1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (C) 2020-2022 Intel Corporation
4  */
5 #include <net/tso.h>
6 #include <linux/tcp.h>
7 
8 #include "iwl-debug.h"
9 #include "iwl-io.h"
10 #include "fw/api/commands.h"
11 #include "fw/api/tx.h"
12 #include "fw/api/datapath.h"
13 #include "queue/tx.h"
14 #include "iwl-fh.h"
15 #include "iwl-scd.h"
16 #include <linux/dmapool.h>
17 
18 /*
19  * iwl_txq_update_byte_tbl - Set up entry in Tx byte-count array
20  */
21 static void iwl_pcie_gen2_update_byte_tbl(struct iwl_trans *trans,
22 					  struct iwl_txq *txq, u16 byte_cnt,
23 					  int num_tbs)
24 {
25 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
26 	u8 filled_tfd_size, num_fetch_chunks;
27 	u16 len = byte_cnt;
28 	__le16 bc_ent;
29 
30 	if (WARN(idx >= txq->n_window, "%d >= %d\n", idx, txq->n_window))
31 		return;
32 
33 	filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) +
34 			  num_tbs * sizeof(struct iwl_tfh_tb);
35 	/*
36 	 * filled_tfd_size contains the number of filled bytes in the TFD.
37 	 * Dividing it by 64 will give the number of chunks to fetch
38 	 * to SRAM- 0 for one chunk, 1 for 2 and so on.
39 	 * If, for example, TFD contains only 3 TBs then 32 bytes
40 	 * of the TFD are used, and only one chunk of 64 bytes should
41 	 * be fetched
42 	 */
43 	num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1;
44 
45 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
46 		struct iwl_gen3_bc_tbl_entry *scd_bc_tbl_gen3 = txq->bc_tbl.addr;
47 
48 		/* Starting from AX210, the HW expects bytes */
49 		WARN_ON(trans->txqs.bc_table_dword);
50 		WARN_ON(len > 0x3FFF);
51 		bc_ent = cpu_to_le16(len | (num_fetch_chunks << 14));
52 		scd_bc_tbl_gen3[idx].tfd_offset = bc_ent;
53 	} else {
54 		struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr;
55 
56 		/* Before AX210, the HW expects DW */
57 		WARN_ON(!trans->txqs.bc_table_dword);
58 		len = DIV_ROUND_UP(len, 4);
59 		WARN_ON(len > 0xFFF);
60 		bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12));
61 		scd_bc_tbl->tfd_offset[idx] = bc_ent;
62 	}
63 }
64 
65 /*
66  * iwl_txq_inc_wr_ptr - Send new write index to hardware
67  */
68 void iwl_txq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_txq *txq)
69 {
70 	lockdep_assert_held(&txq->lock);
71 
72 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr);
73 
74 	/*
75 	 * if not in power-save mode, uCode will never sleep when we're
76 	 * trying to tx (during RFKILL, we're not trying to tx).
77 	 */
78 	iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16));
79 }
80 
81 static u8 iwl_txq_gen2_get_num_tbs(struct iwl_trans *trans,
82 				   struct iwl_tfh_tfd *tfd)
83 {
84 	return le16_to_cpu(tfd->num_tbs) & 0x1f;
85 }
86 
87 void iwl_txq_gen2_tfd_unmap(struct iwl_trans *trans, struct iwl_cmd_meta *meta,
88 			    struct iwl_tfh_tfd *tfd)
89 {
90 	int i, num_tbs;
91 
92 	/* Sanity check on number of chunks */
93 	num_tbs = iwl_txq_gen2_get_num_tbs(trans, tfd);
94 
95 	if (num_tbs > trans->txqs.tfd.max_tbs) {
96 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
97 		return;
98 	}
99 
100 	/* first TB is never freed - it's the bidirectional DMA data */
101 	for (i = 1; i < num_tbs; i++) {
102 		if (meta->tbs & BIT(i))
103 			dma_unmap_page(trans->dev,
104 				       le64_to_cpu(tfd->tbs[i].addr),
105 				       le16_to_cpu(tfd->tbs[i].tb_len),
106 				       DMA_TO_DEVICE);
107 		else
108 			dma_unmap_single(trans->dev,
109 					 le64_to_cpu(tfd->tbs[i].addr),
110 					 le16_to_cpu(tfd->tbs[i].tb_len),
111 					 DMA_TO_DEVICE);
112 	}
113 
114 	tfd->num_tbs = 0;
115 }
116 
117 void iwl_txq_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
118 {
119 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
120 	 * idx is bounded by n_window
121 	 */
122 	int idx = iwl_txq_get_cmd_index(txq, txq->read_ptr);
123 	struct sk_buff *skb;
124 
125 	lockdep_assert_held(&txq->lock);
126 
127 	if (!txq->entries)
128 		return;
129 
130 	iwl_txq_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
131 			       iwl_txq_get_tfd(trans, txq, idx));
132 
133 	skb = txq->entries[idx].skb;
134 
135 	/* Can be called from irqs-disabled context
136 	 * If skb is not NULL, it means that the whole queue is being
137 	 * freed and that the queue is not empty - free the skb
138 	 */
139 	if (skb) {
140 		iwl_op_mode_free_skb(trans->op_mode, skb);
141 		txq->entries[idx].skb = NULL;
142 	}
143 }
144 
145 int iwl_txq_gen2_set_tb(struct iwl_trans *trans, struct iwl_tfh_tfd *tfd,
146 			dma_addr_t addr, u16 len)
147 {
148 	int idx = iwl_txq_gen2_get_num_tbs(trans, tfd);
149 	struct iwl_tfh_tb *tb;
150 
151 	/*
152 	 * Only WARN here so we know about the issue, but we mess up our
153 	 * unmap path because not every place currently checks for errors
154 	 * returned from this function - it can only return an error if
155 	 * there's no more space, and so when we know there is enough we
156 	 * don't always check ...
157 	 */
158 	WARN(iwl_txq_crosses_4g_boundary(addr, len),
159 	     "possible DMA problem with iova:0x%llx, len:%d\n",
160 	     (unsigned long long)addr, len);
161 
162 	if (WARN_ON(idx >= IWL_TFH_NUM_TBS))
163 		return -EINVAL;
164 	tb = &tfd->tbs[idx];
165 
166 	/* Each TFD can point to a maximum max_tbs Tx buffers */
167 	if (le16_to_cpu(tfd->num_tbs) >= trans->txqs.tfd.max_tbs) {
168 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
169 			trans->txqs.tfd.max_tbs);
170 		return -EINVAL;
171 	}
172 
173 	put_unaligned_le64(addr, &tb->addr);
174 	tb->tb_len = cpu_to_le16(len);
175 
176 	tfd->num_tbs = cpu_to_le16(idx + 1);
177 
178 	return idx;
179 }
180 
181 static struct page *get_workaround_page(struct iwl_trans *trans,
182 					struct sk_buff *skb)
183 {
184 	struct page **page_ptr;
185 	struct page *ret;
186 
187 	page_ptr = (void *)((u8 *)skb->cb + trans->txqs.page_offs);
188 
189 	ret = alloc_page(GFP_ATOMIC);
190 	if (!ret)
191 		return NULL;
192 
193 	/* set the chaining pointer to the previous page if there */
194 	*(void **)((u8 *)page_address(ret) + PAGE_SIZE - sizeof(void *)) = *page_ptr;
195 	*page_ptr = ret;
196 
197 	return ret;
198 }
199 
200 /*
201  * Add a TB and if needed apply the FH HW bug workaround;
202  * meta != NULL indicates that it's a page mapping and we
203  * need to dma_unmap_page() and set the meta->tbs bit in
204  * this case.
205  */
206 static int iwl_txq_gen2_set_tb_with_wa(struct iwl_trans *trans,
207 				       struct sk_buff *skb,
208 				       struct iwl_tfh_tfd *tfd,
209 				       dma_addr_t phys, void *virt,
210 				       u16 len, struct iwl_cmd_meta *meta)
211 {
212 	dma_addr_t oldphys = phys;
213 	struct page *page;
214 	int ret;
215 
216 	if (unlikely(dma_mapping_error(trans->dev, phys)))
217 		return -ENOMEM;
218 
219 	if (likely(!iwl_txq_crosses_4g_boundary(phys, len))) {
220 		ret = iwl_txq_gen2_set_tb(trans, tfd, phys, len);
221 
222 		if (ret < 0)
223 			goto unmap;
224 
225 		if (meta)
226 			meta->tbs |= BIT(ret);
227 
228 		ret = 0;
229 		goto trace;
230 	}
231 
232 	/*
233 	 * Work around a hardware bug. If (as expressed in the
234 	 * condition above) the TB ends on a 32-bit boundary,
235 	 * then the next TB may be accessed with the wrong
236 	 * address.
237 	 * To work around it, copy the data elsewhere and make
238 	 * a new mapping for it so the device will not fail.
239 	 */
240 
241 	if (WARN_ON(len > PAGE_SIZE - sizeof(void *))) {
242 		ret = -ENOBUFS;
243 		goto unmap;
244 	}
245 
246 	page = get_workaround_page(trans, skb);
247 	if (!page) {
248 		ret = -ENOMEM;
249 		goto unmap;
250 	}
251 
252 	memcpy(page_address(page), virt, len);
253 
254 	phys = dma_map_single(trans->dev, page_address(page), len,
255 			      DMA_TO_DEVICE);
256 	if (unlikely(dma_mapping_error(trans->dev, phys)))
257 		return -ENOMEM;
258 	ret = iwl_txq_gen2_set_tb(trans, tfd, phys, len);
259 	if (ret < 0) {
260 		/* unmap the new allocation as single */
261 		oldphys = phys;
262 		meta = NULL;
263 		goto unmap;
264 	}
265 	IWL_WARN(trans,
266 		 "TB bug workaround: copied %d bytes from 0x%llx to 0x%llx\n",
267 		 len, (unsigned long long)oldphys, (unsigned long long)phys);
268 
269 	ret = 0;
270 unmap:
271 	if (meta)
272 		dma_unmap_page(trans->dev, oldphys, len, DMA_TO_DEVICE);
273 	else
274 		dma_unmap_single(trans->dev, oldphys, len, DMA_TO_DEVICE);
275 trace:
276 	trace_iwlwifi_dev_tx_tb(trans->dev, skb, virt, phys, len);
277 
278 	return ret;
279 }
280 
281 #ifdef CONFIG_INET
282 struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len,
283 				      struct sk_buff *skb)
284 {
285 	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans->txqs.tso_hdr_page);
286 	struct page **page_ptr;
287 
288 	page_ptr = (void *)((u8 *)skb->cb + trans->txqs.page_offs);
289 
290 	if (WARN_ON(*page_ptr))
291 		return NULL;
292 
293 	if (!p->page)
294 		goto alloc;
295 
296 	/*
297 	 * Check if there's enough room on this page
298 	 *
299 	 * Note that we put a page chaining pointer *last* in the
300 	 * page - we need it somewhere, and if it's there then we
301 	 * avoid DMA mapping the last bits of the page which may
302 	 * trigger the 32-bit boundary hardware bug.
303 	 *
304 	 * (see also get_workaround_page() in tx-gen2.c)
305 	 */
306 	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE -
307 			   sizeof(void *))
308 		goto out;
309 
310 	/* We don't have enough room on this page, get a new one. */
311 	__free_page(p->page);
312 
313 alloc:
314 	p->page = alloc_page(GFP_ATOMIC);
315 	if (!p->page)
316 		return NULL;
317 	p->pos = page_address(p->page);
318 	/* set the chaining pointer to NULL */
319 	*(void **)((u8 *)page_address(p->page) + PAGE_SIZE - sizeof(void *)) = NULL;
320 out:
321 	*page_ptr = p->page;
322 	get_page(p->page);
323 	return p;
324 }
325 #endif
326 
327 static int iwl_txq_gen2_build_amsdu(struct iwl_trans *trans,
328 				    struct sk_buff *skb,
329 				    struct iwl_tfh_tfd *tfd, int start_len,
330 				    u8 hdr_len,
331 				    struct iwl_device_tx_cmd *dev_cmd)
332 {
333 #ifdef CONFIG_INET
334 	struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
335 	struct ieee80211_hdr *hdr = (void *)skb->data;
336 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
337 	unsigned int mss = skb_shinfo(skb)->gso_size;
338 	u16 length, amsdu_pad;
339 	u8 *start_hdr;
340 	struct iwl_tso_hdr_page *hdr_page;
341 	struct tso_t tso;
342 
343 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
344 			     &dev_cmd->hdr, start_len, 0);
345 
346 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
347 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
348 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len;
349 	amsdu_pad = 0;
350 
351 	/* total amount of header we may need for this A-MSDU */
352 	hdr_room = DIV_ROUND_UP(total_len, mss) *
353 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr));
354 
355 	/* Our device supports 9 segments at most, it will fit in 1 page */
356 	hdr_page = get_page_hdr(trans, hdr_room, skb);
357 	if (!hdr_page)
358 		return -ENOMEM;
359 
360 	start_hdr = hdr_page->pos;
361 
362 	/*
363 	 * Pull the ieee80211 header to be able to use TSO core,
364 	 * we will restore it for the tx_status flow.
365 	 */
366 	skb_pull(skb, hdr_len);
367 
368 	/*
369 	 * Remove the length of all the headers that we don't actually
370 	 * have in the MPDU by themselves, but that we duplicate into
371 	 * all the different MSDUs inside the A-MSDU.
372 	 */
373 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
374 
375 	tso_start(skb, &tso);
376 
377 	while (total_len) {
378 		/* this is the data left for this subframe */
379 		unsigned int data_left = min_t(unsigned int, mss, total_len);
380 		unsigned int tb_len;
381 		dma_addr_t tb_phys;
382 		u8 *subf_hdrs_start = hdr_page->pos;
383 
384 		total_len -= data_left;
385 
386 		memset(hdr_page->pos, 0, amsdu_pad);
387 		hdr_page->pos += amsdu_pad;
388 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
389 				  data_left)) & 0x3;
390 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
391 		hdr_page->pos += ETH_ALEN;
392 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
393 		hdr_page->pos += ETH_ALEN;
394 
395 		length = snap_ip_tcp_hdrlen + data_left;
396 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
397 		hdr_page->pos += sizeof(length);
398 
399 		/*
400 		 * This will copy the SNAP as well which will be considered
401 		 * as MAC header.
402 		 */
403 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
404 
405 		hdr_page->pos += snap_ip_tcp_hdrlen;
406 
407 		tb_len = hdr_page->pos - start_hdr;
408 		tb_phys = dma_map_single(trans->dev, start_hdr,
409 					 tb_len, DMA_TO_DEVICE);
410 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
411 			goto out_err;
412 		/*
413 		 * No need for _with_wa, this is from the TSO page and
414 		 * we leave some space at the end of it so can't hit
415 		 * the buggy scenario.
416 		 */
417 		iwl_txq_gen2_set_tb(trans, tfd, tb_phys, tb_len);
418 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
419 					tb_phys, tb_len);
420 		/* add this subframe's headers' length to the tx_cmd */
421 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
422 
423 		/* prepare the start_hdr for the next subframe */
424 		start_hdr = hdr_page->pos;
425 
426 		/* put the payload */
427 		while (data_left) {
428 			int ret;
429 
430 			tb_len = min_t(unsigned int, tso.size, data_left);
431 			tb_phys = dma_map_single(trans->dev, tso.data,
432 						 tb_len, DMA_TO_DEVICE);
433 			ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd,
434 							  tb_phys, tso.data,
435 							  tb_len, NULL);
436 			if (ret)
437 				goto out_err;
438 
439 			data_left -= tb_len;
440 			tso_build_data(skb, &tso, tb_len);
441 		}
442 	}
443 
444 	/* re -add the WiFi header */
445 	skb_push(skb, hdr_len);
446 
447 	return 0;
448 
449 out_err:
450 #endif
451 	return -EINVAL;
452 }
453 
454 static struct
455 iwl_tfh_tfd *iwl_txq_gen2_build_tx_amsdu(struct iwl_trans *trans,
456 					 struct iwl_txq *txq,
457 					 struct iwl_device_tx_cmd *dev_cmd,
458 					 struct sk_buff *skb,
459 					 struct iwl_cmd_meta *out_meta,
460 					 int hdr_len,
461 					 int tx_cmd_len)
462 {
463 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
464 	struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx);
465 	dma_addr_t tb_phys;
466 	int len;
467 	void *tb1_addr;
468 
469 	tb_phys = iwl_txq_get_first_tb_dma(txq, idx);
470 
471 	/*
472 	 * No need for _with_wa, the first TB allocation is aligned up
473 	 * to a 64-byte boundary and thus can't be at the end or cross
474 	 * a page boundary (much less a 2^32 boundary).
475 	 */
476 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
477 
478 	/*
479 	 * The second TB (tb1) points to the remainder of the TX command
480 	 * and the 802.11 header - dword aligned size
481 	 * (This calculation modifies the TX command, so do it before the
482 	 * setup of the first TB)
483 	 */
484 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
485 	      IWL_FIRST_TB_SIZE;
486 
487 	/* do not align A-MSDU to dword as the subframe header aligns it */
488 
489 	/* map the data for TB1 */
490 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
491 	tb_phys = dma_map_single(trans->dev, tb1_addr, len, DMA_TO_DEVICE);
492 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
493 		goto out_err;
494 	/*
495 	 * No need for _with_wa(), we ensure (via alignment) that the data
496 	 * here can never cross or end at a page boundary.
497 	 */
498 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, len);
499 
500 	if (iwl_txq_gen2_build_amsdu(trans, skb, tfd, len + IWL_FIRST_TB_SIZE,
501 				     hdr_len, dev_cmd))
502 		goto out_err;
503 
504 	/* building the A-MSDU might have changed this data, memcpy it now */
505 	memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE);
506 	return tfd;
507 
508 out_err:
509 	iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd);
510 	return NULL;
511 }
512 
513 static int iwl_txq_gen2_tx_add_frags(struct iwl_trans *trans,
514 				     struct sk_buff *skb,
515 				     struct iwl_tfh_tfd *tfd,
516 				     struct iwl_cmd_meta *out_meta)
517 {
518 	int i;
519 
520 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
521 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
522 		dma_addr_t tb_phys;
523 		unsigned int fragsz = skb_frag_size(frag);
524 		int ret;
525 
526 		if (!fragsz)
527 			continue;
528 
529 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
530 					   fragsz, DMA_TO_DEVICE);
531 		ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
532 						  skb_frag_address(frag),
533 						  fragsz, out_meta);
534 		if (ret)
535 			return ret;
536 	}
537 
538 	return 0;
539 }
540 
541 static struct
542 iwl_tfh_tfd *iwl_txq_gen2_build_tx(struct iwl_trans *trans,
543 				   struct iwl_txq *txq,
544 				   struct iwl_device_tx_cmd *dev_cmd,
545 				   struct sk_buff *skb,
546 				   struct iwl_cmd_meta *out_meta,
547 				   int hdr_len,
548 				   int tx_cmd_len,
549 				   bool pad)
550 {
551 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
552 	struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx);
553 	dma_addr_t tb_phys;
554 	int len, tb1_len, tb2_len;
555 	void *tb1_addr;
556 	struct sk_buff *frag;
557 
558 	tb_phys = iwl_txq_get_first_tb_dma(txq, idx);
559 
560 	/* The first TB points to bi-directional DMA data */
561 	memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE);
562 
563 	/*
564 	 * No need for _with_wa, the first TB allocation is aligned up
565 	 * to a 64-byte boundary and thus can't be at the end or cross
566 	 * a page boundary (much less a 2^32 boundary).
567 	 */
568 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
569 
570 	/*
571 	 * The second TB (tb1) points to the remainder of the TX command
572 	 * and the 802.11 header - dword aligned size
573 	 * (This calculation modifies the TX command, so do it before the
574 	 * setup of the first TB)
575 	 */
576 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
577 	      IWL_FIRST_TB_SIZE;
578 
579 	if (pad)
580 		tb1_len = ALIGN(len, 4);
581 	else
582 		tb1_len = len;
583 
584 	/* map the data for TB1 */
585 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
586 	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
587 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
588 		goto out_err;
589 	/*
590 	 * No need for _with_wa(), we ensure (via alignment) that the data
591 	 * here can never cross or end at a page boundary.
592 	 */
593 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
594 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
595 			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
596 
597 	/* set up TFD's third entry to point to remainder of skb's head */
598 	tb2_len = skb_headlen(skb) - hdr_len;
599 
600 	if (tb2_len > 0) {
601 		int ret;
602 
603 		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
604 					 tb2_len, DMA_TO_DEVICE);
605 		ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
606 						  skb->data + hdr_len, tb2_len,
607 						  NULL);
608 		if (ret)
609 			goto out_err;
610 	}
611 
612 	if (iwl_txq_gen2_tx_add_frags(trans, skb, tfd, out_meta))
613 		goto out_err;
614 
615 	skb_walk_frags(skb, frag) {
616 		int ret;
617 
618 		tb_phys = dma_map_single(trans->dev, frag->data,
619 					 skb_headlen(frag), DMA_TO_DEVICE);
620 		ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
621 						  frag->data,
622 						  skb_headlen(frag), NULL);
623 		if (ret)
624 			goto out_err;
625 		if (iwl_txq_gen2_tx_add_frags(trans, frag, tfd, out_meta))
626 			goto out_err;
627 	}
628 
629 	return tfd;
630 
631 out_err:
632 	iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd);
633 	return NULL;
634 }
635 
636 static
637 struct iwl_tfh_tfd *iwl_txq_gen2_build_tfd(struct iwl_trans *trans,
638 					   struct iwl_txq *txq,
639 					   struct iwl_device_tx_cmd *dev_cmd,
640 					   struct sk_buff *skb,
641 					   struct iwl_cmd_meta *out_meta)
642 {
643 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
644 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
645 	struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx);
646 	int len, hdr_len;
647 	bool amsdu;
648 
649 	/* There must be data left over for TB1 or this code must be changed */
650 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
651 
652 	memset(tfd, 0, sizeof(*tfd));
653 
654 	if (trans->trans_cfg->device_family < IWL_DEVICE_FAMILY_AX210)
655 		len = sizeof(struct iwl_tx_cmd_gen2);
656 	else
657 		len = sizeof(struct iwl_tx_cmd_gen3);
658 
659 	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
660 			(*ieee80211_get_qos_ctl(hdr) &
661 			 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
662 
663 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
664 
665 	/*
666 	 * Only build A-MSDUs here if doing so by GSO, otherwise it may be
667 	 * an A-MSDU for other reasons, e.g. NAN or an A-MSDU having been
668 	 * built in the higher layers already.
669 	 */
670 	if (amsdu && skb_shinfo(skb)->gso_size)
671 		return iwl_txq_gen2_build_tx_amsdu(trans, txq, dev_cmd, skb,
672 						    out_meta, hdr_len, len);
673 	return iwl_txq_gen2_build_tx(trans, txq, dev_cmd, skb, out_meta,
674 				      hdr_len, len, !amsdu);
675 }
676 
677 int iwl_txq_space(struct iwl_trans *trans, const struct iwl_txq *q)
678 {
679 	unsigned int max;
680 	unsigned int used;
681 
682 	/*
683 	 * To avoid ambiguity between empty and completely full queues, there
684 	 * should always be less than max_tfd_queue_size elements in the queue.
685 	 * If q->n_window is smaller than max_tfd_queue_size, there is no need
686 	 * to reserve any queue entries for this purpose.
687 	 */
688 	if (q->n_window < trans->trans_cfg->base_params->max_tfd_queue_size)
689 		max = q->n_window;
690 	else
691 		max = trans->trans_cfg->base_params->max_tfd_queue_size - 1;
692 
693 	/*
694 	 * max_tfd_queue_size is a power of 2, so the following is equivalent to
695 	 * modulo by max_tfd_queue_size and is well defined.
696 	 */
697 	used = (q->write_ptr - q->read_ptr) &
698 		(trans->trans_cfg->base_params->max_tfd_queue_size - 1);
699 
700 	if (WARN_ON(used > max))
701 		return 0;
702 
703 	return max - used;
704 }
705 
706 int iwl_txq_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
707 		    struct iwl_device_tx_cmd *dev_cmd, int txq_id)
708 {
709 	struct iwl_cmd_meta *out_meta;
710 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
711 	u16 cmd_len;
712 	int idx;
713 	void *tfd;
714 
715 	if (WARN_ONCE(txq_id >= IWL_MAX_TVQM_QUEUES,
716 		      "queue %d out of range", txq_id))
717 		return -EINVAL;
718 
719 	if (WARN_ONCE(!test_bit(txq_id, trans->txqs.queue_used),
720 		      "TX on unused queue %d\n", txq_id))
721 		return -EINVAL;
722 
723 	if (skb_is_nonlinear(skb) &&
724 	    skb_shinfo(skb)->nr_frags > IWL_TRANS_MAX_FRAGS(trans) &&
725 	    __skb_linearize(skb))
726 		return -ENOMEM;
727 
728 	spin_lock(&txq->lock);
729 
730 	if (iwl_txq_space(trans, txq) < txq->high_mark) {
731 		iwl_txq_stop(trans, txq);
732 
733 		/* don't put the packet on the ring, if there is no room */
734 		if (unlikely(iwl_txq_space(trans, txq) < 3)) {
735 			struct iwl_device_tx_cmd **dev_cmd_ptr;
736 
737 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
738 					       trans->txqs.dev_cmd_offs);
739 
740 			*dev_cmd_ptr = dev_cmd;
741 			__skb_queue_tail(&txq->overflow_q, skb);
742 			spin_unlock(&txq->lock);
743 			return 0;
744 		}
745 	}
746 
747 	idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
748 
749 	/* Set up driver data for this TFD */
750 	txq->entries[idx].skb = skb;
751 	txq->entries[idx].cmd = dev_cmd;
752 
753 	dev_cmd->hdr.sequence =
754 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
755 			    INDEX_TO_SEQ(idx)));
756 
757 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
758 	out_meta = &txq->entries[idx].meta;
759 	out_meta->flags = 0;
760 
761 	tfd = iwl_txq_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
762 	if (!tfd) {
763 		spin_unlock(&txq->lock);
764 		return -1;
765 	}
766 
767 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
768 		struct iwl_tx_cmd_gen3 *tx_cmd_gen3 =
769 			(void *)dev_cmd->payload;
770 
771 		cmd_len = le16_to_cpu(tx_cmd_gen3->len);
772 	} else {
773 		struct iwl_tx_cmd_gen2 *tx_cmd_gen2 =
774 			(void *)dev_cmd->payload;
775 
776 		cmd_len = le16_to_cpu(tx_cmd_gen2->len);
777 	}
778 
779 	/* Set up entry for this TFD in Tx byte-count array */
780 	iwl_pcie_gen2_update_byte_tbl(trans, txq, cmd_len,
781 				      iwl_txq_gen2_get_num_tbs(trans, tfd));
782 
783 	/* start timer if queue currently empty */
784 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
785 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
786 
787 	/* Tell device the write index *just past* this latest filled TFD */
788 	txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr);
789 	iwl_txq_inc_wr_ptr(trans, txq);
790 	/*
791 	 * At this point the frame is "transmitted" successfully
792 	 * and we will get a TX status notification eventually.
793 	 */
794 	spin_unlock(&txq->lock);
795 	return 0;
796 }
797 
798 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
799 
800 /*
801  * iwl_txq_gen2_unmap -  Unmap any remaining DMA mappings and free skb's
802  */
803 void iwl_txq_gen2_unmap(struct iwl_trans *trans, int txq_id)
804 {
805 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
806 
807 	spin_lock_bh(&txq->lock);
808 	while (txq->write_ptr != txq->read_ptr) {
809 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
810 				   txq_id, txq->read_ptr);
811 
812 		if (txq_id != trans->txqs.cmd.q_id) {
813 			int idx = iwl_txq_get_cmd_index(txq, txq->read_ptr);
814 			struct sk_buff *skb = txq->entries[idx].skb;
815 
816 			if (!WARN_ON_ONCE(!skb))
817 				iwl_txq_free_tso_page(trans, skb);
818 		}
819 		iwl_txq_gen2_free_tfd(trans, txq);
820 		txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr);
821 	}
822 
823 	while (!skb_queue_empty(&txq->overflow_q)) {
824 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
825 
826 		iwl_op_mode_free_skb(trans->op_mode, skb);
827 	}
828 
829 	spin_unlock_bh(&txq->lock);
830 
831 	/* just in case - this queue may have been stopped */
832 	iwl_wake_queue(trans, txq);
833 }
834 
835 static void iwl_txq_gen2_free_memory(struct iwl_trans *trans,
836 				     struct iwl_txq *txq)
837 {
838 	struct device *dev = trans->dev;
839 
840 	/* De-alloc circular buffer of TFDs */
841 	if (txq->tfds) {
842 		dma_free_coherent(dev,
843 				  trans->txqs.tfd.size * txq->n_window,
844 				  txq->tfds, txq->dma_addr);
845 		dma_free_coherent(dev,
846 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
847 				  txq->first_tb_bufs, txq->first_tb_dma);
848 	}
849 
850 	kfree(txq->entries);
851 	if (txq->bc_tbl.addr)
852 		dma_pool_free(trans->txqs.bc_pool,
853 			      txq->bc_tbl.addr, txq->bc_tbl.dma);
854 	kfree(txq);
855 }
856 
857 /*
858  * iwl_pcie_txq_free - Deallocate DMA queue.
859  * @txq: Transmit queue to deallocate.
860  *
861  * Empty queue by removing and destroying all BD's.
862  * Free all buffers.
863  * 0-fill, but do not free "txq" descriptor structure.
864  */
865 static void iwl_txq_gen2_free(struct iwl_trans *trans, int txq_id)
866 {
867 	struct iwl_txq *txq;
868 	int i;
869 
870 	if (WARN_ONCE(txq_id >= IWL_MAX_TVQM_QUEUES,
871 		      "queue %d out of range", txq_id))
872 		return;
873 
874 	txq = trans->txqs.txq[txq_id];
875 
876 	if (WARN_ON(!txq))
877 		return;
878 
879 	iwl_txq_gen2_unmap(trans, txq_id);
880 
881 	/* De-alloc array of command/tx buffers */
882 	if (txq_id == trans->txqs.cmd.q_id)
883 		for (i = 0; i < txq->n_window; i++) {
884 			kfree_sensitive(txq->entries[i].cmd);
885 			kfree_sensitive(txq->entries[i].free_buf);
886 		}
887 	del_timer_sync(&txq->stuck_timer);
888 
889 	iwl_txq_gen2_free_memory(trans, txq);
890 
891 	trans->txqs.txq[txq_id] = NULL;
892 
893 	clear_bit(txq_id, trans->txqs.queue_used);
894 }
895 
896 /*
897  * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
898  */
899 static int iwl_queue_init(struct iwl_txq *q, int slots_num)
900 {
901 	q->n_window = slots_num;
902 
903 	/* slots_num must be power-of-two size, otherwise
904 	 * iwl_txq_get_cmd_index is broken. */
905 	if (WARN_ON(!is_power_of_2(slots_num)))
906 		return -EINVAL;
907 
908 	q->low_mark = q->n_window / 4;
909 	if (q->low_mark < 4)
910 		q->low_mark = 4;
911 
912 	q->high_mark = q->n_window / 8;
913 	if (q->high_mark < 2)
914 		q->high_mark = 2;
915 
916 	q->write_ptr = 0;
917 	q->read_ptr = 0;
918 
919 	return 0;
920 }
921 
922 int iwl_txq_init(struct iwl_trans *trans, struct iwl_txq *txq, int slots_num,
923 		 bool cmd_queue)
924 {
925 	int ret;
926 	u32 tfd_queue_max_size =
927 		trans->trans_cfg->base_params->max_tfd_queue_size;
928 
929 	txq->need_update = false;
930 
931 	/* max_tfd_queue_size must be power-of-two size, otherwise
932 	 * iwl_txq_inc_wrap and iwl_txq_dec_wrap are broken. */
933 	if (WARN_ONCE(tfd_queue_max_size & (tfd_queue_max_size - 1),
934 		      "Max tfd queue size must be a power of two, but is %d",
935 		      tfd_queue_max_size))
936 		return -EINVAL;
937 
938 	/* Initialize queue's high/low-water marks, and head/tail indexes */
939 	ret = iwl_queue_init(txq, slots_num);
940 	if (ret)
941 		return ret;
942 
943 	spin_lock_init(&txq->lock);
944 
945 	if (cmd_queue) {
946 		static struct lock_class_key iwl_txq_cmd_queue_lock_class;
947 
948 		lockdep_set_class(&txq->lock, &iwl_txq_cmd_queue_lock_class);
949 	}
950 
951 	__skb_queue_head_init(&txq->overflow_q);
952 
953 	return 0;
954 }
955 
956 void iwl_txq_free_tso_page(struct iwl_trans *trans, struct sk_buff *skb)
957 {
958 	struct page **page_ptr;
959 	struct page *next;
960 
961 	page_ptr = (void *)((u8 *)skb->cb + trans->txqs.page_offs);
962 	next = *page_ptr;
963 	*page_ptr = NULL;
964 
965 	while (next) {
966 		struct page *tmp = next;
967 
968 		next = *(void **)((u8 *)page_address(next) + PAGE_SIZE -
969 				  sizeof(void *));
970 		__free_page(tmp);
971 	}
972 }
973 
974 void iwl_txq_log_scd_error(struct iwl_trans *trans, struct iwl_txq *txq)
975 {
976 	u32 txq_id = txq->id;
977 	u32 status;
978 	bool active;
979 	u8 fifo;
980 
981 	if (trans->trans_cfg->use_tfh) {
982 		IWL_ERR(trans, "Queue %d is stuck %d %d\n", txq_id,
983 			txq->read_ptr, txq->write_ptr);
984 		/* TODO: access new SCD registers and dump them */
985 		return;
986 	}
987 
988 	status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id));
989 	fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7;
990 	active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE));
991 
992 	IWL_ERR(trans,
993 		"Queue %d is %sactive on fifo %d and stuck for %u ms. SW [%d, %d] HW [%d, %d] FH TRB=0x0%x\n",
994 		txq_id, active ? "" : "in", fifo,
995 		jiffies_to_msecs(txq->wd_timeout),
996 		txq->read_ptr, txq->write_ptr,
997 		iwl_read_prph(trans, SCD_QUEUE_RDPTR(txq_id)) &
998 			(trans->trans_cfg->base_params->max_tfd_queue_size - 1),
999 			iwl_read_prph(trans, SCD_QUEUE_WRPTR(txq_id)) &
1000 			(trans->trans_cfg->base_params->max_tfd_queue_size - 1),
1001 			iwl_read_direct32(trans, FH_TX_TRB_REG(fifo)));
1002 }
1003 
1004 static void iwl_txq_stuck_timer(struct timer_list *t)
1005 {
1006 	struct iwl_txq *txq = from_timer(txq, t, stuck_timer);
1007 	struct iwl_trans *trans = txq->trans;
1008 
1009 	spin_lock(&txq->lock);
1010 	/* check if triggered erroneously */
1011 	if (txq->read_ptr == txq->write_ptr) {
1012 		spin_unlock(&txq->lock);
1013 		return;
1014 	}
1015 	spin_unlock(&txq->lock);
1016 
1017 	iwl_txq_log_scd_error(trans, txq);
1018 
1019 	iwl_force_nmi(trans);
1020 }
1021 
1022 int iwl_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq, int slots_num,
1023 		  bool cmd_queue)
1024 {
1025 	size_t tfd_sz = trans->txqs.tfd.size *
1026 		trans->trans_cfg->base_params->max_tfd_queue_size;
1027 	size_t tb0_buf_sz;
1028 	int i;
1029 
1030 	if (WARN_ON(txq->entries || txq->tfds))
1031 		return -EINVAL;
1032 
1033 	if (trans->trans_cfg->use_tfh)
1034 		tfd_sz = trans->txqs.tfd.size * slots_num;
1035 
1036 	timer_setup(&txq->stuck_timer, iwl_txq_stuck_timer, 0);
1037 	txq->trans = trans;
1038 
1039 	txq->n_window = slots_num;
1040 
1041 	txq->entries = kcalloc(slots_num,
1042 			       sizeof(struct iwl_pcie_txq_entry),
1043 			       GFP_KERNEL);
1044 
1045 	if (!txq->entries)
1046 		goto error;
1047 
1048 	if (cmd_queue)
1049 		for (i = 0; i < slots_num; i++) {
1050 			txq->entries[i].cmd =
1051 				kmalloc(sizeof(struct iwl_device_cmd),
1052 					GFP_KERNEL);
1053 			if (!txq->entries[i].cmd)
1054 				goto error;
1055 		}
1056 
1057 	/* Circular buffer of transmit frame descriptors (TFDs),
1058 	 * shared with device */
1059 	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
1060 				       &txq->dma_addr, GFP_KERNEL);
1061 	if (!txq->tfds)
1062 		goto error;
1063 
1064 	BUILD_BUG_ON(sizeof(*txq->first_tb_bufs) != IWL_FIRST_TB_SIZE_ALIGN);
1065 
1066 	tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
1067 
1068 	txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
1069 						&txq->first_tb_dma,
1070 						GFP_KERNEL);
1071 	if (!txq->first_tb_bufs)
1072 		goto err_free_tfds;
1073 
1074 	return 0;
1075 err_free_tfds:
1076 	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
1077 	txq->tfds = NULL;
1078 error:
1079 	if (txq->entries && cmd_queue)
1080 		for (i = 0; i < slots_num; i++)
1081 			kfree(txq->entries[i].cmd);
1082 	kfree(txq->entries);
1083 	txq->entries = NULL;
1084 
1085 	return -ENOMEM;
1086 }
1087 
1088 static struct iwl_txq *
1089 iwl_txq_dyn_alloc_dma(struct iwl_trans *trans, int size, unsigned int timeout)
1090 {
1091 	size_t bc_tbl_size, bc_tbl_entries;
1092 	struct iwl_txq *txq;
1093 	int ret;
1094 
1095 	WARN_ON(!trans->txqs.bc_tbl_size);
1096 
1097 	bc_tbl_size = trans->txqs.bc_tbl_size;
1098 	bc_tbl_entries = bc_tbl_size / sizeof(u16);
1099 
1100 	if (WARN_ON(size > bc_tbl_entries))
1101 		return ERR_PTR(-EINVAL);
1102 
1103 	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1104 	if (!txq)
1105 		return ERR_PTR(-ENOMEM);
1106 
1107 	txq->bc_tbl.addr = dma_pool_alloc(trans->txqs.bc_pool, GFP_KERNEL,
1108 					  &txq->bc_tbl.dma);
1109 	if (!txq->bc_tbl.addr) {
1110 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1111 		kfree(txq);
1112 		return ERR_PTR(-ENOMEM);
1113 	}
1114 
1115 	ret = iwl_txq_alloc(trans, txq, size, false);
1116 	if (ret) {
1117 		IWL_ERR(trans, "Tx queue alloc failed\n");
1118 		goto error;
1119 	}
1120 	ret = iwl_txq_init(trans, txq, size, false);
1121 	if (ret) {
1122 		IWL_ERR(trans, "Tx queue init failed\n");
1123 		goto error;
1124 	}
1125 
1126 	txq->wd_timeout = msecs_to_jiffies(timeout);
1127 
1128 	return txq;
1129 
1130 error:
1131 	iwl_txq_gen2_free_memory(trans, txq);
1132 	return ERR_PTR(ret);
1133 }
1134 
1135 static int iwl_txq_alloc_response(struct iwl_trans *trans, struct iwl_txq *txq,
1136 				  struct iwl_host_cmd *hcmd)
1137 {
1138 	struct iwl_tx_queue_cfg_rsp *rsp;
1139 	int ret, qid;
1140 	u32 wr_ptr;
1141 
1142 	if (WARN_ON(iwl_rx_packet_payload_len(hcmd->resp_pkt) !=
1143 		    sizeof(*rsp))) {
1144 		ret = -EINVAL;
1145 		goto error_free_resp;
1146 	}
1147 
1148 	rsp = (void *)hcmd->resp_pkt->data;
1149 	qid = le16_to_cpu(rsp->queue_number);
1150 	wr_ptr = le16_to_cpu(rsp->write_pointer);
1151 
1152 	if (qid >= ARRAY_SIZE(trans->txqs.txq)) {
1153 		WARN_ONCE(1, "queue index %d unsupported", qid);
1154 		ret = -EIO;
1155 		goto error_free_resp;
1156 	}
1157 
1158 	if (test_and_set_bit(qid, trans->txqs.queue_used)) {
1159 		WARN_ONCE(1, "queue %d already used", qid);
1160 		ret = -EIO;
1161 		goto error_free_resp;
1162 	}
1163 
1164 	if (WARN_ONCE(trans->txqs.txq[qid],
1165 		      "queue %d already allocated\n", qid)) {
1166 		ret = -EIO;
1167 		goto error_free_resp;
1168 	}
1169 
1170 	txq->id = qid;
1171 	trans->txqs.txq[qid] = txq;
1172 	wr_ptr &= (trans->trans_cfg->base_params->max_tfd_queue_size - 1);
1173 
1174 	/* Place first TFD at index corresponding to start sequence number */
1175 	txq->read_ptr = wr_ptr;
1176 	txq->write_ptr = wr_ptr;
1177 
1178 	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1179 
1180 	iwl_free_resp(hcmd);
1181 	return qid;
1182 
1183 error_free_resp:
1184 	iwl_free_resp(hcmd);
1185 	iwl_txq_gen2_free_memory(trans, txq);
1186 	return ret;
1187 }
1188 
1189 int iwl_txq_dyn_alloc(struct iwl_trans *trans, u32 flags, u32 sta_mask,
1190 		      u8 tid, int size, unsigned int timeout)
1191 {
1192 	struct iwl_txq *txq;
1193 	union {
1194 		struct iwl_tx_queue_cfg_cmd old;
1195 		struct iwl_scd_queue_cfg_cmd new;
1196 	} cmd;
1197 	struct iwl_host_cmd hcmd = {
1198 		.flags = CMD_WANT_SKB,
1199 	};
1200 	int ret;
1201 
1202 	if (trans->trans_cfg->device_family == IWL_DEVICE_FAMILY_BZ &&
1203 	    trans->hw_rev_step == SILICON_A_STEP)
1204 		size = 4096;
1205 
1206 	txq = iwl_txq_dyn_alloc_dma(trans, size, timeout);
1207 	if (IS_ERR(txq))
1208 		return PTR_ERR(txq);
1209 
1210 	if (trans->txqs.queue_alloc_cmd_ver == 0) {
1211 		memset(&cmd.old, 0, sizeof(cmd.old));
1212 		cmd.old.tfdq_addr = cpu_to_le64(txq->dma_addr);
1213 		cmd.old.byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1214 		cmd.old.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size));
1215 		cmd.old.flags = cpu_to_le16(flags | TX_QUEUE_CFG_ENABLE_QUEUE);
1216 		cmd.old.tid = tid;
1217 
1218 		if (hweight32(sta_mask) != 1) {
1219 			ret = -EINVAL;
1220 			goto error;
1221 		}
1222 		cmd.old.sta_id = ffs(sta_mask) - 1;
1223 
1224 		hcmd.id = SCD_QUEUE_CFG;
1225 		hcmd.len[0] = sizeof(cmd.old);
1226 		hcmd.data[0] = &cmd.old;
1227 	} else if (trans->txqs.queue_alloc_cmd_ver == 3) {
1228 		memset(&cmd.new, 0, sizeof(cmd.new));
1229 		cmd.new.operation = cpu_to_le32(IWL_SCD_QUEUE_ADD);
1230 		cmd.new.u.add.tfdq_dram_addr = cpu_to_le64(txq->dma_addr);
1231 		cmd.new.u.add.bc_dram_addr = cpu_to_le64(txq->bc_tbl.dma);
1232 		cmd.new.u.add.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size));
1233 		cmd.new.u.add.flags = cpu_to_le32(flags);
1234 		cmd.new.u.add.sta_mask = cpu_to_le32(sta_mask);
1235 		cmd.new.u.add.tid = tid;
1236 
1237 		hcmd.id = WIDE_ID(DATA_PATH_GROUP, SCD_QUEUE_CONFIG_CMD);
1238 		hcmd.len[0] = sizeof(cmd.new);
1239 		hcmd.data[0] = &cmd.new;
1240 	} else {
1241 		ret = -EOPNOTSUPP;
1242 		goto error;
1243 	}
1244 
1245 	ret = iwl_trans_send_cmd(trans, &hcmd);
1246 	if (ret)
1247 		goto error;
1248 
1249 	return iwl_txq_alloc_response(trans, txq, &hcmd);
1250 
1251 error:
1252 	iwl_txq_gen2_free_memory(trans, txq);
1253 	return ret;
1254 }
1255 
1256 void iwl_txq_dyn_free(struct iwl_trans *trans, int queue)
1257 {
1258 	if (WARN(queue >= IWL_MAX_TVQM_QUEUES,
1259 		 "queue %d out of range", queue))
1260 		return;
1261 
1262 	/*
1263 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1264 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1265 	 * allow the op_mode to call txq_disable after it already called
1266 	 * stop_device.
1267 	 */
1268 	if (!test_and_clear_bit(queue, trans->txqs.queue_used)) {
1269 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1270 			  "queue %d not used", queue);
1271 		return;
1272 	}
1273 
1274 	iwl_txq_gen2_free(trans, queue);
1275 
1276 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1277 }
1278 
1279 void iwl_txq_gen2_tx_free(struct iwl_trans *trans)
1280 {
1281 	int i;
1282 
1283 	memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used));
1284 
1285 	/* Free all TX queues */
1286 	for (i = 0; i < ARRAY_SIZE(trans->txqs.txq); i++) {
1287 		if (!trans->txqs.txq[i])
1288 			continue;
1289 
1290 		iwl_txq_gen2_free(trans, i);
1291 	}
1292 }
1293 
1294 int iwl_txq_gen2_init(struct iwl_trans *trans, int txq_id, int queue_size)
1295 {
1296 	struct iwl_txq *queue;
1297 	int ret;
1298 
1299 	/* alloc and init the tx queue */
1300 	if (!trans->txqs.txq[txq_id]) {
1301 		queue = kzalloc(sizeof(*queue), GFP_KERNEL);
1302 		if (!queue) {
1303 			IWL_ERR(trans, "Not enough memory for tx queue\n");
1304 			return -ENOMEM;
1305 		}
1306 		trans->txqs.txq[txq_id] = queue;
1307 		ret = iwl_txq_alloc(trans, queue, queue_size, true);
1308 		if (ret) {
1309 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1310 			goto error;
1311 		}
1312 	} else {
1313 		queue = trans->txqs.txq[txq_id];
1314 	}
1315 
1316 	ret = iwl_txq_init(trans, queue, queue_size,
1317 			   (txq_id == trans->txqs.cmd.q_id));
1318 	if (ret) {
1319 		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1320 		goto error;
1321 	}
1322 	trans->txqs.txq[txq_id]->id = txq_id;
1323 	set_bit(txq_id, trans->txqs.queue_used);
1324 
1325 	return 0;
1326 
1327 error:
1328 	iwl_txq_gen2_tx_free(trans);
1329 	return ret;
1330 }
1331 
1332 static inline dma_addr_t iwl_txq_gen1_tfd_tb_get_addr(struct iwl_trans *trans,
1333 						      void *_tfd, u8 idx)
1334 {
1335 	struct iwl_tfd *tfd;
1336 	struct iwl_tfd_tb *tb;
1337 	dma_addr_t addr;
1338 	dma_addr_t hi_len;
1339 
1340 	if (trans->trans_cfg->use_tfh) {
1341 		struct iwl_tfh_tfd *tfh_tfd = _tfd;
1342 		struct iwl_tfh_tb *tfh_tb = &tfh_tfd->tbs[idx];
1343 
1344 		return (dma_addr_t)(le64_to_cpu(tfh_tb->addr));
1345 	}
1346 
1347 	tfd = _tfd;
1348 	tb = &tfd->tbs[idx];
1349 	addr = get_unaligned_le32(&tb->lo);
1350 
1351 	if (sizeof(dma_addr_t) <= sizeof(u32))
1352 		return addr;
1353 
1354 	hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
1355 
1356 	/*
1357 	 * shift by 16 twice to avoid warnings on 32-bit
1358 	 * (where this code never runs anyway due to the
1359 	 * if statement above)
1360 	 */
1361 	return addr | ((hi_len << 16) << 16);
1362 }
1363 
1364 void iwl_txq_gen1_tfd_unmap(struct iwl_trans *trans,
1365 			    struct iwl_cmd_meta *meta,
1366 			    struct iwl_txq *txq, int index)
1367 {
1368 	int i, num_tbs;
1369 	void *tfd = iwl_txq_get_tfd(trans, txq, index);
1370 
1371 	/* Sanity check on number of chunks */
1372 	num_tbs = iwl_txq_gen1_tfd_get_num_tbs(trans, tfd);
1373 
1374 	if (num_tbs > trans->txqs.tfd.max_tbs) {
1375 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
1376 		/* @todo issue fatal error, it is quite serious situation */
1377 		return;
1378 	}
1379 
1380 	/* first TB is never freed - it's the bidirectional DMA data */
1381 
1382 	for (i = 1; i < num_tbs; i++) {
1383 		if (meta->tbs & BIT(i))
1384 			dma_unmap_page(trans->dev,
1385 				       iwl_txq_gen1_tfd_tb_get_addr(trans,
1386 								    tfd, i),
1387 				       iwl_txq_gen1_tfd_tb_get_len(trans,
1388 								   tfd, i),
1389 				       DMA_TO_DEVICE);
1390 		else
1391 			dma_unmap_single(trans->dev,
1392 					 iwl_txq_gen1_tfd_tb_get_addr(trans,
1393 								      tfd, i),
1394 					 iwl_txq_gen1_tfd_tb_get_len(trans,
1395 								     tfd, i),
1396 					 DMA_TO_DEVICE);
1397 	}
1398 
1399 	meta->tbs = 0;
1400 
1401 	if (trans->trans_cfg->use_tfh) {
1402 		struct iwl_tfh_tfd *tfd_fh = (void *)tfd;
1403 
1404 		tfd_fh->num_tbs = 0;
1405 	} else {
1406 		struct iwl_tfd *tfd_fh = (void *)tfd;
1407 
1408 		tfd_fh->num_tbs = 0;
1409 	}
1410 }
1411 
1412 #define IWL_TX_CRC_SIZE 4
1413 #define IWL_TX_DELIMITER_SIZE 4
1414 
1415 /*
1416  * iwl_txq_gen1_update_byte_cnt_tbl - Set up entry in Tx byte-count array
1417  */
1418 void iwl_txq_gen1_update_byte_cnt_tbl(struct iwl_trans *trans,
1419 				      struct iwl_txq *txq, u16 byte_cnt,
1420 				      int num_tbs)
1421 {
1422 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
1423 	int write_ptr = txq->write_ptr;
1424 	int txq_id = txq->id;
1425 	u8 sec_ctl = 0;
1426 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
1427 	__le16 bc_ent;
1428 	struct iwl_device_tx_cmd *dev_cmd = txq->entries[txq->write_ptr].cmd;
1429 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
1430 	u8 sta_id = tx_cmd->sta_id;
1431 
1432 	scd_bc_tbl = trans->txqs.scd_bc_tbls.addr;
1433 
1434 	sec_ctl = tx_cmd->sec_ctl;
1435 
1436 	switch (sec_ctl & TX_CMD_SEC_MSK) {
1437 	case TX_CMD_SEC_CCM:
1438 		len += IEEE80211_CCMP_MIC_LEN;
1439 		break;
1440 	case TX_CMD_SEC_TKIP:
1441 		len += IEEE80211_TKIP_ICV_LEN;
1442 		break;
1443 	case TX_CMD_SEC_WEP:
1444 		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
1445 		break;
1446 	}
1447 	if (trans->txqs.bc_table_dword)
1448 		len = DIV_ROUND_UP(len, 4);
1449 
1450 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
1451 		return;
1452 
1453 	bc_ent = cpu_to_le16(len | (sta_id << 12));
1454 
1455 	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
1456 
1457 	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
1458 		scd_bc_tbl[txq_id].tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] =
1459 			bc_ent;
1460 }
1461 
1462 void iwl_txq_gen1_inval_byte_cnt_tbl(struct iwl_trans *trans,
1463 				     struct iwl_txq *txq)
1464 {
1465 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans->txqs.scd_bc_tbls.addr;
1466 	int txq_id = txq->id;
1467 	int read_ptr = txq->read_ptr;
1468 	u8 sta_id = 0;
1469 	__le16 bc_ent;
1470 	struct iwl_device_tx_cmd *dev_cmd = txq->entries[read_ptr].cmd;
1471 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
1472 
1473 	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
1474 
1475 	if (txq_id != trans->txqs.cmd.q_id)
1476 		sta_id = tx_cmd->sta_id;
1477 
1478 	bc_ent = cpu_to_le16(1 | (sta_id << 12));
1479 
1480 	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
1481 
1482 	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
1483 		scd_bc_tbl[txq_id].tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] =
1484 			bc_ent;
1485 }
1486 
1487 /*
1488  * iwl_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
1489  * @trans - transport private data
1490  * @txq - tx queue
1491  * @dma_dir - the direction of the DMA mapping
1492  *
1493  * Does NOT advance any TFD circular buffer read/write indexes
1494  * Does NOT free the TFD itself (which is within circular buffer)
1495  */
1496 void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
1497 {
1498 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
1499 	 * idx is bounded by n_window
1500 	 */
1501 	int rd_ptr = txq->read_ptr;
1502 	int idx = iwl_txq_get_cmd_index(txq, rd_ptr);
1503 	struct sk_buff *skb;
1504 
1505 	lockdep_assert_held(&txq->lock);
1506 
1507 	if (!txq->entries)
1508 		return;
1509 
1510 	/* We have only q->n_window txq->entries, but we use
1511 	 * TFD_QUEUE_SIZE_MAX tfds
1512 	 */
1513 	iwl_txq_gen1_tfd_unmap(trans, &txq->entries[idx].meta, txq, rd_ptr);
1514 
1515 	/* free SKB */
1516 	skb = txq->entries[idx].skb;
1517 
1518 	/* Can be called from irqs-disabled context
1519 	 * If skb is not NULL, it means that the whole queue is being
1520 	 * freed and that the queue is not empty - free the skb
1521 	 */
1522 	if (skb) {
1523 		iwl_op_mode_free_skb(trans->op_mode, skb);
1524 		txq->entries[idx].skb = NULL;
1525 	}
1526 }
1527 
1528 void iwl_txq_progress(struct iwl_txq *txq)
1529 {
1530 	lockdep_assert_held(&txq->lock);
1531 
1532 	if (!txq->wd_timeout)
1533 		return;
1534 
1535 	/*
1536 	 * station is asleep and we send data - that must
1537 	 * be uAPSD or PS-Poll. Don't rearm the timer.
1538 	 */
1539 	if (txq->frozen)
1540 		return;
1541 
1542 	/*
1543 	 * if empty delete timer, otherwise move timer forward
1544 	 * since we're making progress on this queue
1545 	 */
1546 	if (txq->read_ptr == txq->write_ptr)
1547 		del_timer(&txq->stuck_timer);
1548 	else
1549 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1550 }
1551 
1552 /* Frees buffers until index _not_ inclusive */
1553 void iwl_txq_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1554 		     struct sk_buff_head *skbs)
1555 {
1556 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
1557 	int tfd_num, read_ptr, last_to_free;
1558 
1559 	/* This function is not meant to release cmd queue*/
1560 	if (WARN_ON(txq_id == trans->txqs.cmd.q_id))
1561 		return;
1562 
1563 	if (WARN_ON(!txq))
1564 		return;
1565 
1566 	tfd_num = iwl_txq_get_cmd_index(txq, ssn);
1567 	read_ptr = iwl_txq_get_cmd_index(txq, txq->read_ptr);
1568 
1569 	spin_lock_bh(&txq->lock);
1570 
1571 	if (!test_bit(txq_id, trans->txqs.queue_used)) {
1572 		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1573 				    txq_id, ssn);
1574 		goto out;
1575 	}
1576 
1577 	if (read_ptr == tfd_num)
1578 		goto out;
1579 
1580 	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1581 			   txq_id, txq->read_ptr, tfd_num, ssn);
1582 
1583 	/*Since we free until index _not_ inclusive, the one before index is
1584 	 * the last we will free. This one must be used */
1585 	last_to_free = iwl_txq_dec_wrap(trans, tfd_num);
1586 
1587 	if (!iwl_txq_used(txq, last_to_free)) {
1588 		IWL_ERR(trans,
1589 			"%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1590 			__func__, txq_id, last_to_free,
1591 			trans->trans_cfg->base_params->max_tfd_queue_size,
1592 			txq->write_ptr, txq->read_ptr);
1593 
1594 		iwl_op_mode_time_point(trans->op_mode,
1595 				       IWL_FW_INI_TIME_POINT_FAKE_TX,
1596 				       NULL);
1597 		goto out;
1598 	}
1599 
1600 	if (WARN_ON(!skb_queue_empty(skbs)))
1601 		goto out;
1602 
1603 	for (;
1604 	     read_ptr != tfd_num;
1605 	     txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr),
1606 	     read_ptr = iwl_txq_get_cmd_index(txq, txq->read_ptr)) {
1607 		struct sk_buff *skb = txq->entries[read_ptr].skb;
1608 
1609 		if (WARN_ON_ONCE(!skb))
1610 			continue;
1611 
1612 		iwl_txq_free_tso_page(trans, skb);
1613 
1614 		__skb_queue_tail(skbs, skb);
1615 
1616 		txq->entries[read_ptr].skb = NULL;
1617 
1618 		if (!trans->trans_cfg->use_tfh)
1619 			iwl_txq_gen1_inval_byte_cnt_tbl(trans, txq);
1620 
1621 		iwl_txq_free_tfd(trans, txq);
1622 	}
1623 
1624 	iwl_txq_progress(txq);
1625 
1626 	if (iwl_txq_space(trans, txq) > txq->low_mark &&
1627 	    test_bit(txq_id, trans->txqs.queue_stopped)) {
1628 		struct sk_buff_head overflow_skbs;
1629 
1630 		__skb_queue_head_init(&overflow_skbs);
1631 		skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1632 
1633 		/*
1634 		 * We are going to transmit from the overflow queue.
1635 		 * Remember this state so that wait_for_txq_empty will know we
1636 		 * are adding more packets to the TFD queue. It cannot rely on
1637 		 * the state of &txq->overflow_q, as we just emptied it, but
1638 		 * haven't TXed the content yet.
1639 		 */
1640 		txq->overflow_tx = true;
1641 
1642 		/*
1643 		 * This is tricky: we are in reclaim path which is non
1644 		 * re-entrant, so noone will try to take the access the
1645 		 * txq data from that path. We stopped tx, so we can't
1646 		 * have tx as well. Bottom line, we can unlock and re-lock
1647 		 * later.
1648 		 */
1649 		spin_unlock_bh(&txq->lock);
1650 
1651 		while (!skb_queue_empty(&overflow_skbs)) {
1652 			struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1653 			struct iwl_device_tx_cmd *dev_cmd_ptr;
1654 
1655 			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1656 						 trans->txqs.dev_cmd_offs);
1657 
1658 			/*
1659 			 * Note that we can very well be overflowing again.
1660 			 * In that case, iwl_txq_space will be small again
1661 			 * and we won't wake mac80211's queue.
1662 			 */
1663 			iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id);
1664 		}
1665 
1666 		if (iwl_txq_space(trans, txq) > txq->low_mark)
1667 			iwl_wake_queue(trans, txq);
1668 
1669 		spin_lock_bh(&txq->lock);
1670 		txq->overflow_tx = false;
1671 	}
1672 
1673 out:
1674 	spin_unlock_bh(&txq->lock);
1675 }
1676 
1677 /* Set wr_ptr of specific device and txq  */
1678 void iwl_txq_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr)
1679 {
1680 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
1681 
1682 	spin_lock_bh(&txq->lock);
1683 
1684 	txq->write_ptr = ptr;
1685 	txq->read_ptr = txq->write_ptr;
1686 
1687 	spin_unlock_bh(&txq->lock);
1688 }
1689 
1690 void iwl_trans_txq_freeze_timer(struct iwl_trans *trans, unsigned long txqs,
1691 				bool freeze)
1692 {
1693 	int queue;
1694 
1695 	for_each_set_bit(queue, &txqs, BITS_PER_LONG) {
1696 		struct iwl_txq *txq = trans->txqs.txq[queue];
1697 		unsigned long now;
1698 
1699 		spin_lock_bh(&txq->lock);
1700 
1701 		now = jiffies;
1702 
1703 		if (txq->frozen == freeze)
1704 			goto next_queue;
1705 
1706 		IWL_DEBUG_TX_QUEUES(trans, "%s TXQ %d\n",
1707 				    freeze ? "Freezing" : "Waking", queue);
1708 
1709 		txq->frozen = freeze;
1710 
1711 		if (txq->read_ptr == txq->write_ptr)
1712 			goto next_queue;
1713 
1714 		if (freeze) {
1715 			if (unlikely(time_after(now,
1716 						txq->stuck_timer.expires))) {
1717 				/*
1718 				 * The timer should have fired, maybe it is
1719 				 * spinning right now on the lock.
1720 				 */
1721 				goto next_queue;
1722 			}
1723 			/* remember how long until the timer fires */
1724 			txq->frozen_expiry_remainder =
1725 				txq->stuck_timer.expires - now;
1726 			del_timer(&txq->stuck_timer);
1727 			goto next_queue;
1728 		}
1729 
1730 		/*
1731 		 * Wake a non-empty queue -> arm timer with the
1732 		 * remainder before it froze
1733 		 */
1734 		mod_timer(&txq->stuck_timer,
1735 			  now + txq->frozen_expiry_remainder);
1736 
1737 next_queue:
1738 		spin_unlock_bh(&txq->lock);
1739 	}
1740 }
1741 
1742 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1743 
1744 static int iwl_trans_txq_send_hcmd_sync(struct iwl_trans *trans,
1745 					struct iwl_host_cmd *cmd)
1746 {
1747 	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
1748 	struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id];
1749 	int cmd_idx;
1750 	int ret;
1751 
1752 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
1753 
1754 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1755 				  &trans->status),
1756 		 "Command %s: a command is already active!\n", cmd_str))
1757 		return -EIO;
1758 
1759 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
1760 
1761 	cmd_idx = trans->ops->send_cmd(trans, cmd);
1762 	if (cmd_idx < 0) {
1763 		ret = cmd_idx;
1764 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1765 		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
1766 			cmd_str, ret);
1767 		return ret;
1768 	}
1769 
1770 	ret = wait_event_timeout(trans->wait_command_queue,
1771 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1772 					   &trans->status),
1773 				 HOST_COMPLETE_TIMEOUT);
1774 	if (!ret) {
1775 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1776 			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1777 
1778 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1779 			txq->read_ptr, txq->write_ptr);
1780 
1781 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1782 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1783 			       cmd_str);
1784 		ret = -ETIMEDOUT;
1785 
1786 		iwl_trans_sync_nmi(trans);
1787 		goto cancel;
1788 	}
1789 
1790 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1791 		if (!test_and_clear_bit(STATUS_SUPPRESS_CMD_ERROR_ONCE,
1792 					&trans->status)) {
1793 			IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
1794 			dump_stack();
1795 		}
1796 		ret = -EIO;
1797 		goto cancel;
1798 	}
1799 
1800 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1801 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1802 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1803 		ret = -ERFKILL;
1804 		goto cancel;
1805 	}
1806 
1807 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1808 		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
1809 		ret = -EIO;
1810 		goto cancel;
1811 	}
1812 
1813 	return 0;
1814 
1815 cancel:
1816 	if (cmd->flags & CMD_WANT_SKB) {
1817 		/*
1818 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1819 		 * TX cmd queue. Otherwise in case the cmd comes
1820 		 * in later, it will possibly set an invalid
1821 		 * address (cmd->meta.source).
1822 		 */
1823 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1824 	}
1825 
1826 	if (cmd->resp_pkt) {
1827 		iwl_free_resp(cmd);
1828 		cmd->resp_pkt = NULL;
1829 	}
1830 
1831 	return ret;
1832 }
1833 
1834 int iwl_trans_txq_send_hcmd(struct iwl_trans *trans,
1835 			    struct iwl_host_cmd *cmd)
1836 {
1837 	/* Make sure the NIC is still alive in the bus */
1838 	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1839 		return -ENODEV;
1840 
1841 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1842 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1843 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1844 				  cmd->id);
1845 		return -ERFKILL;
1846 	}
1847 
1848 	if (unlikely(trans->system_pm_mode == IWL_PLAT_PM_MODE_D3 &&
1849 		     !(cmd->flags & CMD_SEND_IN_D3))) {
1850 		IWL_DEBUG_WOWLAN(trans, "Dropping CMD 0x%x: D3\n", cmd->id);
1851 		return -EHOSTDOWN;
1852 	}
1853 
1854 	if (cmd->flags & CMD_ASYNC) {
1855 		int ret;
1856 
1857 		/* An asynchronous command can not expect an SKB to be set. */
1858 		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1859 			return -EINVAL;
1860 
1861 		ret = trans->ops->send_cmd(trans, cmd);
1862 		if (ret < 0) {
1863 			IWL_ERR(trans,
1864 				"Error sending %s: enqueue_hcmd failed: %d\n",
1865 				iwl_get_cmd_string(trans, cmd->id), ret);
1866 			return ret;
1867 		}
1868 		return 0;
1869 	}
1870 
1871 	return iwl_trans_txq_send_hcmd_sync(trans, cmd);
1872 }
1873 
1874