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 rfc1042_hdr {
641 	u8 llc_dsap;
642 	u8 llc_ssap;
643 	u8 llc_ctrl;
644 	u8 snap_oui[3];
645 	__be16 snap_type;
646 } __packed;
647 
648 struct amsdu_subframe_hdr {
649 	u8 dst[ETH_ALEN];
650 	u8 src[ETH_ALEN];
651 	__be16 len;
652 } __packed;
653 
654 static const u8 rx_legacy_rate_idx[] = {
655 	3,	/* 0x00  - 11Mbps  */
656 	2,	/* 0x01  - 5.5Mbps */
657 	1,	/* 0x02  - 2Mbps   */
658 	0,	/* 0x03  - 1Mbps   */
659 	3,	/* 0x04  - 11Mbps  */
660 	2,	/* 0x05  - 5.5Mbps */
661 	1,	/* 0x06  - 2Mbps   */
662 	0,	/* 0x07  - 1Mbps   */
663 	10,	/* 0x08  - 48Mbps  */
664 	8,	/* 0x09  - 24Mbps  */
665 	6,	/* 0x0A  - 12Mbps  */
666 	4,	/* 0x0B  - 6Mbps   */
667 	11,	/* 0x0C  - 54Mbps  */
668 	9,	/* 0x0D  - 36Mbps  */
669 	7,	/* 0x0E  - 18Mbps  */
670 	5,	/* 0x0F  - 9Mbps   */
671 };
672 
673 static void ath10k_htt_rx_h_rates(struct ath10k *ar,
674 				  struct ieee80211_rx_status *status,
675 				  struct htt_rx_desc *rxd)
676 {
677 	enum ieee80211_band band;
678 	u8 cck, rate, rate_idx, bw, sgi, mcs, nss;
679 	u8 preamble = 0;
680 	u32 info1, info2, info3;
681 
682 	/* Band value can't be set as undefined but freq can be 0 - use that to
683 	 * determine whether band is provided.
684 	 *
685 	 * FIXME: Perhaps this can go away if CCK rate reporting is a little
686 	 * reworked?
687 	 */
688 	if (!status->freq)
689 		return;
690 
691 	band = status->band;
692 	info1 = __le32_to_cpu(rxd->ppdu_start.info1);
693 	info2 = __le32_to_cpu(rxd->ppdu_start.info2);
694 	info3 = __le32_to_cpu(rxd->ppdu_start.info3);
695 
696 	preamble = MS(info1, RX_PPDU_START_INFO1_PREAMBLE_TYPE);
697 
698 	switch (preamble) {
699 	case HTT_RX_LEGACY:
700 		cck = info1 & RX_PPDU_START_INFO1_L_SIG_RATE_SELECT;
701 		rate = MS(info1, RX_PPDU_START_INFO1_L_SIG_RATE);
702 		rate_idx = 0;
703 
704 		if (rate < 0x08 || rate > 0x0F)
705 			break;
706 
707 		switch (band) {
708 		case IEEE80211_BAND_2GHZ:
709 			if (cck)
710 				rate &= ~BIT(3);
711 			rate_idx = rx_legacy_rate_idx[rate];
712 			break;
713 		case IEEE80211_BAND_5GHZ:
714 			rate_idx = rx_legacy_rate_idx[rate];
715 			/* We are using same rate table registering
716 			   HW - ath10k_rates[]. In case of 5GHz skip
717 			   CCK rates, so -4 here */
718 			rate_idx -= 4;
719 			break;
720 		default:
721 			break;
722 		}
723 
724 		status->rate_idx = rate_idx;
725 		break;
726 	case HTT_RX_HT:
727 	case HTT_RX_HT_WITH_TXBF:
728 		/* HT-SIG - Table 20-11 in info2 and info3 */
729 		mcs = info2 & 0x1F;
730 		nss = mcs >> 3;
731 		bw = (info2 >> 7) & 1;
732 		sgi = (info3 >> 7) & 1;
733 
734 		status->rate_idx = mcs;
735 		status->flag |= RX_FLAG_HT;
736 		if (sgi)
737 			status->flag |= RX_FLAG_SHORT_GI;
738 		if (bw)
739 			status->flag |= RX_FLAG_40MHZ;
740 		break;
741 	case HTT_RX_VHT:
742 	case HTT_RX_VHT_WITH_TXBF:
743 		/* VHT-SIG-A1 in info2, VHT-SIG-A2 in info3
744 		   TODO check this */
745 		mcs = (info3 >> 4) & 0x0F;
746 		nss = ((info2 >> 10) & 0x07) + 1;
747 		bw = info2 & 3;
748 		sgi = info3 & 1;
749 
750 		status->rate_idx = mcs;
751 		status->vht_nss = nss;
752 
753 		if (sgi)
754 			status->flag |= RX_FLAG_SHORT_GI;
755 
756 		switch (bw) {
757 		/* 20MHZ */
758 		case 0:
759 			break;
760 		/* 40MHZ */
761 		case 1:
762 			status->flag |= RX_FLAG_40MHZ;
763 			break;
764 		/* 80MHZ */
765 		case 2:
766 			status->vht_flag |= RX_VHT_FLAG_80MHZ;
767 		}
768 
769 		status->flag |= RX_FLAG_VHT;
770 		break;
771 	default:
772 		break;
773 	}
774 }
775 
776 static bool ath10k_htt_rx_h_channel(struct ath10k *ar,
777 				    struct ieee80211_rx_status *status)
778 {
779 	struct ieee80211_channel *ch;
780 
781 	spin_lock_bh(&ar->data_lock);
782 	ch = ar->scan_channel;
783 	if (!ch)
784 		ch = ar->rx_channel;
785 	spin_unlock_bh(&ar->data_lock);
786 
787 	if (!ch)
788 		return false;
789 
790 	status->band = ch->band;
791 	status->freq = ch->center_freq;
792 
793 	return true;
794 }
795 
796 static void ath10k_htt_rx_h_signal(struct ath10k *ar,
797 				   struct ieee80211_rx_status *status,
798 				   struct htt_rx_desc *rxd)
799 {
800 	/* FIXME: Get real NF */
801 	status->signal = ATH10K_DEFAULT_NOISE_FLOOR +
802 			 rxd->ppdu_start.rssi_comb;
803 	status->flag &= ~RX_FLAG_NO_SIGNAL_VAL;
804 }
805 
806 static void ath10k_htt_rx_h_mactime(struct ath10k *ar,
807 				    struct ieee80211_rx_status *status,
808 				    struct htt_rx_desc *rxd)
809 {
810 	/* FIXME: TSF is known only at the end of PPDU, in the last MPDU. This
811 	 * means all prior MSDUs in a PPDU are reported to mac80211 without the
812 	 * TSF. Is it worth holding frames until end of PPDU is known?
813 	 *
814 	 * FIXME: Can we get/compute 64bit TSF?
815 	 */
816 	status->mactime = __le32_to_cpu(rxd->ppdu_end.common.tsf_timestamp);
817 	status->flag |= RX_FLAG_MACTIME_END;
818 }
819 
820 static void ath10k_htt_rx_h_ppdu(struct ath10k *ar,
821 				 struct sk_buff_head *amsdu,
822 				 struct ieee80211_rx_status *status)
823 {
824 	struct sk_buff *first;
825 	struct htt_rx_desc *rxd;
826 	bool is_first_ppdu;
827 	bool is_last_ppdu;
828 
829 	if (skb_queue_empty(amsdu))
830 		return;
831 
832 	first = skb_peek(amsdu);
833 	rxd = (void *)first->data - sizeof(*rxd);
834 
835 	is_first_ppdu = !!(rxd->attention.flags &
836 			   __cpu_to_le32(RX_ATTENTION_FLAGS_FIRST_MPDU));
837 	is_last_ppdu = !!(rxd->attention.flags &
838 			  __cpu_to_le32(RX_ATTENTION_FLAGS_LAST_MPDU));
839 
840 	if (is_first_ppdu) {
841 		/* New PPDU starts so clear out the old per-PPDU status. */
842 		status->freq = 0;
843 		status->rate_idx = 0;
844 		status->vht_nss = 0;
845 		status->vht_flag &= ~RX_VHT_FLAG_80MHZ;
846 		status->flag &= ~(RX_FLAG_HT |
847 				  RX_FLAG_VHT |
848 				  RX_FLAG_SHORT_GI |
849 				  RX_FLAG_40MHZ |
850 				  RX_FLAG_MACTIME_END);
851 		status->flag |= RX_FLAG_NO_SIGNAL_VAL;
852 
853 		ath10k_htt_rx_h_signal(ar, status, rxd);
854 		ath10k_htt_rx_h_channel(ar, status);
855 		ath10k_htt_rx_h_rates(ar, status, rxd);
856 	}
857 
858 	if (is_last_ppdu)
859 		ath10k_htt_rx_h_mactime(ar, status, rxd);
860 }
861 
862 static const char * const tid_to_ac[] = {
863 	"BE",
864 	"BK",
865 	"BK",
866 	"BE",
867 	"VI",
868 	"VI",
869 	"VO",
870 	"VO",
871 };
872 
873 static char *ath10k_get_tid(struct ieee80211_hdr *hdr, char *out, size_t size)
874 {
875 	u8 *qc;
876 	int tid;
877 
878 	if (!ieee80211_is_data_qos(hdr->frame_control))
879 		return "";
880 
881 	qc = ieee80211_get_qos_ctl(hdr);
882 	tid = *qc & IEEE80211_QOS_CTL_TID_MASK;
883 	if (tid < 8)
884 		snprintf(out, size, "tid %d (%s)", tid, tid_to_ac[tid]);
885 	else
886 		snprintf(out, size, "tid %d", tid);
887 
888 	return out;
889 }
890 
891 static void ath10k_process_rx(struct ath10k *ar,
892 			      struct ieee80211_rx_status *rx_status,
893 			      struct sk_buff *skb)
894 {
895 	struct ieee80211_rx_status *status;
896 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
897 	char tid[32];
898 
899 	status = IEEE80211_SKB_RXCB(skb);
900 	*status = *rx_status;
901 
902 	ath10k_dbg(ar, ATH10K_DBG_DATA,
903 		   "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",
904 		   skb,
905 		   skb->len,
906 		   ieee80211_get_SA(hdr),
907 		   ath10k_get_tid(hdr, tid, sizeof(tid)),
908 		   is_multicast_ether_addr(ieee80211_get_DA(hdr)) ?
909 							"mcast" : "ucast",
910 		   (__le16_to_cpu(hdr->seq_ctrl) & IEEE80211_SCTL_SEQ) >> 4,
911 		   status->flag == 0 ? "legacy" : "",
912 		   status->flag & RX_FLAG_HT ? "ht" : "",
913 		   status->flag & RX_FLAG_VHT ? "vht" : "",
914 		   status->flag & RX_FLAG_40MHZ ? "40" : "",
915 		   status->vht_flag & RX_VHT_FLAG_80MHZ ? "80" : "",
916 		   status->flag & RX_FLAG_SHORT_GI ? "sgi " : "",
917 		   status->rate_idx,
918 		   status->vht_nss,
919 		   status->freq,
920 		   status->band, status->flag,
921 		   !!(status->flag & RX_FLAG_FAILED_FCS_CRC),
922 		   !!(status->flag & RX_FLAG_MMIC_ERROR),
923 		   !!(status->flag & RX_FLAG_AMSDU_MORE));
924 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "rx skb: ",
925 			skb->data, skb->len);
926 	trace_ath10k_rx_hdr(ar, skb->data, skb->len);
927 	trace_ath10k_rx_payload(ar, skb->data, skb->len);
928 
929 	ieee80211_rx(ar->hw, skb);
930 }
931 
932 static int ath10k_htt_rx_nwifi_hdrlen(struct ieee80211_hdr *hdr)
933 {
934 	/* nwifi header is padded to 4 bytes. this fixes 4addr rx */
935 	return round_up(ieee80211_hdrlen(hdr->frame_control), 4);
936 }
937 
938 static void ath10k_htt_rx_h_undecap_raw(struct ath10k *ar,
939 					struct sk_buff *msdu,
940 					struct ieee80211_rx_status *status,
941 					enum htt_rx_mpdu_encrypt_type enctype,
942 					bool is_decrypted)
943 {
944 	struct ieee80211_hdr *hdr;
945 	struct htt_rx_desc *rxd;
946 	size_t hdr_len;
947 	size_t crypto_len;
948 	bool is_first;
949 	bool is_last;
950 
951 	rxd = (void *)msdu->data - sizeof(*rxd);
952 	is_first = !!(rxd->msdu_end.info0 &
953 		      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU));
954 	is_last = !!(rxd->msdu_end.info0 &
955 		     __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU));
956 
957 	/* Delivered decapped frame:
958 	 * [802.11 header]
959 	 * [crypto param] <-- can be trimmed if !fcs_err &&
960 	 *                    !decrypt_err && !peer_idx_invalid
961 	 * [amsdu header] <-- only if A-MSDU
962 	 * [rfc1042/llc]
963 	 * [payload]
964 	 * [FCS] <-- at end, needs to be trimmed
965 	 */
966 
967 	/* This probably shouldn't happen but warn just in case */
968 	if (unlikely(WARN_ON_ONCE(!is_first)))
969 		return;
970 
971 	/* This probably shouldn't happen but warn just in case */
972 	if (unlikely(WARN_ON_ONCE(!(is_first && is_last))))
973 		return;
974 
975 	skb_trim(msdu, msdu->len - FCS_LEN);
976 
977 	/* In most cases this will be true for sniffed frames. It makes sense
978 	 * to deliver them as-is without stripping the crypto param. This would
979 	 * also make sense for software based decryption (which is not
980 	 * implemented in ath10k).
981 	 *
982 	 * If there's no error then the frame is decrypted. At least that is
983 	 * the case for frames that come in via fragmented rx indication.
984 	 */
985 	if (!is_decrypted)
986 		return;
987 
988 	/* The payload is decrypted so strip crypto params. Start from tail
989 	 * since hdr is used to compute some stuff.
990 	 */
991 
992 	hdr = (void *)msdu->data;
993 
994 	/* Tail */
995 	skb_trim(msdu, msdu->len - ath10k_htt_rx_crypto_tail_len(ar, enctype));
996 
997 	/* MMIC */
998 	if (!ieee80211_has_morefrags(hdr->frame_control) &&
999 	    enctype == HTT_RX_MPDU_ENCRYPT_TKIP_WPA)
1000 		skb_trim(msdu, msdu->len - 8);
1001 
1002 	/* Head */
1003 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1004 	crypto_len = ath10k_htt_rx_crypto_param_len(ar, enctype);
1005 
1006 	memmove((void *)msdu->data + crypto_len,
1007 		(void *)msdu->data, hdr_len);
1008 	skb_pull(msdu, crypto_len);
1009 }
1010 
1011 static void ath10k_htt_rx_h_undecap_nwifi(struct ath10k *ar,
1012 					  struct sk_buff *msdu,
1013 					  struct ieee80211_rx_status *status,
1014 					  const u8 first_hdr[64])
1015 {
1016 	struct ieee80211_hdr *hdr;
1017 	size_t hdr_len;
1018 	u8 da[ETH_ALEN];
1019 	u8 sa[ETH_ALEN];
1020 
1021 	/* Delivered decapped frame:
1022 	 * [nwifi 802.11 header] <-- replaced with 802.11 hdr
1023 	 * [rfc1042/llc]
1024 	 *
1025 	 * Note: The nwifi header doesn't have QoS Control and is
1026 	 * (always?) a 3addr frame.
1027 	 *
1028 	 * Note2: There's no A-MSDU subframe header. Even if it's part
1029 	 * of an A-MSDU.
1030 	 */
1031 
1032 	/* pull decapped header and copy SA & DA */
1033 	hdr = (struct ieee80211_hdr *)msdu->data;
1034 	hdr_len = ath10k_htt_rx_nwifi_hdrlen(hdr);
1035 	ether_addr_copy(da, ieee80211_get_DA(hdr));
1036 	ether_addr_copy(sa, ieee80211_get_SA(hdr));
1037 	skb_pull(msdu, hdr_len);
1038 
1039 	/* push original 802.11 header */
1040 	hdr = (struct ieee80211_hdr *)first_hdr;
1041 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1042 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1043 
1044 	/* original 802.11 header has a different DA and in
1045 	 * case of 4addr it may also have different SA
1046 	 */
1047 	hdr = (struct ieee80211_hdr *)msdu->data;
1048 	ether_addr_copy(ieee80211_get_DA(hdr), da);
1049 	ether_addr_copy(ieee80211_get_SA(hdr), sa);
1050 }
1051 
1052 static void *ath10k_htt_rx_h_find_rfc1042(struct ath10k *ar,
1053 					  struct sk_buff *msdu,
1054 					  enum htt_rx_mpdu_encrypt_type enctype)
1055 {
1056 	struct ieee80211_hdr *hdr;
1057 	struct htt_rx_desc *rxd;
1058 	size_t hdr_len, crypto_len;
1059 	void *rfc1042;
1060 	bool is_first, is_last, is_amsdu;
1061 
1062 	rxd = (void *)msdu->data - sizeof(*rxd);
1063 	hdr = (void *)rxd->rx_hdr_status;
1064 
1065 	is_first = !!(rxd->msdu_end.info0 &
1066 		      __cpu_to_le32(RX_MSDU_END_INFO0_FIRST_MSDU));
1067 	is_last = !!(rxd->msdu_end.info0 &
1068 		     __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU));
1069 	is_amsdu = !(is_first && is_last);
1070 
1071 	rfc1042 = hdr;
1072 
1073 	if (is_first) {
1074 		hdr_len = ieee80211_hdrlen(hdr->frame_control);
1075 		crypto_len = ath10k_htt_rx_crypto_param_len(ar, enctype);
1076 
1077 		rfc1042 += round_up(hdr_len, 4) +
1078 			   round_up(crypto_len, 4);
1079 	}
1080 
1081 	if (is_amsdu)
1082 		rfc1042 += sizeof(struct amsdu_subframe_hdr);
1083 
1084 	return rfc1042;
1085 }
1086 
1087 static void ath10k_htt_rx_h_undecap_eth(struct ath10k *ar,
1088 					struct sk_buff *msdu,
1089 					struct ieee80211_rx_status *status,
1090 					const u8 first_hdr[64],
1091 					enum htt_rx_mpdu_encrypt_type enctype)
1092 {
1093 	struct ieee80211_hdr *hdr;
1094 	struct ethhdr *eth;
1095 	size_t hdr_len;
1096 	void *rfc1042;
1097 	u8 da[ETH_ALEN];
1098 	u8 sa[ETH_ALEN];
1099 
1100 	/* Delivered decapped frame:
1101 	 * [eth header] <-- replaced with 802.11 hdr & rfc1042/llc
1102 	 * [payload]
1103 	 */
1104 
1105 	rfc1042 = ath10k_htt_rx_h_find_rfc1042(ar, msdu, enctype);
1106 	if (WARN_ON_ONCE(!rfc1042))
1107 		return;
1108 
1109 	/* pull decapped header and copy SA & DA */
1110 	eth = (struct ethhdr *)msdu->data;
1111 	ether_addr_copy(da, eth->h_dest);
1112 	ether_addr_copy(sa, eth->h_source);
1113 	skb_pull(msdu, sizeof(struct ethhdr));
1114 
1115 	/* push rfc1042/llc/snap */
1116 	memcpy(skb_push(msdu, sizeof(struct rfc1042_hdr)), rfc1042,
1117 	       sizeof(struct rfc1042_hdr));
1118 
1119 	/* push original 802.11 header */
1120 	hdr = (struct ieee80211_hdr *)first_hdr;
1121 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1122 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1123 
1124 	/* original 802.11 header has a different DA and in
1125 	 * case of 4addr it may also have different SA
1126 	 */
1127 	hdr = (struct ieee80211_hdr *)msdu->data;
1128 	ether_addr_copy(ieee80211_get_DA(hdr), da);
1129 	ether_addr_copy(ieee80211_get_SA(hdr), sa);
1130 }
1131 
1132 static void ath10k_htt_rx_h_undecap_snap(struct ath10k *ar,
1133 					 struct sk_buff *msdu,
1134 					 struct ieee80211_rx_status *status,
1135 					 const u8 first_hdr[64])
1136 {
1137 	struct ieee80211_hdr *hdr;
1138 	size_t hdr_len;
1139 
1140 	/* Delivered decapped frame:
1141 	 * [amsdu header] <-- replaced with 802.11 hdr
1142 	 * [rfc1042/llc]
1143 	 * [payload]
1144 	 */
1145 
1146 	skb_pull(msdu, sizeof(struct amsdu_subframe_hdr));
1147 
1148 	hdr = (struct ieee80211_hdr *)first_hdr;
1149 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1150 	memcpy(skb_push(msdu, hdr_len), hdr, hdr_len);
1151 }
1152 
1153 static void ath10k_htt_rx_h_undecap(struct ath10k *ar,
1154 				    struct sk_buff *msdu,
1155 				    struct ieee80211_rx_status *status,
1156 				    u8 first_hdr[64],
1157 				    enum htt_rx_mpdu_encrypt_type enctype,
1158 				    bool is_decrypted)
1159 {
1160 	struct htt_rx_desc *rxd;
1161 	enum rx_msdu_decap_format decap;
1162 	struct ieee80211_hdr *hdr;
1163 
1164 	/* First msdu's decapped header:
1165 	 * [802.11 header] <-- padded to 4 bytes long
1166 	 * [crypto param] <-- padded to 4 bytes long
1167 	 * [amsdu header] <-- only if A-MSDU
1168 	 * [rfc1042/llc]
1169 	 *
1170 	 * Other (2nd, 3rd, ..) msdu's decapped header:
1171 	 * [amsdu header] <-- only if A-MSDU
1172 	 * [rfc1042/llc]
1173 	 */
1174 
1175 	rxd = (void *)msdu->data - sizeof(*rxd);
1176 	hdr = (void *)rxd->rx_hdr_status;
1177 	decap = MS(__le32_to_cpu(rxd->msdu_start.info1),
1178 		   RX_MSDU_START_INFO1_DECAP_FORMAT);
1179 
1180 	switch (decap) {
1181 	case RX_MSDU_DECAP_RAW:
1182 		ath10k_htt_rx_h_undecap_raw(ar, msdu, status, enctype,
1183 					    is_decrypted);
1184 		break;
1185 	case RX_MSDU_DECAP_NATIVE_WIFI:
1186 		ath10k_htt_rx_h_undecap_nwifi(ar, msdu, status, first_hdr);
1187 		break;
1188 	case RX_MSDU_DECAP_ETHERNET2_DIX:
1189 		ath10k_htt_rx_h_undecap_eth(ar, msdu, status, first_hdr, enctype);
1190 		break;
1191 	case RX_MSDU_DECAP_8023_SNAP_LLC:
1192 		ath10k_htt_rx_h_undecap_snap(ar, msdu, status, first_hdr);
1193 		break;
1194 	}
1195 }
1196 
1197 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb)
1198 {
1199 	struct htt_rx_desc *rxd;
1200 	u32 flags, info;
1201 	bool is_ip4, is_ip6;
1202 	bool is_tcp, is_udp;
1203 	bool ip_csum_ok, tcpudp_csum_ok;
1204 
1205 	rxd = (void *)skb->data - sizeof(*rxd);
1206 	flags = __le32_to_cpu(rxd->attention.flags);
1207 	info = __le32_to_cpu(rxd->msdu_start.info1);
1208 
1209 	is_ip4 = !!(info & RX_MSDU_START_INFO1_IPV4_PROTO);
1210 	is_ip6 = !!(info & RX_MSDU_START_INFO1_IPV6_PROTO);
1211 	is_tcp = !!(info & RX_MSDU_START_INFO1_TCP_PROTO);
1212 	is_udp = !!(info & RX_MSDU_START_INFO1_UDP_PROTO);
1213 	ip_csum_ok = !(flags & RX_ATTENTION_FLAGS_IP_CHKSUM_FAIL);
1214 	tcpudp_csum_ok = !(flags & RX_ATTENTION_FLAGS_TCP_UDP_CHKSUM_FAIL);
1215 
1216 	if (!is_ip4 && !is_ip6)
1217 		return CHECKSUM_NONE;
1218 	if (!is_tcp && !is_udp)
1219 		return CHECKSUM_NONE;
1220 	if (!ip_csum_ok)
1221 		return CHECKSUM_NONE;
1222 	if (!tcpudp_csum_ok)
1223 		return CHECKSUM_NONE;
1224 
1225 	return CHECKSUM_UNNECESSARY;
1226 }
1227 
1228 static void ath10k_htt_rx_h_csum_offload(struct sk_buff *msdu)
1229 {
1230 	msdu->ip_summed = ath10k_htt_rx_get_csum_state(msdu);
1231 }
1232 
1233 static void ath10k_htt_rx_h_mpdu(struct ath10k *ar,
1234 				 struct sk_buff_head *amsdu,
1235 				 struct ieee80211_rx_status *status)
1236 {
1237 	struct sk_buff *first;
1238 	struct sk_buff *last;
1239 	struct sk_buff *msdu;
1240 	struct htt_rx_desc *rxd;
1241 	struct ieee80211_hdr *hdr;
1242 	enum htt_rx_mpdu_encrypt_type enctype;
1243 	u8 first_hdr[64];
1244 	u8 *qos;
1245 	size_t hdr_len;
1246 	bool has_fcs_err;
1247 	bool has_crypto_err;
1248 	bool has_tkip_err;
1249 	bool has_peer_idx_invalid;
1250 	bool is_decrypted;
1251 	u32 attention;
1252 
1253 	if (skb_queue_empty(amsdu))
1254 		return;
1255 
1256 	first = skb_peek(amsdu);
1257 	rxd = (void *)first->data - sizeof(*rxd);
1258 
1259 	enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
1260 		     RX_MPDU_START_INFO0_ENCRYPT_TYPE);
1261 
1262 	/* First MSDU's Rx descriptor in an A-MSDU contains full 802.11
1263 	 * decapped header. It'll be used for undecapping of each MSDU.
1264 	 */
1265 	hdr = (void *)rxd->rx_hdr_status;
1266 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
1267 	memcpy(first_hdr, hdr, hdr_len);
1268 
1269 	/* Each A-MSDU subframe will use the original header as the base and be
1270 	 * reported as a separate MSDU so strip the A-MSDU bit from QoS Ctl.
1271 	 */
1272 	hdr = (void *)first_hdr;
1273 	qos = ieee80211_get_qos_ctl(hdr);
1274 	qos[0] &= ~IEEE80211_QOS_CTL_A_MSDU_PRESENT;
1275 
1276 	/* Some attention flags are valid only in the last MSDU. */
1277 	last = skb_peek_tail(amsdu);
1278 	rxd = (void *)last->data - sizeof(*rxd);
1279 	attention = __le32_to_cpu(rxd->attention.flags);
1280 
1281 	has_fcs_err = !!(attention & RX_ATTENTION_FLAGS_FCS_ERR);
1282 	has_crypto_err = !!(attention & RX_ATTENTION_FLAGS_DECRYPT_ERR);
1283 	has_tkip_err = !!(attention & RX_ATTENTION_FLAGS_TKIP_MIC_ERR);
1284 	has_peer_idx_invalid = !!(attention & RX_ATTENTION_FLAGS_PEER_IDX_INVALID);
1285 
1286 	/* Note: If hardware captures an encrypted frame that it can't decrypt,
1287 	 * e.g. due to fcs error, missing peer or invalid key data it will
1288 	 * report the frame as raw.
1289 	 */
1290 	is_decrypted = (enctype != HTT_RX_MPDU_ENCRYPT_NONE &&
1291 			!has_fcs_err &&
1292 			!has_crypto_err &&
1293 			!has_peer_idx_invalid);
1294 
1295 	/* Clear per-MPDU flags while leaving per-PPDU flags intact. */
1296 	status->flag &= ~(RX_FLAG_FAILED_FCS_CRC |
1297 			  RX_FLAG_MMIC_ERROR |
1298 			  RX_FLAG_DECRYPTED |
1299 			  RX_FLAG_IV_STRIPPED |
1300 			  RX_FLAG_MMIC_STRIPPED);
1301 
1302 	if (has_fcs_err)
1303 		status->flag |= RX_FLAG_FAILED_FCS_CRC;
1304 
1305 	if (has_tkip_err)
1306 		status->flag |= RX_FLAG_MMIC_ERROR;
1307 
1308 	if (is_decrypted)
1309 		status->flag |= RX_FLAG_DECRYPTED |
1310 				RX_FLAG_IV_STRIPPED |
1311 				RX_FLAG_MMIC_STRIPPED;
1312 
1313 	skb_queue_walk(amsdu, msdu) {
1314 		ath10k_htt_rx_h_csum_offload(msdu);
1315 		ath10k_htt_rx_h_undecap(ar, msdu, status, first_hdr, enctype,
1316 					is_decrypted);
1317 
1318 		/* Undecapping involves copying the original 802.11 header back
1319 		 * to sk_buff. If frame is protected and hardware has decrypted
1320 		 * it then remove the protected bit.
1321 		 */
1322 		if (!is_decrypted)
1323 			continue;
1324 
1325 		hdr = (void *)msdu->data;
1326 		hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
1327 	}
1328 }
1329 
1330 static void ath10k_htt_rx_h_deliver(struct ath10k *ar,
1331 				    struct sk_buff_head *amsdu,
1332 				    struct ieee80211_rx_status *status)
1333 {
1334 	struct sk_buff *msdu;
1335 
1336 	while ((msdu = __skb_dequeue(amsdu))) {
1337 		/* Setup per-MSDU flags */
1338 		if (skb_queue_empty(amsdu))
1339 			status->flag &= ~RX_FLAG_AMSDU_MORE;
1340 		else
1341 			status->flag |= RX_FLAG_AMSDU_MORE;
1342 
1343 		ath10k_process_rx(ar, status, msdu);
1344 	}
1345 }
1346 
1347 static int ath10k_unchain_msdu(struct sk_buff_head *amsdu)
1348 {
1349 	struct sk_buff *skb, *first;
1350 	int space;
1351 	int total_len = 0;
1352 
1353 	/* TODO:  Might could optimize this by using
1354 	 * skb_try_coalesce or similar method to
1355 	 * decrease copying, or maybe get mac80211 to
1356 	 * provide a way to just receive a list of
1357 	 * skb?
1358 	 */
1359 
1360 	first = __skb_dequeue(amsdu);
1361 
1362 	/* Allocate total length all at once. */
1363 	skb_queue_walk(amsdu, skb)
1364 		total_len += skb->len;
1365 
1366 	space = total_len - skb_tailroom(first);
1367 	if ((space > 0) &&
1368 	    (pskb_expand_head(first, 0, space, GFP_ATOMIC) < 0)) {
1369 		/* TODO:  bump some rx-oom error stat */
1370 		/* put it back together so we can free the
1371 		 * whole list at once.
1372 		 */
1373 		__skb_queue_head(amsdu, first);
1374 		return -1;
1375 	}
1376 
1377 	/* Walk list again, copying contents into
1378 	 * msdu_head
1379 	 */
1380 	while ((skb = __skb_dequeue(amsdu))) {
1381 		skb_copy_from_linear_data(skb, skb_put(first, skb->len),
1382 					  skb->len);
1383 		dev_kfree_skb_any(skb);
1384 	}
1385 
1386 	__skb_queue_head(amsdu, first);
1387 	return 0;
1388 }
1389 
1390 static void ath10k_htt_rx_h_unchain(struct ath10k *ar,
1391 				    struct sk_buff_head *amsdu,
1392 				    bool chained)
1393 {
1394 	struct sk_buff *first;
1395 	struct htt_rx_desc *rxd;
1396 	enum rx_msdu_decap_format decap;
1397 
1398 	first = skb_peek(amsdu);
1399 	rxd = (void *)first->data - sizeof(*rxd);
1400 	decap = MS(__le32_to_cpu(rxd->msdu_start.info1),
1401 		   RX_MSDU_START_INFO1_DECAP_FORMAT);
1402 
1403 	if (!chained)
1404 		return;
1405 
1406 	/* FIXME: Current unchaining logic can only handle simple case of raw
1407 	 * msdu chaining. If decapping is other than raw the chaining may be
1408 	 * more complex and this isn't handled by the current code. Don't even
1409 	 * try re-constructing such frames - it'll be pretty much garbage.
1410 	 */
1411 	if (decap != RX_MSDU_DECAP_RAW ||
1412 	    skb_queue_len(amsdu) != 1 + rxd->frag_info.ring2_more_count) {
1413 		__skb_queue_purge(amsdu);
1414 		return;
1415 	}
1416 
1417 	ath10k_unchain_msdu(amsdu);
1418 }
1419 
1420 static bool ath10k_htt_rx_amsdu_allowed(struct ath10k *ar,
1421 					struct sk_buff_head *amsdu,
1422 					struct ieee80211_rx_status *rx_status)
1423 {
1424 	struct sk_buff *msdu;
1425 	struct htt_rx_desc *rxd;
1426 	bool is_mgmt;
1427 	bool has_fcs_err;
1428 
1429 	msdu = skb_peek(amsdu);
1430 	rxd = (void *)msdu->data - sizeof(*rxd);
1431 
1432 	/* FIXME: It might be a good idea to do some fuzzy-testing to drop
1433 	 * invalid/dangerous frames.
1434 	 */
1435 
1436 	if (!rx_status->freq) {
1437 		ath10k_warn(ar, "no channel configured; ignoring frame(s)!\n");
1438 		return false;
1439 	}
1440 
1441 	is_mgmt = !!(rxd->attention.flags &
1442 		     __cpu_to_le32(RX_ATTENTION_FLAGS_MGMT_TYPE));
1443 	has_fcs_err = !!(rxd->attention.flags &
1444 			 __cpu_to_le32(RX_ATTENTION_FLAGS_FCS_ERR));
1445 
1446 	/* Management frames are handled via WMI events. The pros of such
1447 	 * approach is that channel is explicitly provided in WMI events
1448 	 * whereas HTT doesn't provide channel information for Rxed frames.
1449 	 *
1450 	 * However some firmware revisions don't report corrupted frames via
1451 	 * WMI so don't drop them.
1452 	 */
1453 	if (is_mgmt && !has_fcs_err) {
1454 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx mgmt ctrl\n");
1455 		return false;
1456 	}
1457 
1458 	if (test_bit(ATH10K_CAC_RUNNING, &ar->dev_flags)) {
1459 		ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx cac running\n");
1460 		return false;
1461 	}
1462 
1463 	return true;
1464 }
1465 
1466 static void ath10k_htt_rx_h_filter(struct ath10k *ar,
1467 				   struct sk_buff_head *amsdu,
1468 				   struct ieee80211_rx_status *rx_status)
1469 {
1470 	if (skb_queue_empty(amsdu))
1471 		return;
1472 
1473 	if (ath10k_htt_rx_amsdu_allowed(ar, amsdu, rx_status))
1474 		return;
1475 
1476 	__skb_queue_purge(amsdu);
1477 }
1478 
1479 static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
1480 				  struct htt_rx_indication *rx)
1481 {
1482 	struct ath10k *ar = htt->ar;
1483 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1484 	struct htt_rx_indication_mpdu_range *mpdu_ranges;
1485 	struct sk_buff_head amsdu;
1486 	int num_mpdu_ranges;
1487 	int fw_desc_len;
1488 	u8 *fw_desc;
1489 	int i, ret, mpdu_count = 0;
1490 
1491 	lockdep_assert_held(&htt->rx_ring.lock);
1492 
1493 	if (htt->rx_confused)
1494 		return;
1495 
1496 	fw_desc_len = __le16_to_cpu(rx->prefix.fw_rx_desc_bytes);
1497 	fw_desc = (u8 *)&rx->fw_desc;
1498 
1499 	num_mpdu_ranges = MS(__le32_to_cpu(rx->hdr.info1),
1500 			     HTT_RX_INDICATION_INFO1_NUM_MPDU_RANGES);
1501 	mpdu_ranges = htt_rx_ind_get_mpdu_ranges(rx);
1502 
1503 	ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt rx ind: ",
1504 			rx, sizeof(*rx) +
1505 			(sizeof(struct htt_rx_indication_mpdu_range) *
1506 				num_mpdu_ranges));
1507 
1508 	for (i = 0; i < num_mpdu_ranges; i++)
1509 		mpdu_count += mpdu_ranges[i].mpdu_count;
1510 
1511 	while (mpdu_count--) {
1512 		__skb_queue_head_init(&amsdu);
1513 		ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc,
1514 					      &fw_desc_len, &amsdu);
1515 		if (ret < 0) {
1516 			ath10k_warn(ar, "rx ring became corrupted: %d\n", ret);
1517 			__skb_queue_purge(&amsdu);
1518 			/* FIXME: It's probably a good idea to reboot the
1519 			 * device instead of leaving it inoperable.
1520 			 */
1521 			htt->rx_confused = true;
1522 			break;
1523 		}
1524 
1525 		ath10k_htt_rx_h_ppdu(ar, &amsdu, rx_status);
1526 		ath10k_htt_rx_h_unchain(ar, &amsdu, ret > 0);
1527 		ath10k_htt_rx_h_filter(ar, &amsdu, rx_status);
1528 		ath10k_htt_rx_h_mpdu(ar, &amsdu, rx_status);
1529 		ath10k_htt_rx_h_deliver(ar, &amsdu, rx_status);
1530 	}
1531 
1532 	tasklet_schedule(&htt->rx_replenish_task);
1533 }
1534 
1535 static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
1536 				       struct htt_rx_fragment_indication *frag)
1537 {
1538 	struct ath10k *ar = htt->ar;
1539 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1540 	struct sk_buff_head amsdu;
1541 	int ret;
1542 	u8 *fw_desc;
1543 	int fw_desc_len;
1544 
1545 	fw_desc_len = __le16_to_cpu(frag->fw_rx_desc_bytes);
1546 	fw_desc = (u8 *)frag->fw_msdu_rx_desc;
1547 
1548 	__skb_queue_head_init(&amsdu);
1549 
1550 	spin_lock_bh(&htt->rx_ring.lock);
1551 	ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc, &fw_desc_len,
1552 				      &amsdu);
1553 	spin_unlock_bh(&htt->rx_ring.lock);
1554 
1555 	tasklet_schedule(&htt->rx_replenish_task);
1556 
1557 	ath10k_dbg(ar, ATH10K_DBG_HTT_DUMP, "htt rx frag ahead\n");
1558 
1559 	if (ret) {
1560 		ath10k_warn(ar, "failed to pop amsdu from httr rx ring for fragmented rx %d\n",
1561 			    ret);
1562 		__skb_queue_purge(&amsdu);
1563 		return;
1564 	}
1565 
1566 	if (skb_queue_len(&amsdu) != 1) {
1567 		ath10k_warn(ar, "failed to pop frag amsdu: too many msdus\n");
1568 		__skb_queue_purge(&amsdu);
1569 		return;
1570 	}
1571 
1572 	ath10k_htt_rx_h_ppdu(ar, &amsdu, rx_status);
1573 	ath10k_htt_rx_h_filter(ar, &amsdu, rx_status);
1574 	ath10k_htt_rx_h_mpdu(ar, &amsdu, rx_status);
1575 	ath10k_htt_rx_h_deliver(ar, &amsdu, rx_status);
1576 
1577 	if (fw_desc_len > 0) {
1578 		ath10k_dbg(ar, ATH10K_DBG_HTT,
1579 			   "expecting more fragmented rx in one indication %d\n",
1580 			   fw_desc_len);
1581 	}
1582 }
1583 
1584 static void ath10k_htt_rx_frm_tx_compl(struct ath10k *ar,
1585 				       struct sk_buff *skb)
1586 {
1587 	struct ath10k_htt *htt = &ar->htt;
1588 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1589 	struct htt_tx_done tx_done = {};
1590 	int status = MS(resp->data_tx_completion.flags, HTT_DATA_TX_STATUS);
1591 	__le16 msdu_id;
1592 	int i;
1593 
1594 	lockdep_assert_held(&htt->tx_lock);
1595 
1596 	switch (status) {
1597 	case HTT_DATA_TX_STATUS_NO_ACK:
1598 		tx_done.no_ack = true;
1599 		break;
1600 	case HTT_DATA_TX_STATUS_OK:
1601 		break;
1602 	case HTT_DATA_TX_STATUS_DISCARD:
1603 	case HTT_DATA_TX_STATUS_POSTPONE:
1604 	case HTT_DATA_TX_STATUS_DOWNLOAD_FAIL:
1605 		tx_done.discard = true;
1606 		break;
1607 	default:
1608 		ath10k_warn(ar, "unhandled tx completion status %d\n", status);
1609 		tx_done.discard = true;
1610 		break;
1611 	}
1612 
1613 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx completion num_msdus %d\n",
1614 		   resp->data_tx_completion.num_msdus);
1615 
1616 	for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
1617 		msdu_id = resp->data_tx_completion.msdus[i];
1618 		tx_done.msdu_id = __le16_to_cpu(msdu_id);
1619 		ath10k_txrx_tx_unref(htt, &tx_done);
1620 	}
1621 }
1622 
1623 static void ath10k_htt_rx_addba(struct ath10k *ar, struct htt_resp *resp)
1624 {
1625 	struct htt_rx_addba *ev = &resp->rx_addba;
1626 	struct ath10k_peer *peer;
1627 	struct ath10k_vif *arvif;
1628 	u16 info0, tid, peer_id;
1629 
1630 	info0 = __le16_to_cpu(ev->info0);
1631 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1632 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1633 
1634 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1635 		   "htt rx addba tid %hu peer_id %hu size %hhu\n",
1636 		   tid, peer_id, ev->window_size);
1637 
1638 	spin_lock_bh(&ar->data_lock);
1639 	peer = ath10k_peer_find_by_id(ar, peer_id);
1640 	if (!peer) {
1641 		ath10k_warn(ar, "received addba event for invalid peer_id: %hu\n",
1642 			    peer_id);
1643 		spin_unlock_bh(&ar->data_lock);
1644 		return;
1645 	}
1646 
1647 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1648 	if (!arvif) {
1649 		ath10k_warn(ar, "received addba event for invalid vdev_id: %u\n",
1650 			    peer->vdev_id);
1651 		spin_unlock_bh(&ar->data_lock);
1652 		return;
1653 	}
1654 
1655 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1656 		   "htt rx start rx ba session sta %pM tid %hu size %hhu\n",
1657 		   peer->addr, tid, ev->window_size);
1658 
1659 	ieee80211_start_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1660 	spin_unlock_bh(&ar->data_lock);
1661 }
1662 
1663 static void ath10k_htt_rx_delba(struct ath10k *ar, struct htt_resp *resp)
1664 {
1665 	struct htt_rx_delba *ev = &resp->rx_delba;
1666 	struct ath10k_peer *peer;
1667 	struct ath10k_vif *arvif;
1668 	u16 info0, tid, peer_id;
1669 
1670 	info0 = __le16_to_cpu(ev->info0);
1671 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1672 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1673 
1674 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1675 		   "htt rx delba tid %hu peer_id %hu\n",
1676 		   tid, peer_id);
1677 
1678 	spin_lock_bh(&ar->data_lock);
1679 	peer = ath10k_peer_find_by_id(ar, peer_id);
1680 	if (!peer) {
1681 		ath10k_warn(ar, "received addba event for invalid peer_id: %hu\n",
1682 			    peer_id);
1683 		spin_unlock_bh(&ar->data_lock);
1684 		return;
1685 	}
1686 
1687 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1688 	if (!arvif) {
1689 		ath10k_warn(ar, "received addba event for invalid vdev_id: %u\n",
1690 			    peer->vdev_id);
1691 		spin_unlock_bh(&ar->data_lock);
1692 		return;
1693 	}
1694 
1695 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1696 		   "htt rx stop rx ba session sta %pM tid %hu\n",
1697 		   peer->addr, tid);
1698 
1699 	ieee80211_stop_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1700 	spin_unlock_bh(&ar->data_lock);
1701 }
1702 
1703 static int ath10k_htt_rx_extract_amsdu(struct sk_buff_head *list,
1704 				       struct sk_buff_head *amsdu)
1705 {
1706 	struct sk_buff *msdu;
1707 	struct htt_rx_desc *rxd;
1708 
1709 	if (skb_queue_empty(list))
1710 		return -ENOBUFS;
1711 
1712 	if (WARN_ON(!skb_queue_empty(amsdu)))
1713 		return -EINVAL;
1714 
1715 	while ((msdu = __skb_dequeue(list))) {
1716 		__skb_queue_tail(amsdu, msdu);
1717 
1718 		rxd = (void *)msdu->data - sizeof(*rxd);
1719 		if (rxd->msdu_end.info0 &
1720 		    __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))
1721 			break;
1722 	}
1723 
1724 	msdu = skb_peek_tail(amsdu);
1725 	rxd = (void *)msdu->data - sizeof(*rxd);
1726 	if (!(rxd->msdu_end.info0 &
1727 	      __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))) {
1728 		skb_queue_splice_init(amsdu, list);
1729 		return -EAGAIN;
1730 	}
1731 
1732 	return 0;
1733 }
1734 
1735 static void ath10k_htt_rx_h_rx_offload_prot(struct ieee80211_rx_status *status,
1736 					    struct sk_buff *skb)
1737 {
1738 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1739 
1740 	if (!ieee80211_has_protected(hdr->frame_control))
1741 		return;
1742 
1743 	/* Offloaded frames are already decrypted but firmware insists they are
1744 	 * protected in the 802.11 header. Strip the flag.  Otherwise mac80211
1745 	 * will drop the frame.
1746 	 */
1747 
1748 	hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
1749 	status->flag |= RX_FLAG_DECRYPTED |
1750 			RX_FLAG_IV_STRIPPED |
1751 			RX_FLAG_MMIC_STRIPPED;
1752 }
1753 
1754 static void ath10k_htt_rx_h_rx_offload(struct ath10k *ar,
1755 				       struct sk_buff_head *list)
1756 {
1757 	struct ath10k_htt *htt = &ar->htt;
1758 	struct ieee80211_rx_status *status = &htt->rx_status;
1759 	struct htt_rx_offload_msdu *rx;
1760 	struct sk_buff *msdu;
1761 	size_t offset;
1762 
1763 	while ((msdu = __skb_dequeue(list))) {
1764 		/* Offloaded frames don't have Rx descriptor. Instead they have
1765 		 * a short meta information header.
1766 		 */
1767 
1768 		rx = (void *)msdu->data;
1769 
1770 		skb_put(msdu, sizeof(*rx));
1771 		skb_pull(msdu, sizeof(*rx));
1772 
1773 		if (skb_tailroom(msdu) < __le16_to_cpu(rx->msdu_len)) {
1774 			ath10k_warn(ar, "dropping frame: offloaded rx msdu is too long!\n");
1775 			dev_kfree_skb_any(msdu);
1776 			continue;
1777 		}
1778 
1779 		skb_put(msdu, __le16_to_cpu(rx->msdu_len));
1780 
1781 		/* Offloaded rx header length isn't multiple of 2 nor 4 so the
1782 		 * actual payload is unaligned. Align the frame.  Otherwise
1783 		 * mac80211 complains.  This shouldn't reduce performance much
1784 		 * because these offloaded frames are rare.
1785 		 */
1786 		offset = 4 - ((unsigned long)msdu->data & 3);
1787 		skb_put(msdu, offset);
1788 		memmove(msdu->data + offset, msdu->data, msdu->len);
1789 		skb_pull(msdu, offset);
1790 
1791 		/* FIXME: The frame is NWifi. Re-construct QoS Control
1792 		 * if possible later.
1793 		 */
1794 
1795 		memset(status, 0, sizeof(*status));
1796 		status->flag |= RX_FLAG_NO_SIGNAL_VAL;
1797 
1798 		ath10k_htt_rx_h_rx_offload_prot(status, msdu);
1799 		ath10k_htt_rx_h_channel(ar, status);
1800 		ath10k_process_rx(ar, status, msdu);
1801 	}
1802 }
1803 
1804 static void ath10k_htt_rx_in_ord_ind(struct ath10k *ar, struct sk_buff *skb)
1805 {
1806 	struct ath10k_htt *htt = &ar->htt;
1807 	struct htt_resp *resp = (void *)skb->data;
1808 	struct ieee80211_rx_status *status = &htt->rx_status;
1809 	struct sk_buff_head list;
1810 	struct sk_buff_head amsdu;
1811 	u16 peer_id;
1812 	u16 msdu_count;
1813 	u8 vdev_id;
1814 	u8 tid;
1815 	bool offload;
1816 	bool frag;
1817 	int ret;
1818 
1819 	lockdep_assert_held(&htt->rx_ring.lock);
1820 
1821 	if (htt->rx_confused)
1822 		return;
1823 
1824 	skb_pull(skb, sizeof(resp->hdr));
1825 	skb_pull(skb, sizeof(resp->rx_in_ord_ind));
1826 
1827 	peer_id = __le16_to_cpu(resp->rx_in_ord_ind.peer_id);
1828 	msdu_count = __le16_to_cpu(resp->rx_in_ord_ind.msdu_count);
1829 	vdev_id = resp->rx_in_ord_ind.vdev_id;
1830 	tid = SM(resp->rx_in_ord_ind.info, HTT_RX_IN_ORD_IND_INFO_TID);
1831 	offload = !!(resp->rx_in_ord_ind.info &
1832 			HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK);
1833 	frag = !!(resp->rx_in_ord_ind.info & HTT_RX_IN_ORD_IND_INFO_FRAG_MASK);
1834 
1835 	ath10k_dbg(ar, ATH10K_DBG_HTT,
1836 		   "htt rx in ord vdev %i peer %i tid %i offload %i frag %i msdu count %i\n",
1837 		   vdev_id, peer_id, tid, offload, frag, msdu_count);
1838 
1839 	if (skb->len < msdu_count * sizeof(*resp->rx_in_ord_ind.msdu_descs)) {
1840 		ath10k_warn(ar, "dropping invalid in order rx indication\n");
1841 		return;
1842 	}
1843 
1844 	/* The event can deliver more than 1 A-MSDU. Each A-MSDU is later
1845 	 * extracted and processed.
1846 	 */
1847 	__skb_queue_head_init(&list);
1848 	ret = ath10k_htt_rx_pop_paddr_list(htt, &resp->rx_in_ord_ind, &list);
1849 	if (ret < 0) {
1850 		ath10k_warn(ar, "failed to pop paddr list: %d\n", ret);
1851 		htt->rx_confused = true;
1852 		return;
1853 	}
1854 
1855 	/* Offloaded frames are very different and need to be handled
1856 	 * separately.
1857 	 */
1858 	if (offload)
1859 		ath10k_htt_rx_h_rx_offload(ar, &list);
1860 
1861 	while (!skb_queue_empty(&list)) {
1862 		__skb_queue_head_init(&amsdu);
1863 		ret = ath10k_htt_rx_extract_amsdu(&list, &amsdu);
1864 		switch (ret) {
1865 		case 0:
1866 			/* Note: The in-order indication may report interleaved
1867 			 * frames from different PPDUs meaning reported rx rate
1868 			 * to mac80211 isn't accurate/reliable. It's still
1869 			 * better to report something than nothing though. This
1870 			 * should still give an idea about rx rate to the user.
1871 			 */
1872 			ath10k_htt_rx_h_ppdu(ar, &amsdu, status);
1873 			ath10k_htt_rx_h_filter(ar, &amsdu, status);
1874 			ath10k_htt_rx_h_mpdu(ar, &amsdu, status);
1875 			ath10k_htt_rx_h_deliver(ar, &amsdu, status);
1876 			break;
1877 		case -EAGAIN:
1878 			/* fall through */
1879 		default:
1880 			/* Should not happen. */
1881 			ath10k_warn(ar, "failed to extract amsdu: %d\n", ret);
1882 			htt->rx_confused = true;
1883 			__skb_queue_purge(&list);
1884 			return;
1885 		}
1886 	}
1887 
1888 	tasklet_schedule(&htt->rx_replenish_task);
1889 }
1890 
1891 void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
1892 {
1893 	struct ath10k_htt *htt = &ar->htt;
1894 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1895 
1896 	/* confirm alignment */
1897 	if (!IS_ALIGNED((unsigned long)skb->data, 4))
1898 		ath10k_warn(ar, "unaligned htt message, expect trouble\n");
1899 
1900 	ath10k_dbg(ar, ATH10K_DBG_HTT, "htt rx, msg_type: 0x%0X\n",
1901 		   resp->hdr.msg_type);
1902 	switch (resp->hdr.msg_type) {
1903 	case HTT_T2H_MSG_TYPE_VERSION_CONF: {
1904 		htt->target_version_major = resp->ver_resp.major;
1905 		htt->target_version_minor = resp->ver_resp.minor;
1906 		complete(&htt->target_version_received);
1907 		break;
1908 	}
1909 	case HTT_T2H_MSG_TYPE_RX_IND:
1910 		spin_lock_bh(&htt->rx_ring.lock);
1911 		__skb_queue_tail(&htt->rx_compl_q, skb);
1912 		spin_unlock_bh(&htt->rx_ring.lock);
1913 		tasklet_schedule(&htt->txrx_compl_task);
1914 		return;
1915 	case HTT_T2H_MSG_TYPE_PEER_MAP: {
1916 		struct htt_peer_map_event ev = {
1917 			.vdev_id = resp->peer_map.vdev_id,
1918 			.peer_id = __le16_to_cpu(resp->peer_map.peer_id),
1919 		};
1920 		memcpy(ev.addr, resp->peer_map.addr, sizeof(ev.addr));
1921 		ath10k_peer_map_event(htt, &ev);
1922 		break;
1923 	}
1924 	case HTT_T2H_MSG_TYPE_PEER_UNMAP: {
1925 		struct htt_peer_unmap_event ev = {
1926 			.peer_id = __le16_to_cpu(resp->peer_unmap.peer_id),
1927 		};
1928 		ath10k_peer_unmap_event(htt, &ev);
1929 		break;
1930 	}
1931 	case HTT_T2H_MSG_TYPE_MGMT_TX_COMPLETION: {
1932 		struct htt_tx_done tx_done = {};
1933 		int status = __le32_to_cpu(resp->mgmt_tx_completion.status);
1934 
1935 		tx_done.msdu_id =
1936 			__le32_to_cpu(resp->mgmt_tx_completion.desc_id);
1937 
1938 		switch (status) {
1939 		case HTT_MGMT_TX_STATUS_OK:
1940 			break;
1941 		case HTT_MGMT_TX_STATUS_RETRY:
1942 			tx_done.no_ack = true;
1943 			break;
1944 		case HTT_MGMT_TX_STATUS_DROP:
1945 			tx_done.discard = true;
1946 			break;
1947 		}
1948 
1949 		spin_lock_bh(&htt->tx_lock);
1950 		ath10k_txrx_tx_unref(htt, &tx_done);
1951 		spin_unlock_bh(&htt->tx_lock);
1952 		break;
1953 	}
1954 	case HTT_T2H_MSG_TYPE_TX_COMPL_IND:
1955 		spin_lock_bh(&htt->tx_lock);
1956 		__skb_queue_tail(&htt->tx_compl_q, skb);
1957 		spin_unlock_bh(&htt->tx_lock);
1958 		tasklet_schedule(&htt->txrx_compl_task);
1959 		return;
1960 	case HTT_T2H_MSG_TYPE_SEC_IND: {
1961 		struct ath10k *ar = htt->ar;
1962 		struct htt_security_indication *ev = &resp->security_indication;
1963 
1964 		ath10k_dbg(ar, ATH10K_DBG_HTT,
1965 			   "sec ind peer_id %d unicast %d type %d\n",
1966 			  __le16_to_cpu(ev->peer_id),
1967 			  !!(ev->flags & HTT_SECURITY_IS_UNICAST),
1968 			  MS(ev->flags, HTT_SECURITY_TYPE));
1969 		complete(&ar->install_key_done);
1970 		break;
1971 	}
1972 	case HTT_T2H_MSG_TYPE_RX_FRAG_IND: {
1973 		ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
1974 				skb->data, skb->len);
1975 		ath10k_htt_rx_frag_handler(htt, &resp->rx_frag_ind);
1976 		break;
1977 	}
1978 	case HTT_T2H_MSG_TYPE_TEST:
1979 		/* FIX THIS */
1980 		break;
1981 	case HTT_T2H_MSG_TYPE_STATS_CONF:
1982 		trace_ath10k_htt_stats(ar, skb->data, skb->len);
1983 		break;
1984 	case HTT_T2H_MSG_TYPE_TX_INSPECT_IND:
1985 		/* Firmware can return tx frames if it's unable to fully
1986 		 * process them and suspects host may be able to fix it. ath10k
1987 		 * sends all tx frames as already inspected so this shouldn't
1988 		 * happen unless fw has a bug.
1989 		 */
1990 		ath10k_warn(ar, "received an unexpected htt tx inspect event\n");
1991 		break;
1992 	case HTT_T2H_MSG_TYPE_RX_ADDBA:
1993 		ath10k_htt_rx_addba(ar, resp);
1994 		break;
1995 	case HTT_T2H_MSG_TYPE_RX_DELBA:
1996 		ath10k_htt_rx_delba(ar, resp);
1997 		break;
1998 	case HTT_T2H_MSG_TYPE_PKTLOG: {
1999 		struct ath10k_pktlog_hdr *hdr =
2000 			(struct ath10k_pktlog_hdr *)resp->pktlog_msg.payload;
2001 
2002 		trace_ath10k_htt_pktlog(ar, resp->pktlog_msg.payload,
2003 					sizeof(*hdr) +
2004 					__le16_to_cpu(hdr->size));
2005 		break;
2006 	}
2007 	case HTT_T2H_MSG_TYPE_RX_FLUSH: {
2008 		/* Ignore this event because mac80211 takes care of Rx
2009 		 * aggregation reordering.
2010 		 */
2011 		break;
2012 	}
2013 	case HTT_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND: {
2014 		spin_lock_bh(&htt->rx_ring.lock);
2015 		__skb_queue_tail(&htt->rx_in_ord_compl_q, skb);
2016 		spin_unlock_bh(&htt->rx_ring.lock);
2017 		tasklet_schedule(&htt->txrx_compl_task);
2018 		return;
2019 	}
2020 	case HTT_T2H_MSG_TYPE_TX_CREDIT_UPDATE_IND:
2021 		/* FIXME: This WMI-TLV event is overlapping with 10.2
2022 		 * CHAN_CHANGE - both being 0xF. Neither is being used in
2023 		 * practice so no immediate action is necessary. Nevertheless
2024 		 * HTT may need an abstraction layer like WMI has one day.
2025 		 */
2026 		break;
2027 	default:
2028 		ath10k_warn(ar, "htt event (%d) not handled\n",
2029 			    resp->hdr.msg_type);
2030 		ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
2031 				skb->data, skb->len);
2032 		break;
2033 	};
2034 
2035 	/* Free the indication buffer */
2036 	dev_kfree_skb_any(skb);
2037 }
2038 
2039 static void ath10k_htt_txrx_compl_task(unsigned long ptr)
2040 {
2041 	struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
2042 	struct ath10k *ar = htt->ar;
2043 	struct htt_resp *resp;
2044 	struct sk_buff *skb;
2045 
2046 	spin_lock_bh(&htt->tx_lock);
2047 	while ((skb = __skb_dequeue(&htt->tx_compl_q))) {
2048 		ath10k_htt_rx_frm_tx_compl(htt->ar, skb);
2049 		dev_kfree_skb_any(skb);
2050 	}
2051 	spin_unlock_bh(&htt->tx_lock);
2052 
2053 	spin_lock_bh(&htt->rx_ring.lock);
2054 	while ((skb = __skb_dequeue(&htt->rx_compl_q))) {
2055 		resp = (struct htt_resp *)skb->data;
2056 		ath10k_htt_rx_handler(htt, &resp->rx_ind);
2057 		dev_kfree_skb_any(skb);
2058 	}
2059 
2060 	while ((skb = __skb_dequeue(&htt->rx_in_ord_compl_q))) {
2061 		ath10k_htt_rx_in_ord_ind(ar, skb);
2062 		dev_kfree_skb_any(skb);
2063 	}
2064 	spin_unlock_bh(&htt->rx_ring.lock);
2065 }
2066