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 "fw/api/tx.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 idx = iwl_pcie_get_cmd_index(txq, 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 || idx >= txq->n_window))
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[idx] = 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 idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
180 
181 	lockdep_assert_held(&txq->lock);
182 
183 	iwl_pcie_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
184 				iwl_pcie_get_tfd(trans_pcie, txq, idx));
185 
186 	/* free SKB */
187 	if (txq->entries) {
188 		struct sk_buff *skb;
189 
190 		skb = txq->entries[idx].skb;
191 
192 		/* Can be called from irqs-disabled context
193 		 * If skb is not NULL, it means that the whole queue is being
194 		 * freed and that the queue is not empty - free the skb
195 		 */
196 		if (skb) {
197 			iwl_op_mode_free_skb(trans->op_mode, skb);
198 			txq->entries[idx].skb = NULL;
199 		}
200 	}
201 }
202 
203 static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
204 				struct iwl_tfh_tfd *tfd, dma_addr_t addr,
205 				u16 len)
206 {
207 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
208 	int idx = iwl_pcie_gen2_get_num_tbs(trans, tfd);
209 	struct iwl_tfh_tb *tb = &tfd->tbs[idx];
210 
211 	/* Each TFD can point to a maximum max_tbs Tx buffers */
212 	if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) {
213 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
214 			trans_pcie->max_tbs);
215 		return -EINVAL;
216 	}
217 
218 	put_unaligned_le64(addr, &tb->addr);
219 	tb->tb_len = cpu_to_le16(len);
220 
221 	tfd->num_tbs = cpu_to_le16(idx + 1);
222 
223 	return idx;
224 }
225 
226 static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
227 				     struct sk_buff *skb,
228 				     struct iwl_tfh_tfd *tfd, int start_len,
229 				     u8 hdr_len, struct iwl_device_cmd *dev_cmd)
230 {
231 #ifdef CONFIG_INET
232 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
233 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
234 	struct ieee80211_hdr *hdr = (void *)skb->data;
235 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
236 	unsigned int mss = skb_shinfo(skb)->gso_size;
237 	u16 length, iv_len, amsdu_pad;
238 	u8 *start_hdr;
239 	struct iwl_tso_hdr_page *hdr_page;
240 	struct page **page_ptr;
241 	struct tso_t tso;
242 
243 	/* if the packet is protected, then it must be CCMP or GCMP */
244 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
245 		IEEE80211_CCMP_HDR_LEN : 0;
246 
247 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
248 			     &dev_cmd->hdr, start_len, 0);
249 
250 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
251 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
252 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
253 	amsdu_pad = 0;
254 
255 	/* total amount of header we may need for this A-MSDU */
256 	hdr_room = DIV_ROUND_UP(total_len, mss) *
257 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
258 
259 	/* Our device supports 9 segments at most, it will fit in 1 page */
260 	hdr_page = get_page_hdr(trans, hdr_room);
261 	if (!hdr_page)
262 		return -ENOMEM;
263 
264 	get_page(hdr_page->page);
265 	start_hdr = hdr_page->pos;
266 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
267 	*page_ptr = hdr_page->page;
268 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
269 	hdr_page->pos += iv_len;
270 
271 	/*
272 	 * Pull the ieee80211 header + IV to be able to use TSO core,
273 	 * we will restore it for the tx_status flow.
274 	 */
275 	skb_pull(skb, hdr_len + iv_len);
276 
277 	/*
278 	 * Remove the length of all the headers that we don't actually
279 	 * have in the MPDU by themselves, but that we duplicate into
280 	 * all the different MSDUs inside the A-MSDU.
281 	 */
282 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
283 
284 	tso_start(skb, &tso);
285 
286 	while (total_len) {
287 		/* this is the data left for this subframe */
288 		unsigned int data_left = min_t(unsigned int, mss, total_len);
289 		struct sk_buff *csum_skb = NULL;
290 		unsigned int tb_len;
291 		dma_addr_t tb_phys;
292 		struct tcphdr *tcph;
293 		u8 *iph, *subf_hdrs_start = hdr_page->pos;
294 
295 		total_len -= data_left;
296 
297 		memset(hdr_page->pos, 0, amsdu_pad);
298 		hdr_page->pos += amsdu_pad;
299 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
300 				  data_left)) & 0x3;
301 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
302 		hdr_page->pos += ETH_ALEN;
303 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
304 		hdr_page->pos += ETH_ALEN;
305 
306 		length = snap_ip_tcp_hdrlen + data_left;
307 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
308 		hdr_page->pos += sizeof(length);
309 
310 		/*
311 		 * This will copy the SNAP as well which will be considered
312 		 * as MAC header.
313 		 */
314 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
315 		iph = hdr_page->pos + 8;
316 		tcph = (void *)(iph + ip_hdrlen);
317 
318 		hdr_page->pos += snap_ip_tcp_hdrlen;
319 
320 		tb_len = hdr_page->pos - start_hdr;
321 		tb_phys = dma_map_single(trans->dev, start_hdr,
322 					 tb_len, DMA_TO_DEVICE);
323 		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
324 			dev_kfree_skb(csum_skb);
325 			goto out_err;
326 		}
327 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
328 		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr, tb_len);
329 		/* add this subframe's headers' length to the tx_cmd */
330 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
331 
332 		/* prepare the start_hdr for the next subframe */
333 		start_hdr = hdr_page->pos;
334 
335 		/* put the payload */
336 		while (data_left) {
337 			tb_len = min_t(unsigned int, tso.size, data_left);
338 			tb_phys = dma_map_single(trans->dev, tso.data,
339 						 tb_len, DMA_TO_DEVICE);
340 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
341 				dev_kfree_skb(csum_skb);
342 				goto out_err;
343 			}
344 			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
345 			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
346 						       tb_len);
347 
348 			data_left -= tb_len;
349 			tso_build_data(skb, &tso, tb_len);
350 		}
351 	}
352 
353 	/* re -add the WiFi header and IV */
354 	skb_push(skb, hdr_len + iv_len);
355 
356 	return 0;
357 
358 out_err:
359 #endif
360 	return -EINVAL;
361 }
362 
363 static
364 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
365 					    struct iwl_txq *txq,
366 					    struct iwl_device_cmd *dev_cmd,
367 					    struct sk_buff *skb,
368 					    struct iwl_cmd_meta *out_meta)
369 {
370 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
371 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
372 	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
373 	struct iwl_tfh_tfd *tfd =
374 		iwl_pcie_get_tfd(trans_pcie, txq, idx);
375 	dma_addr_t tb_phys;
376 	bool amsdu;
377 	int i, len, tb1_len, tb2_len, hdr_len;
378 	void *tb1_addr;
379 
380 	memset(tfd, 0, sizeof(*tfd));
381 
382 	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
383 		(*ieee80211_get_qos_ctl(hdr) &
384 		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
385 
386 	tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
387 	/* The first TB points to bi-directional DMA data */
388 	if (!amsdu)
389 		memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr,
390 		       IWL_FIRST_TB_SIZE);
391 
392 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
393 
394 	/* there must be data left over for TB1 or this code must be changed */
395 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
396 
397 	/*
398 	 * The second TB (tb1) points to the remainder of the TX command
399 	 * and the 802.11 header - dword aligned size
400 	 * (This calculation modifies the TX command, so do it before the
401 	 * setup of the first TB)
402 	 */
403 	len = sizeof(struct iwl_tx_cmd_gen2) + sizeof(struct iwl_cmd_header) +
404 	      ieee80211_hdrlen(hdr->frame_control) - IWL_FIRST_TB_SIZE;
405 
406 	/* do not align A-MSDU to dword as the subframe header aligns it */
407 	if (amsdu)
408 		tb1_len = len;
409 	else
410 		tb1_len = ALIGN(len, 4);
411 
412 	/* map the data for TB1 */
413 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
414 	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
415 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
416 		goto out_err;
417 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
418 
419 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
420 
421 	if (amsdu) {
422 		if (iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
423 					      tb1_len + IWL_FIRST_TB_SIZE,
424 					      hdr_len, dev_cmd))
425 			goto out_err;
426 
427 		/*
428 		 * building the A-MSDU might have changed this data, so memcpy
429 		 * it now
430 		 */
431 		memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr,
432 		       IWL_FIRST_TB_SIZE);
433 		return tfd;
434 	}
435 
436 	/* set up TFD's third entry to point to remainder of skb's head */
437 	tb2_len = skb_headlen(skb) - hdr_len;
438 
439 	if (tb2_len > 0) {
440 		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
441 					 tb2_len, DMA_TO_DEVICE);
442 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
443 			goto out_err;
444 		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
445 	}
446 
447 	/* set up the remaining entries to point to the data */
448 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
449 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
450 		int tb_idx;
451 
452 		if (!skb_frag_size(frag))
453 			continue;
454 
455 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
456 					   skb_frag_size(frag), DMA_TO_DEVICE);
457 
458 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
459 			goto out_err;
460 		tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
461 					      skb_frag_size(frag));
462 
463 		out_meta->tbs |= BIT(tb_idx);
464 	}
465 
466 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
467 			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
468 	trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
469 
470 	return tfd;
471 
472 out_err:
473 	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
474 	return NULL;
475 }
476 
477 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
478 			   struct iwl_device_cmd *dev_cmd, int txq_id)
479 {
480 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
481 	struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
482 	struct iwl_cmd_meta *out_meta;
483 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
484 	int idx;
485 	void *tfd;
486 
487 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
488 		      "TX on unused queue %d\n", txq_id))
489 		return -EINVAL;
490 
491 	if (skb_is_nonlinear(skb) &&
492 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
493 	    __skb_linearize(skb))
494 		return -ENOMEM;
495 
496 	spin_lock(&txq->lock);
497 
498 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
499 
500 	/* Set up driver data for this TFD */
501 	txq->entries[idx].skb = skb;
502 	txq->entries[idx].cmd = dev_cmd;
503 
504 	dev_cmd->hdr.sequence =
505 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
506 			    INDEX_TO_SEQ(idx)));
507 
508 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
509 	out_meta = &txq->entries[idx].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 i, cmd_pos, idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
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 	out_cmd = txq->entries[idx].cmd;
655 	out_meta = &txq->entries[idx].meta;
656 
657 	/* re-initialize to NULL */
658 	memset(out_meta, 0, sizeof(*out_meta));
659 	if (cmd->flags & CMD_WANT_SKB)
660 		out_meta->source = cmd;
661 
662 	/* set up the header */
663 	out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
664 	out_cmd->hdr_wide.group_id = group_id;
665 	out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
666 	out_cmd->hdr_wide.length =
667 		cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
668 	out_cmd->hdr_wide.reserved = 0;
669 	out_cmd->hdr_wide.sequence =
670 		cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
671 					 INDEX_TO_SEQ(txq->write_ptr));
672 
673 	cmd_pos = sizeof(struct iwl_cmd_header_wide);
674 	copy_size = sizeof(struct iwl_cmd_header_wide);
675 
676 	/* and copy the data that needs to be copied */
677 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
678 		int copy;
679 
680 		if (!cmd->len[i])
681 			continue;
682 
683 		/* copy everything if not nocopy/dup */
684 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
685 					   IWL_HCMD_DFL_DUP))) {
686 			copy = cmd->len[i];
687 
688 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
689 			cmd_pos += copy;
690 			copy_size += copy;
691 			continue;
692 		}
693 
694 		/*
695 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
696 		 * in total (for bi-directional DMA), but copy up to what
697 		 * we can fit into the payload for debug dump purposes.
698 		 */
699 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
700 
701 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
702 		cmd_pos += copy;
703 
704 		/* However, treat copy_size the proper way, we need it below */
705 		if (copy_size < IWL_FIRST_TB_SIZE) {
706 			copy = IWL_FIRST_TB_SIZE - copy_size;
707 
708 			if (copy > cmd->len[i])
709 				copy = cmd->len[i];
710 			copy_size += copy;
711 		}
712 	}
713 
714 	IWL_DEBUG_HC(trans,
715 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
716 		     iwl_get_cmd_string(trans, cmd->id), group_id,
717 		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
718 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
719 
720 	/* start the TFD with the minimum copy bytes */
721 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
722 	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
723 	iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx),
724 			     tb0_size);
725 
726 	/* map first command fragment, if any remains */
727 	if (copy_size > tb0_size) {
728 		phys_addr = dma_map_single(trans->dev,
729 					   ((u8 *)&out_cmd->hdr) + tb0_size,
730 					   copy_size - tb0_size,
731 					   DMA_TO_DEVICE);
732 		if (dma_mapping_error(trans->dev, phys_addr)) {
733 			idx = -ENOMEM;
734 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
735 			goto out;
736 		}
737 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr,
738 				     copy_size - tb0_size);
739 	}
740 
741 	/* map the remaining (adjusted) nocopy/dup fragments */
742 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
743 		const void *data = cmddata[i];
744 
745 		if (!cmdlen[i])
746 			continue;
747 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
748 					   IWL_HCMD_DFL_DUP)))
749 			continue;
750 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
751 			data = dup_buf;
752 		phys_addr = dma_map_single(trans->dev, (void *)data,
753 					   cmdlen[i], DMA_TO_DEVICE);
754 		if (dma_mapping_error(trans->dev, phys_addr)) {
755 			idx = -ENOMEM;
756 			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
757 			goto out;
758 		}
759 		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
760 	}
761 
762 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
763 	out_meta->flags = cmd->flags;
764 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
765 		kzfree(txq->entries[idx].free_buf);
766 	txq->entries[idx].free_buf = dup_buf;
767 
768 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
769 
770 	/* start timer if queue currently empty */
771 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
772 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
773 
774 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
775 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
776 	    !trans_pcie->ref_cmd_in_flight) {
777 		trans_pcie->ref_cmd_in_flight = true;
778 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
779 		iwl_trans_ref(trans);
780 	}
781 	/* Increment and update queue's write index */
782 	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
783 	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
784 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
785 
786 out:
787 	spin_unlock_bh(&txq->lock);
788 free_dup_buf:
789 	if (idx < 0)
790 		kfree(dup_buf);
791 	return idx;
792 }
793 
794 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
795 
796 static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans,
797 					struct iwl_host_cmd *cmd)
798 {
799 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
800 	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
801 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
802 	int cmd_idx;
803 	int ret;
804 
805 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
806 
807 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
808 				  &trans->status),
809 		 "Command %s: a command is already active!\n", cmd_str))
810 		return -EIO;
811 
812 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
813 
814 	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
815 		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
816 				 pm_runtime_active(&trans_pcie->pci_dev->dev),
817 				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
818 		if (!ret) {
819 			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
820 			return -ETIMEDOUT;
821 		}
822 	}
823 
824 	cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
825 	if (cmd_idx < 0) {
826 		ret = cmd_idx;
827 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
828 		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
829 			cmd_str, ret);
830 		return ret;
831 	}
832 
833 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
834 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
835 					   &trans->status),
836 				 HOST_COMPLETE_TIMEOUT);
837 	if (!ret) {
838 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
839 			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
840 
841 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
842 			txq->read_ptr, txq->write_ptr);
843 
844 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
845 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
846 			       cmd_str);
847 		ret = -ETIMEDOUT;
848 
849 		iwl_force_nmi(trans);
850 		iwl_trans_fw_error(trans);
851 
852 		goto cancel;
853 	}
854 
855 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
856 		IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
857 		dump_stack();
858 		ret = -EIO;
859 		goto cancel;
860 	}
861 
862 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
863 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
864 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
865 		ret = -ERFKILL;
866 		goto cancel;
867 	}
868 
869 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
870 		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
871 		ret = -EIO;
872 		goto cancel;
873 	}
874 
875 	return 0;
876 
877 cancel:
878 	if (cmd->flags & CMD_WANT_SKB) {
879 		/*
880 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
881 		 * TX cmd queue. Otherwise in case the cmd comes
882 		 * in later, it will possibly set an invalid
883 		 * address (cmd->meta.source).
884 		 */
885 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
886 	}
887 
888 	if (cmd->resp_pkt) {
889 		iwl_free_resp(cmd);
890 		cmd->resp_pkt = NULL;
891 	}
892 
893 	return ret;
894 }
895 
896 int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans,
897 				  struct iwl_host_cmd *cmd)
898 {
899 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
900 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
901 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
902 				  cmd->id);
903 		return -ERFKILL;
904 	}
905 
906 	if (cmd->flags & CMD_ASYNC) {
907 		int ret;
908 
909 		/* An asynchronous command can not expect an SKB to be set. */
910 		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
911 			return -EINVAL;
912 
913 		ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
914 		if (ret < 0) {
915 			IWL_ERR(trans,
916 				"Error sending %s: enqueue_hcmd failed: %d\n",
917 				iwl_get_cmd_string(trans, cmd->id), ret);
918 			return ret;
919 		}
920 		return 0;
921 	}
922 
923 	return iwl_pcie_gen2_send_hcmd_sync(trans, cmd);
924 }
925 
926 /*
927  * iwl_pcie_gen2_txq_unmap -  Unmap any remaining DMA mappings and free skb's
928  */
929 void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id)
930 {
931 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
932 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
933 
934 	spin_lock_bh(&txq->lock);
935 	while (txq->write_ptr != txq->read_ptr) {
936 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
937 				   txq_id, txq->read_ptr);
938 
939 		if (txq_id != trans_pcie->cmd_queue) {
940 			int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
941 			struct sk_buff *skb = txq->entries[idx].skb;
942 
943 			if (WARN_ON_ONCE(!skb))
944 				continue;
945 
946 			iwl_pcie_free_tso_page(trans_pcie, skb);
947 		}
948 		iwl_pcie_gen2_free_tfd(trans, txq);
949 		txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr);
950 
951 		if (txq->read_ptr == txq->write_ptr) {
952 			unsigned long flags;
953 
954 			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
955 			if (txq_id != trans_pcie->cmd_queue) {
956 				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
957 					      txq->id);
958 				iwl_trans_unref(trans);
959 			} else if (trans_pcie->ref_cmd_in_flight) {
960 				trans_pcie->ref_cmd_in_flight = false;
961 				IWL_DEBUG_RPM(trans,
962 					      "clear ref_cmd_in_flight\n");
963 				iwl_trans_unref(trans);
964 			}
965 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
966 		}
967 	}
968 	spin_unlock_bh(&txq->lock);
969 
970 	/* just in case - this queue may have been stopped */
971 	iwl_wake_queue(trans, txq);
972 }
973 
974 static void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans,
975 					  struct iwl_txq *txq)
976 {
977 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
978 	struct device *dev = trans->dev;
979 
980 	/* De-alloc circular buffer of TFDs */
981 	if (txq->tfds) {
982 		dma_free_coherent(dev,
983 				  trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX,
984 				  txq->tfds, txq->dma_addr);
985 		dma_free_coherent(dev,
986 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
987 				  txq->first_tb_bufs, txq->first_tb_dma);
988 	}
989 
990 	kfree(txq->entries);
991 	iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl);
992 	kfree(txq);
993 }
994 
995 /*
996  * iwl_pcie_txq_free - Deallocate DMA queue.
997  * @txq: Transmit queue to deallocate.
998  *
999  * Empty queue by removing and destroying all BD's.
1000  * Free all buffers.
1001  * 0-fill, but do not free "txq" descriptor structure.
1002  */
1003 static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id)
1004 {
1005 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1006 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1007 	int i;
1008 
1009 	if (WARN_ON(!txq))
1010 		return;
1011 
1012 	iwl_pcie_gen2_txq_unmap(trans, txq_id);
1013 
1014 	/* De-alloc array of command/tx buffers */
1015 	if (txq_id == trans_pcie->cmd_queue)
1016 		for (i = 0; i < txq->n_window; i++) {
1017 			kzfree(txq->entries[i].cmd);
1018 			kzfree(txq->entries[i].free_buf);
1019 		}
1020 	del_timer_sync(&txq->stuck_timer);
1021 
1022 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1023 
1024 	trans_pcie->txq[txq_id] = NULL;
1025 
1026 	clear_bit(txq_id, trans_pcie->queue_used);
1027 }
1028 
1029 int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
1030 				 struct iwl_tx_queue_cfg_cmd *cmd,
1031 				 int cmd_id,
1032 				 unsigned int timeout)
1033 {
1034 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1035 	struct iwl_tx_queue_cfg_rsp *rsp;
1036 	struct iwl_txq *txq;
1037 	struct iwl_host_cmd hcmd = {
1038 		.id = cmd_id,
1039 		.len = { sizeof(*cmd) },
1040 		.data = { cmd, },
1041 		.flags = CMD_WANT_SKB,
1042 	};
1043 	int ret, qid;
1044 	u32 wr_ptr;
1045 
1046 	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1047 	if (!txq)
1048 		return -ENOMEM;
1049 	ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl,
1050 				     sizeof(struct iwlagn_scd_bc_tbl));
1051 	if (ret) {
1052 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1053 		kfree(txq);
1054 		return -ENOMEM;
1055 	}
1056 
1057 	ret = iwl_pcie_txq_alloc(trans, txq, TFD_TX_CMD_SLOTS, false);
1058 	if (ret) {
1059 		IWL_ERR(trans, "Tx queue alloc failed\n");
1060 		goto error;
1061 	}
1062 	ret = iwl_pcie_txq_init(trans, txq, TFD_TX_CMD_SLOTS, false);
1063 	if (ret) {
1064 		IWL_ERR(trans, "Tx queue init failed\n");
1065 		goto error;
1066 	}
1067 
1068 	txq->wd_timeout = msecs_to_jiffies(timeout);
1069 
1070 	cmd->tfdq_addr = cpu_to_le64(txq->dma_addr);
1071 	cmd->byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1072 	cmd->cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(TFD_TX_CMD_SLOTS));
1073 
1074 	ret = iwl_trans_send_cmd(trans, &hcmd);
1075 	if (ret)
1076 		goto error;
1077 
1078 	if (WARN_ON(iwl_rx_packet_payload_len(hcmd.resp_pkt) != sizeof(*rsp))) {
1079 		ret = -EINVAL;
1080 		goto error_free_resp;
1081 	}
1082 
1083 	rsp = (void *)hcmd.resp_pkt->data;
1084 	qid = le16_to_cpu(rsp->queue_number);
1085 	wr_ptr = le16_to_cpu(rsp->write_pointer);
1086 
1087 	if (qid >= ARRAY_SIZE(trans_pcie->txq)) {
1088 		WARN_ONCE(1, "queue index %d unsupported", qid);
1089 		ret = -EIO;
1090 		goto error_free_resp;
1091 	}
1092 
1093 	if (test_and_set_bit(qid, trans_pcie->queue_used)) {
1094 		WARN_ONCE(1, "queue %d already used", qid);
1095 		ret = -EIO;
1096 		goto error_free_resp;
1097 	}
1098 
1099 	txq->id = qid;
1100 	trans_pcie->txq[qid] = txq;
1101 	wr_ptr &= (TFD_QUEUE_SIZE_MAX - 1);
1102 
1103 	/* Place first TFD at index corresponding to start sequence number */
1104 	txq->read_ptr = wr_ptr;
1105 	txq->write_ptr = wr_ptr;
1106 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1107 			   (txq->write_ptr) | (qid << 16));
1108 	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1109 
1110 	iwl_free_resp(&hcmd);
1111 	return qid;
1112 
1113 error_free_resp:
1114 	iwl_free_resp(&hcmd);
1115 error:
1116 	iwl_pcie_gen2_txq_free_memory(trans, txq);
1117 	return ret;
1118 }
1119 
1120 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue)
1121 {
1122 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1123 
1124 	/*
1125 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1126 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1127 	 * allow the op_mode to call txq_disable after it already called
1128 	 * stop_device.
1129 	 */
1130 	if (!test_and_clear_bit(queue, trans_pcie->queue_used)) {
1131 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1132 			  "queue %d not used", queue);
1133 		return;
1134 	}
1135 
1136 	iwl_pcie_gen2_txq_unmap(trans, queue);
1137 
1138 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1139 }
1140 
1141 void iwl_pcie_gen2_tx_free(struct iwl_trans *trans)
1142 {
1143 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1144 	int i;
1145 
1146 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
1147 
1148 	/* Free all TX queues */
1149 	for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) {
1150 		if (!trans_pcie->txq[i])
1151 			continue;
1152 
1153 		iwl_pcie_gen2_txq_free(trans, i);
1154 	}
1155 }
1156 
1157 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans)
1158 {
1159 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1160 	struct iwl_txq *cmd_queue;
1161 	int txq_id = trans_pcie->cmd_queue, ret;
1162 
1163 	/* alloc and init the command queue */
1164 	if (!trans_pcie->txq[txq_id]) {
1165 		cmd_queue = kzalloc(sizeof(*cmd_queue), GFP_KERNEL);
1166 		if (!cmd_queue) {
1167 			IWL_ERR(trans, "Not enough memory for command queue\n");
1168 			return -ENOMEM;
1169 		}
1170 		trans_pcie->txq[txq_id] = cmd_queue;
1171 		ret = iwl_pcie_txq_alloc(trans, cmd_queue, TFD_CMD_SLOTS, true);
1172 		if (ret) {
1173 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1174 			goto error;
1175 		}
1176 	} else {
1177 		cmd_queue = trans_pcie->txq[txq_id];
1178 	}
1179 
1180 	ret = iwl_pcie_txq_init(trans, cmd_queue, TFD_CMD_SLOTS, true);
1181 	if (ret) {
1182 		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1183 		goto error;
1184 	}
1185 	trans_pcie->txq[txq_id]->id = txq_id;
1186 	set_bit(txq_id, trans_pcie->queue_used);
1187 
1188 	return 0;
1189 
1190 error:
1191 	iwl_pcie_gen2_tx_free(trans);
1192 	return ret;
1193 }
1194 
1195