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        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        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 static 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 static 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 = &tfd->tbs[idx];
218 
219 	/* Each TFD can point to a maximum max_tbs Tx buffers */
220 	if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) {
221 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
222 			trans_pcie->max_tbs);
223 		return -EINVAL;
224 	}
225 
226 	put_unaligned_le64(addr, &tb->addr);
227 	tb->tb_len = cpu_to_le16(len);
228 
229 	tfd->num_tbs = cpu_to_le16(idx + 1);
230 
231 	return idx;
232 }
233 
234 static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
235 				     struct sk_buff *skb,
236 				     struct iwl_tfh_tfd *tfd, int start_len,
237 				     u8 hdr_len, struct iwl_device_cmd *dev_cmd)
238 {
239 #ifdef CONFIG_INET
240 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
241 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
242 	struct ieee80211_hdr *hdr = (void *)skb->data;
243 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
244 	unsigned int mss = skb_shinfo(skb)->gso_size;
245 	u16 length, iv_len, amsdu_pad;
246 	u8 *start_hdr;
247 	struct iwl_tso_hdr_page *hdr_page;
248 	struct page **page_ptr;
249 	struct tso_t tso;
250 
251 	/* if the packet is protected, then it must be CCMP or GCMP */
252 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
253 		IEEE80211_CCMP_HDR_LEN : 0;
254 
255 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
256 			     &dev_cmd->hdr, start_len, 0);
257 
258 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
259 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
260 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
261 	amsdu_pad = 0;
262 
263 	/* total amount of header we may need for this A-MSDU */
264 	hdr_room = DIV_ROUND_UP(total_len, mss) *
265 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
266 
267 	/* Our device supports 9 segments at most, it will fit in 1 page */
268 	hdr_page = get_page_hdr(trans, hdr_room);
269 	if (!hdr_page)
270 		return -ENOMEM;
271 
272 	get_page(hdr_page->page);
273 	start_hdr = hdr_page->pos;
274 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
275 	*page_ptr = hdr_page->page;
276 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
277 	hdr_page->pos += iv_len;
278 
279 	/*
280 	 * Pull the ieee80211 header + IV to be able to use TSO core,
281 	 * we will restore it for the tx_status flow.
282 	 */
283 	skb_pull(skb, hdr_len + iv_len);
284 
285 	/*
286 	 * Remove the length of all the headers that we don't actually
287 	 * have in the MPDU by themselves, but that we duplicate into
288 	 * all the different MSDUs inside the A-MSDU.
289 	 */
290 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
291 
292 	tso_start(skb, &tso);
293 
294 	while (total_len) {
295 		/* this is the data left for this subframe */
296 		unsigned int data_left = min_t(unsigned int, mss, total_len);
297 		struct sk_buff *csum_skb = NULL;
298 		unsigned int tb_len;
299 		dma_addr_t tb_phys;
300 		u8 *subf_hdrs_start = hdr_page->pos;
301 
302 		total_len -= data_left;
303 
304 		memset(hdr_page->pos, 0, amsdu_pad);
305 		hdr_page->pos += amsdu_pad;
306 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
307 				  data_left)) & 0x3;
308 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
309 		hdr_page->pos += ETH_ALEN;
310 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
311 		hdr_page->pos += ETH_ALEN;
312 
313 		length = snap_ip_tcp_hdrlen + data_left;
314 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
315 		hdr_page->pos += sizeof(length);
316 
317 		/*
318 		 * This will copy the SNAP as well which will be considered
319 		 * as MAC header.
320 		 */
321 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
322 
323 		hdr_page->pos += snap_ip_tcp_hdrlen;
324 
325 		tb_len = hdr_page->pos - start_hdr;
326 		tb_phys = dma_map_single(trans->dev, start_hdr,
327 					 tb_len, DMA_TO_DEVICE);
328 		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
329 			dev_kfree_skb(csum_skb);
330 			goto out_err;
331 		}
332 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
333 		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr, tb_len);
334 		/* add this subframe's headers' length to the tx_cmd */
335 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
336 
337 		/* prepare the start_hdr for the next subframe */
338 		start_hdr = hdr_page->pos;
339 
340 		/* put the payload */
341 		while (data_left) {
342 			tb_len = min_t(unsigned int, tso.size, data_left);
343 			tb_phys = dma_map_single(trans->dev, tso.data,
344 						 tb_len, DMA_TO_DEVICE);
345 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
346 				dev_kfree_skb(csum_skb);
347 				goto out_err;
348 			}
349 			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
350 			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
351 						       tb_len);
352 
353 			data_left -= tb_len;
354 			tso_build_data(skb, &tso, tb_len);
355 		}
356 	}
357 
358 	/* re -add the WiFi header and IV */
359 	skb_push(skb, hdr_len + iv_len);
360 
361 	return 0;
362 
363 out_err:
364 #endif
365 	return -EINVAL;
366 }
367 
368 static struct
369 iwl_tfh_tfd *iwl_pcie_gen2_build_tx_amsdu(struct iwl_trans *trans,
370 					  struct iwl_txq *txq,
371 					  struct iwl_device_cmd *dev_cmd,
372 					  struct sk_buff *skb,
373 					  struct iwl_cmd_meta *out_meta,
374 					  int hdr_len,
375 					  int tx_cmd_len)
376 {
377 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
378 	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
379 	dma_addr_t tb_phys;
380 	int len;
381 	void *tb1_addr;
382 
383 	tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
384 
385 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
386 
387 	/*
388 	 * The second TB (tb1) points to the remainder of the TX command
389 	 * and the 802.11 header - dword aligned size
390 	 * (This calculation modifies the TX command, so do it before the
391 	 * setup of the first TB)
392 	 */
393 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
394 	      IWL_FIRST_TB_SIZE;
395 
396 	/* do not align A-MSDU to dword as the subframe header aligns it */
397 
398 	/* map the data for TB1 */
399 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
400 	tb_phys = dma_map_single(trans->dev, tb1_addr, len, DMA_TO_DEVICE);
401 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
402 		goto out_err;
403 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, len);
404 
405 	if (iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
406 				      len + IWL_FIRST_TB_SIZE,
407 				      hdr_len, dev_cmd))
408 		goto out_err;
409 
410 	/* building the A-MSDU might have changed this data, memcpy it now */
411 	memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr, IWL_FIRST_TB_SIZE);
412 	return tfd;
413 
414 out_err:
415 	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
416 	return NULL;
417 }
418 
419 static struct
420 iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
421 				    struct iwl_txq *txq,
422 				    struct iwl_device_cmd *dev_cmd,
423 				    struct sk_buff *skb,
424 				    struct iwl_cmd_meta *out_meta,
425 				    int hdr_len,
426 				    int tx_cmd_len)
427 {
428 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
429 	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
430 	dma_addr_t tb_phys;
431 	int i, len, tb1_len, tb2_len;
432 	void *tb1_addr;
433 
434 	tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
435 
436 	/* The first TB points to bi-directional DMA data */
437 	memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr, IWL_FIRST_TB_SIZE);
438 
439 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
440 
441 	/*
442 	 * The second TB (tb1) points to the remainder of the TX command
443 	 * and the 802.11 header - dword aligned size
444 	 * (This calculation modifies the TX command, so do it before the
445 	 * setup of the first TB)
446 	 */
447 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
448 	      IWL_FIRST_TB_SIZE;
449 
450 	tb1_len = ALIGN(len, 4);
451 
452 	/* map the data for TB1 */
453 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
454 	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
455 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
456 		goto out_err;
457 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
458 
459 	/* set up TFD's third entry to point to remainder of skb's head */
460 	tb2_len = skb_headlen(skb) - hdr_len;
461 
462 	if (tb2_len > 0) {
463 		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
464 					 tb2_len, DMA_TO_DEVICE);
465 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
466 			goto out_err;
467 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
468 	}
469 
470 	/* set up the remaining entries to point to the data */
471 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
472 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
473 		int tb_idx;
474 
475 		if (!skb_frag_size(frag))
476 			continue;
477 
478 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
479 					   skb_frag_size(frag), DMA_TO_DEVICE);
480 
481 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
482 			goto out_err;
483 		tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
484 					      skb_frag_size(frag));
485 
486 		out_meta->tbs |= BIT(tb_idx);
487 	}
488 
489 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
490 			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
491 	trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
492 
493 	return tfd;
494 
495 out_err:
496 	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
497 	return NULL;
498 }
499 
500 static
501 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
502 					    struct iwl_txq *txq,
503 					    struct iwl_device_cmd *dev_cmd,
504 					    struct sk_buff *skb,
505 					    struct iwl_cmd_meta *out_meta)
506 {
507 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
508 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
509 	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
510 	int len, hdr_len;
511 	bool amsdu;
512 
513 	/* There must be data left over for TB1 or this code must be changed */
514 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
515 
516 	memset(tfd, 0, sizeof(*tfd));
517 
518 	if (trans->cfg->device_family < IWL_DEVICE_FAMILY_22560)
519 		len = sizeof(struct iwl_tx_cmd_gen2);
520 	else
521 		len = sizeof(struct iwl_tx_cmd_gen3);
522 
523 	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
524 			(*ieee80211_get_qos_ctl(hdr) &
525 			 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
526 
527 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
528 
529 	if (amsdu)
530 		return iwl_pcie_gen2_build_tx_amsdu(trans, txq, dev_cmd, skb,
531 						    out_meta, hdr_len, len);
532 
533 	return iwl_pcie_gen2_build_tx(trans, txq, dev_cmd, skb, out_meta,
534 				      hdr_len, len);
535 }
536 
537 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
538 			   struct iwl_device_cmd *dev_cmd, int txq_id)
539 {
540 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
541 	struct iwl_cmd_meta *out_meta;
542 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
543 	u16 cmd_len;
544 	int idx;
545 	void *tfd;
546 
547 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
548 		      "TX on unused queue %d\n", txq_id))
549 		return -EINVAL;
550 
551 	if (skb_is_nonlinear(skb) &&
552 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
553 	    __skb_linearize(skb))
554 		return -ENOMEM;
555 
556 	spin_lock(&txq->lock);
557 
558 	if (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560) {
559 		struct iwl_tx_cmd_gen3 *tx_cmd_gen3 =
560 			(void *)dev_cmd->payload;
561 
562 		cmd_len = le16_to_cpu(tx_cmd_gen3->len);
563 	} else {
564 		struct iwl_tx_cmd_gen2 *tx_cmd_gen2 =
565 			(void *)dev_cmd->payload;
566 
567 		cmd_len = le16_to_cpu(tx_cmd_gen2->len);
568 	}
569 
570 	if (iwl_queue_space(trans, txq) < txq->high_mark) {
571 		iwl_stop_queue(trans, txq);
572 
573 		/* don't put the packet on the ring, if there is no room */
574 		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
575 			struct iwl_device_cmd **dev_cmd_ptr;
576 
577 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
578 					       trans_pcie->dev_cmd_offs);
579 
580 			*dev_cmd_ptr = dev_cmd;
581 			__skb_queue_tail(&txq->overflow_q, skb);
582 			spin_unlock(&txq->lock);
583 			return 0;
584 		}
585 	}
586 
587 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
588 
589 	/* Set up driver data for this TFD */
590 	txq->entries[idx].skb = skb;
591 	txq->entries[idx].cmd = dev_cmd;
592 
593 	dev_cmd->hdr.sequence =
594 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
595 			    INDEX_TO_SEQ(idx)));
596 
597 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
598 	out_meta = &txq->entries[idx].meta;
599 	out_meta->flags = 0;
600 
601 	tfd = iwl_pcie_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
602 	if (!tfd) {
603 		spin_unlock(&txq->lock);
604 		return -1;
605 	}
606 
607 	/* Set up entry for this TFD in Tx byte-count array */
608 	iwl_pcie_gen2_update_byte_tbl(trans_pcie, txq, cmd_len,
609 				      iwl_pcie_gen2_get_num_tbs(trans, tfd));
610 
611 	/* start timer if queue currently empty */
612 	if (txq->read_ptr == txq->write_ptr) {
613 		if (txq->wd_timeout)
614 			mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
615 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
616 		iwl_trans_ref(trans);
617 	}
618 
619 	/* Tell device the write index *just past* this latest filled TFD */
620 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
621 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
622 	/*
623 	 * At this point the frame is "transmitted" successfully
624 	 * and we will get a TX status notification eventually.
625 	 */
626 	spin_unlock(&txq->lock);
627 	return 0;
628 }
629 
630 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
631 
632 /*
633  * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command
634  * @priv: device private data point
635  * @cmd: a pointer to the ucode command structure
636  *
637  * The function returns < 0 values to indicate the operation
638  * failed. On success, it returns the index (>= 0) of command in the
639  * command queue.
640  */
641 static int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans,
642 				      struct iwl_host_cmd *cmd)
643 {
644 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
645 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
646 	struct iwl_device_cmd *out_cmd;
647 	struct iwl_cmd_meta *out_meta;
648 	unsigned long flags;
649 	void *dup_buf = NULL;
650 	dma_addr_t phys_addr;
651 	int i, cmd_pos, idx;
652 	u16 copy_size, cmd_size, tb0_size;
653 	bool had_nocopy = false;
654 	u8 group_id = iwl_cmd_groupid(cmd->id);
655 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
656 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
657 	struct iwl_tfh_tfd *tfd;
658 
659 	copy_size = sizeof(struct iwl_cmd_header_wide);
660 	cmd_size = sizeof(struct iwl_cmd_header_wide);
661 
662 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
663 		cmddata[i] = cmd->data[i];
664 		cmdlen[i] = cmd->len[i];
665 
666 		if (!cmd->len[i])
667 			continue;
668 
669 		/* need at least IWL_FIRST_TB_SIZE copied */
670 		if (copy_size < IWL_FIRST_TB_SIZE) {
671 			int copy = IWL_FIRST_TB_SIZE - copy_size;
672 
673 			if (copy > cmdlen[i])
674 				copy = cmdlen[i];
675 			cmdlen[i] -= copy;
676 			cmddata[i] += copy;
677 			copy_size += copy;
678 		}
679 
680 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
681 			had_nocopy = true;
682 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
683 				idx = -EINVAL;
684 				goto free_dup_buf;
685 			}
686 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
687 			/*
688 			 * This is also a chunk that isn't copied
689 			 * to the static buffer so set had_nocopy.
690 			 */
691 			had_nocopy = true;
692 
693 			/* only allowed once */
694 			if (WARN_ON(dup_buf)) {
695 				idx = -EINVAL;
696 				goto free_dup_buf;
697 			}
698 
699 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
700 					  GFP_ATOMIC);
701 			if (!dup_buf)
702 				return -ENOMEM;
703 		} else {
704 			/* NOCOPY must not be followed by normal! */
705 			if (WARN_ON(had_nocopy)) {
706 				idx = -EINVAL;
707 				goto free_dup_buf;
708 			}
709 			copy_size += cmdlen[i];
710 		}
711 		cmd_size += cmd->len[i];
712 	}
713 
714 	/*
715 	 * If any of the command structures end up being larger than the
716 	 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into
717 	 * separate TFDs, then we will need to increase the size of the buffers
718 	 */
719 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
720 		 "Command %s (%#x) is too large (%d bytes)\n",
721 		 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) {
722 		idx = -EINVAL;
723 		goto free_dup_buf;
724 	}
725 
726 	spin_lock_bh(&txq->lock);
727 
728 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
729 	tfd = iwl_pcie_get_tfd(trans, txq, txq->write_ptr);
730 	memset(tfd, 0, sizeof(*tfd));
731 
732 	if (iwl_queue_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
733 		spin_unlock_bh(&txq->lock);
734 
735 		IWL_ERR(trans, "No space in command queue\n");
736 		iwl_op_mode_cmd_queue_full(trans->op_mode);
737 		idx = -ENOSPC;
738 		goto free_dup_buf;
739 	}
740 
741 	out_cmd = txq->entries[idx].cmd;
742 	out_meta = &txq->entries[idx].meta;
743 
744 	/* re-initialize to NULL */
745 	memset(out_meta, 0, sizeof(*out_meta));
746 	if (cmd->flags & CMD_WANT_SKB)
747 		out_meta->source = cmd;
748 
749 	/* set up the header */
750 	out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
751 	out_cmd->hdr_wide.group_id = group_id;
752 	out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
753 	out_cmd->hdr_wide.length =
754 		cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
755 	out_cmd->hdr_wide.reserved = 0;
756 	out_cmd->hdr_wide.sequence =
757 		cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
758 					 INDEX_TO_SEQ(txq->write_ptr));
759 
760 	cmd_pos = sizeof(struct iwl_cmd_header_wide);
761 	copy_size = sizeof(struct iwl_cmd_header_wide);
762 
763 	/* and copy the data that needs to be copied */
764 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
765 		int copy;
766 
767 		if (!cmd->len[i])
768 			continue;
769 
770 		/* copy everything if not nocopy/dup */
771 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
772 					   IWL_HCMD_DFL_DUP))) {
773 			copy = cmd->len[i];
774 
775 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
776 			cmd_pos += copy;
777 			copy_size += copy;
778 			continue;
779 		}
780 
781 		/*
782 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
783 		 * in total (for bi-directional DMA), but copy up to what
784 		 * we can fit into the payload for debug dump purposes.
785 		 */
786 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
787 
788 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
789 		cmd_pos += copy;
790 
791 		/* However, treat copy_size the proper way, we need it below */
792 		if (copy_size < IWL_FIRST_TB_SIZE) {
793 			copy = IWL_FIRST_TB_SIZE - copy_size;
794 
795 			if (copy > cmd->len[i])
796 				copy = cmd->len[i];
797 			copy_size += copy;
798 		}
799 	}
800 
801 	IWL_DEBUG_HC(trans,
802 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
803 		     iwl_get_cmd_string(trans, cmd->id), group_id,
804 		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
805 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
806 
807 	/* start the TFD with the minimum copy bytes */
808 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
809 	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
810 	iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx),
811 			     tb0_size);
812 
813 	/* map first command fragment, if any remains */
814 	if (copy_size > tb0_size) {
815 		phys_addr = dma_map_single(trans->dev,
816 					   ((u8 *)&out_cmd->hdr) + tb0_size,
817 					   copy_size - tb0_size,
818 					   DMA_TO_DEVICE);
819 		if (dma_mapping_error(trans->dev, phys_addr)) {
820 			idx = -ENOMEM;
821 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
822 			goto out;
823 		}
824 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr,
825 				     copy_size - tb0_size);
826 	}
827 
828 	/* map the remaining (adjusted) nocopy/dup fragments */
829 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
830 		const void *data = cmddata[i];
831 
832 		if (!cmdlen[i])
833 			continue;
834 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
835 					   IWL_HCMD_DFL_DUP)))
836 			continue;
837 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
838 			data = dup_buf;
839 		phys_addr = dma_map_single(trans->dev, (void *)data,
840 					   cmdlen[i], DMA_TO_DEVICE);
841 		if (dma_mapping_error(trans->dev, phys_addr)) {
842 			idx = -ENOMEM;
843 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
844 			goto out;
845 		}
846 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
847 	}
848 
849 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
850 	out_meta->flags = cmd->flags;
851 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
852 		kzfree(txq->entries[idx].free_buf);
853 	txq->entries[idx].free_buf = dup_buf;
854 
855 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
856 
857 	/* start timer if queue currently empty */
858 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
859 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
860 
861 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
862 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
863 	    !trans_pcie->ref_cmd_in_flight) {
864 		trans_pcie->ref_cmd_in_flight = true;
865 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
866 		iwl_trans_ref(trans);
867 	}
868 	/* Increment and update queue's write index */
869 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
870 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
871 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
872 
873 out:
874 	spin_unlock_bh(&txq->lock);
875 free_dup_buf:
876 	if (idx < 0)
877 		kfree(dup_buf);
878 	return idx;
879 }
880 
881 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
882 
883 static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans,
884 					struct iwl_host_cmd *cmd)
885 {
886 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
887 	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
888 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
889 	int cmd_idx;
890 	int ret;
891 
892 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
893 
894 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
895 				  &trans->status),
896 		 "Command %s: a command is already active!\n", cmd_str))
897 		return -EIO;
898 
899 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
900 
901 	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
902 		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
903 				 pm_runtime_active(&trans_pcie->pci_dev->dev),
904 				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
905 		if (!ret) {
906 			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
907 			return -ETIMEDOUT;
908 		}
909 	}
910 
911 	cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
912 	if (cmd_idx < 0) {
913 		ret = cmd_idx;
914 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
915 		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
916 			cmd_str, ret);
917 		return ret;
918 	}
919 
920 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
921 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
922 					   &trans->status),
923 				 HOST_COMPLETE_TIMEOUT);
924 	if (!ret) {
925 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
926 			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
927 
928 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
929 			txq->read_ptr, txq->write_ptr);
930 
931 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
932 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
933 			       cmd_str);
934 		ret = -ETIMEDOUT;
935 
936 		iwl_force_nmi(trans);
937 		iwl_trans_fw_error(trans);
938 
939 		goto cancel;
940 	}
941 
942 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
943 		IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
944 		dump_stack();
945 		ret = -EIO;
946 		goto cancel;
947 	}
948 
949 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
950 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
951 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
952 		ret = -ERFKILL;
953 		goto cancel;
954 	}
955 
956 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
957 		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
958 		ret = -EIO;
959 		goto cancel;
960 	}
961 
962 	return 0;
963 
964 cancel:
965 	if (cmd->flags & CMD_WANT_SKB) {
966 		/*
967 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
968 		 * TX cmd queue. Otherwise in case the cmd comes
969 		 * in later, it will possibly set an invalid
970 		 * address (cmd->meta.source).
971 		 */
972 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
973 	}
974 
975 	if (cmd->resp_pkt) {
976 		iwl_free_resp(cmd);
977 		cmd->resp_pkt = NULL;
978 	}
979 
980 	return ret;
981 }
982 
983 int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans,
984 				  struct iwl_host_cmd *cmd)
985 {
986 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
987 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
988 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
989 				  cmd->id);
990 		return -ERFKILL;
991 	}
992 
993 	if (cmd->flags & CMD_ASYNC) {
994 		int ret;
995 
996 		/* An asynchronous command can not expect an SKB to be set. */
997 		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
998 			return -EINVAL;
999 
1000 		ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
1001 		if (ret < 0) {
1002 			IWL_ERR(trans,
1003 				"Error sending %s: enqueue_hcmd failed: %d\n",
1004 				iwl_get_cmd_string(trans, cmd->id), ret);
1005 			return ret;
1006 		}
1007 		return 0;
1008 	}
1009 
1010 	return iwl_pcie_gen2_send_hcmd_sync(trans, cmd);
1011 }
1012 
1013 /*
1014  * iwl_pcie_gen2_txq_unmap -  Unmap any remaining DMA mappings and free skb's
1015  */
1016 void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id)
1017 {
1018 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1019 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1020 
1021 	spin_lock_bh(&txq->lock);
1022 	while (txq->write_ptr != txq->read_ptr) {
1023 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
1024 				   txq_id, txq->read_ptr);
1025 
1026 		if (txq_id != trans_pcie->cmd_queue) {
1027 			int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1028 			struct sk_buff *skb = txq->entries[idx].skb;
1029 
1030 			if (WARN_ON_ONCE(!skb))
1031 				continue;
1032 
1033 			iwl_pcie_free_tso_page(trans_pcie, skb);
1034 		}
1035 		iwl_pcie_gen2_free_tfd(trans, txq);
1036 		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
1037 
1038 		if (txq->read_ptr == txq->write_ptr) {
1039 			unsigned long flags;
1040 
1041 			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1042 			if (txq_id != trans_pcie->cmd_queue) {
1043 				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
1044 					      txq->id);
1045 				iwl_trans_unref(trans);
1046 			} else if (trans_pcie->ref_cmd_in_flight) {
1047 				trans_pcie->ref_cmd_in_flight = false;
1048 				IWL_DEBUG_RPM(trans,
1049 					      "clear ref_cmd_in_flight\n");
1050 				iwl_trans_unref(trans);
1051 			}
1052 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1053 		}
1054 	}
1055 
1056 	while (!skb_queue_empty(&txq->overflow_q)) {
1057 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
1058 
1059 		iwl_op_mode_free_skb(trans->op_mode, skb);
1060 	}
1061 
1062 	spin_unlock_bh(&txq->lock);
1063 
1064 	/* just in case - this queue may have been stopped */
1065 	iwl_wake_queue(trans, txq);
1066 }
1067 
1068 static void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans,
1069 					  struct iwl_txq *txq)
1070 {
1071 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1072 	struct device *dev = trans->dev;
1073 
1074 	/* De-alloc circular buffer of TFDs */
1075 	if (txq->tfds) {
1076 		dma_free_coherent(dev,
1077 				  trans_pcie->tfd_size * txq->n_window,
1078 				  txq->tfds, txq->dma_addr);
1079 		dma_free_coherent(dev,
1080 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
1081 				  txq->first_tb_bufs, txq->first_tb_dma);
1082 	}
1083 
1084 	kfree(txq->entries);
1085 	iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl);
1086 	kfree(txq);
1087 }
1088 
1089 /*
1090  * iwl_pcie_txq_free - Deallocate DMA queue.
1091  * @txq: Transmit queue to deallocate.
1092  *
1093  * Empty queue by removing and destroying all BD's.
1094  * Free all buffers.
1095  * 0-fill, but do not free "txq" descriptor structure.
1096  */
1097 static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id)
1098 {
1099 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1100 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1101 	int i;
1102 
1103 	if (WARN_ON(!txq))
1104 		return;
1105 
1106 	iwl_pcie_gen2_txq_unmap(trans, txq_id);
1107 
1108 	/* De-alloc array of command/tx buffers */
1109 	if (txq_id == trans_pcie->cmd_queue)
1110 		for (i = 0; i < txq->n_window; i++) {
1111 			kzfree(txq->entries[i].cmd);
1112 			kzfree(txq->entries[i].free_buf);
1113 		}
1114 	del_timer_sync(&txq->stuck_timer);
1115 
1116 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1117 
1118 	trans_pcie->txq[txq_id] = NULL;
1119 
1120 	clear_bit(txq_id, trans_pcie->queue_used);
1121 }
1122 
1123 int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
1124 				 struct iwl_tx_queue_cfg_cmd *cmd,
1125 				 int cmd_id, int size,
1126 				 unsigned int timeout)
1127 {
1128 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1129 	struct iwl_tx_queue_cfg_rsp *rsp;
1130 	struct iwl_txq *txq;
1131 	struct iwl_host_cmd hcmd = {
1132 		.id = cmd_id,
1133 		.len = { sizeof(*cmd) },
1134 		.data = { cmd, },
1135 		.flags = CMD_WANT_SKB,
1136 	};
1137 	int ret, qid;
1138 	u32 wr_ptr;
1139 
1140 	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1141 	if (!txq)
1142 		return -ENOMEM;
1143 	ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl,
1144 				     (trans->cfg->device_family >=
1145 				      IWL_DEVICE_FAMILY_22560) ?
1146 				     sizeof(struct iwl_gen3_bc_tbl) :
1147 				     sizeof(struct iwlagn_scd_bc_tbl));
1148 	if (ret) {
1149 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1150 		kfree(txq);
1151 		return -ENOMEM;
1152 	}
1153 
1154 	ret = iwl_pcie_txq_alloc(trans, txq, size, false);
1155 	if (ret) {
1156 		IWL_ERR(trans, "Tx queue alloc failed\n");
1157 		goto error;
1158 	}
1159 	ret = iwl_pcie_txq_init(trans, txq, size, false);
1160 	if (ret) {
1161 		IWL_ERR(trans, "Tx queue init failed\n");
1162 		goto error;
1163 	}
1164 
1165 	txq->wd_timeout = msecs_to_jiffies(timeout);
1166 
1167 	cmd->tfdq_addr = cpu_to_le64(txq->dma_addr);
1168 	cmd->byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1169 	cmd->cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size));
1170 
1171 	ret = iwl_trans_send_cmd(trans, &hcmd);
1172 	if (ret)
1173 		goto error;
1174 
1175 	if (WARN_ON(iwl_rx_packet_payload_len(hcmd.resp_pkt) != sizeof(*rsp))) {
1176 		ret = -EINVAL;
1177 		goto error_free_resp;
1178 	}
1179 
1180 	rsp = (void *)hcmd.resp_pkt->data;
1181 	qid = le16_to_cpu(rsp->queue_number);
1182 	wr_ptr = le16_to_cpu(rsp->write_pointer);
1183 
1184 	if (qid >= ARRAY_SIZE(trans_pcie->txq)) {
1185 		WARN_ONCE(1, "queue index %d unsupported", qid);
1186 		ret = -EIO;
1187 		goto error_free_resp;
1188 	}
1189 
1190 	if (test_and_set_bit(qid, trans_pcie->queue_used)) {
1191 		WARN_ONCE(1, "queue %d already used", qid);
1192 		ret = -EIO;
1193 		goto error_free_resp;
1194 	}
1195 
1196 	txq->id = qid;
1197 	trans_pcie->txq[qid] = txq;
1198 	wr_ptr &= (trans->cfg->base_params->max_tfd_queue_size - 1);
1199 
1200 	/* Place first TFD at index corresponding to start sequence number */
1201 	txq->read_ptr = wr_ptr;
1202 	txq->write_ptr = wr_ptr;
1203 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1204 			   (txq->write_ptr) | (qid << 16));
1205 	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1206 
1207 	iwl_free_resp(&hcmd);
1208 	return qid;
1209 
1210 error_free_resp:
1211 	iwl_free_resp(&hcmd);
1212 error:
1213 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1214 	return ret;
1215 }
1216 
1217 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue)
1218 {
1219 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1220 
1221 	/*
1222 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1223 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1224 	 * allow the op_mode to call txq_disable after it already called
1225 	 * stop_device.
1226 	 */
1227 	if (!test_and_clear_bit(queue, trans_pcie->queue_used)) {
1228 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1229 			  "queue %d not used", queue);
1230 		return;
1231 	}
1232 
1233 	iwl_pcie_gen2_txq_unmap(trans, queue);
1234 
1235 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1236 }
1237 
1238 void iwl_pcie_gen2_tx_free(struct iwl_trans *trans)
1239 {
1240 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1241 	int i;
1242 
1243 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
1244 
1245 	/* Free all TX queues */
1246 	for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) {
1247 		if (!trans_pcie->txq[i])
1248 			continue;
1249 
1250 		iwl_pcie_gen2_txq_free(trans, i);
1251 	}
1252 }
1253 
1254 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans)
1255 {
1256 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1257 	struct iwl_txq *cmd_queue;
1258 	int txq_id = trans_pcie->cmd_queue, ret;
1259 
1260 	/* alloc and init the command queue */
1261 	if (!trans_pcie->txq[txq_id]) {
1262 		cmd_queue = kzalloc(sizeof(*cmd_queue), GFP_KERNEL);
1263 		if (!cmd_queue) {
1264 			IWL_ERR(trans, "Not enough memory for command queue\n");
1265 			return -ENOMEM;
1266 		}
1267 		trans_pcie->txq[txq_id] = cmd_queue;
1268 		ret = iwl_pcie_txq_alloc(trans, cmd_queue, TFD_CMD_SLOTS, true);
1269 		if (ret) {
1270 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1271 			goto error;
1272 		}
1273 	} else {
1274 		cmd_queue = trans_pcie->txq[txq_id];
1275 	}
1276 
1277 	ret = iwl_pcie_txq_init(trans, cmd_queue, TFD_CMD_SLOTS, true);
1278 	if (ret) {
1279 		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1280 		goto error;
1281 	}
1282 	trans_pcie->txq[txq_id]->id = txq_id;
1283 	set_bit(txq_id, trans_pcie->queue_used);
1284 
1285 	return 0;
1286 
1287 error:
1288 	iwl_pcie_gen2_tx_free(trans);
1289 	return ret;
1290 }
1291 
1292