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 /* slightly larger than one large A-MPDU */
29 #define HTT_RX_RING_SIZE_MIN 128
30 
31 /* roughly 20 ms @ 1 Gbps of 1500B MSDUs */
32 #define HTT_RX_RING_SIZE_MAX 2048
33 
34 #define HTT_RX_AVG_FRM_BYTES 1000
35 
36 /* ms, very conservative */
37 #define HTT_RX_HOST_LATENCY_MAX_MS 20
38 
39 /* ms, conservative */
40 #define HTT_RX_HOST_LATENCY_WORST_LIKELY_MS 10
41 
42 /* when under memory pressure rx ring refill may fail and needs a retry */
43 #define HTT_RX_RING_REFILL_RETRY_MS 50
44 
45 
46 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb);
47 static void ath10k_htt_txrx_compl_task(unsigned long ptr);
48 
49 static int ath10k_htt_rx_ring_size(struct ath10k_htt *htt)
50 {
51 	int size;
52 
53 	/*
54 	 * It is expected that the host CPU will typically be able to
55 	 * service the rx indication from one A-MPDU before the rx
56 	 * indication from the subsequent A-MPDU happens, roughly 1-2 ms
57 	 * later. However, the rx ring should be sized very conservatively,
58 	 * to accomodate the worst reasonable delay before the host CPU
59 	 * services a rx indication interrupt.
60 	 *
61 	 * The rx ring need not be kept full of empty buffers. In theory,
62 	 * the htt host SW can dynamically track the low-water mark in the
63 	 * rx ring, and dynamically adjust the level to which the rx ring
64 	 * is filled with empty buffers, to dynamically meet the desired
65 	 * low-water mark.
66 	 *
67 	 * In contrast, it's difficult to resize the rx ring itself, once
68 	 * it's in use. Thus, the ring itself should be sized very
69 	 * conservatively, while the degree to which the ring is filled
70 	 * with empty buffers should be sized moderately conservatively.
71 	 */
72 
73 	/* 1e6 bps/mbps / 1e3 ms per sec = 1000 */
74 	size =
75 	    htt->max_throughput_mbps +
76 	    1000  /
77 	    (8 * HTT_RX_AVG_FRM_BYTES) * HTT_RX_HOST_LATENCY_MAX_MS;
78 
79 	if (size < HTT_RX_RING_SIZE_MIN)
80 		size = HTT_RX_RING_SIZE_MIN;
81 
82 	if (size > HTT_RX_RING_SIZE_MAX)
83 		size = HTT_RX_RING_SIZE_MAX;
84 
85 	size = roundup_pow_of_two(size);
86 
87 	return size;
88 }
89 
90 static int ath10k_htt_rx_ring_fill_level(struct ath10k_htt *htt)
91 {
92 	int size;
93 
94 	/* 1e6 bps/mbps / 1e3 ms per sec = 1000 */
95 	size =
96 	    htt->max_throughput_mbps *
97 	    1000  /
98 	    (8 * HTT_RX_AVG_FRM_BYTES) * HTT_RX_HOST_LATENCY_WORST_LIKELY_MS;
99 
100 	/*
101 	 * Make sure the fill level is at least 1 less than the ring size.
102 	 * Leaving 1 element empty allows the SW to easily distinguish
103 	 * between a full ring vs. an empty ring.
104 	 */
105 	if (size >= htt->rx_ring.size)
106 		size = htt->rx_ring.size - 1;
107 
108 	return size;
109 }
110 
111 static void ath10k_htt_rx_ring_free(struct ath10k_htt *htt)
112 {
113 	struct sk_buff *skb;
114 	struct ath10k_skb_cb *cb;
115 	int i;
116 
117 	for (i = 0; i < htt->rx_ring.fill_cnt; i++) {
118 		skb = htt->rx_ring.netbufs_ring[i];
119 		cb = ATH10K_SKB_CB(skb);
120 		dma_unmap_single(htt->ar->dev, cb->paddr,
121 				 skb->len + skb_tailroom(skb),
122 				 DMA_FROM_DEVICE);
123 		dev_kfree_skb_any(skb);
124 	}
125 
126 	htt->rx_ring.fill_cnt = 0;
127 }
128 
129 static int __ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
130 {
131 	struct htt_rx_desc *rx_desc;
132 	struct sk_buff *skb;
133 	dma_addr_t paddr;
134 	int ret = 0, idx;
135 
136 	idx = __le32_to_cpu(*(htt->rx_ring.alloc_idx.vaddr));
137 	while (num > 0) {
138 		skb = dev_alloc_skb(HTT_RX_BUF_SIZE + HTT_RX_DESC_ALIGN);
139 		if (!skb) {
140 			ret = -ENOMEM;
141 			goto fail;
142 		}
143 
144 		if (!IS_ALIGNED((unsigned long)skb->data, HTT_RX_DESC_ALIGN))
145 			skb_pull(skb,
146 				 PTR_ALIGN(skb->data, HTT_RX_DESC_ALIGN) -
147 				 skb->data);
148 
149 		/* Clear rx_desc attention word before posting to Rx ring */
150 		rx_desc = (struct htt_rx_desc *)skb->data;
151 		rx_desc->attention.flags = __cpu_to_le32(0);
152 
153 		paddr = dma_map_single(htt->ar->dev, skb->data,
154 				       skb->len + skb_tailroom(skb),
155 				       DMA_FROM_DEVICE);
156 
157 		if (unlikely(dma_mapping_error(htt->ar->dev, paddr))) {
158 			dev_kfree_skb_any(skb);
159 			ret = -ENOMEM;
160 			goto fail;
161 		}
162 
163 		ATH10K_SKB_CB(skb)->paddr = paddr;
164 		htt->rx_ring.netbufs_ring[idx] = skb;
165 		htt->rx_ring.paddrs_ring[idx] = __cpu_to_le32(paddr);
166 		htt->rx_ring.fill_cnt++;
167 
168 		num--;
169 		idx++;
170 		idx &= htt->rx_ring.size_mask;
171 	}
172 
173 fail:
174 	*(htt->rx_ring.alloc_idx.vaddr) = __cpu_to_le32(idx);
175 	return ret;
176 }
177 
178 static int ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
179 {
180 	lockdep_assert_held(&htt->rx_ring.lock);
181 	return __ath10k_htt_rx_ring_fill_n(htt, num);
182 }
183 
184 static void ath10k_htt_rx_msdu_buff_replenish(struct ath10k_htt *htt)
185 {
186 	int ret, num_deficit, num_to_fill;
187 
188 	/* Refilling the whole RX ring buffer proves to be a bad idea. The
189 	 * reason is RX may take up significant amount of CPU cycles and starve
190 	 * other tasks, e.g. TX on an ethernet device while acting as a bridge
191 	 * with ath10k wlan interface. This ended up with very poor performance
192 	 * once CPU the host system was overwhelmed with RX on ath10k.
193 	 *
194 	 * By limiting the number of refills the replenishing occurs
195 	 * progressively. This in turns makes use of the fact tasklets are
196 	 * processed in FIFO order. This means actual RX processing can starve
197 	 * out refilling. If there's not enough buffers on RX ring FW will not
198 	 * report RX until it is refilled with enough buffers. This
199 	 * automatically balances load wrt to CPU power.
200 	 *
201 	 * This probably comes at a cost of lower maximum throughput but
202 	 * improves the avarage and stability. */
203 	spin_lock_bh(&htt->rx_ring.lock);
204 	num_deficit = htt->rx_ring.fill_level - htt->rx_ring.fill_cnt;
205 	num_to_fill = min(ATH10K_HTT_MAX_NUM_REFILL, num_deficit);
206 	num_deficit -= num_to_fill;
207 	ret = ath10k_htt_rx_ring_fill_n(htt, num_to_fill);
208 	if (ret == -ENOMEM) {
209 		/*
210 		 * Failed to fill it to the desired level -
211 		 * we'll start a timer and try again next time.
212 		 * As long as enough buffers are left in the ring for
213 		 * another A-MPDU rx, no special recovery is needed.
214 		 */
215 		mod_timer(&htt->rx_ring.refill_retry_timer, jiffies +
216 			  msecs_to_jiffies(HTT_RX_RING_REFILL_RETRY_MS));
217 	} else if (num_deficit > 0) {
218 		tasklet_schedule(&htt->rx_replenish_task);
219 	}
220 	spin_unlock_bh(&htt->rx_ring.lock);
221 }
222 
223 static void ath10k_htt_rx_ring_refill_retry(unsigned long arg)
224 {
225 	struct ath10k_htt *htt = (struct ath10k_htt *)arg;
226 	ath10k_htt_rx_msdu_buff_replenish(htt);
227 }
228 
229 static void ath10k_htt_rx_ring_clean_up(struct ath10k_htt *htt)
230 {
231 	struct sk_buff *skb;
232 	int i;
233 
234 	for (i = 0; i < htt->rx_ring.size; i++) {
235 		skb = htt->rx_ring.netbufs_ring[i];
236 		if (!skb)
237 			continue;
238 
239 		dma_unmap_single(htt->ar->dev, ATH10K_SKB_CB(skb)->paddr,
240 				 skb->len + skb_tailroom(skb),
241 				 DMA_FROM_DEVICE);
242 		dev_kfree_skb_any(skb);
243 		htt->rx_ring.netbufs_ring[i] = NULL;
244 	}
245 }
246 
247 void ath10k_htt_rx_free(struct ath10k_htt *htt)
248 {
249 	del_timer_sync(&htt->rx_ring.refill_retry_timer);
250 	tasklet_kill(&htt->rx_replenish_task);
251 	tasklet_kill(&htt->txrx_compl_task);
252 
253 	skb_queue_purge(&htt->tx_compl_q);
254 	skb_queue_purge(&htt->rx_compl_q);
255 
256 	ath10k_htt_rx_ring_clean_up(htt);
257 
258 	dma_free_coherent(htt->ar->dev,
259 			  (htt->rx_ring.size *
260 			   sizeof(htt->rx_ring.paddrs_ring)),
261 			  htt->rx_ring.paddrs_ring,
262 			  htt->rx_ring.base_paddr);
263 
264 	dma_free_coherent(htt->ar->dev,
265 			  sizeof(*htt->rx_ring.alloc_idx.vaddr),
266 			  htt->rx_ring.alloc_idx.vaddr,
267 			  htt->rx_ring.alloc_idx.paddr);
268 
269 	kfree(htt->rx_ring.netbufs_ring);
270 }
271 
272 static inline struct sk_buff *ath10k_htt_rx_netbuf_pop(struct ath10k_htt *htt)
273 {
274 	int idx;
275 	struct sk_buff *msdu;
276 
277 	lockdep_assert_held(&htt->rx_ring.lock);
278 
279 	if (htt->rx_ring.fill_cnt == 0) {
280 		ath10k_warn("tried to pop sk_buff from an empty rx ring\n");
281 		return NULL;
282 	}
283 
284 	idx = htt->rx_ring.sw_rd_idx.msdu_payld;
285 	msdu = htt->rx_ring.netbufs_ring[idx];
286 	htt->rx_ring.netbufs_ring[idx] = NULL;
287 
288 	idx++;
289 	idx &= htt->rx_ring.size_mask;
290 	htt->rx_ring.sw_rd_idx.msdu_payld = idx;
291 	htt->rx_ring.fill_cnt--;
292 
293 	return msdu;
294 }
295 
296 static void ath10k_htt_rx_free_msdu_chain(struct sk_buff *skb)
297 {
298 	struct sk_buff *next;
299 
300 	while (skb) {
301 		next = skb->next;
302 		dev_kfree_skb_any(skb);
303 		skb = next;
304 	}
305 }
306 
307 /* return: < 0 fatal error, 0 - non chained msdu, 1 chained msdu */
308 static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
309 				   u8 **fw_desc, int *fw_desc_len,
310 				   struct sk_buff **head_msdu,
311 				   struct sk_buff **tail_msdu,
312 				   u32 *attention)
313 {
314 	int msdu_len, msdu_chaining = 0;
315 	struct sk_buff *msdu;
316 	struct htt_rx_desc *rx_desc;
317 
318 	lockdep_assert_held(&htt->rx_ring.lock);
319 
320 	if (htt->rx_confused) {
321 		ath10k_warn("htt is confused. refusing rx\n");
322 		return -1;
323 	}
324 
325 	msdu = *head_msdu = ath10k_htt_rx_netbuf_pop(htt);
326 	while (msdu) {
327 		int last_msdu, msdu_len_invalid, msdu_chained;
328 
329 		dma_unmap_single(htt->ar->dev,
330 				 ATH10K_SKB_CB(msdu)->paddr,
331 				 msdu->len + skb_tailroom(msdu),
332 				 DMA_FROM_DEVICE);
333 
334 		ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt rx pop: ",
335 				msdu->data, msdu->len + skb_tailroom(msdu));
336 
337 		rx_desc = (struct htt_rx_desc *)msdu->data;
338 
339 		/* FIXME: we must report msdu payload since this is what caller
340 		 *        expects now */
341 		skb_put(msdu, offsetof(struct htt_rx_desc, msdu_payload));
342 		skb_pull(msdu, offsetof(struct htt_rx_desc, msdu_payload));
343 
344 		/*
345 		 * Sanity check - confirm the HW is finished filling in the
346 		 * rx data.
347 		 * If the HW and SW are working correctly, then it's guaranteed
348 		 * that the HW's MAC DMA is done before this point in the SW.
349 		 * To prevent the case that we handle a stale Rx descriptor,
350 		 * just assert for now until we have a way to recover.
351 		 */
352 		if (!(__le32_to_cpu(rx_desc->attention.flags)
353 				& RX_ATTENTION_FLAGS_MSDU_DONE)) {
354 			ath10k_htt_rx_free_msdu_chain(*head_msdu);
355 			*head_msdu = NULL;
356 			msdu = NULL;
357 			ath10k_err("htt rx stopped. cannot recover\n");
358 			htt->rx_confused = true;
359 			break;
360 		}
361 
362 		*attention |= __le32_to_cpu(rx_desc->attention.flags) &
363 					    (RX_ATTENTION_FLAGS_TKIP_MIC_ERR |
364 					     RX_ATTENTION_FLAGS_DECRYPT_ERR |
365 					     RX_ATTENTION_FLAGS_FCS_ERR |
366 					     RX_ATTENTION_FLAGS_MGMT_TYPE);
367 		/*
368 		 * Copy the FW rx descriptor for this MSDU from the rx
369 		 * indication message into the MSDU's netbuf. HL uses the
370 		 * same rx indication message definition as LL, and simply
371 		 * appends new info (fields from the HW rx desc, and the
372 		 * MSDU payload itself). So, the offset into the rx
373 		 * indication message only has to account for the standard
374 		 * offset of the per-MSDU FW rx desc info within the
375 		 * message, and how many bytes of the per-MSDU FW rx desc
376 		 * info have already been consumed. (And the endianness of
377 		 * the host, since for a big-endian host, the rx ind
378 		 * message contents, including the per-MSDU rx desc bytes,
379 		 * were byteswapped during upload.)
380 		 */
381 		if (*fw_desc_len > 0) {
382 			rx_desc->fw_desc.info0 = **fw_desc;
383 			/*
384 			 * The target is expected to only provide the basic
385 			 * per-MSDU rx descriptors. Just to be sure, verify
386 			 * that the target has not attached extension data
387 			 * (e.g. LRO flow ID).
388 			 */
389 
390 			/* or more, if there's extension data */
391 			(*fw_desc)++;
392 			(*fw_desc_len)--;
393 		} else {
394 			/*
395 			 * When an oversized AMSDU happened, FW will lost
396 			 * some of MSDU status - in this case, the FW
397 			 * descriptors provided will be less than the
398 			 * actual MSDUs inside this MPDU. Mark the FW
399 			 * descriptors so that it will still deliver to
400 			 * upper stack, if no CRC error for this MPDU.
401 			 *
402 			 * FIX THIS - the FW descriptors are actually for
403 			 * MSDUs in the end of this A-MSDU instead of the
404 			 * beginning.
405 			 */
406 			rx_desc->fw_desc.info0 = 0;
407 		}
408 
409 		msdu_len_invalid = !!(__le32_to_cpu(rx_desc->attention.flags)
410 					& (RX_ATTENTION_FLAGS_MPDU_LENGTH_ERR |
411 					   RX_ATTENTION_FLAGS_MSDU_LENGTH_ERR));
412 		msdu_len = MS(__le32_to_cpu(rx_desc->msdu_start.info0),
413 			      RX_MSDU_START_INFO0_MSDU_LENGTH);
414 		msdu_chained = rx_desc->frag_info.ring2_more_count;
415 
416 		if (msdu_len_invalid)
417 			msdu_len = 0;
418 
419 		skb_trim(msdu, 0);
420 		skb_put(msdu, min(msdu_len, HTT_RX_MSDU_SIZE));
421 		msdu_len -= msdu->len;
422 
423 		/* FIXME: Do chained buffers include htt_rx_desc or not? */
424 		while (msdu_chained--) {
425 			struct sk_buff *next = ath10k_htt_rx_netbuf_pop(htt);
426 
427 			dma_unmap_single(htt->ar->dev,
428 					 ATH10K_SKB_CB(next)->paddr,
429 					 next->len + skb_tailroom(next),
430 					 DMA_FROM_DEVICE);
431 
432 			ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL,
433 					"htt rx chained: ", next->data,
434 					next->len + skb_tailroom(next));
435 
436 			skb_trim(next, 0);
437 			skb_put(next, min(msdu_len, HTT_RX_BUF_SIZE));
438 			msdu_len -= next->len;
439 
440 			msdu->next = next;
441 			msdu = next;
442 			msdu_chaining = 1;
443 		}
444 
445 		last_msdu = __le32_to_cpu(rx_desc->msdu_end.info0) &
446 				RX_MSDU_END_INFO0_LAST_MSDU;
447 
448 		if (last_msdu) {
449 			msdu->next = NULL;
450 			break;
451 		} else {
452 			struct sk_buff *next = ath10k_htt_rx_netbuf_pop(htt);
453 			msdu->next = next;
454 			msdu = next;
455 		}
456 	}
457 	*tail_msdu = msdu;
458 
459 	if (*head_msdu == NULL)
460 		msdu_chaining = -1;
461 
462 	/*
463 	 * Don't refill the ring yet.
464 	 *
465 	 * First, the elements popped here are still in use - it is not
466 	 * safe to overwrite them until the matching call to
467 	 * mpdu_desc_list_next. Second, for efficiency it is preferable to
468 	 * refill the rx ring with 1 PPDU's worth of rx buffers (something
469 	 * like 32 x 3 buffers), rather than one MPDU's worth of rx buffers
470 	 * (something like 3 buffers). Consequently, we'll rely on the txrx
471 	 * SW to tell us when it is done pulling all the PPDU's rx buffers
472 	 * out of the rx ring, and then refill it just once.
473 	 */
474 
475 	return msdu_chaining;
476 }
477 
478 static void ath10k_htt_rx_replenish_task(unsigned long ptr)
479 {
480 	struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
481 	ath10k_htt_rx_msdu_buff_replenish(htt);
482 }
483 
484 int ath10k_htt_rx_alloc(struct ath10k_htt *htt)
485 {
486 	dma_addr_t paddr;
487 	void *vaddr;
488 	struct timer_list *timer = &htt->rx_ring.refill_retry_timer;
489 
490 	htt->rx_ring.size = ath10k_htt_rx_ring_size(htt);
491 	if (!is_power_of_2(htt->rx_ring.size)) {
492 		ath10k_warn("htt rx ring size is not power of 2\n");
493 		return -EINVAL;
494 	}
495 
496 	htt->rx_ring.size_mask = htt->rx_ring.size - 1;
497 
498 	/*
499 	 * Set the initial value for the level to which the rx ring
500 	 * should be filled, based on the max throughput and the
501 	 * worst likely latency for the host to fill the rx ring
502 	 * with new buffers. In theory, this fill level can be
503 	 * dynamically adjusted from the initial value set here, to
504 	 * reflect the actual host latency rather than a
505 	 * conservative assumption about the host latency.
506 	 */
507 	htt->rx_ring.fill_level = ath10k_htt_rx_ring_fill_level(htt);
508 
509 	htt->rx_ring.netbufs_ring =
510 		kzalloc(htt->rx_ring.size * sizeof(struct sk_buff *),
511 			GFP_KERNEL);
512 	if (!htt->rx_ring.netbufs_ring)
513 		goto err_netbuf;
514 
515 	vaddr = dma_alloc_coherent(htt->ar->dev,
516 		   (htt->rx_ring.size * sizeof(htt->rx_ring.paddrs_ring)),
517 		   &paddr, GFP_DMA);
518 	if (!vaddr)
519 		goto err_dma_ring;
520 
521 	htt->rx_ring.paddrs_ring = vaddr;
522 	htt->rx_ring.base_paddr = paddr;
523 
524 	vaddr = dma_alloc_coherent(htt->ar->dev,
525 				   sizeof(*htt->rx_ring.alloc_idx.vaddr),
526 				   &paddr, GFP_DMA);
527 	if (!vaddr)
528 		goto err_dma_idx;
529 
530 	htt->rx_ring.alloc_idx.vaddr = vaddr;
531 	htt->rx_ring.alloc_idx.paddr = paddr;
532 	htt->rx_ring.sw_rd_idx.msdu_payld = 0;
533 	*htt->rx_ring.alloc_idx.vaddr = 0;
534 
535 	/* Initialize the Rx refill retry timer */
536 	setup_timer(timer, ath10k_htt_rx_ring_refill_retry, (unsigned long)htt);
537 
538 	spin_lock_init(&htt->rx_ring.lock);
539 
540 	htt->rx_ring.fill_cnt = 0;
541 	if (__ath10k_htt_rx_ring_fill_n(htt, htt->rx_ring.fill_level))
542 		goto err_fill_ring;
543 
544 	tasklet_init(&htt->rx_replenish_task, ath10k_htt_rx_replenish_task,
545 		     (unsigned long)htt);
546 
547 	skb_queue_head_init(&htt->tx_compl_q);
548 	skb_queue_head_init(&htt->rx_compl_q);
549 
550 	tasklet_init(&htt->txrx_compl_task, ath10k_htt_txrx_compl_task,
551 		     (unsigned long)htt);
552 
553 	ath10k_dbg(ATH10K_DBG_BOOT, "htt rx ring size %d fill_level %d\n",
554 		   htt->rx_ring.size, htt->rx_ring.fill_level);
555 	return 0;
556 
557 err_fill_ring:
558 	ath10k_htt_rx_ring_free(htt);
559 	dma_free_coherent(htt->ar->dev,
560 			  sizeof(*htt->rx_ring.alloc_idx.vaddr),
561 			  htt->rx_ring.alloc_idx.vaddr,
562 			  htt->rx_ring.alloc_idx.paddr);
563 err_dma_idx:
564 	dma_free_coherent(htt->ar->dev,
565 			  (htt->rx_ring.size *
566 			   sizeof(htt->rx_ring.paddrs_ring)),
567 			  htt->rx_ring.paddrs_ring,
568 			  htt->rx_ring.base_paddr);
569 err_dma_ring:
570 	kfree(htt->rx_ring.netbufs_ring);
571 err_netbuf:
572 	return -ENOMEM;
573 }
574 
575 static int ath10k_htt_rx_crypto_param_len(enum htt_rx_mpdu_encrypt_type type)
576 {
577 	switch (type) {
578 	case HTT_RX_MPDU_ENCRYPT_WEP40:
579 	case HTT_RX_MPDU_ENCRYPT_WEP104:
580 		return 4;
581 	case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
582 	case HTT_RX_MPDU_ENCRYPT_WEP128: /* not tested */
583 	case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
584 	case HTT_RX_MPDU_ENCRYPT_WAPI: /* not tested */
585 	case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
586 		return 8;
587 	case HTT_RX_MPDU_ENCRYPT_NONE:
588 		return 0;
589 	}
590 
591 	ath10k_warn("unknown encryption type %d\n", type);
592 	return 0;
593 }
594 
595 static int ath10k_htt_rx_crypto_tail_len(enum htt_rx_mpdu_encrypt_type type)
596 {
597 	switch (type) {
598 	case HTT_RX_MPDU_ENCRYPT_NONE:
599 	case HTT_RX_MPDU_ENCRYPT_WEP40:
600 	case HTT_RX_MPDU_ENCRYPT_WEP104:
601 	case HTT_RX_MPDU_ENCRYPT_WEP128:
602 	case HTT_RX_MPDU_ENCRYPT_WAPI:
603 		return 0;
604 	case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
605 	case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
606 		return 4;
607 	case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
608 		return 8;
609 	}
610 
611 	ath10k_warn("unknown encryption type %d\n", type);
612 	return 0;
613 }
614 
615 /* Applies for first msdu in chain, before altering it. */
616 static struct ieee80211_hdr *ath10k_htt_rx_skb_get_hdr(struct sk_buff *skb)
617 {
618 	struct htt_rx_desc *rxd;
619 	enum rx_msdu_decap_format fmt;
620 
621 	rxd = (void *)skb->data - sizeof(*rxd);
622 	fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
623 			RX_MSDU_START_INFO1_DECAP_FORMAT);
624 
625 	if (fmt == RX_MSDU_DECAP_RAW)
626 		return (void *)skb->data;
627 	else
628 		return (void *)skb->data - RX_HTT_HDR_STATUS_LEN;
629 }
630 
631 /* This function only applies for first msdu in an msdu chain */
632 static bool ath10k_htt_rx_hdr_is_amsdu(struct ieee80211_hdr *hdr)
633 {
634 	if (ieee80211_is_data_qos(hdr->frame_control)) {
635 		u8 *qc = ieee80211_get_qos_ctl(hdr);
636 		if (qc[0] & 0x80)
637 			return true;
638 	}
639 	return false;
640 }
641 
642 struct rfc1042_hdr {
643 	u8 llc_dsap;
644 	u8 llc_ssap;
645 	u8 llc_ctrl;
646 	u8 snap_oui[3];
647 	__be16 snap_type;
648 } __packed;
649 
650 struct amsdu_subframe_hdr {
651 	u8 dst[ETH_ALEN];
652 	u8 src[ETH_ALEN];
653 	__be16 len;
654 } __packed;
655 
656 static const u8 rx_legacy_rate_idx[] = {
657 	3,	/* 0x00  - 11Mbps  */
658 	2,	/* 0x01  - 5.5Mbps */
659 	1,	/* 0x02  - 2Mbps   */
660 	0,	/* 0x03  - 1Mbps   */
661 	3,	/* 0x04  - 11Mbps  */
662 	2,	/* 0x05  - 5.5Mbps */
663 	1,	/* 0x06  - 2Mbps   */
664 	0,	/* 0x07  - 1Mbps   */
665 	10,	/* 0x08  - 48Mbps  */
666 	8,	/* 0x09  - 24Mbps  */
667 	6,	/* 0x0A  - 12Mbps  */
668 	4,	/* 0x0B  - 6Mbps   */
669 	11,	/* 0x0C  - 54Mbps  */
670 	9,	/* 0x0D  - 36Mbps  */
671 	7,	/* 0x0E  - 18Mbps  */
672 	5,	/* 0x0F  - 9Mbps   */
673 };
674 
675 static void ath10k_htt_rx_h_rates(struct ath10k *ar,
676 				  enum ieee80211_band band,
677 				  u8 info0, u32 info1, u32 info2,
678 				  struct ieee80211_rx_status *status)
679 {
680 	u8 cck, rate, rate_idx, bw, sgi, mcs, nss;
681 	u8 preamble = 0;
682 
683 	/* Check if valid fields */
684 	if (!(info0 & HTT_RX_INDICATION_INFO0_START_VALID))
685 		return;
686 
687 	preamble = MS(info1, HTT_RX_INDICATION_INFO1_PREAMBLE_TYPE);
688 
689 	switch (preamble) {
690 	case HTT_RX_LEGACY:
691 		cck = info0 & HTT_RX_INDICATION_INFO0_LEGACY_RATE_CCK;
692 		rate = MS(info0, HTT_RX_INDICATION_INFO0_LEGACY_RATE);
693 		rate_idx = 0;
694 
695 		if (rate < 0x08 || rate > 0x0F)
696 			break;
697 
698 		switch (band) {
699 		case IEEE80211_BAND_2GHZ:
700 			if (cck)
701 				rate &= ~BIT(3);
702 			rate_idx = rx_legacy_rate_idx[rate];
703 			break;
704 		case IEEE80211_BAND_5GHZ:
705 			rate_idx = rx_legacy_rate_idx[rate];
706 			/* We are using same rate table registering
707 			   HW - ath10k_rates[]. In case of 5GHz skip
708 			   CCK rates, so -4 here */
709 			rate_idx -= 4;
710 			break;
711 		default:
712 			break;
713 		}
714 
715 		status->rate_idx = rate_idx;
716 		break;
717 	case HTT_RX_HT:
718 	case HTT_RX_HT_WITH_TXBF:
719 		/* HT-SIG - Table 20-11 in info1 and info2 */
720 		mcs = info1 & 0x1F;
721 		nss = mcs >> 3;
722 		bw = (info1 >> 7) & 1;
723 		sgi = (info2 >> 7) & 1;
724 
725 		status->rate_idx = mcs;
726 		status->flag |= RX_FLAG_HT;
727 		if (sgi)
728 			status->flag |= RX_FLAG_SHORT_GI;
729 		if (bw)
730 			status->flag |= RX_FLAG_40MHZ;
731 		break;
732 	case HTT_RX_VHT:
733 	case HTT_RX_VHT_WITH_TXBF:
734 		/* VHT-SIG-A1 in info 1, VHT-SIG-A2 in info2
735 		   TODO check this */
736 		mcs = (info2 >> 4) & 0x0F;
737 		nss = ((info1 >> 10) & 0x07) + 1;
738 		bw = info1 & 3;
739 		sgi = info2 & 1;
740 
741 		status->rate_idx = mcs;
742 		status->vht_nss = nss;
743 
744 		if (sgi)
745 			status->flag |= RX_FLAG_SHORT_GI;
746 
747 		switch (bw) {
748 		/* 20MHZ */
749 		case 0:
750 			break;
751 		/* 40MHZ */
752 		case 1:
753 			status->flag |= RX_FLAG_40MHZ;
754 			break;
755 		/* 80MHZ */
756 		case 2:
757 			status->vht_flag |= RX_VHT_FLAG_80MHZ;
758 		}
759 
760 		status->flag |= RX_FLAG_VHT;
761 		break;
762 	default:
763 		break;
764 	}
765 }
766 
767 static void ath10k_htt_rx_h_protected(struct ath10k_htt *htt,
768 				      struct ieee80211_rx_status *rx_status,
769 				      struct sk_buff *skb,
770 				      enum htt_rx_mpdu_encrypt_type enctype,
771 				      enum rx_msdu_decap_format fmt,
772 				      bool dot11frag)
773 {
774 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
775 
776 	rx_status->flag &= ~(RX_FLAG_DECRYPTED |
777 			     RX_FLAG_IV_STRIPPED |
778 			     RX_FLAG_MMIC_STRIPPED);
779 
780 	if (enctype == HTT_RX_MPDU_ENCRYPT_NONE)
781 		return;
782 
783 	/*
784 	 * There's no explicit rx descriptor flag to indicate whether a given
785 	 * frame has been decrypted or not. We're forced to use the decap
786 	 * format as an implicit indication. However fragmentation rx is always
787 	 * raw and it probably never reports undecrypted raws.
788 	 *
789 	 * This makes sure sniffed frames are reported as-is without stripping
790 	 * the protected flag.
791 	 */
792 	if (fmt == RX_MSDU_DECAP_RAW && !dot11frag)
793 		return;
794 
795 	rx_status->flag |= RX_FLAG_DECRYPTED |
796 			   RX_FLAG_IV_STRIPPED |
797 			   RX_FLAG_MMIC_STRIPPED;
798 	hdr->frame_control = __cpu_to_le16(__le16_to_cpu(hdr->frame_control) &
799 					   ~IEEE80211_FCTL_PROTECTED);
800 }
801 
802 static bool ath10k_htt_rx_h_channel(struct ath10k *ar,
803 				    struct ieee80211_rx_status *status)
804 {
805 	struct ieee80211_channel *ch;
806 
807 	spin_lock_bh(&ar->data_lock);
808 	ch = ar->scan_channel;
809 	if (!ch)
810 		ch = ar->rx_channel;
811 	spin_unlock_bh(&ar->data_lock);
812 
813 	if (!ch)
814 		return false;
815 
816 	status->band = ch->band;
817 	status->freq = ch->center_freq;
818 
819 	return true;
820 }
821 
822 static void ath10k_process_rx(struct ath10k *ar,
823 			      struct ieee80211_rx_status *rx_status,
824 			      struct sk_buff *skb)
825 {
826 	struct ieee80211_rx_status *status;
827 
828 	status = IEEE80211_SKB_RXCB(skb);
829 	*status = *rx_status;
830 
831 	ath10k_dbg(ATH10K_DBG_DATA,
832 		   "rx skb %p len %u %s%s%s%s%s %srate_idx %u vht_nss %u freq %u band %u flag 0x%x fcs-err %imic-err %i\n",
833 		   skb,
834 		   skb->len,
835 		   status->flag == 0 ? "legacy" : "",
836 		   status->flag & RX_FLAG_HT ? "ht" : "",
837 		   status->flag & RX_FLAG_VHT ? "vht" : "",
838 		   status->flag & RX_FLAG_40MHZ ? "40" : "",
839 		   status->vht_flag & RX_VHT_FLAG_80MHZ ? "80" : "",
840 		   status->flag & RX_FLAG_SHORT_GI ? "sgi " : "",
841 		   status->rate_idx,
842 		   status->vht_nss,
843 		   status->freq,
844 		   status->band, status->flag,
845 		   !!(status->flag & RX_FLAG_FAILED_FCS_CRC),
846 		   !!(status->flag & RX_FLAG_MMIC_ERROR));
847 	ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "rx skb: ",
848 			skb->data, skb->len);
849 
850 	ieee80211_rx(ar->hw, skb);
851 }
852 
853 static int ath10k_htt_rx_nwifi_hdrlen(struct ieee80211_hdr *hdr)
854 {
855 	/* nwifi header is padded to 4 bytes. this fixes 4addr rx */
856 	return round_up(ieee80211_hdrlen(hdr->frame_control), 4);
857 }
858 
859 static void ath10k_htt_rx_amsdu(struct ath10k_htt *htt,
860 				struct ieee80211_rx_status *rx_status,
861 				struct sk_buff *skb_in)
862 {
863 	struct htt_rx_desc *rxd;
864 	struct sk_buff *skb = skb_in;
865 	struct sk_buff *first;
866 	enum rx_msdu_decap_format fmt;
867 	enum htt_rx_mpdu_encrypt_type enctype;
868 	struct ieee80211_hdr *hdr;
869 	u8 hdr_buf[64], addr[ETH_ALEN], *qos;
870 	unsigned int hdr_len;
871 
872 	rxd = (void *)skb->data - sizeof(*rxd);
873 	enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
874 			RX_MPDU_START_INFO0_ENCRYPT_TYPE);
875 
876 	hdr = (struct ieee80211_hdr *)rxd->rx_hdr_status;
877 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
878 	memcpy(hdr_buf, hdr, hdr_len);
879 	hdr = (struct ieee80211_hdr *)hdr_buf;
880 
881 	first = skb;
882 	while (skb) {
883 		void *decap_hdr;
884 		int len;
885 
886 		rxd = (void *)skb->data - sizeof(*rxd);
887 		fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
888 			 RX_MSDU_START_INFO1_DECAP_FORMAT);
889 		decap_hdr = (void *)rxd->rx_hdr_status;
890 
891 		skb->ip_summed = ath10k_htt_rx_get_csum_state(skb);
892 
893 		/* First frame in an A-MSDU chain has more decapped data. */
894 		if (skb == first) {
895 			len = round_up(ieee80211_hdrlen(hdr->frame_control), 4);
896 			len += round_up(ath10k_htt_rx_crypto_param_len(enctype),
897 					4);
898 			decap_hdr += len;
899 		}
900 
901 		switch (fmt) {
902 		case RX_MSDU_DECAP_RAW:
903 			/* remove trailing FCS */
904 			skb_trim(skb, skb->len - FCS_LEN);
905 			break;
906 		case RX_MSDU_DECAP_NATIVE_WIFI:
907 			/* pull decapped header and copy DA */
908 			hdr = (struct ieee80211_hdr *)skb->data;
909 			hdr_len = ath10k_htt_rx_nwifi_hdrlen(hdr);
910 			memcpy(addr, ieee80211_get_DA(hdr), ETH_ALEN);
911 			skb_pull(skb, hdr_len);
912 
913 			/* push original 802.11 header */
914 			hdr = (struct ieee80211_hdr *)hdr_buf;
915 			hdr_len = ieee80211_hdrlen(hdr->frame_control);
916 			memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
917 
918 			/* original A-MSDU header has the bit set but we're
919 			 * not including A-MSDU subframe header */
920 			hdr = (struct ieee80211_hdr *)skb->data;
921 			qos = ieee80211_get_qos_ctl(hdr);
922 			qos[0] &= ~IEEE80211_QOS_CTL_A_MSDU_PRESENT;
923 
924 			/* original 802.11 header has a different DA */
925 			memcpy(ieee80211_get_DA(hdr), addr, ETH_ALEN);
926 			break;
927 		case RX_MSDU_DECAP_ETHERNET2_DIX:
928 			/* strip ethernet header and insert decapped 802.11
929 			 * header, amsdu subframe header and rfc1042 header */
930 
931 			len = 0;
932 			len += sizeof(struct rfc1042_hdr);
933 			len += sizeof(struct amsdu_subframe_hdr);
934 
935 			skb_pull(skb, sizeof(struct ethhdr));
936 			memcpy(skb_push(skb, len), decap_hdr, len);
937 			memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
938 			break;
939 		case RX_MSDU_DECAP_8023_SNAP_LLC:
940 			/* insert decapped 802.11 header making a singly
941 			 * A-MSDU */
942 			memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
943 			break;
944 		}
945 
946 		skb_in = skb;
947 		ath10k_htt_rx_h_protected(htt, rx_status, skb_in, enctype, fmt,
948 					  false);
949 		skb = skb->next;
950 		skb_in->next = NULL;
951 
952 		if (skb)
953 			rx_status->flag |= RX_FLAG_AMSDU_MORE;
954 		else
955 			rx_status->flag &= ~RX_FLAG_AMSDU_MORE;
956 
957 		ath10k_process_rx(htt->ar, rx_status, skb_in);
958 	}
959 
960 	/* FIXME: It might be nice to re-assemble the A-MSDU when there's a
961 	 * monitor interface active for sniffing purposes. */
962 }
963 
964 static void ath10k_htt_rx_msdu(struct ath10k_htt *htt,
965 			       struct ieee80211_rx_status *rx_status,
966 			       struct sk_buff *skb)
967 {
968 	struct htt_rx_desc *rxd;
969 	struct ieee80211_hdr *hdr;
970 	enum rx_msdu_decap_format fmt;
971 	enum htt_rx_mpdu_encrypt_type enctype;
972 	int hdr_len;
973 	void *rfc1042;
974 
975 	/* This shouldn't happen. If it does than it may be a FW bug. */
976 	if (skb->next) {
977 		ath10k_warn("htt rx received chained non A-MSDU frame\n");
978 		ath10k_htt_rx_free_msdu_chain(skb->next);
979 		skb->next = NULL;
980 	}
981 
982 	rxd = (void *)skb->data - sizeof(*rxd);
983 	fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
984 			RX_MSDU_START_INFO1_DECAP_FORMAT);
985 	enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
986 			RX_MPDU_START_INFO0_ENCRYPT_TYPE);
987 	hdr = (struct ieee80211_hdr *)rxd->rx_hdr_status;
988 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
989 
990 	skb->ip_summed = ath10k_htt_rx_get_csum_state(skb);
991 
992 	switch (fmt) {
993 	case RX_MSDU_DECAP_RAW:
994 		/* remove trailing FCS */
995 		skb_trim(skb, skb->len - FCS_LEN);
996 		break;
997 	case RX_MSDU_DECAP_NATIVE_WIFI:
998 		/* Pull decapped header */
999 		hdr = (struct ieee80211_hdr *)skb->data;
1000 		hdr_len = ath10k_htt_rx_nwifi_hdrlen(hdr);
1001 		skb_pull(skb, hdr_len);
1002 
1003 		/* Push original header */
1004 		hdr = (struct ieee80211_hdr *)rxd->rx_hdr_status;
1005 		hdr_len = ieee80211_hdrlen(hdr->frame_control);
1006 		memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
1007 		break;
1008 	case RX_MSDU_DECAP_ETHERNET2_DIX:
1009 		/* strip ethernet header and insert decapped 802.11 header and
1010 		 * rfc1042 header */
1011 
1012 		rfc1042 = hdr;
1013 		rfc1042 += roundup(hdr_len, 4);
1014 		rfc1042 += roundup(ath10k_htt_rx_crypto_param_len(enctype), 4);
1015 
1016 		skb_pull(skb, sizeof(struct ethhdr));
1017 		memcpy(skb_push(skb, sizeof(struct rfc1042_hdr)),
1018 		       rfc1042, sizeof(struct rfc1042_hdr));
1019 		memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
1020 		break;
1021 	case RX_MSDU_DECAP_8023_SNAP_LLC:
1022 		/* remove A-MSDU subframe header and insert
1023 		 * decapped 802.11 header. rfc1042 header is already there */
1024 
1025 		skb_pull(skb, sizeof(struct amsdu_subframe_hdr));
1026 		memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
1027 		break;
1028 	}
1029 
1030 	ath10k_htt_rx_h_protected(htt, rx_status, skb, enctype, fmt, false);
1031 
1032 	ath10k_process_rx(htt->ar, rx_status, skb);
1033 }
1034 
1035 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb)
1036 {
1037 	struct htt_rx_desc *rxd;
1038 	u32 flags, info;
1039 	bool is_ip4, is_ip6;
1040 	bool is_tcp, is_udp;
1041 	bool ip_csum_ok, tcpudp_csum_ok;
1042 
1043 	rxd = (void *)skb->data - sizeof(*rxd);
1044 	flags = __le32_to_cpu(rxd->attention.flags);
1045 	info = __le32_to_cpu(rxd->msdu_start.info1);
1046 
1047 	is_ip4 = !!(info & RX_MSDU_START_INFO1_IPV4_PROTO);
1048 	is_ip6 = !!(info & RX_MSDU_START_INFO1_IPV6_PROTO);
1049 	is_tcp = !!(info & RX_MSDU_START_INFO1_TCP_PROTO);
1050 	is_udp = !!(info & RX_MSDU_START_INFO1_UDP_PROTO);
1051 	ip_csum_ok = !(flags & RX_ATTENTION_FLAGS_IP_CHKSUM_FAIL);
1052 	tcpudp_csum_ok = !(flags & RX_ATTENTION_FLAGS_TCP_UDP_CHKSUM_FAIL);
1053 
1054 	if (!is_ip4 && !is_ip6)
1055 		return CHECKSUM_NONE;
1056 	if (!is_tcp && !is_udp)
1057 		return CHECKSUM_NONE;
1058 	if (!ip_csum_ok)
1059 		return CHECKSUM_NONE;
1060 	if (!tcpudp_csum_ok)
1061 		return CHECKSUM_NONE;
1062 
1063 	return CHECKSUM_UNNECESSARY;
1064 }
1065 
1066 static int ath10k_unchain_msdu(struct sk_buff *msdu_head)
1067 {
1068 	struct sk_buff *next = msdu_head->next;
1069 	struct sk_buff *to_free = next;
1070 	int space;
1071 	int total_len = 0;
1072 
1073 	/* TODO:  Might could optimize this by using
1074 	 * skb_try_coalesce or similar method to
1075 	 * decrease copying, or maybe get mac80211 to
1076 	 * provide a way to just receive a list of
1077 	 * skb?
1078 	 */
1079 
1080 	msdu_head->next = NULL;
1081 
1082 	/* Allocate total length all at once. */
1083 	while (next) {
1084 		total_len += next->len;
1085 		next = next->next;
1086 	}
1087 
1088 	space = total_len - skb_tailroom(msdu_head);
1089 	if ((space > 0) &&
1090 	    (pskb_expand_head(msdu_head, 0, space, GFP_ATOMIC) < 0)) {
1091 		/* TODO:  bump some rx-oom error stat */
1092 		/* put it back together so we can free the
1093 		 * whole list at once.
1094 		 */
1095 		msdu_head->next = to_free;
1096 		return -1;
1097 	}
1098 
1099 	/* Walk list again, copying contents into
1100 	 * msdu_head
1101 	 */
1102 	next = to_free;
1103 	while (next) {
1104 		skb_copy_from_linear_data(next, skb_put(msdu_head, next->len),
1105 					  next->len);
1106 		next = next->next;
1107 	}
1108 
1109 	/* If here, we have consolidated skb.  Free the
1110 	 * fragments and pass the main skb on up the
1111 	 * stack.
1112 	 */
1113 	ath10k_htt_rx_free_msdu_chain(to_free);
1114 	return 0;
1115 }
1116 
1117 static bool ath10k_htt_rx_amsdu_allowed(struct ath10k_htt *htt,
1118 					struct sk_buff *head,
1119 					enum htt_rx_mpdu_status status,
1120 					bool channel_set,
1121 					u32 attention)
1122 {
1123 	if (head->len == 0) {
1124 		ath10k_dbg(ATH10K_DBG_HTT,
1125 			   "htt rx dropping due to zero-len\n");
1126 		return false;
1127 	}
1128 
1129 	if (attention & RX_ATTENTION_FLAGS_DECRYPT_ERR) {
1130 		ath10k_dbg(ATH10K_DBG_HTT,
1131 			   "htt rx dropping due to decrypt-err\n");
1132 		return false;
1133 	}
1134 
1135 	if (!channel_set) {
1136 		ath10k_warn("no channel configured; ignoring frame!\n");
1137 		return false;
1138 	}
1139 
1140 	/* Skip mgmt frames while we handle this in WMI */
1141 	if (status == HTT_RX_IND_MPDU_STATUS_MGMT_CTRL ||
1142 	    attention & RX_ATTENTION_FLAGS_MGMT_TYPE) {
1143 		ath10k_dbg(ATH10K_DBG_HTT, "htt rx mgmt ctrl\n");
1144 		return false;
1145 	}
1146 
1147 	if (status != HTT_RX_IND_MPDU_STATUS_OK &&
1148 	    status != HTT_RX_IND_MPDU_STATUS_TKIP_MIC_ERR &&
1149 	    status != HTT_RX_IND_MPDU_STATUS_ERR_INV_PEER &&
1150 	    !htt->ar->monitor_started) {
1151 		ath10k_dbg(ATH10K_DBG_HTT,
1152 			   "htt rx ignoring frame w/ status %d\n",
1153 			   status);
1154 		return false;
1155 	}
1156 
1157 	if (test_bit(ATH10K_CAC_RUNNING, &htt->ar->dev_flags)) {
1158 		ath10k_dbg(ATH10K_DBG_HTT,
1159 			   "htt rx CAC running\n");
1160 		return false;
1161 	}
1162 
1163 	return true;
1164 }
1165 
1166 static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
1167 				  struct htt_rx_indication *rx)
1168 {
1169 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1170 	struct htt_rx_indication_mpdu_range *mpdu_ranges;
1171 	struct htt_rx_desc *rxd;
1172 	enum htt_rx_mpdu_status status;
1173 	struct ieee80211_hdr *hdr;
1174 	int num_mpdu_ranges;
1175 	u32 attention;
1176 	int fw_desc_len;
1177 	u8 *fw_desc;
1178 	bool channel_set;
1179 	int i, j;
1180 	int ret;
1181 
1182 	lockdep_assert_held(&htt->rx_ring.lock);
1183 
1184 	fw_desc_len = __le16_to_cpu(rx->prefix.fw_rx_desc_bytes);
1185 	fw_desc = (u8 *)&rx->fw_desc;
1186 
1187 	num_mpdu_ranges = MS(__le32_to_cpu(rx->hdr.info1),
1188 			     HTT_RX_INDICATION_INFO1_NUM_MPDU_RANGES);
1189 	mpdu_ranges = htt_rx_ind_get_mpdu_ranges(rx);
1190 
1191 	/* Fill this once, while this is per-ppdu */
1192 	if (rx->ppdu.info0 & HTT_RX_INDICATION_INFO0_START_VALID) {
1193 		memset(rx_status, 0, sizeof(*rx_status));
1194 		rx_status->signal  = ATH10K_DEFAULT_NOISE_FLOOR +
1195 				     rx->ppdu.combined_rssi;
1196 	}
1197 
1198 	if (rx->ppdu.info0 & HTT_RX_INDICATION_INFO0_END_VALID) {
1199 		/* TSF available only in 32-bit */
1200 		rx_status->mactime = __le32_to_cpu(rx->ppdu.tsf) & 0xffffffff;
1201 		rx_status->flag |= RX_FLAG_MACTIME_END;
1202 	}
1203 
1204 	channel_set = ath10k_htt_rx_h_channel(htt->ar, rx_status);
1205 
1206 	if (channel_set) {
1207 		ath10k_htt_rx_h_rates(htt->ar, rx_status->band,
1208 				      rx->ppdu.info0,
1209 				      __le32_to_cpu(rx->ppdu.info1),
1210 				      __le32_to_cpu(rx->ppdu.info2),
1211 				      rx_status);
1212 	}
1213 
1214 	ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt rx ind: ",
1215 			rx, sizeof(*rx) +
1216 			(sizeof(struct htt_rx_indication_mpdu_range) *
1217 				num_mpdu_ranges));
1218 
1219 	for (i = 0; i < num_mpdu_ranges; i++) {
1220 		status = mpdu_ranges[i].mpdu_range_status;
1221 
1222 		for (j = 0; j < mpdu_ranges[i].mpdu_count; j++) {
1223 			struct sk_buff *msdu_head, *msdu_tail;
1224 
1225 			attention = 0;
1226 			msdu_head = NULL;
1227 			msdu_tail = NULL;
1228 			ret = ath10k_htt_rx_amsdu_pop(htt,
1229 						      &fw_desc,
1230 						      &fw_desc_len,
1231 						      &msdu_head,
1232 						      &msdu_tail,
1233 						      &attention);
1234 
1235 			if (ret < 0) {
1236 				ath10k_warn("failed to pop amsdu from htt rx ring %d\n",
1237 					    ret);
1238 				ath10k_htt_rx_free_msdu_chain(msdu_head);
1239 				continue;
1240 			}
1241 
1242 			rxd = container_of((void *)msdu_head->data,
1243 					   struct htt_rx_desc,
1244 					   msdu_payload);
1245 
1246 			if (!ath10k_htt_rx_amsdu_allowed(htt, msdu_head,
1247 							 status,
1248 							 channel_set,
1249 							 attention)) {
1250 				ath10k_htt_rx_free_msdu_chain(msdu_head);
1251 				continue;
1252 			}
1253 
1254 			if (ret > 0 &&
1255 			    ath10k_unchain_msdu(msdu_head) < 0) {
1256 				ath10k_htt_rx_free_msdu_chain(msdu_head);
1257 				continue;
1258 			}
1259 
1260 			if (attention & RX_ATTENTION_FLAGS_FCS_ERR)
1261 				rx_status->flag |= RX_FLAG_FAILED_FCS_CRC;
1262 			else
1263 				rx_status->flag &= ~RX_FLAG_FAILED_FCS_CRC;
1264 
1265 			if (attention & RX_ATTENTION_FLAGS_TKIP_MIC_ERR)
1266 				rx_status->flag |= RX_FLAG_MMIC_ERROR;
1267 			else
1268 				rx_status->flag &= ~RX_FLAG_MMIC_ERROR;
1269 
1270 			hdr = ath10k_htt_rx_skb_get_hdr(msdu_head);
1271 
1272 			if (ath10k_htt_rx_hdr_is_amsdu(hdr))
1273 				ath10k_htt_rx_amsdu(htt, rx_status, msdu_head);
1274 			else
1275 				ath10k_htt_rx_msdu(htt, rx_status, msdu_head);
1276 		}
1277 	}
1278 
1279 	tasklet_schedule(&htt->rx_replenish_task);
1280 }
1281 
1282 static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
1283 				struct htt_rx_fragment_indication *frag)
1284 {
1285 	struct sk_buff *msdu_head, *msdu_tail;
1286 	enum htt_rx_mpdu_encrypt_type enctype;
1287 	struct htt_rx_desc *rxd;
1288 	enum rx_msdu_decap_format fmt;
1289 	struct ieee80211_rx_status *rx_status = &htt->rx_status;
1290 	struct ieee80211_hdr *hdr;
1291 	int ret;
1292 	bool tkip_mic_err;
1293 	bool decrypt_err;
1294 	u8 *fw_desc;
1295 	int fw_desc_len, hdrlen, paramlen;
1296 	int trim;
1297 	u32 attention = 0;
1298 
1299 	fw_desc_len = __le16_to_cpu(frag->fw_rx_desc_bytes);
1300 	fw_desc = (u8 *)frag->fw_msdu_rx_desc;
1301 
1302 	msdu_head = NULL;
1303 	msdu_tail = NULL;
1304 
1305 	spin_lock_bh(&htt->rx_ring.lock);
1306 	ret = ath10k_htt_rx_amsdu_pop(htt, &fw_desc, &fw_desc_len,
1307 				      &msdu_head, &msdu_tail,
1308 				      &attention);
1309 	spin_unlock_bh(&htt->rx_ring.lock);
1310 
1311 	ath10k_dbg(ATH10K_DBG_HTT_DUMP, "htt rx frag ahead\n");
1312 
1313 	if (ret) {
1314 		ath10k_warn("failed to pop amsdu from httr rx ring for fragmented rx %d\n",
1315 			    ret);
1316 		ath10k_htt_rx_free_msdu_chain(msdu_head);
1317 		return;
1318 	}
1319 
1320 	/* FIXME: implement signal strength */
1321 	rx_status->flag |= RX_FLAG_NO_SIGNAL_VAL;
1322 
1323 	hdr = (struct ieee80211_hdr *)msdu_head->data;
1324 	rxd = (void *)msdu_head->data - sizeof(*rxd);
1325 	tkip_mic_err = !!(attention & RX_ATTENTION_FLAGS_TKIP_MIC_ERR);
1326 	decrypt_err = !!(attention & RX_ATTENTION_FLAGS_DECRYPT_ERR);
1327 	fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
1328 			RX_MSDU_START_INFO1_DECAP_FORMAT);
1329 
1330 	if (fmt != RX_MSDU_DECAP_RAW) {
1331 		ath10k_warn("we dont support non-raw fragmented rx yet\n");
1332 		dev_kfree_skb_any(msdu_head);
1333 		goto end;
1334 	}
1335 
1336 	enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
1337 		     RX_MPDU_START_INFO0_ENCRYPT_TYPE);
1338 	ath10k_htt_rx_h_protected(htt, rx_status, msdu_head, enctype, fmt,
1339 				  true);
1340 	msdu_head->ip_summed = ath10k_htt_rx_get_csum_state(msdu_head);
1341 
1342 	if (tkip_mic_err)
1343 		ath10k_warn("tkip mic error\n");
1344 
1345 	if (decrypt_err) {
1346 		ath10k_warn("decryption err in fragmented rx\n");
1347 		dev_kfree_skb_any(msdu_head);
1348 		goto end;
1349 	}
1350 
1351 	if (enctype != HTT_RX_MPDU_ENCRYPT_NONE) {
1352 		hdrlen = ieee80211_hdrlen(hdr->frame_control);
1353 		paramlen = ath10k_htt_rx_crypto_param_len(enctype);
1354 
1355 		/* It is more efficient to move the header than the payload */
1356 		memmove((void *)msdu_head->data + paramlen,
1357 			(void *)msdu_head->data,
1358 			hdrlen);
1359 		skb_pull(msdu_head, paramlen);
1360 		hdr = (struct ieee80211_hdr *)msdu_head->data;
1361 	}
1362 
1363 	/* remove trailing FCS */
1364 	trim  = 4;
1365 
1366 	/* remove crypto trailer */
1367 	trim += ath10k_htt_rx_crypto_tail_len(enctype);
1368 
1369 	/* last fragment of TKIP frags has MIC */
1370 	if (!ieee80211_has_morefrags(hdr->frame_control) &&
1371 	    enctype == HTT_RX_MPDU_ENCRYPT_TKIP_WPA)
1372 		trim += 8;
1373 
1374 	if (trim > msdu_head->len) {
1375 		ath10k_warn("htt rx fragment: trailer longer than the frame itself? drop\n");
1376 		dev_kfree_skb_any(msdu_head);
1377 		goto end;
1378 	}
1379 
1380 	skb_trim(msdu_head, msdu_head->len - trim);
1381 
1382 	ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt rx frag mpdu: ",
1383 			msdu_head->data, msdu_head->len);
1384 	ath10k_process_rx(htt->ar, rx_status, msdu_head);
1385 
1386 end:
1387 	if (fw_desc_len > 0) {
1388 		ath10k_dbg(ATH10K_DBG_HTT,
1389 			   "expecting more fragmented rx in one indication %d\n",
1390 			   fw_desc_len);
1391 	}
1392 }
1393 
1394 static void ath10k_htt_rx_frm_tx_compl(struct ath10k *ar,
1395 				       struct sk_buff *skb)
1396 {
1397 	struct ath10k_htt *htt = &ar->htt;
1398 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1399 	struct htt_tx_done tx_done = {};
1400 	int status = MS(resp->data_tx_completion.flags, HTT_DATA_TX_STATUS);
1401 	__le16 msdu_id;
1402 	int i;
1403 
1404 	lockdep_assert_held(&htt->tx_lock);
1405 
1406 	switch (status) {
1407 	case HTT_DATA_TX_STATUS_NO_ACK:
1408 		tx_done.no_ack = true;
1409 		break;
1410 	case HTT_DATA_TX_STATUS_OK:
1411 		break;
1412 	case HTT_DATA_TX_STATUS_DISCARD:
1413 	case HTT_DATA_TX_STATUS_POSTPONE:
1414 	case HTT_DATA_TX_STATUS_DOWNLOAD_FAIL:
1415 		tx_done.discard = true;
1416 		break;
1417 	default:
1418 		ath10k_warn("unhandled tx completion status %d\n", status);
1419 		tx_done.discard = true;
1420 		break;
1421 	}
1422 
1423 	ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion num_msdus %d\n",
1424 		   resp->data_tx_completion.num_msdus);
1425 
1426 	for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
1427 		msdu_id = resp->data_tx_completion.msdus[i];
1428 		tx_done.msdu_id = __le16_to_cpu(msdu_id);
1429 		ath10k_txrx_tx_unref(htt, &tx_done);
1430 	}
1431 }
1432 
1433 static void ath10k_htt_rx_addba(struct ath10k *ar, struct htt_resp *resp)
1434 {
1435 	struct htt_rx_addba *ev = &resp->rx_addba;
1436 	struct ath10k_peer *peer;
1437 	struct ath10k_vif *arvif;
1438 	u16 info0, tid, peer_id;
1439 
1440 	info0 = __le16_to_cpu(ev->info0);
1441 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1442 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1443 
1444 	ath10k_dbg(ATH10K_DBG_HTT,
1445 		   "htt rx addba tid %hu peer_id %hu size %hhu\n",
1446 		   tid, peer_id, ev->window_size);
1447 
1448 	spin_lock_bh(&ar->data_lock);
1449 	peer = ath10k_peer_find_by_id(ar, peer_id);
1450 	if (!peer) {
1451 		ath10k_warn("received addba event for invalid peer_id: %hu\n",
1452 			    peer_id);
1453 		spin_unlock_bh(&ar->data_lock);
1454 		return;
1455 	}
1456 
1457 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1458 	if (!arvif) {
1459 		ath10k_warn("received addba event for invalid vdev_id: %u\n",
1460 			    peer->vdev_id);
1461 		spin_unlock_bh(&ar->data_lock);
1462 		return;
1463 	}
1464 
1465 	ath10k_dbg(ATH10K_DBG_HTT,
1466 		   "htt rx start rx ba session sta %pM tid %hu size %hhu\n",
1467 		   peer->addr, tid, ev->window_size);
1468 
1469 	ieee80211_start_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1470 	spin_unlock_bh(&ar->data_lock);
1471 }
1472 
1473 static void ath10k_htt_rx_delba(struct ath10k *ar, struct htt_resp *resp)
1474 {
1475 	struct htt_rx_delba *ev = &resp->rx_delba;
1476 	struct ath10k_peer *peer;
1477 	struct ath10k_vif *arvif;
1478 	u16 info0, tid, peer_id;
1479 
1480 	info0 = __le16_to_cpu(ev->info0);
1481 	tid = MS(info0, HTT_RX_BA_INFO0_TID);
1482 	peer_id = MS(info0, HTT_RX_BA_INFO0_PEER_ID);
1483 
1484 	ath10k_dbg(ATH10K_DBG_HTT,
1485 		   "htt rx delba tid %hu peer_id %hu\n",
1486 		   tid, peer_id);
1487 
1488 	spin_lock_bh(&ar->data_lock);
1489 	peer = ath10k_peer_find_by_id(ar, peer_id);
1490 	if (!peer) {
1491 		ath10k_warn("received addba event for invalid peer_id: %hu\n",
1492 			    peer_id);
1493 		spin_unlock_bh(&ar->data_lock);
1494 		return;
1495 	}
1496 
1497 	arvif = ath10k_get_arvif(ar, peer->vdev_id);
1498 	if (!arvif) {
1499 		ath10k_warn("received addba event for invalid vdev_id: %u\n",
1500 			    peer->vdev_id);
1501 		spin_unlock_bh(&ar->data_lock);
1502 		return;
1503 	}
1504 
1505 	ath10k_dbg(ATH10K_DBG_HTT,
1506 		   "htt rx stop rx ba session sta %pM tid %hu\n",
1507 		   peer->addr, tid);
1508 
1509 	ieee80211_stop_rx_ba_session_offl(arvif->vif, peer->addr, tid);
1510 	spin_unlock_bh(&ar->data_lock);
1511 }
1512 
1513 void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
1514 {
1515 	struct ath10k_htt *htt = &ar->htt;
1516 	struct htt_resp *resp = (struct htt_resp *)skb->data;
1517 
1518 	/* confirm alignment */
1519 	if (!IS_ALIGNED((unsigned long)skb->data, 4))
1520 		ath10k_warn("unaligned htt message, expect trouble\n");
1521 
1522 	ath10k_dbg(ATH10K_DBG_HTT, "htt rx, msg_type: 0x%0X\n",
1523 		   resp->hdr.msg_type);
1524 	switch (resp->hdr.msg_type) {
1525 	case HTT_T2H_MSG_TYPE_VERSION_CONF: {
1526 		htt->target_version_major = resp->ver_resp.major;
1527 		htt->target_version_minor = resp->ver_resp.minor;
1528 		complete(&htt->target_version_received);
1529 		break;
1530 	}
1531 	case HTT_T2H_MSG_TYPE_RX_IND:
1532 		spin_lock_bh(&htt->rx_ring.lock);
1533 		__skb_queue_tail(&htt->rx_compl_q, skb);
1534 		spin_unlock_bh(&htt->rx_ring.lock);
1535 		tasklet_schedule(&htt->txrx_compl_task);
1536 		return;
1537 	case HTT_T2H_MSG_TYPE_PEER_MAP: {
1538 		struct htt_peer_map_event ev = {
1539 			.vdev_id = resp->peer_map.vdev_id,
1540 			.peer_id = __le16_to_cpu(resp->peer_map.peer_id),
1541 		};
1542 		memcpy(ev.addr, resp->peer_map.addr, sizeof(ev.addr));
1543 		ath10k_peer_map_event(htt, &ev);
1544 		break;
1545 	}
1546 	case HTT_T2H_MSG_TYPE_PEER_UNMAP: {
1547 		struct htt_peer_unmap_event ev = {
1548 			.peer_id = __le16_to_cpu(resp->peer_unmap.peer_id),
1549 		};
1550 		ath10k_peer_unmap_event(htt, &ev);
1551 		break;
1552 	}
1553 	case HTT_T2H_MSG_TYPE_MGMT_TX_COMPLETION: {
1554 		struct htt_tx_done tx_done = {};
1555 		int status = __le32_to_cpu(resp->mgmt_tx_completion.status);
1556 
1557 		tx_done.msdu_id =
1558 			__le32_to_cpu(resp->mgmt_tx_completion.desc_id);
1559 
1560 		switch (status) {
1561 		case HTT_MGMT_TX_STATUS_OK:
1562 			break;
1563 		case HTT_MGMT_TX_STATUS_RETRY:
1564 			tx_done.no_ack = true;
1565 			break;
1566 		case HTT_MGMT_TX_STATUS_DROP:
1567 			tx_done.discard = true;
1568 			break;
1569 		}
1570 
1571 		spin_lock_bh(&htt->tx_lock);
1572 		ath10k_txrx_tx_unref(htt, &tx_done);
1573 		spin_unlock_bh(&htt->tx_lock);
1574 		break;
1575 	}
1576 	case HTT_T2H_MSG_TYPE_TX_COMPL_IND:
1577 		spin_lock_bh(&htt->tx_lock);
1578 		__skb_queue_tail(&htt->tx_compl_q, skb);
1579 		spin_unlock_bh(&htt->tx_lock);
1580 		tasklet_schedule(&htt->txrx_compl_task);
1581 		return;
1582 	case HTT_T2H_MSG_TYPE_SEC_IND: {
1583 		struct ath10k *ar = htt->ar;
1584 		struct htt_security_indication *ev = &resp->security_indication;
1585 
1586 		ath10k_dbg(ATH10K_DBG_HTT,
1587 			   "sec ind peer_id %d unicast %d type %d\n",
1588 			  __le16_to_cpu(ev->peer_id),
1589 			  !!(ev->flags & HTT_SECURITY_IS_UNICAST),
1590 			  MS(ev->flags, HTT_SECURITY_TYPE));
1591 		complete(&ar->install_key_done);
1592 		break;
1593 	}
1594 	case HTT_T2H_MSG_TYPE_RX_FRAG_IND: {
1595 		ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
1596 				skb->data, skb->len);
1597 		ath10k_htt_rx_frag_handler(htt, &resp->rx_frag_ind);
1598 		break;
1599 	}
1600 	case HTT_T2H_MSG_TYPE_TEST:
1601 		/* FIX THIS */
1602 		break;
1603 	case HTT_T2H_MSG_TYPE_STATS_CONF:
1604 		trace_ath10k_htt_stats(skb->data, skb->len);
1605 		break;
1606 	case HTT_T2H_MSG_TYPE_TX_INSPECT_IND:
1607 		/* Firmware can return tx frames if it's unable to fully
1608 		 * process them and suspects host may be able to fix it. ath10k
1609 		 * sends all tx frames as already inspected so this shouldn't
1610 		 * happen unless fw has a bug.
1611 		 */
1612 		ath10k_warn("received an unexpected htt tx inspect event\n");
1613 		break;
1614 	case HTT_T2H_MSG_TYPE_RX_ADDBA:
1615 		ath10k_htt_rx_addba(ar, resp);
1616 		break;
1617 	case HTT_T2H_MSG_TYPE_RX_DELBA:
1618 		ath10k_htt_rx_delba(ar, resp);
1619 		break;
1620 	case HTT_T2H_MSG_TYPE_RX_FLUSH: {
1621 		/* Ignore this event because mac80211 takes care of Rx
1622 		 * aggregation reordering.
1623 		 */
1624 		break;
1625 	}
1626 	default:
1627 		ath10k_dbg(ATH10K_DBG_HTT, "htt event (%d) not handled\n",
1628 			   resp->hdr.msg_type);
1629 		ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
1630 				skb->data, skb->len);
1631 		break;
1632 	};
1633 
1634 	/* Free the indication buffer */
1635 	dev_kfree_skb_any(skb);
1636 }
1637 
1638 static void ath10k_htt_txrx_compl_task(unsigned long ptr)
1639 {
1640 	struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
1641 	struct htt_resp *resp;
1642 	struct sk_buff *skb;
1643 
1644 	spin_lock_bh(&htt->tx_lock);
1645 	while ((skb = __skb_dequeue(&htt->tx_compl_q))) {
1646 		ath10k_htt_rx_frm_tx_compl(htt->ar, skb);
1647 		dev_kfree_skb_any(skb);
1648 	}
1649 	spin_unlock_bh(&htt->tx_lock);
1650 
1651 	spin_lock_bh(&htt->rx_ring.lock);
1652 	while ((skb = __skb_dequeue(&htt->rx_compl_q))) {
1653 		resp = (struct htt_resp *)skb->data;
1654 		ath10k_htt_rx_handler(htt, &resp->rx_ind);
1655 		dev_kfree_skb_any(skb);
1656 	}
1657 	spin_unlock_bh(&htt->rx_ring.lock);
1658 }
1659