 	if (skb_queue_empty(list))
1753 		return -ENOBUFS;
1754 
1755 	if (WARN_ON(!skb_queue_empty(amsdu)))
1756 		return -EINVAL;
1757 
1758 	while ((msdu = __skb_dequeue(list))) {
1759 		__skb_queue_tail(amsdu, msdu);
1760 
1761 		rxd = (void *)msdu->data - sizeof(*rxd);
1762 		if (rxd->msdu_end.info0 &
1763 		    __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))
1764 			break;
1765 	}
1766 
1767 	msdu = skb_peek_tail(amsdu);
1768 	rxd = (void *)msdu->data - sizeof(*rxd);
1769 	if (!(rxd->msdu_end.info0 &
1770 	      __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))) {
1771 		skb_queue_splice_init(amsdu, list);
1772 		return -EAGAIN;
1773 	}
1774 
1775 	return 0;
1776 }
1777 
1778 static void ath10k_htt_rx_h_rx_offload_prot(struct ieee80211_rx_status *status,
1779 					    struct sk_buff *skb)
1780 {
1781 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1782 
1783 	if (!ieee80211_has_protected(hdr->frame_control))
1784 		return;
1785 
1786 	/* Offloaded frames are already decrypted but firmware insists they are
1787 	 * protected in the 802.11 header. Strip the flag.  Otherwise mac80211
1788 	 * will drop the frame.
1789 	 */
1790 
1791 	hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
1792 	status->flag |= RX_FLAG_DECRYPTED |
1793 			RX_FLAG_IV_STRIPPED |
1794 			RX_FLAG_MMIC_STRIPPED;
1795 }
1796 
1797 static void ath10k_htt_rx_h_rx_offload(struct ath10k *ar,
1798 				       struct sk_buff_head *list)
1799 {
1800 	struct ath10k_htt *htt = &ar->htt;
1801 	struct ieee80211_rx_status *status = &htt->rx_status;
1802 	struct htt_rx_offload_msdu *rx;
1803 	struct sk_buff *msdu;
1804 	size_t offset;
1805 
1806 	while ((msdu = __skb_dequeue(list))) {
1807 		/* Offloaded frames don't have Rx descriptor. Instead they have
1808 		 * a short meta information header.
1809 		 */
1810 
1811 		rx = (void *)msdu->data;
1812 
1813 		skb_put(msdu, sizeof(*rx));
1814 		skb_pull(msdu, sizeof(*rx));
1815 
1816 		if (skb_tailroom(msdu) < __le16_to_cpu(rx->msdu_len)) {
1817 			ath10k_warn(ar, "dropping frame: offloaded rx msdu is too long!\n");
1818 			dev_kfree_skb_any(msdu);
1819 			continue;
1820 		}
1821 
1822 		skb_put(msdu, __le16_to_cpu(rx->msdu_len));
1823 
1824 		/* Offloaded rx header length isn't multiple of 2 nor 4 so the
1825 		 * actual payload is unaligned. Align the frame.  Otherwise
1826 		 * mac80211 complains.  This shouldn't reduce performance much
1827 		 * because these offloaded frames are rare.
1828 		 */
1829 		offset = 4 - ((unsigned long)msdu->data & 3);
1830 		skb_put(msdu, offset);
1831 		memmove(msdu->data + offset, msdu->data, msdu->len);
1832 		skb_pull(msdu, offset);
1833 
1834 		/* FIXME: The frame is NWifi. Re-construct QoS Control
1835 		 * if possible later.
1836 		 */
1837 
1838 		memset(status, 0, sizeof(*status));
1839 		status->flag |= RX_FLAG_NO_SIGNAL_VAL;
1840 
1841 		ath10k_htt_rx_h_rx_offload_prot(status, msdu);
1842 		ath10k_htt_rx_h_channel(ar, status, NULL, rx->vdev_id);
1843 		ath10k_process_rx(ar, status, msdu);
1844 	}
1845 }
1846 
1847 static void ath10k_htt_rx_in_ord_ind(struct ath10k *ar, struct sk_buff *skb)
1848 {
1849 	struct ath10k_htt *htt = &ar->htt;
1850 	struct htt_resp *resp = (void *)skb->data;
1851 	struct ieee80211_rx_status *status = &htt->rx_status;
1852 	struct sk_buff_head list;
1853 	struct sk_buff_head amsdu;
1854 	u16 peer_id;
1855 	u16 msdu_count;
1856 	u8 vdev_id;
1857 	u8 tid;
1858 	bool offload;
1859 	bool frag;
1860 	int ret;
1861 
1862 	lockdep_assert_held(&htt->rx_ring.lock);
1863 
1864 	if (htt->rx_confused)
1865 		return;
1866 
1867 	skb_pull(skb, sizeof(resp->hdr));
1868 	skb_pull(skb, sizeof(resp->rx_in_ord_ind));
1869 
1870 	peer_id = __le16_to_cpu(resp->rx_in_ord_ind.peer_id);
1871 	msdu_count = __le16_to_cpu(resp->rx_in_ord_ind.msdu_count);
1872 	vdev_id = resp->rx_in_ord_ind.vdev_id;
1873 	tid = SM(resp->rx_in_ord_ind.info, HTT_RX_IN_ORD_IND_INFO_TID);
1874 	offload = !!(resp->rx_in_ord_ind.info &
1875 			HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK);
1876 	frag = !!(resp->rx_in_ord_ind.info & HTT_RX_IN_ORD_IND_INFO_FRAG_MASK);
1877 
1878 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1879 		   "htt rx in ord vdev %i peer %i tid %i offload %i frag %i msdu count %i\n",
1880 		   vdev_id, peer_id, tid, offload, frag, msdu_count);
1881 
1882 	if (skb->len < msdu_count * sizeof(*resp->rx_in_ord_ind.msdu_descs)) {
1883 		ath10k_warn(ar, "dropping invalid in order rx indication\n");
1884 		return;
1885 	}
1886 
1887 	/* The event can deliver more than 1 A-MSDU. Each A-MSDU is later
1888 	 * extracted and processed.
1889 	 */
1890 	__skb_queue_head_init(&list);
1891 	ret = ath10k_htt_rx_pop_paddr_list(htt, &resp->rx_in_ord_ind, &list);
1892 	if (ret < 0) {
1893 		ath10k_warn(ar, "failed to pop paddr list: %d\n", ret);
1894 		htt->rx_confused = true;
1895 		return;
1896 	}
1897 
1898 	/* Offloaded frames are very different and need to be handled
1899 	 * separately.
1900 	 */
1901 	if (offload)
1902 		ath10k_htt_rx_h_rx_offload(ar, &list);
1903 
1904 	while (!skb_queue_empty(&list)) {
1905 		__skb_queue_head_init(&amsdu);
1906 		ret = ath10k_htt_rx_extract_amsdu(&list, &amsdu);
1907 		switch (ret) {
1908 		case 0:
1909 			/* Note: The in-order indication may report interleaved
1910 			 * frames from different PPDUs meaning reported rx rate
1911 			 * to mac80211 isn't accurate/reliable. It's still
1912 			 * better to report something than nothing though. This
1913 			 * should still give an idea about rx rate to the user.
1914 			 */
1915 			ath10k_htt_rx_h_ppdu(ar, &amsdu, status, vdev_id);
1916 			ath10k_htt_rx_h_filter(ar, &amsdu, status);
1917 			ath10k_htt_rx_h_mpdu(ar, &amsdu, status);
1918 			ath10k_htt_rx_h_deliver(ar, &amsdu, status);
1919 			break;
1920 		case -EAGAIN:
1921 			/* fall through */
1922 		default:
1923 			/* Should not happen. */
1924 			ath10k_warn(ar, "failed to extract amsdu: %d\n", ret);
1925 			htt->rx_confused = true;
1926 			__skb_queue_purge(&list);
1927 			return;
1928 		}
1929 	}
1930 
1931 	tasklet_schedule(&htt->rx_replenish_task);
1932 }
1933 
1934 void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
1935 {
1936 	struct ath10k_htt *htt = &ar->htt;
1937 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1938 	enum htt_t2h_msg_type type;
1939 
1940 	/* confirm alignment */
1941 	if (!IS_ALIGNED((unsigned long)skb->data, 4))
1942 		ath10k_warn(ar, "unaligned htt message, expect trouble\n");
1943 
1944 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx, msg_type: 0x%0X\n",
1945 		   resp->hdr.msg_type);
1946 
1947 	if (resp->hdr.msg_type >= ar->htt.t2h_msg_types_max) {
1948 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx, unsupported msg_type: 0x%0X\n max: 0x%0X",
1949 			   resp->hdr.msg_type, ar->htt.t2h_msg_types_max);
1950 		dev_kfree_skb_any(skb);
1951 		return;
1952 	}
1953 	type = ar->htt.t2h_msg_types[resp->hdr.msg_type];
1954 
1955 	switch (type) {
1956 	case HTT_T2H_MSG_TYPE_VERSION_CONF: {
1957 		htt->target_version_major = resp->ver_resp.major;
1958 		htt->target_version_minor = resp->ver_resp.minor;
1959 		complete(&htt->target_version_received);
1960 		break;
1961 	}
1962 	case HTT_T2H_MSG_TYPE_RX_IND:
1963 		spin_lock_bh(&htt->rx_ring.lock);
1964 		__skb_queue_tail(&htt->rx_compl_q, skb);
1965 		spin_unlock_bh(&htt->rx_ring.lock);
1966 		tasklet_schedule(&htt->txrx_compl_task);
1967 		return;
1968 	case HTT_T2H_MSG_TYPE_PEER_MAP: {
1969 		struct htt_peer_map_event ev = {
1970 			.vdev_id = resp->peer_map.vdev_id,
1971 			.peer_id = __le16_to_cpu(resp->peer_map.peer_id),
1972 		};
1973 		memcpy(ev.addr, resp->peer_map.addr, sizeof(ev.addr));
1974 		ath10k_peer_map_event(htt, &ev);
1975 		break;
1976 	}
1977 	case HTT_T2H_MSG_TYPE_PEER_UNMAP: {
1978 		struct htt_peer_unmap_event ev = {
1979 			.peer_id = __le16_to_cpu(resp->peer_unmap.peer_id),
1980 		};
1981 		ath10k_peer_unmap_event(htt, &ev);
1982 		break;
1983 	}
1984 	case HTT_T2H_MSG_TYPE_MGMT_TX_COMPLETION: {
1985 		struct htt_tx_done tx_done = {};
1986 		int status = __le32_to_cpu(resp->mgmt_tx_completion.status);
1987 
1988 		tx_done.msdu_id =
1989 			__le32_to_cpu(resp->mgmt_tx_completion.desc_id);
1990 
1991 		switch (status) {
1992 		case HTT_MGMT_TX_STATUS_OK:
1993 			tx_done.success = true;
1994 			break;
1995 		case HTT_MGMT_TX_STATUS_RETRY:
1996 			tx_done.no_ack = true;
1997 			break;
1998 		case HTT_MGMT_TX_STATUS_DROP:
1999 			tx_done.discard = true;
2000 			break;
2001 		}
2002 
2003 		spin_lock_bh(&htt->tx_lock);
2004 		ath10k_txrx_tx_unref(htt, &tx_done);
2005 		spin_unlock_bh(&htt->tx_lock);
2006 		break;
2007 	}
2008 	case HTT_T2H_MSG_TYPE_TX_COMPL_IND:
2009 		spin_lock_bh(&htt->tx_lock);
2010 		__skb_queue_tail(&htt->tx_compl_q, skb);
2011 		spin_unlock_bh(&htt->tx_lock);
2012 		tasklet_schedule(&htt->txrx_compl_task);
2013 		return;
2014 	case HTT_T2H_MSG_TYPE_SEC_IND: {
2015 		struct ath10k *ar = htt->ar;
2016 		struct htt_security_indication *ev = &resp->security_indication;
2017 
2018 		ath10k_dbg(ar, ATH10K_DBG_HTT,
2019 			   "sec ind peer_id %d unicast %d type %d\n",
2020 			  __le16_to_cpu(ev->peer_id),
2021 			  !!(ev->flags & HTT_SECURITY_IS_UNICAST),
2022 			  MS(ev->flags, HTT_SECURITY_TYPE));
2023 		complete(&ar->install_key_done);
2024 		break;
2025 	}
2026 	case HTT_T2H_MSG_TYPE_RX_FRAG_IND: {
2027 		ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
2028 				skb->data, skb->len);
2029 		ath10k_htt_rx_frag_handler(htt, &resp->rx_frag_ind);
2030 		break;
2031 	}
2032 	case HTT_T2H_MSG_TYPE_TEST:
2033 		break;
2034 	case HTT_T2H_MSG_TYPE_STATS_CONF:
2035 		trace_ath10k_htt_stats(ar, skb->data, skb->len);
2036 		break;
2037 	case HTT_T2H_MSG_TYPE_TX_INSPECT_IND:
2038 		/* Firmware can return tx frames if it's unable to fully
2039 		 * process them and suspects host may be able to fix it. ath10k
2040 		 * sends all tx frames as already inspected so this shouldn't
2041 		 * happen unless fw has a bug.
2042 		 */
2043 		ath10k_warn(ar, "received an unexpected htt tx inspect event\n");
2044 		break;
2045 	case HTT_T2H_MSG_TYPE_RX_ADDBA:
2046 		ath10k_htt_rx_addba(ar, resp);
2047 		break;
2048 	case HTT_T2H_MSG_TYPE_RX_DELBA:
2049 		ath10k_htt_rx_delba(ar, resp);
2050 		break;
2051 	case HTT_T2H_MSG_TYPE_PKTLOG: {
2052 		struct ath10k_pktlog_hdr *hdr =
2053 			(struct ath10k_pktlog_hdr *)resp->pktlog_msg.payload;
2054 
2055 		trace_ath10k_htt_pktlog(ar, resp->pktlog_msg.payload,
2056 					sizeof(*hdr) +
2057 					__le16_to_cpu(hdr->size));
2058 		break;
2059 	}
2060 	case HTT_T2H_MSG_TYPE_RX_FLUSH: {
2061 		/* Ignore this event because mac80211 takes care of Rx
2062 		 * aggregation reordering.
2063 		 */
2064 		break;
2065 	}
2066 	case HTT_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND: {
2067 		spin_lock_bh(&htt->rx_ring.lock);
2068 		__skb_queue_tail(&htt->rx_in_ord_compl_q, skb);
2069 		spin_unlock_bh(&htt->rx_ring.lock);
2070 		tasklet_schedule(&htt->txrx_compl_task);
2071 		return;
2072 	}
2073 	case HTT_T2H_MSG_TYPE_TX_CREDIT_UPDATE_IND:
2074 		break;
2075 	case HTT_T2H_MSG_TYPE_CHAN_CHANGE:
2076 		break;
2077 	default:
2078 		ath10k_warn(ar, "htt event (%d) not handled\n",
2079 			    resp->hdr.msg_type);
2080 		ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
2081 				skb->data, skb->len);
2082 		break;
2083 	};
2084 
2085 	/* Free the indication buffer */
2086 	dev_kfree_skb_any(skb);
2087 }
2088 
2089 static void ath10k_htt_txrx_compl_task(unsigned long ptr)
2090 {
2091 	struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
2092 	struct ath10k *ar = htt->ar;
2093 	struct htt_resp *resp;
2094 	struct sk_buff *skb;
2095 
2096 	spin_lock_bh(&htt->tx_lock);
2097 	while ((skb = __skb_dequeue(&htt->tx_compl_q))) {
2098 		ath10k_htt_rx_frm_tx_compl(htt->ar, skb);
2099 		dev_kfree_skb_any(skb);
2100 	}
2101 	spin_unlock_bh(&htt->tx_lock);
2102 
2103 	spin_lock_bh(&htt->rx_ring.lock);
2104 	while ((skb = __skb_dequeue(&htt->rx_compl_q))) {
2105 		resp = (struct htt_resp *)skb->data;
2106 		ath10k_htt_rx_handler(htt, &resp->rx_ind);
2107 		dev_kfree_skb_any(skb);
2108 	}
2109 
2110 	while ((skb = __skb_dequeue(&htt->rx_in_ord_compl_q))) {
2111 		ath10k_htt_rx_in_ord_ind(ar, skb);
2112 		dev_kfree_skb_any(skb);
2113 	}
2114 	spin_unlock_bh(&htt->rx_ring.lock);
2115 }
2116