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  * Copyright(c) 2018 - 2019 Intel Corporation
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of version 2 of the GNU General Public License as
13  * published by the Free Software Foundation.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * BSD LICENSE
21  *
22  * Copyright(c) 2017 Intel Deutschland GmbH
23  * Copyright(c) 2018 - 2019 Intel Corporation
24  * All rights reserved.
25  *
26  * Redistribution and use in source and binary forms, with or without
27  * modification, are permitted provided that the following conditions
28  * are met:
29  *
30  *  * Redistributions of source code must retain the above copyright
31  *    notice, this list of conditions and the following disclaimer.
32  *  * Redistributions in binary form must reproduce the above copyright
33  *    notice, this list of conditions and the following disclaimer in
34  *    the documentation and/or other materials provided with the
35  *    distribution.
36  *  * Neither the name Intel Corporation nor the names of its
37  *    contributors may be used to endorse or promote products derived
38  *    from this software without specific prior written permission.
39  *
40  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
41  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
42  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
43  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
44  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
45  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
46  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
47  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
48  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
49  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
50  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
51  *
52  *****************************************************************************/
53 #include <linux/pm_runtime.h>
54 #include <net/tso.h>
55 #include <linux/tcp.h>
56 
57 #include "iwl-debug.h"
58 #include "iwl-csr.h"
59 #include "iwl-io.h"
60 #include "internal.h"
61 #include "fw/api/tx.h"
62 
63  /*
64  * iwl_pcie_gen2_tx_stop - Stop all Tx DMA channels
65  */
66 void iwl_pcie_gen2_tx_stop(struct iwl_trans *trans)
67 {
68 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
69 	int txq_id;
70 
71 	/*
72 	 * This function can be called before the op_mode disabled the
73 	 * queues. This happens when we have an rfkill interrupt.
74 	 * Since we stop Tx altogether - mark the queues as stopped.
75 	 */
76 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
77 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
78 
79 	/* Unmap DMA from host system and free skb's */
80 	for (txq_id = 0; txq_id < ARRAY_SIZE(trans_pcie->txq); txq_id++) {
81 		if (!trans_pcie->txq[txq_id])
82 			continue;
83 		iwl_pcie_gen2_txq_unmap(trans, txq_id);
84 	}
85 }
86 
87 /*
88  * iwl_pcie_txq_update_byte_tbl - Set up entry in Tx byte-count array
89  */
90 void iwl_pcie_gen2_update_byte_tbl(struct iwl_trans_pcie *trans_pcie,
91 				   struct iwl_txq *txq, u16 byte_cnt,
92 				   int num_tbs)
93 {
94 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr;
95 	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
96 	struct iwl_gen3_bc_tbl *scd_bc_tbl_gen3 = txq->bc_tbl.addr;
97 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
98 	u8 filled_tfd_size, num_fetch_chunks;
99 	u16 len = byte_cnt;
100 	__le16 bc_ent;
101 
102 	if (trans_pcie->bc_table_dword)
103 		len = DIV_ROUND_UP(len, 4);
104 
105 	if (WARN_ON(len > 0xFFF || idx >= txq->n_window))
106 		return;
107 
108 	filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) +
109 				   num_tbs * sizeof(struct iwl_tfh_tb);
110 	/*
111 	 * filled_tfd_size contains the number of filled bytes in the TFD.
112 	 * Dividing it by 64 will give the number of chunks to fetch
113 	 * to SRAM- 0 for one chunk, 1 for 2 and so on.
114 	 * If, for example, TFD contains only 3 TBs then 32 bytes
115 	 * of the TFD are used, and only one chunk of 64 bytes should
116 	 * be fetched
117 	 */
118 	num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1;
119 
120 	bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12));
121 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560)
122 		scd_bc_tbl_gen3->tfd_offset[idx] = bc_ent;
123 	else
124 		scd_bc_tbl->tfd_offset[idx] = bc_ent;
125 }
126 
127 /*
128  * iwl_pcie_gen2_txq_inc_wr_ptr - Send new write index to hardware
129  */
130 void iwl_pcie_gen2_txq_inc_wr_ptr(struct iwl_trans *trans,
131 				  struct iwl_txq *txq)
132 {
133 	lockdep_assert_held(&txq->lock);
134 
135 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr);
136 
137 	/*
138 	 * if not in power-save mode, uCode will never sleep when we're
139 	 * trying to tx (during RFKILL, we're not trying to tx).
140 	 */
141 	iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16));
142 }
143 
144 static u8 iwl_pcie_gen2_get_num_tbs(struct iwl_trans *trans,
145 				    struct iwl_tfh_tfd *tfd)
146 {
147 	return le16_to_cpu(tfd->num_tbs) & 0x1f;
148 }
149 
150 static void iwl_pcie_gen2_tfd_unmap(struct iwl_trans *trans,
151 				    struct iwl_cmd_meta *meta,
152 				    struct iwl_tfh_tfd *tfd)
153 {
154 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
155 	int i, num_tbs;
156 
157 	/* Sanity check on number of chunks */
158 	num_tbs = iwl_pcie_gen2_get_num_tbs(trans, tfd);
159 
160 	if (num_tbs > trans_pcie->max_tbs) {
161 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
162 		return;
163 	}
164 
165 	/* first TB is never freed - it's the bidirectional DMA data */
166 	for (i = 1; i < num_tbs; i++) {
167 		if (meta->tbs & BIT(i))
168 			dma_unmap_page(trans->dev,
169 				       le64_to_cpu(tfd->tbs[i].addr),
170 				       le16_to_cpu(tfd->tbs[i].tb_len),
171 				       DMA_TO_DEVICE);
172 		else
173 			dma_unmap_single(trans->dev,
174 					 le64_to_cpu(tfd->tbs[i].addr),
175 					 le16_to_cpu(tfd->tbs[i].tb_len),
176 					 DMA_TO_DEVICE);
177 	}
178 
179 	tfd->num_tbs = 0;
180 }
181 
182 static void iwl_pcie_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
183 {
184 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
185 	 * idx is bounded by n_window
186 	 */
187 	int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
188 
189 	lockdep_assert_held(&txq->lock);
190 
191 	iwl_pcie_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
192 				iwl_pcie_get_tfd(trans, txq, idx));
193 
194 	/* free SKB */
195 	if (txq->entries) {
196 		struct sk_buff *skb;
197 
198 		skb = txq->entries[idx].skb;
199 
200 		/* Can be called from irqs-disabled context
201 		 * If skb is not NULL, it means that the whole queue is being
202 		 * freed and that the queue is not empty - free the skb
203 		 */
204 		if (skb) {
205 			iwl_op_mode_free_skb(trans->op_mode, skb);
206 			txq->entries[idx].skb = NULL;
207 		}
208 	}
209 }
210 
211 static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
212 				struct iwl_tfh_tfd *tfd, dma_addr_t addr,
213 				u16 len)
214 {
215 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
216 	int idx = iwl_pcie_gen2_get_num_tbs(trans, tfd);
217 	struct iwl_tfh_tb *tb;
218 
219 	if (WARN_ON(idx >= IWL_TFH_NUM_TBS))
220 		return -EINVAL;
221 	tb = &tfd->tbs[idx];
222 
223 	/* Each TFD can point to a maximum max_tbs Tx buffers */
224 	if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) {
225 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
226 			trans_pcie->max_tbs);
227 		return -EINVAL;
228 	}
229 
230 	put_unaligned_le64(addr, &tb->addr);
231 	tb->tb_len = cpu_to_le16(len);
232 
233 	tfd->num_tbs = cpu_to_le16(idx + 1);
234 
235 	return idx;
236 }
237 
238 static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
239 				     struct sk_buff *skb,
240 				     struct iwl_tfh_tfd *tfd, int start_len,
241 				     u8 hdr_len, struct iwl_device_cmd *dev_cmd)
242 {
243 #ifdef CONFIG_INET
244 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
245 	struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
246 	struct ieee80211_hdr *hdr = (void *)skb->data;
247 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
248 	unsigned int mss = skb_shinfo(skb)->gso_size;
249 	u16 length, iv_len, amsdu_pad;
250 	u8 *start_hdr;
251 	struct iwl_tso_hdr_page *hdr_page;
252 	struct page **page_ptr;
253 	struct tso_t tso;
254 
255 	/* if the packet is protected, then it must be CCMP or GCMP */
256 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
257 		IEEE80211_CCMP_HDR_LEN : 0;
258 
259 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
260 			     &dev_cmd->hdr, start_len, 0);
261 
262 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
263 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
264 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
265 	amsdu_pad = 0;
266 
267 	/* total amount of header we may need for this A-MSDU */
268 	hdr_room = DIV_ROUND_UP(total_len, mss) *
269 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
270 
271 	/* Our device supports 9 segments at most, it will fit in 1 page */
272 	hdr_page = get_page_hdr(trans, hdr_room);
273 	if (!hdr_page)
274 		return -ENOMEM;
275 
276 	get_page(hdr_page->page);
277 	start_hdr = hdr_page->pos;
278 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
279 	*page_ptr = hdr_page->page;
280 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
281 	hdr_page->pos += iv_len;
282 
283 	/*
284 	 * Pull the ieee80211 header + IV to be able to use TSO core,
285 	 * we will restore it for the tx_status flow.
286 	 */
287 	skb_pull(skb, hdr_len + iv_len);
288 
289 	/*
290 	 * Remove the length of all the headers that we don't actually
291 	 * have in the MPDU by themselves, but that we duplicate into
292 	 * all the different MSDUs inside the A-MSDU.
293 	 */
294 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
295 
296 	tso_start(skb, &tso);
297 
298 	while (total_len) {
299 		/* this is the data left for this subframe */
300 		unsigned int data_left = min_t(unsigned int, mss, total_len);
301 		struct sk_buff *csum_skb = NULL;
302 		unsigned int tb_len;
303 		dma_addr_t tb_phys;
304 		u8 *subf_hdrs_start = hdr_page->pos;
305 
306 		total_len -= data_left;
307 
308 		memset(hdr_page->pos, 0, amsdu_pad);
309 		hdr_page->pos += amsdu_pad;
310 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
311 				  data_left)) & 0x3;
312 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
313 		hdr_page->pos += ETH_ALEN;
314 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
315 		hdr_page->pos += ETH_ALEN;
316 
317 		length = snap_ip_tcp_hdrlen + data_left;
318 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
319 		hdr_page->pos += sizeof(length);
320 
321 		/*
322 		 * This will copy the SNAP as well which will be considered
323 		 * as MAC header.
324 		 */
325 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
326 
327 		hdr_page->pos += snap_ip_tcp_hdrlen;
328 
329 		tb_len = hdr_page->pos - start_hdr;
330 		tb_phys = dma_map_single(trans->dev, start_hdr,
331 					 tb_len, DMA_TO_DEVICE);
332 		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
333 			dev_kfree_skb(csum_skb);
334 			goto out_err;
335 		}
336 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
337 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr, tb_len);
338 		/* add this subframe's headers' length to the tx_cmd */
339 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
340 
341 		/* prepare the start_hdr for the next subframe */
342 		start_hdr = hdr_page->pos;
343 
344 		/* put the payload */
345 		while (data_left) {
346 			tb_len = min_t(unsigned int, tso.size, data_left);
347 			tb_phys = dma_map_single(trans->dev, tso.data,
348 						 tb_len, DMA_TO_DEVICE);
349 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
350 				dev_kfree_skb(csum_skb);
351 				goto out_err;
352 			}
353 			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
354 			trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data,
355 						tb_len);
356 
357 			data_left -= tb_len;
358 			tso_build_data(skb, &tso, tb_len);
359 		}
360 	}
361 
362 	/* re -add the WiFi header and IV */
363 	skb_push(skb, hdr_len + iv_len);
364 
365 	return 0;
366 
367 out_err:
368 #endif
369 	return -EINVAL;
370 }
371 
372 static struct
373 iwl_tfh_tfd *iwl_pcie_gen2_build_tx_amsdu(struct iwl_trans *trans,
374 					  struct iwl_txq *txq,
375 					  struct iwl_device_cmd *dev_cmd,
376 					  struct sk_buff *skb,
377 					  struct iwl_cmd_meta *out_meta,
378 					  int hdr_len,
379 					  int tx_cmd_len)
380 {
381 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
382 	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
383 	dma_addr_t tb_phys;
384 	int len;
385 	void *tb1_addr;
386 
387 	tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
388 
389 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
390 
391 	/*
392 	 * The second TB (tb1) points to the remainder of the TX command
393 	 * and the 802.11 header - dword aligned size
394 	 * (This calculation modifies the TX command, so do it before the
395 	 * setup of the first TB)
396 	 */
397 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
398 	      IWL_FIRST_TB_SIZE;
399 
400 	/* do not align A-MSDU to dword as the subframe header aligns it */
401 
402 	/* map the data for TB1 */
403 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
404 	tb_phys = dma_map_single(trans->dev, tb1_addr, len, DMA_TO_DEVICE);
405 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
406 		goto out_err;
407 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, len);
408 
409 	if (iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
410 				      len + IWL_FIRST_TB_SIZE,
411 				      hdr_len, dev_cmd))
412 		goto out_err;
413 
414 	/* building the A-MSDU might have changed this data, memcpy it now */
415 	memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE);
416 	return tfd;
417 
418 out_err:
419 	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
420 	return NULL;
421 }
422 
423 static int iwl_pcie_gen2_tx_add_frags(struct iwl_trans *trans,
424 				      struct sk_buff *skb,
425 				      struct iwl_tfh_tfd *tfd,
426 				      struct iwl_cmd_meta *out_meta)
427 {
428 	int i;
429 
430 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
431 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
432 		dma_addr_t tb_phys;
433 		int tb_idx;
434 
435 		if (!skb_frag_size(frag))
436 			continue;
437 
438 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
439 					   skb_frag_size(frag), DMA_TO_DEVICE);
440 
441 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
442 			return -ENOMEM;
443 		tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
444 					      skb_frag_size(frag));
445 		trace_iwlwifi_dev_tx_tb(trans->dev, skb,
446 					skb_frag_address(frag),
447 					skb_frag_size(frag));
448 		if (tb_idx < 0)
449 			return tb_idx;
450 
451 		out_meta->tbs |= BIT(tb_idx);
452 	}
453 
454 	return 0;
455 }
456 
457 static struct
458 iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
459 				    struct iwl_txq *txq,
460 				    struct iwl_device_cmd *dev_cmd,
461 				    struct sk_buff *skb,
462 				    struct iwl_cmd_meta *out_meta,
463 				    int hdr_len,
464 				    int tx_cmd_len,
465 				    bool pad)
466 {
467 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
468 	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
469 	dma_addr_t tb_phys;
470 	int len, tb1_len, tb2_len;
471 	void *tb1_addr;
472 
473 	tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
474 
475 	/* The first TB points to bi-directional DMA data */
476 	memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE);
477 
478 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
479 
480 	/*
481 	 * The second TB (tb1) points to the remainder of the TX command
482 	 * and the 802.11 header - dword aligned size
483 	 * (This calculation modifies the TX command, so do it before the
484 	 * setup of the first TB)
485 	 */
486 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
487 	      IWL_FIRST_TB_SIZE;
488 
489 	if (pad)
490 		tb1_len = ALIGN(len, 4);
491 	else
492 		tb1_len = len;
493 
494 	/* map the data for TB1 */
495 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
496 	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
497 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
498 		goto out_err;
499 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
500 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
501 			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
502 
503 	/* set up TFD's third entry to point to remainder of skb's head */
504 	tb2_len = skb_headlen(skb) - hdr_len;
505 
506 	if (tb2_len > 0) {
507 		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
508 					 tb2_len, DMA_TO_DEVICE);
509 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
510 			goto out_err;
511 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
512 		trace_iwlwifi_dev_tx_tb(trans->dev, skb,
513 					skb->data + hdr_len,
514 					tb2_len);
515 	}
516 
517 	if (iwl_pcie_gen2_tx_add_frags(trans, skb, tfd, out_meta))
518 		goto out_err;
519 
520 	return tfd;
521 
522 out_err:
523 	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
524 	return NULL;
525 }
526 
527 static
528 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
529 					    struct iwl_txq *txq,
530 					    struct iwl_device_cmd *dev_cmd,
531 					    struct sk_buff *skb,
532 					    struct iwl_cmd_meta *out_meta)
533 {
534 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
535 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
536 	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
537 	int len, hdr_len;
538 	bool amsdu;
539 
540 	/* There must be data left over for TB1 or this code must be changed */
541 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
542 
543 	memset(tfd, 0, sizeof(*tfd));
544 
545 	if (trans->cfg->device_family < IWL_DEVICE_FAMILY_22560)
546 		len = sizeof(struct iwl_tx_cmd_gen2);
547 	else
548 		len = sizeof(struct iwl_tx_cmd_gen3);
549 
550 	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
551 			(*ieee80211_get_qos_ctl(hdr) &
552 			 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
553 
554 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
555 
556 	/*
557 	 * Only build A-MSDUs here if doing so by GSO, otherwise it may be
558 	 * an A-MSDU for other reasons, e.g. NAN or an A-MSDU having been
559 	 * built in the higher layers already.
560 	 */
561 	if (amsdu && skb_shinfo(skb)->gso_size)
562 		return iwl_pcie_gen2_build_tx_amsdu(trans, txq, dev_cmd, skb,
563 						    out_meta, hdr_len, len);
564 
565 	return iwl_pcie_gen2_build_tx(trans, txq, dev_cmd, skb, out_meta,
566 				      hdr_len, len, !amsdu);
567 }
568 
569 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
570 			   struct iwl_device_cmd *dev_cmd, int txq_id)
571 {
572 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
573 	struct iwl_cmd_meta *out_meta;
574 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
575 	u16 cmd_len;
576 	int idx;
577 	void *tfd;
578 
579 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
580 		      "TX on unused queue %d\n", txq_id))
581 		return -EINVAL;
582 
583 	if (skb_is_nonlinear(skb) &&
584 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
585 	    __skb_linearize(skb))
586 		return -ENOMEM;
587 
588 	spin_lock(&txq->lock);
589 
590 	if (iwl_queue_space(trans, txq) < txq->high_mark) {
591 		iwl_stop_queue(trans, txq);
592 
593 		/* don't put the packet on the ring, if there is no room */
594 		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
595 			struct iwl_device_cmd **dev_cmd_ptr;
596 
597 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
598 					       trans_pcie->dev_cmd_offs);
599 
600 			*dev_cmd_ptr = dev_cmd;
601 			__skb_queue_tail(&txq->overflow_q, skb);
602 			spin_unlock(&txq->lock);
603 			return 0;
604 		}
605 	}
606 
607 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
608 
609 	/* Set up driver data for this TFD */
610 	txq->entries[idx].skb = skb;
611 	txq->entries[idx].cmd = dev_cmd;
612 
613 	dev_cmd->hdr.sequence =
614 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
615 			    INDEX_TO_SEQ(idx)));
616 
617 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
618 	out_meta = &txq->entries[idx].meta;
619 	out_meta->flags = 0;
620 
621 	tfd = iwl_pcie_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
622 	if (!tfd) {
623 		spin_unlock(&txq->lock);
624 		return -1;
625 	}
626 
627 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560) {
628 		struct iwl_tx_cmd_gen3 *tx_cmd_gen3 =
629 			(void *)dev_cmd->payload;
630 
631 		cmd_len = le16_to_cpu(tx_cmd_gen3->len);
632 	} else {
633 		struct iwl_tx_cmd_gen2 *tx_cmd_gen2 =
634 			(void *)dev_cmd->payload;
635 
636 		cmd_len = le16_to_cpu(tx_cmd_gen2->len);
637 	}
638 
639 	/* Set up entry for this TFD in Tx byte-count array */
640 	iwl_pcie_gen2_update_byte_tbl(trans_pcie, txq, cmd_len,
641 				      iwl_pcie_gen2_get_num_tbs(trans, tfd));
642 
643 	/* start timer if queue currently empty */
644 	if (txq->read_ptr == txq->write_ptr) {
645 		if (txq->wd_timeout)
646 			mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
647 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
648 		iwl_trans_ref(trans);
649 	}
650 
651 	/* Tell device the write index *just past* this latest filled TFD */
652 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
653 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
654 	/*
655 	 * At this point the frame is "transmitted" successfully
656 	 * and we will get a TX status notification eventually.
657 	 */
658 	spin_unlock(&txq->lock);
659 	return 0;
660 }
661 
662 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
663 
664 /*
665  * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command
666  * @priv: device private data point
667  * @cmd: a pointer to the ucode command structure
668  *
669  * The function returns < 0 values to indicate the operation
670  * failed. On success, it returns the index (>= 0) of command in the
671  * command queue.
672  */
673 static int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans,
674 				      struct iwl_host_cmd *cmd)
675 {
676 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
677 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
678 	struct iwl_device_cmd *out_cmd;
679 	struct iwl_cmd_meta *out_meta;
680 	unsigned long flags;
681 	void *dup_buf = NULL;
682 	dma_addr_t phys_addr;
683 	int i, cmd_pos, idx;
684 	u16 copy_size, cmd_size, tb0_size;
685 	bool had_nocopy = false;
686 	u8 group_id = iwl_cmd_groupid(cmd->id);
687 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
688 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
689 	struct iwl_tfh_tfd *tfd;
690 
691 	copy_size = sizeof(struct iwl_cmd_header_wide);
692 	cmd_size = sizeof(struct iwl_cmd_header_wide);
693 
694 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
695 		cmddata[i] = cmd->data[i];
696 		cmdlen[i] = cmd->len[i];
697 
698 		if (!cmd->len[i])
699 			continue;
700 
701 		/* need at least IWL_FIRST_TB_SIZE copied */
702 		if (copy_size < IWL_FIRST_TB_SIZE) {
703 			int copy = IWL_FIRST_TB_SIZE - copy_size;
704 
705 			if (copy > cmdlen[i])
706 				copy = cmdlen[i];
707 			cmdlen[i] -= copy;
708 			cmddata[i] += copy;
709 			copy_size += copy;
710 		}
711 
712 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
713 			had_nocopy = true;
714 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
715 				idx = -EINVAL;
716 				goto free_dup_buf;
717 			}
718 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
719 			/*
720 			 * This is also a chunk that isn't copied
721 			 * to the static buffer so set had_nocopy.
722 			 */
723 			had_nocopy = true;
724 
725 			/* only allowed once */
726 			if (WARN_ON(dup_buf)) {
727 				idx = -EINVAL;
728 				goto free_dup_buf;
729 			}
730 
731 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
732 					  GFP_ATOMIC);
733 			if (!dup_buf)
734 				return -ENOMEM;
735 		} else {
736 			/* NOCOPY must not be followed by normal! */
737 			if (WARN_ON(had_nocopy)) {
738 				idx = -EINVAL;
739 				goto free_dup_buf;
740 			}
741 			copy_size += cmdlen[i];
742 		}
743 		cmd_size += cmd->len[i];
744 	}
745 
746 	/*
747 	 * If any of the command structures end up being larger than the
748 	 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into
749 	 * separate TFDs, then we will need to increase the size of the buffers
750 	 */
751 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
752 		 "Command %s (%#x) is too large (%d bytes)\n",
753 		 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) {
754 		idx = -EINVAL;
755 		goto free_dup_buf;
756 	}
757 
758 	spin_lock_bh(&txq->lock);
759 
760 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
761 	tfd = iwl_pcie_get_tfd(trans, txq, txq->write_ptr);
762 	memset(tfd, 0, sizeof(*tfd));
763 
764 	if (iwl_queue_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
765 		spin_unlock_bh(&txq->lock);
766 
767 		IWL_ERR(trans, "No space in command queue\n");
768 		iwl_op_mode_cmd_queue_full(trans->op_mode);
769 		idx = -ENOSPC;
770 		goto free_dup_buf;
771 	}
772 
773 	out_cmd = txq->entries[idx].cmd;
774 	out_meta = &txq->entries[idx].meta;
775 
776 	/* re-initialize to NULL */
777 	memset(out_meta, 0, sizeof(*out_meta));
778 	if (cmd->flags & CMD_WANT_SKB)
779 		out_meta->source = cmd;
780 
781 	/* set up the header */
782 	out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
783 	out_cmd->hdr_wide.group_id = group_id;
784 	out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
785 	out_cmd->hdr_wide.length =
786 		cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
787 	out_cmd->hdr_wide.reserved = 0;
788 	out_cmd->hdr_wide.sequence =
789 		cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
790 					 INDEX_TO_SEQ(txq->write_ptr));
791 
792 	cmd_pos = sizeof(struct iwl_cmd_header_wide);
793 	copy_size = sizeof(struct iwl_cmd_header_wide);
794 
795 	/* and copy the data that needs to be copied */
796 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
797 		int copy;
798 
799 		if (!cmd->len[i])
800 			continue;
801 
802 		/* copy everything if not nocopy/dup */
803 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
804 					   IWL_HCMD_DFL_DUP))) {
805 			copy = cmd->len[i];
806 
807 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
808 			cmd_pos += copy;
809 			copy_size += copy;
810 			continue;
811 		}
812 
813 		/*
814 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
815 		 * in total (for bi-directional DMA), but copy up to what
816 		 * we can fit into the payload for debug dump purposes.
817 		 */
818 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
819 
820 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
821 		cmd_pos += copy;
822 
823 		/* However, treat copy_size the proper way, we need it below */
824 		if (copy_size < IWL_FIRST_TB_SIZE) {
825 			copy = IWL_FIRST_TB_SIZE - copy_size;
826 
827 			if (copy > cmd->len[i])
828 				copy = cmd->len[i];
829 			copy_size += copy;
830 		}
831 	}
832 
833 	IWL_DEBUG_HC(trans,
834 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
835 		     iwl_get_cmd_string(trans, cmd->id), group_id,
836 		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
837 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
838 
839 	/* start the TFD with the minimum copy bytes */
840 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
841 	memcpy(&txq->first_tb_bufs[idx], out_cmd, tb0_size);
842 	iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx),
843 			     tb0_size);
844 
845 	/* map first command fragment, if any remains */
846 	if (copy_size > tb0_size) {
847 		phys_addr = dma_map_single(trans->dev,
848 					   (u8 *)out_cmd + tb0_size,
849 					   copy_size - tb0_size,
850 					   DMA_TO_DEVICE);
851 		if (dma_mapping_error(trans->dev, phys_addr)) {
852 			idx = -ENOMEM;
853 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
854 			goto out;
855 		}
856 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr,
857 				     copy_size - tb0_size);
858 	}
859 
860 	/* map the remaining (adjusted) nocopy/dup fragments */
861 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
862 		const void *data = cmddata[i];
863 
864 		if (!cmdlen[i])
865 			continue;
866 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
867 					   IWL_HCMD_DFL_DUP)))
868 			continue;
869 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
870 			data = dup_buf;
871 		phys_addr = dma_map_single(trans->dev, (void *)data,
872 					   cmdlen[i], DMA_TO_DEVICE);
873 		if (dma_mapping_error(trans->dev, phys_addr)) {
874 			idx = -ENOMEM;
875 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
876 			goto out;
877 		}
878 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
879 	}
880 
881 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
882 	out_meta->flags = cmd->flags;
883 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
884 		kzfree(txq->entries[idx].free_buf);
885 	txq->entries[idx].free_buf = dup_buf;
886 
887 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
888 
889 	/* start timer if queue currently empty */
890 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
891 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
892 
893 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
894 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
895 	    !trans_pcie->ref_cmd_in_flight) {
896 		trans_pcie->ref_cmd_in_flight = true;
897 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
898 		iwl_trans_ref(trans);
899 	}
900 	/* Increment and update queue's write index */
901 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
902 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
903 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
904 
905 out:
906 	spin_unlock_bh(&txq->lock);
907 free_dup_buf:
908 	if (idx < 0)
909 		kfree(dup_buf);
910 	return idx;
911 }
912 
913 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
914 
915 static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans,
916 					struct iwl_host_cmd *cmd)
917 {
918 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
919 	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
920 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
921 	int cmd_idx;
922 	int ret;
923 
924 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
925 
926 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
927 				  &trans->status),
928 		 "Command %s: a command is already active!\n", cmd_str))
929 		return -EIO;
930 
931 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
932 
933 	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
934 		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
935 				 pm_runtime_active(&trans_pcie->pci_dev->dev),
936 				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
937 		if (!ret) {
938 			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
939 			return -ETIMEDOUT;
940 		}
941 	}
942 
943 	cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
944 	if (cmd_idx < 0) {
945 		ret = cmd_idx;
946 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
947 		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
948 			cmd_str, ret);
949 		return ret;
950 	}
951 
952 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
953 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
954 					   &trans->status),
955 				 HOST_COMPLETE_TIMEOUT);
956 	if (!ret) {
957 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
958 			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
959 
960 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
961 			txq->read_ptr, txq->write_ptr);
962 
963 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
964 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
965 			       cmd_str);
966 		ret = -ETIMEDOUT;
967 
968 		iwl_trans_pcie_sync_nmi(trans);
969 		goto cancel;
970 	}
971 
972 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
973 		IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
974 		dump_stack();
975 		ret = -EIO;
976 		goto cancel;
977 	}
978 
979 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
980 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
981 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
982 		ret = -ERFKILL;
983 		goto cancel;
984 	}
985 
986 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
987 		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
988 		ret = -EIO;
989 		goto cancel;
990 	}
991 
992 	return 0;
993 
994 cancel:
995 	if (cmd->flags & CMD_WANT_SKB) {
996 		/*
997 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
998 		 * TX cmd queue. Otherwise in case the cmd comes
999 		 * in later, it will possibly set an invalid
1000 		 * address (cmd->meta.source).
1001 		 */
1002 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1003 	}
1004 
1005 	if (cmd->resp_pkt) {
1006 		iwl_free_resp(cmd);
1007 		cmd->resp_pkt = NULL;
1008 	}
1009 
1010 	return ret;
1011 }
1012 
1013 int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans,
1014 				  struct iwl_host_cmd *cmd)
1015 {
1016 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1017 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1018 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1019 				  cmd->id);
1020 		return -ERFKILL;
1021 	}
1022 
1023 	if (cmd->flags & CMD_ASYNC) {
1024 		int ret;
1025 
1026 		/* An asynchronous command can not expect an SKB to be set. */
1027 		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1028 			return -EINVAL;
1029 
1030 		ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
1031 		if (ret < 0) {
1032 			IWL_ERR(trans,
1033 				"Error sending %s: enqueue_hcmd failed: %d\n",
1034 				iwl_get_cmd_string(trans, cmd->id), ret);
1035 			return ret;
1036 		}
1037 		return 0;
1038 	}
1039 
1040 	return iwl_pcie_gen2_send_hcmd_sync(trans, cmd);
1041 }
1042 
1043 /*
1044  * iwl_pcie_gen2_txq_unmap -  Unmap any remaining DMA mappings and free skb's
1045  */
1046 void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id)
1047 {
1048 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1049 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1050 
1051 	spin_lock_bh(&txq->lock);
1052 	while (txq->write_ptr != txq->read_ptr) {
1053 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
1054 				   txq_id, txq->read_ptr);
1055 
1056 		if (txq_id != trans_pcie->cmd_queue) {
1057 			int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1058 			struct sk_buff *skb = txq->entries[idx].skb;
1059 
1060 			if (WARN_ON_ONCE(!skb))
1061 				continue;
1062 
1063 			iwl_pcie_free_tso_page(trans_pcie, skb);
1064 		}
1065 		iwl_pcie_gen2_free_tfd(trans, txq);
1066 		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
1067 
1068 		if (txq->read_ptr == txq->write_ptr) {
1069 			unsigned long flags;
1070 
1071 			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1072 			if (txq_id != trans_pcie->cmd_queue) {
1073 				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
1074 					      txq->id);
1075 				iwl_trans_unref(trans);
1076 			} else if (trans_pcie->ref_cmd_in_flight) {
1077 				trans_pcie->ref_cmd_in_flight = false;
1078 				IWL_DEBUG_RPM(trans,
1079 					      "clear ref_cmd_in_flight\n");
1080 				iwl_trans_unref(trans);
1081 			}
1082 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1083 		}
1084 	}
1085 
1086 	while (!skb_queue_empty(&txq->overflow_q)) {
1087 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
1088 
1089 		iwl_op_mode_free_skb(trans->op_mode, skb);
1090 	}
1091 
1092 	spin_unlock_bh(&txq->lock);
1093 
1094 	/* just in case - this queue may have been stopped */
1095 	iwl_wake_queue(trans, txq);
1096 }
1097 
1098 void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans,
1099 				   struct iwl_txq *txq)
1100 {
1101 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1102 	struct device *dev = trans->dev;
1103 
1104 	/* De-alloc circular buffer of TFDs */
1105 	if (txq->tfds) {
1106 		dma_free_coherent(dev,
1107 				  trans_pcie->tfd_size * txq->n_window,
1108 				  txq->tfds, txq->dma_addr);
1109 		dma_free_coherent(dev,
1110 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
1111 				  txq->first_tb_bufs, txq->first_tb_dma);
1112 	}
1113 
1114 	kfree(txq->entries);
1115 	iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl);
1116 	kfree(txq);
1117 }
1118 
1119 /*
1120  * iwl_pcie_txq_free - Deallocate DMA queue.
1121  * @txq: Transmit queue to deallocate.
1122  *
1123  * Empty queue by removing and destroying all BD's.
1124  * Free all buffers.
1125  * 0-fill, but do not free "txq" descriptor structure.
1126  */
1127 static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id)
1128 {
1129 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1130 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1131 	int i;
1132 
1133 	if (WARN_ON(!txq))
1134 		return;
1135 
1136 	iwl_pcie_gen2_txq_unmap(trans, txq_id);
1137 
1138 	/* De-alloc array of command/tx buffers */
1139 	if (txq_id == trans_pcie->cmd_queue)
1140 		for (i = 0; i < txq->n_window; i++) {
1141 			kzfree(txq->entries[i].cmd);
1142 			kzfree(txq->entries[i].free_buf);
1143 		}
1144 	del_timer_sync(&txq->stuck_timer);
1145 
1146 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1147 
1148 	trans_pcie->txq[txq_id] = NULL;
1149 
1150 	clear_bit(txq_id, trans_pcie->queue_used);
1151 }
1152 
1153 int iwl_trans_pcie_dyn_txq_alloc_dma(struct iwl_trans *trans,
1154 				     struct iwl_txq **intxq, int size,
1155 				     unsigned int timeout)
1156 {
1157 	int ret;
1158 
1159 	struct iwl_txq *txq;
1160 	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1161 	if (!txq)
1162 		return -ENOMEM;
1163 	ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl,
1164 				     (trans->cfg->device_family >=
1165 				      IWL_DEVICE_FAMILY_22560) ?
1166 				     sizeof(struct iwl_gen3_bc_tbl) :
1167 				     sizeof(struct iwlagn_scd_bc_tbl));
1168 	if (ret) {
1169 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1170 		kfree(txq);
1171 		return -ENOMEM;
1172 	}
1173 
1174 	ret = iwl_pcie_txq_alloc(trans, txq, size, false);
1175 	if (ret) {
1176 		IWL_ERR(trans, "Tx queue alloc failed\n");
1177 		goto error;
1178 	}
1179 	ret = iwl_pcie_txq_init(trans, txq, size, false);
1180 	if (ret) {
1181 		IWL_ERR(trans, "Tx queue init failed\n");
1182 		goto error;
1183 	}
1184 
1185 	txq->wd_timeout = msecs_to_jiffies(timeout);
1186 
1187 	*intxq = txq;
1188 	return 0;
1189 
1190 error:
1191 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1192 	return ret;
1193 }
1194 
1195 int iwl_trans_pcie_txq_alloc_response(struct iwl_trans *trans,
1196 				      struct iwl_txq *txq,
1197 				      struct iwl_host_cmd *hcmd)
1198 {
1199 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1200 	struct iwl_tx_queue_cfg_rsp *rsp;
1201 	int ret, qid;
1202 	u32 wr_ptr;
1203 
1204 	if (WARN_ON(iwl_rx_packet_payload_len(hcmd->resp_pkt) !=
1205 		    sizeof(*rsp))) {
1206 		ret = -EINVAL;
1207 		goto error_free_resp;
1208 	}
1209 
1210 	rsp = (void *)hcmd->resp_pkt->data;
1211 	qid = le16_to_cpu(rsp->queue_number);
1212 	wr_ptr = le16_to_cpu(rsp->write_pointer);
1213 
1214 	if (qid >= ARRAY_SIZE(trans_pcie->txq)) {
1215 		WARN_ONCE(1, "queue index %d unsupported", qid);
1216 		ret = -EIO;
1217 		goto error_free_resp;
1218 	}
1219 
1220 	if (test_and_set_bit(qid, trans_pcie->queue_used)) {
1221 		WARN_ONCE(1, "queue %d already used", qid);
1222 		ret = -EIO;
1223 		goto error_free_resp;
1224 	}
1225 
1226 	txq->id = qid;
1227 	trans_pcie->txq[qid] = txq;
1228 	wr_ptr &= (trans->cfg->base_params->max_tfd_queue_size - 1);
1229 
1230 	/* Place first TFD at index corresponding to start sequence number */
1231 	txq->read_ptr = wr_ptr;
1232 	txq->write_ptr = wr_ptr;
1233 
1234 	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1235 
1236 	iwl_free_resp(hcmd);
1237 	return qid;
1238 
1239 error_free_resp:
1240 	iwl_free_resp(hcmd);
1241 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1242 	return ret;
1243 }
1244 
1245 int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
1246 				 __le16 flags, u8 sta_id, u8 tid,
1247 				 int cmd_id, int size,
1248 				 unsigned int timeout)
1249 {
1250 	struct iwl_txq *txq = NULL;
1251 	struct iwl_tx_queue_cfg_cmd cmd = {
1252 		.flags = flags,
1253 		.sta_id = sta_id,
1254 		.tid = tid,
1255 	};
1256 	struct iwl_host_cmd hcmd = {
1257 		.id = cmd_id,
1258 		.len = { sizeof(cmd) },
1259 		.data = { &cmd, },
1260 		.flags = CMD_WANT_SKB,
1261 	};
1262 	int ret;
1263 
1264 	ret = iwl_trans_pcie_dyn_txq_alloc_dma(trans, &txq, size, timeout);
1265 	if (ret)
1266 		return ret;
1267 
1268 	cmd.tfdq_addr = cpu_to_le64(txq->dma_addr);
1269 	cmd.byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1270 	cmd.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size));
1271 
1272 	ret = iwl_trans_send_cmd(trans, &hcmd);
1273 	if (ret)
1274 		goto error;
1275 
1276 	return iwl_trans_pcie_txq_alloc_response(trans, txq, &hcmd);
1277 
1278 error:
1279 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1280 	return ret;
1281 }
1282 
1283 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue)
1284 {
1285 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1286 
1287 	/*
1288 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1289 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1290 	 * allow the op_mode to call txq_disable after it already called
1291 	 * stop_device.
1292 	 */
1293 	if (!test_and_clear_bit(queue, trans_pcie->queue_used)) {
1294 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1295 			  "queue %d not used", queue);
1296 		return;
1297 	}
1298 
1299 	iwl_pcie_gen2_txq_unmap(trans, queue);
1300 
1301 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1302 }
1303 
1304 void iwl_pcie_gen2_tx_free(struct iwl_trans *trans)
1305 {
1306 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1307 	int i;
1308 
1309 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
1310 
1311 	/* Free all TX queues */
1312 	for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) {
1313 		if (!trans_pcie->txq[i])
1314 			continue;
1315 
1316 		iwl_pcie_gen2_txq_free(trans, i);
1317 	}
1318 }
1319 
1320 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans, int txq_id, int queue_size)
1321 {
1322 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1323 	struct iwl_txq *queue;
1324 	int ret;
1325 
1326 	/* alloc and init the tx queue */
1327 	if (!trans_pcie->txq[txq_id]) {
1328 		queue = kzalloc(sizeof(*queue), GFP_KERNEL);
1329 		if (!queue) {
1330 			IWL_ERR(trans, "Not enough memory for tx queue\n");
1331 			return -ENOMEM;
1332 		}
1333 		trans_pcie->txq[txq_id] = queue;
1334 		ret = iwl_pcie_txq_alloc(trans, queue, queue_size, true);
1335 		if (ret) {
1336 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1337 			goto error;
1338 		}
1339 	} else {
1340 		queue = trans_pcie->txq[txq_id];
1341 	}
1342 
1343 	ret = iwl_pcie_txq_init(trans, queue, queue_size,
1344 				(txq_id == trans_pcie->cmd_queue));
1345 	if (ret) {
1346 		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1347 		goto error;
1348 	}
1349 	trans_pcie->txq[txq_id]->id = txq_id;
1350 	set_bit(txq_id, trans_pcie->queue_used);
1351 
1352 	return 0;
1353 
1354 error:
1355 	iwl_pcie_gen2_tx_free(trans);
1356 	return ret;
1357 }
1358 
1359