1 /******************************************************************************
2  *
3  * This file is provided under a dual BSD/GPLv2 license.  When using or
4  * redistributing this file, you may do so under either license.
5  *
6  * GPL LICENSE SUMMARY
7  *
8  * Copyright(c) 2017 Intel Deutschland GmbH
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of version 2 of the GNU General Public License as
12  * published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * BSD LICENSE
20  *
21  * Copyright(c) 2017 Intel Deutschland GmbH
22  * All rights reserved.
23  *
24  * Redistribution and use in source and binary forms, with or without
25  * modification, are permitted provided that the following conditions
26  * are met:
27  *
28  *  * Redistributions of source code must retain the above copyright
29  *    notice, this list of conditions and the following disclaimer.
30  *  * Redistributions in binary form must reproduce the above copyright
31  *    notice, this list of conditions and the following disclaimer in
32  *    the documentation and/or other materials provided with the
33  *    distribution.
34  *  * Neither the name Intel Corporation nor the names of its
35  *    contributors may be used to endorse or promote products derived
36  *    from this software without specific prior written permission.
37  *
38  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
39  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
40  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
41  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
42  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
43  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
44  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
45  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
46  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
47  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
48  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
49  *
50  *****************************************************************************/
51 #include <linux/pm_runtime.h>
52 #include <net/tso.h>
53 
54 #include "iwl-debug.h"
55 #include "iwl-csr.h"
56 #include "iwl-io.h"
57 #include "internal.h"
58 #include "mvm/fw-api.h"
59 
60  /*
61  * iwl_pcie_gen2_tx_stop - Stop all Tx DMA channels
62  */
63 void iwl_pcie_gen2_tx_stop(struct iwl_trans *trans)
64 {
65 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
66 	int txq_id;
67 
68 	/*
69 	 * This function can be called before the op_mode disabled the
70 	 * queues. This happens when we have an rfkill interrupt.
71 	 * Since we stop Tx altogether - mark the queues as stopped.
72 	 */
73 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
74 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
75 
76 	/* Unmap DMA from host system and free skb's */
77 	for (txq_id = 0; txq_id < ARRAY_SIZE(trans_pcie->txq); txq_id++) {
78 		if (!trans_pcie->txq[txq_id])
79 			continue;
80 		iwl_pcie_gen2_txq_unmap(trans, txq_id);
81 	}
82 }
83 
84 /*
85  * iwl_pcie_txq_update_byte_tbl - Set up entry in Tx byte-count array
86  */
87 static void iwl_pcie_gen2_update_byte_tbl(struct iwl_txq *txq, u16 byte_cnt,
88 					  int num_tbs)
89 {
90 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr;
91 	int write_ptr = txq->write_ptr;
92 	u8 filled_tfd_size, num_fetch_chunks;
93 	u16 len = byte_cnt;
94 	__le16 bc_ent;
95 
96 	len = DIV_ROUND_UP(len, 4);
97 
98 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
99 		return;
100 
101 	filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) +
102 				   num_tbs * sizeof(struct iwl_tfh_tb);
103 	/*
104 	 * filled_tfd_size contains the number of filled bytes in the TFD.
105 	 * Dividing it by 64 will give the number of chunks to fetch
106 	 * to SRAM- 0 for one chunk, 1 for 2 and so on.
107 	 * If, for example, TFD contains only 3 TBs then 32 bytes
108 	 * of the TFD are used, and only one chunk of 64 bytes should
109 	 * be fetched
110 	 */
111 	num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1;
112 
113 	bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12));
114 	scd_bc_tbl->tfd_offset[write_ptr] = bc_ent;
115 }
116 
117 /*
118  * iwl_pcie_gen2_txq_inc_wr_ptr - Send new write index to hardware
119  */
120 static void iwl_pcie_gen2_txq_inc_wr_ptr(struct iwl_trans *trans,
121 					 struct iwl_txq *txq)
122 {
123 	lockdep_assert_held(&txq->lock);
124 
125 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr);
126 
127 	/*
128 	 * if not in power-save mode, uCode will never sleep when we're
129 	 * trying to tx (during RFKILL, we're not trying to tx).
130 	 */
131 	iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16));
132 }
133 
134 static u8 iwl_pcie_gen2_get_num_tbs(struct iwl_trans *trans,
135 				    struct iwl_tfh_tfd *tfd)
136 {
137 	return le16_to_cpu(tfd->num_tbs) & 0x1f;
138 }
139 
140 static void iwl_pcie_gen2_tfd_unmap(struct iwl_trans *trans,
141 				    struct iwl_cmd_meta *meta,
142 				    struct iwl_tfh_tfd *tfd)
143 {
144 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
145 	int i, num_tbs;
146 
147 	/* Sanity check on number of chunks */
148 	num_tbs = iwl_pcie_gen2_get_num_tbs(trans, tfd);
149 
150 	if (num_tbs >= trans_pcie->max_tbs) {
151 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
152 		return;
153 	}
154 
155 	/* first TB is never freed - it's the bidirectional DMA data */
156 	for (i = 1; i < num_tbs; i++) {
157 		if (meta->tbs & BIT(i))
158 			dma_unmap_page(trans->dev,
159 				       le64_to_cpu(tfd->tbs[i].addr),
160 				       le16_to_cpu(tfd->tbs[i].tb_len),
161 				       DMA_TO_DEVICE);
162 		else
163 			dma_unmap_single(trans->dev,
164 					 le64_to_cpu(tfd->tbs[i].addr),
165 					 le16_to_cpu(tfd->tbs[i].tb_len),
166 					 DMA_TO_DEVICE);
167 	}
168 
169 	tfd->num_tbs = 0;
170 }
171 
172 static void iwl_pcie_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
173 {
174 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
175 
176 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
177 	 * idx is bounded by n_window
178 	 */
179 	int rd_ptr = txq->read_ptr;
180 	int idx = get_cmd_index(txq, rd_ptr);
181 
182 	lockdep_assert_held(&txq->lock);
183 
184 	/* We have only q->n_window txq->entries, but we use
185 	 * TFD_QUEUE_SIZE_MAX tfds
186 	 */
187 	iwl_pcie_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
188 				iwl_pcie_get_tfd(trans_pcie, txq, rd_ptr));
189 
190 	/* free SKB */
191 	if (txq->entries) {
192 		struct sk_buff *skb;
193 
194 		skb = txq->entries[idx].skb;
195 
196 		/* Can be called from irqs-disabled context
197 		 * If skb is not NULL, it means that the whole queue is being
198 		 * freed and that the queue is not empty - free the skb
199 		 */
200 		if (skb) {
201 			iwl_op_mode_free_skb(trans->op_mode, skb);
202 			txq->entries[idx].skb = NULL;
203 		}
204 	}
205 }
206 
207 static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
208 				struct iwl_tfh_tfd *tfd, dma_addr_t addr,
209 				u16 len)
210 {
211 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
212 	int idx = iwl_pcie_gen2_get_num_tbs(trans, tfd);
213 	struct iwl_tfh_tb *tb = &tfd->tbs[idx];
214 
215 	/* Each TFD can point to a maximum max_tbs Tx buffers */
216 	if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) {
217 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
218 			trans_pcie->max_tbs);
219 		return -EINVAL;
220 	}
221 
222 	put_unaligned_le64(addr, &tb->addr);
223 	tb->tb_len = cpu_to_le16(len);
224 
225 	tfd->num_tbs = cpu_to_le16(idx + 1);
226 
227 	return idx;
228 }
229 
230 static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
231 				     struct sk_buff *skb,
232 				     struct iwl_tfh_tfd *tfd, int start_len,
233 				     u8 hdr_len, struct iwl_device_cmd *dev_cmd)
234 {
235 #ifdef CONFIG_INET
236 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
237 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
238 	struct ieee80211_hdr *hdr = (void *)skb->data;
239 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
240 	unsigned int mss = skb_shinfo(skb)->gso_size;
241 	u16 length, iv_len, amsdu_pad;
242 	u8 *start_hdr;
243 	struct iwl_tso_hdr_page *hdr_page;
244 	struct page **page_ptr;
245 	struct tso_t tso;
246 
247 	/* if the packet is protected, then it must be CCMP or GCMP */
248 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
249 		IEEE80211_CCMP_HDR_LEN : 0;
250 
251 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
252 			     &dev_cmd->hdr, start_len, 0);
253 
254 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
255 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
256 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
257 	amsdu_pad = 0;
258 
259 	/* total amount of header we may need for this A-MSDU */
260 	hdr_room = DIV_ROUND_UP(total_len, mss) *
261 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
262 
263 	/* Our device supports 9 segments at most, it will fit in 1 page */
264 	hdr_page = get_page_hdr(trans, hdr_room);
265 	if (!hdr_page)
266 		return -ENOMEM;
267 
268 	get_page(hdr_page->page);
269 	start_hdr = hdr_page->pos;
270 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
271 	*page_ptr = hdr_page->page;
272 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
273 	hdr_page->pos += iv_len;
274 
275 	/*
276 	 * Pull the ieee80211 header + IV to be able to use TSO core,
277 	 * we will restore it for the tx_status flow.
278 	 */
279 	skb_pull(skb, hdr_len + iv_len);
280 
281 	/*
282 	 * Remove the length of all the headers that we don't actually
283 	 * have in the MPDU by themselves, but that we duplicate into
284 	 * all the different MSDUs inside the A-MSDU.
285 	 */
286 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
287 
288 	tso_start(skb, &tso);
289 
290 	while (total_len) {
291 		/* this is the data left for this subframe */
292 		unsigned int data_left = min_t(unsigned int, mss, total_len);
293 		struct sk_buff *csum_skb = NULL;
294 		unsigned int tb_len;
295 		dma_addr_t tb_phys;
296 		struct tcphdr *tcph;
297 		u8 *iph, *subf_hdrs_start = hdr_page->pos;
298 
299 		total_len -= data_left;
300 
301 		memset(hdr_page->pos, 0, amsdu_pad);
302 		hdr_page->pos += amsdu_pad;
303 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
304 				  data_left)) & 0x3;
305 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
306 		hdr_page->pos += ETH_ALEN;
307 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
308 		hdr_page->pos += ETH_ALEN;
309 
310 		length = snap_ip_tcp_hdrlen + data_left;
311 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
312 		hdr_page->pos += sizeof(length);
313 
314 		/*
315 		 * This will copy the SNAP as well which will be considered
316 		 * as MAC header.
317 		 */
318 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
319 		iph = hdr_page->pos + 8;
320 		tcph = (void *)(iph + ip_hdrlen);
321 
322 		hdr_page->pos += snap_ip_tcp_hdrlen;
323 
324 		tb_len = hdr_page->pos - start_hdr;
325 		tb_phys = dma_map_single(trans->dev, start_hdr,
326 					 tb_len, DMA_TO_DEVICE);
327 		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
328 			dev_kfree_skb(csum_skb);
329 			goto out_err;
330 		}
331 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
332 		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr, tb_len);
333 		/* add this subframe's headers' length to the tx_cmd */
334 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
335 
336 		/* prepare the start_hdr for the next subframe */
337 		start_hdr = hdr_page->pos;
338 
339 		/* put the payload */
340 		while (data_left) {
341 			tb_len = min_t(unsigned int, tso.size, data_left);
342 			tb_phys = dma_map_single(trans->dev, tso.data,
343 						 tb_len, DMA_TO_DEVICE);
344 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
345 				dev_kfree_skb(csum_skb);
346 				goto out_err;
347 			}
348 			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
349 			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
350 						       tb_len);
351 
352 			data_left -= tb_len;
353 			tso_build_data(skb, &tso, tb_len);
354 		}
355 	}
356 
357 	/* re -add the WiFi header and IV */
358 	skb_push(skb, hdr_len + iv_len);
359 
360 	return 0;
361 
362 out_err:
363 #endif
364 	return -EINVAL;
365 }
366 
367 static
368 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
369 					    struct iwl_txq *txq,
370 					    struct iwl_device_cmd *dev_cmd,
371 					    struct sk_buff *skb,
372 					    struct iwl_cmd_meta *out_meta)
373 {
374 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
375 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
376 	struct iwl_tfh_tfd *tfd =
377 		iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr);
378 	dma_addr_t tb_phys;
379 	bool amsdu;
380 	int i, len, tb1_len, tb2_len, hdr_len;
381 	void *tb1_addr;
382 
383 	memset(tfd, 0, sizeof(*tfd));
384 
385 	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
386 		(*ieee80211_get_qos_ctl(hdr) &
387 		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
388 
389 	tb_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
390 	/* The first TB points to bi-directional DMA data */
391 	if (!amsdu)
392 		memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
393 		       IWL_FIRST_TB_SIZE);
394 
395 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
396 
397 	/* there must be data left over for TB1 or this code must be changed */
398 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
399 
400 	/*
401 	 * The second TB (tb1) points to the remainder of the TX command
402 	 * and the 802.11 header - dword aligned size
403 	 * (This calculation modifies the TX command, so do it before the
404 	 * setup of the first TB)
405 	 */
406 	len = sizeof(struct iwl_tx_cmd_gen2) + sizeof(struct iwl_cmd_header) +
407 	      ieee80211_hdrlen(hdr->frame_control) - IWL_FIRST_TB_SIZE;
408 
409 	/* do not align A-MSDU to dword as the subframe header aligns it */
410 	if (amsdu)
411 		tb1_len = len;
412 	else
413 		tb1_len = ALIGN(len, 4);
414 
415 	/* map the data for TB1 */
416 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
417 	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
418 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
419 		goto out_err;
420 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
421 
422 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
423 
424 	if (amsdu) {
425 		if (!iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
426 					       tb1_len + IWL_FIRST_TB_SIZE,
427 					       hdr_len, dev_cmd))
428 			goto out_err;
429 
430 		/*
431 		 * building the A-MSDU might have changed this data, so memcpy
432 		 * it now
433 		 */
434 		memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
435 		       IWL_FIRST_TB_SIZE);
436 		return tfd;
437 	}
438 
439 	/* set up TFD's third entry to point to remainder of skb's head */
440 	tb2_len = skb_headlen(skb) - hdr_len;
441 
442 	if (tb2_len > 0) {
443 		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
444 					 tb2_len, DMA_TO_DEVICE);
445 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
446 			goto out_err;
447 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
448 	}
449 
450 	/* set up the remaining entries to point to the data */
451 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
452 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
453 		int tb_idx;
454 
455 		if (!skb_frag_size(frag))
456 			continue;
457 
458 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
459 					   skb_frag_size(frag), DMA_TO_DEVICE);
460 
461 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
462 			goto out_err;
463 		tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
464 					      skb_frag_size(frag));
465 
466 		out_meta->tbs |= BIT(tb_idx);
467 	}
468 
469 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
470 			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
471 	trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
472 
473 	return tfd;
474 
475 out_err:
476 	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
477 	return NULL;
478 }
479 
480 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
481 			   struct iwl_device_cmd *dev_cmd, int txq_id)
482 {
483 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
484 	struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
485 	struct iwl_cmd_meta *out_meta;
486 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
487 	void *tfd;
488 
489 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
490 		      "TX on unused queue %d\n", txq_id))
491 		return -EINVAL;
492 
493 	if (skb_is_nonlinear(skb) &&
494 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
495 	    __skb_linearize(skb))
496 		return -ENOMEM;
497 
498 	spin_lock(&txq->lock);
499 
500 	/* Set up driver data for this TFD */
501 	txq->entries[txq->write_ptr].skb = skb;
502 	txq->entries[txq->write_ptr].cmd = dev_cmd;
503 
504 	dev_cmd->hdr.sequence =
505 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
506 			    INDEX_TO_SEQ(txq->write_ptr)));
507 
508 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
509 	out_meta = &txq->entries[txq->write_ptr].meta;
510 	out_meta->flags = 0;
511 
512 	tfd = iwl_pcie_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
513 	if (!tfd) {
514 		spin_unlock(&txq->lock);
515 		return -1;
516 	}
517 
518 	/* Set up entry for this TFD in Tx byte-count array */
519 	iwl_pcie_gen2_update_byte_tbl(txq, le16_to_cpu(tx_cmd->len),
520 				      iwl_pcie_gen2_get_num_tbs(trans, tfd));
521 
522 	/* start timer if queue currently empty */
523 	if (txq->read_ptr == txq->write_ptr) {
524 		if (txq->wd_timeout)
525 			mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
526 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
527 		iwl_trans_ref(trans);
528 	}
529 
530 	/* Tell device the write index *just past* this latest filled TFD */
531 	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
532 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
533 	if (iwl_queue_space(txq) < txq->high_mark)
534 		iwl_stop_queue(trans, txq);
535 
536 	/*
537 	 * At this point the frame is "transmitted" successfully
538 	 * and we will get a TX status notification eventually.
539 	 */
540 	spin_unlock(&txq->lock);
541 	return 0;
542 }
543 
544 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
545 
546 /*
547  * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command
548  * @priv: device private data point
549  * @cmd: a pointer to the ucode command structure
550  *
551  * The function returns < 0 values to indicate the operation
552  * failed. On success, it returns the index (>= 0) of command in the
553  * command queue.
554  */
555 static int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans,
556 				      struct iwl_host_cmd *cmd)
557 {
558 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
559 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
560 	struct iwl_device_cmd *out_cmd;
561 	struct iwl_cmd_meta *out_meta;
562 	unsigned long flags;
563 	void *dup_buf = NULL;
564 	dma_addr_t phys_addr;
565 	int idx, i, cmd_pos;
566 	u16 copy_size, cmd_size, tb0_size;
567 	bool had_nocopy = false;
568 	u8 group_id = iwl_cmd_groupid(cmd->id);
569 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
570 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
571 	struct iwl_tfh_tfd *tfd =
572 		iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr);
573 
574 	memset(tfd, 0, sizeof(*tfd));
575 
576 	copy_size = sizeof(struct iwl_cmd_header_wide);
577 	cmd_size = sizeof(struct iwl_cmd_header_wide);
578 
579 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
580 		cmddata[i] = cmd->data[i];
581 		cmdlen[i] = cmd->len[i];
582 
583 		if (!cmd->len[i])
584 			continue;
585 
586 		/* need at least IWL_FIRST_TB_SIZE copied */
587 		if (copy_size < IWL_FIRST_TB_SIZE) {
588 			int copy = IWL_FIRST_TB_SIZE - copy_size;
589 
590 			if (copy > cmdlen[i])
591 				copy = cmdlen[i];
592 			cmdlen[i] -= copy;
593 			cmddata[i] += copy;
594 			copy_size += copy;
595 		}
596 
597 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
598 			had_nocopy = true;
599 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
600 				idx = -EINVAL;
601 				goto free_dup_buf;
602 			}
603 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
604 			/*
605 			 * This is also a chunk that isn't copied
606 			 * to the static buffer so set had_nocopy.
607 			 */
608 			had_nocopy = true;
609 
610 			/* only allowed once */
611 			if (WARN_ON(dup_buf)) {
612 				idx = -EINVAL;
613 				goto free_dup_buf;
614 			}
615 
616 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
617 					  GFP_ATOMIC);
618 			if (!dup_buf)
619 				return -ENOMEM;
620 		} else {
621 			/* NOCOPY must not be followed by normal! */
622 			if (WARN_ON(had_nocopy)) {
623 				idx = -EINVAL;
624 				goto free_dup_buf;
625 			}
626 			copy_size += cmdlen[i];
627 		}
628 		cmd_size += cmd->len[i];
629 	}
630 
631 	/*
632 	 * If any of the command structures end up being larger than the
633 	 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into
634 	 * separate TFDs, then we will need to increase the size of the buffers
635 	 */
636 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
637 		 "Command %s (%#x) is too large (%d bytes)\n",
638 		 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) {
639 		idx = -EINVAL;
640 		goto free_dup_buf;
641 	}
642 
643 	spin_lock_bh(&txq->lock);
644 
645 	if (iwl_queue_space(txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
646 		spin_unlock_bh(&txq->lock);
647 
648 		IWL_ERR(trans, "No space in command queue\n");
649 		iwl_op_mode_cmd_queue_full(trans->op_mode);
650 		idx = -ENOSPC;
651 		goto free_dup_buf;
652 	}
653 
654 	idx = get_cmd_index(txq, txq->write_ptr);
655 	out_cmd = txq->entries[idx].cmd;
656 	out_meta = &txq->entries[idx].meta;
657 
658 	/* re-initialize to NULL */
659 	memset(out_meta, 0, sizeof(*out_meta));
660 	if (cmd->flags & CMD_WANT_SKB)
661 		out_meta->source = cmd;
662 
663 	/* set up the header */
664 	out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
665 	out_cmd->hdr_wide.group_id = group_id;
666 	out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
667 	out_cmd->hdr_wide.length =
668 		cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
669 	out_cmd->hdr_wide.reserved = 0;
670 	out_cmd->hdr_wide.sequence =
671 		cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
672 					 INDEX_TO_SEQ(txq->write_ptr));
673 
674 	cmd_pos = sizeof(struct iwl_cmd_header_wide);
675 	copy_size = sizeof(struct iwl_cmd_header_wide);
676 
677 	/* and copy the data that needs to be copied */
678 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
679 		int copy;
680 
681 		if (!cmd->len[i])
682 			continue;
683 
684 		/* copy everything if not nocopy/dup */
685 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
686 					   IWL_HCMD_DFL_DUP))) {
687 			copy = cmd->len[i];
688 
689 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
690 			cmd_pos += copy;
691 			copy_size += copy;
692 			continue;
693 		}
694 
695 		/*
696 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
697 		 * in total (for bi-directional DMA), but copy up to what
698 		 * we can fit into the payload for debug dump purposes.
699 		 */
700 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
701 
702 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
703 		cmd_pos += copy;
704 
705 		/* However, treat copy_size the proper way, we need it below */
706 		if (copy_size < IWL_FIRST_TB_SIZE) {
707 			copy = IWL_FIRST_TB_SIZE - copy_size;
708 
709 			if (copy > cmd->len[i])
710 				copy = cmd->len[i];
711 			copy_size += copy;
712 		}
713 	}
714 
715 	IWL_DEBUG_HC(trans,
716 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
717 		     iwl_get_cmd_string(trans, cmd->id), group_id,
718 		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
719 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
720 
721 	/* start the TFD with the minimum copy bytes */
722 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
723 	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
724 	iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx),
725 			     tb0_size);
726 
727 	/* map first command fragment, if any remains */
728 	if (copy_size > tb0_size) {
729 		phys_addr = dma_map_single(trans->dev,
730 					   ((u8 *)&out_cmd->hdr) + tb0_size,
731 					   copy_size - tb0_size,
732 					   DMA_TO_DEVICE);
733 		if (dma_mapping_error(trans->dev, phys_addr)) {
734 			idx = -ENOMEM;
735 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
736 			goto out;
737 		}
738 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr,
739 				     copy_size - tb0_size);
740 	}
741 
742 	/* map the remaining (adjusted) nocopy/dup fragments */
743 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
744 		const void *data = cmddata[i];
745 
746 		if (!cmdlen[i])
747 			continue;
748 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
749 					   IWL_HCMD_DFL_DUP)))
750 			continue;
751 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
752 			data = dup_buf;
753 		phys_addr = dma_map_single(trans->dev, (void *)data,
754 					   cmdlen[i], DMA_TO_DEVICE);
755 		if (dma_mapping_error(trans->dev, phys_addr)) {
756 			idx = -ENOMEM;
757 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
758 			goto out;
759 		}
760 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
761 	}
762 
763 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
764 	out_meta->flags = cmd->flags;
765 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
766 		kzfree(txq->entries[idx].free_buf);
767 	txq->entries[idx].free_buf = dup_buf;
768 
769 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
770 
771 	/* start timer if queue currently empty */
772 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
773 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
774 
775 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
776 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
777 	    !trans_pcie->ref_cmd_in_flight) {
778 		trans_pcie->ref_cmd_in_flight = true;
779 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
780 		iwl_trans_ref(trans);
781 	}
782 	/* Increment and update queue's write index */
783 	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
784 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
785 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
786 
787 out:
788 	spin_unlock_bh(&txq->lock);
789 free_dup_buf:
790 	if (idx < 0)
791 		kfree(dup_buf);
792 	return idx;
793 }
794 
795 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
796 
797 static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans,
798 					struct iwl_host_cmd *cmd)
799 {
800 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
801 	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
802 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
803 	int cmd_idx;
804 	int ret;
805 
806 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
807 
808 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
809 				  &trans->status),
810 		 "Command %s: a command is already active!\n", cmd_str))
811 		return -EIO;
812 
813 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
814 
815 	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
816 		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
817 				 pm_runtime_active(&trans_pcie->pci_dev->dev),
818 				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
819 		if (!ret) {
820 			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
821 			return -ETIMEDOUT;
822 		}
823 	}
824 
825 	cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
826 	if (cmd_idx < 0) {
827 		ret = cmd_idx;
828 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
829 		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
830 			cmd_str, ret);
831 		return ret;
832 	}
833 
834 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
835 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
836 					   &trans->status),
837 				 HOST_COMPLETE_TIMEOUT);
838 	if (!ret) {
839 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
840 			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
841 
842 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
843 			txq->read_ptr, txq->write_ptr);
844 
845 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
846 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
847 			       cmd_str);
848 		ret = -ETIMEDOUT;
849 
850 		iwl_force_nmi(trans);
851 		iwl_trans_fw_error(trans);
852 
853 		goto cancel;
854 	}
855 
856 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
857 		IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
858 		dump_stack();
859 		ret = -EIO;
860 		goto cancel;
861 	}
862 
863 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
864 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
865 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
866 		ret = -ERFKILL;
867 		goto cancel;
868 	}
869 
870 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
871 		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
872 		ret = -EIO;
873 		goto cancel;
874 	}
875 
876 	return 0;
877 
878 cancel:
879 	if (cmd->flags & CMD_WANT_SKB) {
880 		/*
881 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
882 		 * TX cmd queue. Otherwise in case the cmd comes
883 		 * in later, it will possibly set an invalid
884 		 * address (cmd->meta.source).
885 		 */
886 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
887 	}
888 
889 	if (cmd->resp_pkt) {
890 		iwl_free_resp(cmd);
891 		cmd->resp_pkt = NULL;
892 	}
893 
894 	return ret;
895 }
896 
897 int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans,
898 				  struct iwl_host_cmd *cmd)
899 {
900 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
901 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
902 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
903 				  cmd->id);
904 		return -ERFKILL;
905 	}
906 
907 	if (cmd->flags & CMD_ASYNC) {
908 		int ret;
909 
910 		/* An asynchronous command can not expect an SKB to be set. */
911 		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
912 			return -EINVAL;
913 
914 		ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
915 		if (ret < 0) {
916 			IWL_ERR(trans,
917 				"Error sending %s: enqueue_hcmd failed: %d\n",
918 				iwl_get_cmd_string(trans, cmd->id), ret);
919 			return ret;
920 		}
921 		return 0;
922 	}
923 
924 	return iwl_pcie_gen2_send_hcmd_sync(trans, cmd);
925 }
926 
927 /*
928  * iwl_pcie_gen2_txq_unmap -  Unmap any remaining DMA mappings and free skb's
929  */
930 void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id)
931 {
932 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
933 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
934 
935 	spin_lock_bh(&txq->lock);
936 	while (txq->write_ptr != txq->read_ptr) {
937 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
938 				   txq_id, txq->read_ptr);
939 
940 		iwl_pcie_gen2_free_tfd(trans, txq);
941 		txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr);
942 
943 		if (txq->read_ptr == txq->write_ptr) {
944 			unsigned long flags;
945 
946 			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
947 			if (txq_id != trans_pcie->cmd_queue) {
948 				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
949 					      txq->id);
950 				iwl_trans_unref(trans);
951 			} else if (trans_pcie->ref_cmd_in_flight) {
952 				trans_pcie->ref_cmd_in_flight = false;
953 				IWL_DEBUG_RPM(trans,
954 					      "clear ref_cmd_in_flight\n");
955 				iwl_trans_unref(trans);
956 			}
957 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
958 		}
959 	}
960 	spin_unlock_bh(&txq->lock);
961 
962 	/* just in case - this queue may have been stopped */
963 	iwl_wake_queue(trans, txq);
964 }
965 
966 static void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans,
967 					  struct iwl_txq *txq)
968 {
969 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
970 	struct device *dev = trans->dev;
971 
972 	/* De-alloc circular buffer of TFDs */
973 	if (txq->tfds) {
974 		dma_free_coherent(dev,
975 				  trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX,
976 				  txq->tfds, txq->dma_addr);
977 		dma_free_coherent(dev,
978 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
979 				  txq->first_tb_bufs, txq->first_tb_dma);
980 	}
981 
982 	kfree(txq->entries);
983 	iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl);
984 	kfree(txq);
985 }
986 
987 /*
988  * iwl_pcie_txq_free - Deallocate DMA queue.
989  * @txq: Transmit queue to deallocate.
990  *
991  * Empty queue by removing and destroying all BD's.
992  * Free all buffers.
993  * 0-fill, but do not free "txq" descriptor structure.
994  */
995 static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id)
996 {
997 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
998 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
999 	int i;
1000 
1001 	if (WARN_ON(!txq))
1002 		return;
1003 
1004 	iwl_pcie_gen2_txq_unmap(trans, txq_id);
1005 
1006 	/* De-alloc array of command/tx buffers */
1007 	if (txq_id == trans_pcie->cmd_queue)
1008 		for (i = 0; i < txq->n_window; i++) {
1009 			kzfree(txq->entries[i].cmd);
1010 			kzfree(txq->entries[i].free_buf);
1011 		}
1012 	del_timer_sync(&txq->stuck_timer);
1013 
1014 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1015 
1016 	trans_pcie->txq[txq_id] = NULL;
1017 
1018 	clear_bit(txq_id, trans_pcie->queue_used);
1019 }
1020 
1021 int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
1022 				 struct iwl_tx_queue_cfg_cmd *cmd,
1023 				 int cmd_id,
1024 				 unsigned int timeout)
1025 {
1026 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1027 	struct iwl_tx_queue_cfg_rsp *rsp;
1028 	struct iwl_txq *txq;
1029 	struct iwl_host_cmd hcmd = {
1030 		.id = cmd_id,
1031 		.len = { sizeof(*cmd) },
1032 		.data = { cmd, },
1033 		.flags = CMD_WANT_SKB,
1034 	};
1035 	int ret, qid;
1036 
1037 	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1038 	if (!txq)
1039 		return -ENOMEM;
1040 	ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl,
1041 				     sizeof(struct iwlagn_scd_bc_tbl));
1042 	if (ret) {
1043 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1044 		kfree(txq);
1045 		return -ENOMEM;
1046 	}
1047 
1048 	ret = iwl_pcie_txq_alloc(trans, txq, TFD_TX_CMD_SLOTS, false);
1049 	if (ret) {
1050 		IWL_ERR(trans, "Tx queue alloc failed\n");
1051 		goto error;
1052 	}
1053 	ret = iwl_pcie_txq_init(trans, txq, TFD_TX_CMD_SLOTS, false);
1054 	if (ret) {
1055 		IWL_ERR(trans, "Tx queue init failed\n");
1056 		goto error;
1057 	}
1058 
1059 	txq->wd_timeout = msecs_to_jiffies(timeout);
1060 
1061 	cmd->tfdq_addr = cpu_to_le64(txq->dma_addr);
1062 	cmd->byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1063 	cmd->cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(TFD_QUEUE_SIZE_MAX));
1064 
1065 	ret = iwl_trans_send_cmd(trans, &hcmd);
1066 	if (ret)
1067 		goto error;
1068 
1069 	if (WARN_ON(iwl_rx_packet_payload_len(hcmd.resp_pkt) != sizeof(*rsp))) {
1070 		ret = -EINVAL;
1071 		goto error_free_resp;
1072 	}
1073 
1074 	rsp = (void *)hcmd.resp_pkt->data;
1075 	qid = le16_to_cpu(rsp->queue_number);
1076 
1077 	if (qid >= ARRAY_SIZE(trans_pcie->txq)) {
1078 		WARN_ONCE(1, "queue index %d unsupported", qid);
1079 		ret = -EIO;
1080 		goto error_free_resp;
1081 	}
1082 
1083 	if (test_and_set_bit(qid, trans_pcie->queue_used)) {
1084 		WARN_ONCE(1, "queue %d already used", qid);
1085 		ret = -EIO;
1086 		goto error_free_resp;
1087 	}
1088 
1089 	txq->id = qid;
1090 	trans_pcie->txq[qid] = txq;
1091 
1092 	/* Place first TFD at index corresponding to start sequence number */
1093 	txq->read_ptr = le16_to_cpu(rsp->write_pointer);
1094 	txq->write_ptr = le16_to_cpu(rsp->write_pointer);
1095 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1096 			   (txq->write_ptr) | (qid << 16));
1097 	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1098 
1099 	iwl_free_resp(&hcmd);
1100 	return qid;
1101 
1102 error_free_resp:
1103 	iwl_free_resp(&hcmd);
1104 error:
1105 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1106 	return ret;
1107 }
1108 
1109 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue)
1110 {
1111 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1112 
1113 	/*
1114 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1115 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1116 	 * allow the op_mode to call txq_disable after it already called
1117 	 * stop_device.
1118 	 */
1119 	if (!test_and_clear_bit(queue, trans_pcie->queue_used)) {
1120 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1121 			  "queue %d not used", queue);
1122 		return;
1123 	}
1124 
1125 	iwl_pcie_gen2_txq_unmap(trans, queue);
1126 
1127 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1128 }
1129 
1130 void iwl_pcie_gen2_tx_free(struct iwl_trans *trans)
1131 {
1132 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1133 	int i;
1134 
1135 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
1136 
1137 	/* Free all TX queues */
1138 	for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) {
1139 		if (!trans_pcie->txq[i])
1140 			continue;
1141 
1142 		iwl_pcie_gen2_txq_free(trans, i);
1143 	}
1144 }
1145 
1146 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans)
1147 {
1148 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1149 	struct iwl_txq *cmd_queue;
1150 	int txq_id = trans_pcie->cmd_queue, ret;
1151 
1152 	/* alloc and init the command queue */
1153 	if (!trans_pcie->txq[txq_id]) {
1154 		cmd_queue = kzalloc(sizeof(*cmd_queue), GFP_KERNEL);
1155 		if (!cmd_queue) {
1156 			IWL_ERR(trans, "Not enough memory for command queue\n");
1157 			return -ENOMEM;
1158 		}
1159 		trans_pcie->txq[txq_id] = cmd_queue;
1160 		ret = iwl_pcie_txq_alloc(trans, cmd_queue, TFD_CMD_SLOTS, true);
1161 		if (ret) {
1162 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1163 			goto error;
1164 		}
1165 	} else {
1166 		cmd_queue = trans_pcie->txq[txq_id];
1167 	}
1168 
1169 	ret = iwl_pcie_txq_init(trans, cmd_queue, TFD_CMD_SLOTS, true);
1170 	if (ret) {
1171 		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1172 		goto error;
1173 	}
1174 	trans_pcie->txq[txq_id]->id = txq_id;
1175 	set_bit(txq_id, trans_pcie->queue_used);
1176 
1177 	return 0;
1178 
1179 error:
1180 	iwl_pcie_gen2_tx_free(trans);
1181 	return ret;
1182 }
1183 
1184