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) 2003 - 2014 Intel Corporation. All rights reserved.
9  * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
10  * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
11  * Copyright(c) 2018 - 2019 Intel Corporation
12  *
13  * This program is free software; you can redistribute it and/or modify it
14  * under the terms of version 2 of the GNU General Public License as
15  * published by the Free Software Foundation.
16  *
17  * This program is distributed in the hope that it will be useful, but WITHOUT
18  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
19  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
20  * more details.
21  *
22  * The full GNU General Public License is included in this distribution in the
23  * file called COPYING.
24  *
25  * Contact Information:
26  *  Intel Linux Wireless <linuxwifi@intel.com>
27  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
28  *
29  * BSD LICENSE
30  *
31  * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
32  * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
33  * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
34  * Copyright(c) 2018 - 2019 Intel Corporation
35  * All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  *
41  *  * Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  *  * Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in
45  *    the documentation and/or other materials provided with the
46  *    distribution.
47  *  * Neither the name Intel Corporation nor the names of its
48  *    contributors may be used to endorse or promote products derived
49  *    from this software without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
52  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
53  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
54  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
55  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
56  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
57  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
58  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
59  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
60  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
61  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
62  *
63  *****************************************************************************/
64 #include <linux/etherdevice.h>
65 #include <linux/ieee80211.h>
66 #include <linux/slab.h>
67 #include <linux/sched.h>
68 #include <linux/pm_runtime.h>
69 #include <net/ip6_checksum.h>
70 #include <net/tso.h>
71 
72 #include "iwl-debug.h"
73 #include "iwl-csr.h"
74 #include "iwl-prph.h"
75 #include "iwl-io.h"
76 #include "iwl-scd.h"
77 #include "iwl-op-mode.h"
78 #include "internal.h"
79 #include "fw/api/tx.h"
80 
81 #define IWL_TX_CRC_SIZE 4
82 #define IWL_TX_DELIMITER_SIZE 4
83 
84 /*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
85  * DMA services
86  *
87  * Theory of operation
88  *
89  * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
90  * of buffer descriptors, each of which points to one or more data buffers for
91  * the device to read from or fill.  Driver and device exchange status of each
92  * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
93  * entries in each circular buffer, to protect against confusing empty and full
94  * queue states.
95  *
96  * The device reads or writes the data in the queues via the device's several
97  * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
98  *
99  * For Tx queue, there are low mark and high mark limits. If, after queuing
100  * the packet for Tx, free space become < low mark, Tx queue stopped. When
101  * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
102  * Tx queue resumed.
103  *
104  ***************************************************/
105 
106 int iwl_queue_space(struct iwl_trans *trans, const struct iwl_txq *q)
107 {
108 	unsigned int max;
109 	unsigned int used;
110 
111 	/*
112 	 * To avoid ambiguity between empty and completely full queues, there
113 	 * should always be less than max_tfd_queue_size elements in the queue.
114 	 * If q->n_window is smaller than max_tfd_queue_size, there is no need
115 	 * to reserve any queue entries for this purpose.
116 	 */
117 	if (q->n_window < trans->cfg->base_params->max_tfd_queue_size)
118 		max = q->n_window;
119 	else
120 		max = trans->cfg->base_params->max_tfd_queue_size - 1;
121 
122 	/*
123 	 * max_tfd_queue_size is a power of 2, so the following is equivalent to
124 	 * modulo by max_tfd_queue_size and is well defined.
125 	 */
126 	used = (q->write_ptr - q->read_ptr) &
127 		(trans->cfg->base_params->max_tfd_queue_size - 1);
128 
129 	if (WARN_ON(used > max))
130 		return 0;
131 
132 	return max - used;
133 }
134 
135 /*
136  * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
137  */
138 static int iwl_queue_init(struct iwl_txq *q, int slots_num)
139 {
140 	q->n_window = slots_num;
141 
142 	/* slots_num must be power-of-two size, otherwise
143 	 * iwl_pcie_get_cmd_index is broken. */
144 	if (WARN_ON(!is_power_of_2(slots_num)))
145 		return -EINVAL;
146 
147 	q->low_mark = q->n_window / 4;
148 	if (q->low_mark < 4)
149 		q->low_mark = 4;
150 
151 	q->high_mark = q->n_window / 8;
152 	if (q->high_mark < 2)
153 		q->high_mark = 2;
154 
155 	q->write_ptr = 0;
156 	q->read_ptr = 0;
157 
158 	return 0;
159 }
160 
161 int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
162 			   struct iwl_dma_ptr *ptr, size_t size)
163 {
164 	if (WARN_ON(ptr->addr))
165 		return -EINVAL;
166 
167 	ptr->addr = dma_alloc_coherent(trans->dev, size,
168 				       &ptr->dma, GFP_KERNEL);
169 	if (!ptr->addr)
170 		return -ENOMEM;
171 	ptr->size = size;
172 	return 0;
173 }
174 
175 void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr)
176 {
177 	if (unlikely(!ptr->addr))
178 		return;
179 
180 	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
181 	memset(ptr, 0, sizeof(*ptr));
182 }
183 
184 static void iwl_pcie_txq_stuck_timer(struct timer_list *t)
185 {
186 	struct iwl_txq *txq = from_timer(txq, t, stuck_timer);
187 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
188 	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
189 
190 	spin_lock(&txq->lock);
191 	/* check if triggered erroneously */
192 	if (txq->read_ptr == txq->write_ptr) {
193 		spin_unlock(&txq->lock);
194 		return;
195 	}
196 	spin_unlock(&txq->lock);
197 
198 	iwl_trans_pcie_log_scd_error(trans, txq);
199 
200 	iwl_force_nmi(trans);
201 }
202 
203 /*
204  * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
205  */
206 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
207 					     struct iwl_txq *txq, u16 byte_cnt,
208 					     int num_tbs)
209 {
210 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
211 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
212 	int write_ptr = txq->write_ptr;
213 	int txq_id = txq->id;
214 	u8 sec_ctl = 0;
215 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
216 	__le16 bc_ent;
217 	struct iwl_tx_cmd *tx_cmd =
218 		(void *)txq->entries[txq->write_ptr].cmd->payload;
219 	u8 sta_id = tx_cmd->sta_id;
220 
221 	scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
222 
223 	sec_ctl = tx_cmd->sec_ctl;
224 
225 	switch (sec_ctl & TX_CMD_SEC_MSK) {
226 	case TX_CMD_SEC_CCM:
227 		len += IEEE80211_CCMP_MIC_LEN;
228 		break;
229 	case TX_CMD_SEC_TKIP:
230 		len += IEEE80211_TKIP_ICV_LEN;
231 		break;
232 	case TX_CMD_SEC_WEP:
233 		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
234 		break;
235 	}
236 	if (trans_pcie->bc_table_dword)
237 		len = DIV_ROUND_UP(len, 4);
238 
239 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
240 		return;
241 
242 	bc_ent = cpu_to_le16(len | (sta_id << 12));
243 
244 	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
245 
246 	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
247 		scd_bc_tbl[txq_id].
248 			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
249 }
250 
251 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
252 					    struct iwl_txq *txq)
253 {
254 	struct iwl_trans_pcie *trans_pcie =
255 		IWL_TRANS_GET_PCIE_TRANS(trans);
256 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
257 	int txq_id = txq->id;
258 	int read_ptr = txq->read_ptr;
259 	u8 sta_id = 0;
260 	__le16 bc_ent;
261 	struct iwl_tx_cmd *tx_cmd =
262 		(void *)txq->entries[read_ptr].cmd->payload;
263 
264 	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
265 
266 	if (txq_id != trans_pcie->cmd_queue)
267 		sta_id = tx_cmd->sta_id;
268 
269 	bc_ent = cpu_to_le16(1 | (sta_id << 12));
270 
271 	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
272 
273 	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
274 		scd_bc_tbl[txq_id].
275 			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
276 }
277 
278 /*
279  * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
280  */
281 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
282 				    struct iwl_txq *txq)
283 {
284 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
285 	u32 reg = 0;
286 	int txq_id = txq->id;
287 
288 	lockdep_assert_held(&txq->lock);
289 
290 	/*
291 	 * explicitly wake up the NIC if:
292 	 * 1. shadow registers aren't enabled
293 	 * 2. NIC is woken up for CMD regardless of shadow outside this function
294 	 * 3. there is a chance that the NIC is asleep
295 	 */
296 	if (!trans->cfg->base_params->shadow_reg_enable &&
297 	    txq_id != trans_pcie->cmd_queue &&
298 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
299 		/*
300 		 * wake up nic if it's powered down ...
301 		 * uCode will wake up, and interrupt us again, so next
302 		 * time we'll skip this part.
303 		 */
304 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
305 
306 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
307 			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
308 				       txq_id, reg);
309 			iwl_set_bit(trans, CSR_GP_CNTRL,
310 				    BIT(trans->cfg->csr->flag_mac_access_req));
311 			txq->need_update = true;
312 			return;
313 		}
314 	}
315 
316 	/*
317 	 * if not in power-save mode, uCode will never sleep when we're
318 	 * trying to tx (during RFKILL, we're not trying to tx).
319 	 */
320 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr);
321 	if (!txq->block)
322 		iwl_write32(trans, HBUS_TARG_WRPTR,
323 			    txq->write_ptr | (txq_id << 8));
324 }
325 
326 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
327 {
328 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
329 	int i;
330 
331 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
332 		struct iwl_txq *txq = trans_pcie->txq[i];
333 
334 		if (!test_bit(i, trans_pcie->queue_used))
335 			continue;
336 
337 		spin_lock_bh(&txq->lock);
338 		if (txq->need_update) {
339 			iwl_pcie_txq_inc_wr_ptr(trans, txq);
340 			txq->need_update = false;
341 		}
342 		spin_unlock_bh(&txq->lock);
343 	}
344 }
345 
346 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_trans *trans,
347 						  void *_tfd, u8 idx)
348 {
349 
350 	if (trans->cfg->use_tfh) {
351 		struct iwl_tfh_tfd *tfd = _tfd;
352 		struct iwl_tfh_tb *tb = &tfd->tbs[idx];
353 
354 		return (dma_addr_t)(le64_to_cpu(tb->addr));
355 	} else {
356 		struct iwl_tfd *tfd = _tfd;
357 		struct iwl_tfd_tb *tb = &tfd->tbs[idx];
358 		dma_addr_t addr = get_unaligned_le32(&tb->lo);
359 		dma_addr_t hi_len;
360 
361 		if (sizeof(dma_addr_t) <= sizeof(u32))
362 			return addr;
363 
364 		hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
365 
366 		/*
367 		 * shift by 16 twice to avoid warnings on 32-bit
368 		 * (where this code never runs anyway due to the
369 		 * if statement above)
370 		 */
371 		return addr | ((hi_len << 16) << 16);
372 	}
373 }
374 
375 static inline void iwl_pcie_tfd_set_tb(struct iwl_trans *trans, void *tfd,
376 				       u8 idx, dma_addr_t addr, u16 len)
377 {
378 	struct iwl_tfd *tfd_fh = (void *)tfd;
379 	struct iwl_tfd_tb *tb = &tfd_fh->tbs[idx];
380 
381 	u16 hi_n_len = len << 4;
382 
383 	put_unaligned_le32(addr, &tb->lo);
384 	hi_n_len |= iwl_get_dma_hi_addr(addr);
385 
386 	tb->hi_n_len = cpu_to_le16(hi_n_len);
387 
388 	tfd_fh->num_tbs = idx + 1;
389 }
390 
391 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_trans *trans, void *_tfd)
392 {
393 	if (trans->cfg->use_tfh) {
394 		struct iwl_tfh_tfd *tfd = _tfd;
395 
396 		return le16_to_cpu(tfd->num_tbs) & 0x1f;
397 	} else {
398 		struct iwl_tfd *tfd = _tfd;
399 
400 		return tfd->num_tbs & 0x1f;
401 	}
402 }
403 
404 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
405 			       struct iwl_cmd_meta *meta,
406 			       struct iwl_txq *txq, int index)
407 {
408 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
409 	int i, num_tbs;
410 	void *tfd = iwl_pcie_get_tfd(trans, txq, index);
411 
412 	/* Sanity check on number of chunks */
413 	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
414 
415 	if (num_tbs > trans_pcie->max_tbs) {
416 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
417 		/* @todo issue fatal error, it is quite serious situation */
418 		return;
419 	}
420 
421 	/* first TB is never freed - it's the bidirectional DMA data */
422 
423 	for (i = 1; i < num_tbs; i++) {
424 		if (meta->tbs & BIT(i))
425 			dma_unmap_page(trans->dev,
426 				       iwl_pcie_tfd_tb_get_addr(trans, tfd, i),
427 				       iwl_pcie_tfd_tb_get_len(trans, tfd, i),
428 				       DMA_TO_DEVICE);
429 		else
430 			dma_unmap_single(trans->dev,
431 					 iwl_pcie_tfd_tb_get_addr(trans, tfd,
432 								  i),
433 					 iwl_pcie_tfd_tb_get_len(trans, tfd,
434 								 i),
435 					 DMA_TO_DEVICE);
436 	}
437 
438 	meta->tbs = 0;
439 
440 	if (trans->cfg->use_tfh) {
441 		struct iwl_tfh_tfd *tfd_fh = (void *)tfd;
442 
443 		tfd_fh->num_tbs = 0;
444 	} else {
445 		struct iwl_tfd *tfd_fh = (void *)tfd;
446 
447 		tfd_fh->num_tbs = 0;
448 	}
449 
450 }
451 
452 /*
453  * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
454  * @trans - transport private data
455  * @txq - tx queue
456  * @dma_dir - the direction of the DMA mapping
457  *
458  * Does NOT advance any TFD circular buffer read/write indexes
459  * Does NOT free the TFD itself (which is within circular buffer)
460  */
461 void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
462 {
463 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
464 	 * idx is bounded by n_window
465 	 */
466 	int rd_ptr = txq->read_ptr;
467 	int idx = iwl_pcie_get_cmd_index(txq, rd_ptr);
468 
469 	lockdep_assert_held(&txq->lock);
470 
471 	/* We have only q->n_window txq->entries, but we use
472 	 * TFD_QUEUE_SIZE_MAX tfds
473 	 */
474 	iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, txq, rd_ptr);
475 
476 	/* free SKB */
477 	if (txq->entries) {
478 		struct sk_buff *skb;
479 
480 		skb = txq->entries[idx].skb;
481 
482 		/* Can be called from irqs-disabled context
483 		 * If skb is not NULL, it means that the whole queue is being
484 		 * freed and that the queue is not empty - free the skb
485 		 */
486 		if (skb) {
487 			iwl_op_mode_free_skb(trans->op_mode, skb);
488 			txq->entries[idx].skb = NULL;
489 		}
490 	}
491 }
492 
493 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
494 				  dma_addr_t addr, u16 len, bool reset)
495 {
496 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
497 	void *tfd;
498 	u32 num_tbs;
499 
500 	tfd = txq->tfds + trans_pcie->tfd_size * txq->write_ptr;
501 
502 	if (reset)
503 		memset(tfd, 0, trans_pcie->tfd_size);
504 
505 	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
506 
507 	/* Each TFD can point to a maximum max_tbs Tx buffers */
508 	if (num_tbs >= trans_pcie->max_tbs) {
509 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
510 			trans_pcie->max_tbs);
511 		return -EINVAL;
512 	}
513 
514 	if (WARN(addr & ~IWL_TX_DMA_MASK,
515 		 "Unaligned address = %llx\n", (unsigned long long)addr))
516 		return -EINVAL;
517 
518 	iwl_pcie_tfd_set_tb(trans, tfd, num_tbs, addr, len);
519 
520 	return num_tbs;
521 }
522 
523 int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq,
524 		       int slots_num, bool cmd_queue)
525 {
526 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
527 	size_t tfd_sz = trans_pcie->tfd_size *
528 		trans->cfg->base_params->max_tfd_queue_size;
529 	size_t tb0_buf_sz;
530 	int i;
531 
532 	if (WARN_ON(txq->entries || txq->tfds))
533 		return -EINVAL;
534 
535 	if (trans->cfg->use_tfh)
536 		tfd_sz = trans_pcie->tfd_size * slots_num;
537 
538 	timer_setup(&txq->stuck_timer, iwl_pcie_txq_stuck_timer, 0);
539 	txq->trans_pcie = trans_pcie;
540 
541 	txq->n_window = slots_num;
542 
543 	txq->entries = kcalloc(slots_num,
544 			       sizeof(struct iwl_pcie_txq_entry),
545 			       GFP_KERNEL);
546 
547 	if (!txq->entries)
548 		goto error;
549 
550 	if (cmd_queue)
551 		for (i = 0; i < slots_num; i++) {
552 			txq->entries[i].cmd =
553 				kmalloc(sizeof(struct iwl_device_cmd),
554 					GFP_KERNEL);
555 			if (!txq->entries[i].cmd)
556 				goto error;
557 		}
558 
559 	/* Circular buffer of transmit frame descriptors (TFDs),
560 	 * shared with device */
561 	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
562 				       &txq->dma_addr, GFP_KERNEL);
563 	if (!txq->tfds)
564 		goto error;
565 
566 	BUILD_BUG_ON(IWL_FIRST_TB_SIZE_ALIGN != sizeof(*txq->first_tb_bufs));
567 
568 	tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
569 
570 	txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
571 					      &txq->first_tb_dma,
572 					      GFP_KERNEL);
573 	if (!txq->first_tb_bufs)
574 		goto err_free_tfds;
575 
576 	return 0;
577 err_free_tfds:
578 	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
579 error:
580 	if (txq->entries && cmd_queue)
581 		for (i = 0; i < slots_num; i++)
582 			kfree(txq->entries[i].cmd);
583 	kfree(txq->entries);
584 	txq->entries = NULL;
585 
586 	return -ENOMEM;
587 
588 }
589 
590 int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
591 		      int slots_num, bool cmd_queue)
592 {
593 	int ret;
594 	u32 tfd_queue_max_size = trans->cfg->base_params->max_tfd_queue_size;
595 
596 	txq->need_update = false;
597 
598 	/* max_tfd_queue_size must be power-of-two size, otherwise
599 	 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
600 	if (WARN_ONCE(tfd_queue_max_size & (tfd_queue_max_size - 1),
601 		      "Max tfd queue size must be a power of two, but is %d",
602 		      tfd_queue_max_size))
603 		return -EINVAL;
604 
605 	/* Initialize queue's high/low-water marks, and head/tail indexes */
606 	ret = iwl_queue_init(txq, slots_num);
607 	if (ret)
608 		return ret;
609 
610 	spin_lock_init(&txq->lock);
611 
612 	if (cmd_queue) {
613 		static struct lock_class_key iwl_pcie_cmd_queue_lock_class;
614 
615 		lockdep_set_class(&txq->lock, &iwl_pcie_cmd_queue_lock_class);
616 	}
617 
618 	__skb_queue_head_init(&txq->overflow_q);
619 
620 	return 0;
621 }
622 
623 void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
624 			    struct sk_buff *skb)
625 {
626 	struct page **page_ptr;
627 
628 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
629 
630 	if (*page_ptr) {
631 		__free_page(*page_ptr);
632 		*page_ptr = NULL;
633 	}
634 }
635 
636 static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
637 {
638 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
639 
640 	lockdep_assert_held(&trans_pcie->reg_lock);
641 
642 	if (trans_pcie->ref_cmd_in_flight) {
643 		trans_pcie->ref_cmd_in_flight = false;
644 		IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
645 		iwl_trans_unref(trans);
646 	}
647 
648 	if (!trans->cfg->base_params->apmg_wake_up_wa)
649 		return;
650 	if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
651 		return;
652 
653 	trans_pcie->cmd_hold_nic_awake = false;
654 	__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
655 				   BIT(trans->cfg->csr->flag_mac_access_req));
656 }
657 
658 /*
659  * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
660  */
661 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
662 {
663 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
664 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
665 
666 	spin_lock_bh(&txq->lock);
667 	while (txq->write_ptr != txq->read_ptr) {
668 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
669 				   txq_id, txq->read_ptr);
670 
671 		if (txq_id != trans_pcie->cmd_queue) {
672 			struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
673 
674 			if (WARN_ON_ONCE(!skb))
675 				continue;
676 
677 			iwl_pcie_free_tso_page(trans_pcie, skb);
678 		}
679 		iwl_pcie_txq_free_tfd(trans, txq);
680 		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
681 
682 		if (txq->read_ptr == txq->write_ptr) {
683 			unsigned long flags;
684 
685 			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
686 			if (txq_id != trans_pcie->cmd_queue) {
687 				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
688 					      txq->id);
689 				iwl_trans_unref(trans);
690 			} else {
691 				iwl_pcie_clear_cmd_in_flight(trans);
692 			}
693 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
694 		}
695 	}
696 
697 	while (!skb_queue_empty(&txq->overflow_q)) {
698 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
699 
700 		iwl_op_mode_free_skb(trans->op_mode, skb);
701 	}
702 
703 	spin_unlock_bh(&txq->lock);
704 
705 	/* just in case - this queue may have been stopped */
706 	iwl_wake_queue(trans, txq);
707 }
708 
709 /*
710  * iwl_pcie_txq_free - Deallocate DMA queue.
711  * @txq: Transmit queue to deallocate.
712  *
713  * Empty queue by removing and destroying all BD's.
714  * Free all buffers.
715  * 0-fill, but do not free "txq" descriptor structure.
716  */
717 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
718 {
719 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
720 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
721 	struct device *dev = trans->dev;
722 	int i;
723 
724 	if (WARN_ON(!txq))
725 		return;
726 
727 	iwl_pcie_txq_unmap(trans, txq_id);
728 
729 	/* De-alloc array of command/tx buffers */
730 	if (txq_id == trans_pcie->cmd_queue)
731 		for (i = 0; i < txq->n_window; i++) {
732 			kzfree(txq->entries[i].cmd);
733 			kzfree(txq->entries[i].free_buf);
734 		}
735 
736 	/* De-alloc circular buffer of TFDs */
737 	if (txq->tfds) {
738 		dma_free_coherent(dev,
739 				  trans_pcie->tfd_size *
740 				  trans->cfg->base_params->max_tfd_queue_size,
741 				  txq->tfds, txq->dma_addr);
742 		txq->dma_addr = 0;
743 		txq->tfds = NULL;
744 
745 		dma_free_coherent(dev,
746 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
747 				  txq->first_tb_bufs, txq->first_tb_dma);
748 	}
749 
750 	kfree(txq->entries);
751 	txq->entries = NULL;
752 
753 	del_timer_sync(&txq->stuck_timer);
754 
755 	/* 0-fill queue descriptor structure */
756 	memset(txq, 0, sizeof(*txq));
757 }
758 
759 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
760 {
761 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
762 	int nq = trans->cfg->base_params->num_of_queues;
763 	int chan;
764 	u32 reg_val;
765 	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
766 				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
767 
768 	/* make sure all queue are not stopped/used */
769 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
770 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
771 
772 	trans_pcie->scd_base_addr =
773 		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
774 
775 	WARN_ON(scd_base_addr != 0 &&
776 		scd_base_addr != trans_pcie->scd_base_addr);
777 
778 	/* reset context data, TX status and translation data */
779 	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
780 				   SCD_CONTEXT_MEM_LOWER_BOUND,
781 			    NULL, clear_dwords);
782 
783 	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
784 		       trans_pcie->scd_bc_tbls.dma >> 10);
785 
786 	/* The chain extension of the SCD doesn't work well. This feature is
787 	 * enabled by default by the HW, so we need to disable it manually.
788 	 */
789 	if (trans->cfg->base_params->scd_chain_ext_wa)
790 		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
791 
792 	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
793 				trans_pcie->cmd_fifo,
794 				trans_pcie->cmd_q_wdg_timeout);
795 
796 	/* Activate all Tx DMA/FIFO channels */
797 	iwl_scd_activate_fifos(trans);
798 
799 	/* Enable DMA channel */
800 	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
801 		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
802 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
803 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
804 
805 	/* Update FH chicken bits */
806 	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
807 	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
808 			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
809 
810 	/* Enable L1-Active */
811 	if (trans->cfg->device_family < IWL_DEVICE_FAMILY_8000)
812 		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
813 				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
814 }
815 
816 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
817 {
818 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
819 	int txq_id;
820 
821 	/*
822 	 * we should never get here in gen2 trans mode return early to avoid
823 	 * having invalid accesses
824 	 */
825 	if (WARN_ON_ONCE(trans->cfg->gen2))
826 		return;
827 
828 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
829 	     txq_id++) {
830 		struct iwl_txq *txq = trans_pcie->txq[txq_id];
831 		if (trans->cfg->use_tfh)
832 			iwl_write_direct64(trans,
833 					   FH_MEM_CBBC_QUEUE(trans, txq_id),
834 					   txq->dma_addr);
835 		else
836 			iwl_write_direct32(trans,
837 					   FH_MEM_CBBC_QUEUE(trans, txq_id),
838 					   txq->dma_addr >> 8);
839 		iwl_pcie_txq_unmap(trans, txq_id);
840 		txq->read_ptr = 0;
841 		txq->write_ptr = 0;
842 	}
843 
844 	/* Tell NIC where to find the "keep warm" buffer */
845 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
846 			   trans_pcie->kw.dma >> 4);
847 
848 	/*
849 	 * Send 0 as the scd_base_addr since the device may have be reset
850 	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
851 	 * contain garbage.
852 	 */
853 	iwl_pcie_tx_start(trans, 0);
854 }
855 
856 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
857 {
858 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
859 	unsigned long flags;
860 	int ch, ret;
861 	u32 mask = 0;
862 
863 	spin_lock(&trans_pcie->irq_lock);
864 
865 	if (!iwl_trans_grab_nic_access(trans, &flags))
866 		goto out;
867 
868 	/* Stop each Tx DMA channel */
869 	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
870 		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
871 		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
872 	}
873 
874 	/* Wait for DMA channels to be idle */
875 	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
876 	if (ret < 0)
877 		IWL_ERR(trans,
878 			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
879 			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
880 
881 	iwl_trans_release_nic_access(trans, &flags);
882 
883 out:
884 	spin_unlock(&trans_pcie->irq_lock);
885 }
886 
887 /*
888  * iwl_pcie_tx_stop - Stop all Tx DMA channels
889  */
890 int iwl_pcie_tx_stop(struct iwl_trans *trans)
891 {
892 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
893 	int txq_id;
894 
895 	/* Turn off all Tx DMA fifos */
896 	iwl_scd_deactivate_fifos(trans);
897 
898 	/* Turn off all Tx DMA channels */
899 	iwl_pcie_tx_stop_fh(trans);
900 
901 	/*
902 	 * This function can be called before the op_mode disabled the
903 	 * queues. This happens when we have an rfkill interrupt.
904 	 * Since we stop Tx altogether - mark the queues as stopped.
905 	 */
906 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
907 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
908 
909 	/* This can happen: start_hw, stop_device */
910 	if (!trans_pcie->txq_memory)
911 		return 0;
912 
913 	/* Unmap DMA from host system and free skb's */
914 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
915 	     txq_id++)
916 		iwl_pcie_txq_unmap(trans, txq_id);
917 
918 	return 0;
919 }
920 
921 /*
922  * iwl_trans_tx_free - Free TXQ Context
923  *
924  * Destroy all TX DMA queues and structures
925  */
926 void iwl_pcie_tx_free(struct iwl_trans *trans)
927 {
928 	int txq_id;
929 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
930 
931 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
932 
933 	/* Tx queues */
934 	if (trans_pcie->txq_memory) {
935 		for (txq_id = 0;
936 		     txq_id < trans->cfg->base_params->num_of_queues;
937 		     txq_id++) {
938 			iwl_pcie_txq_free(trans, txq_id);
939 			trans_pcie->txq[txq_id] = NULL;
940 		}
941 	}
942 
943 	kfree(trans_pcie->txq_memory);
944 	trans_pcie->txq_memory = NULL;
945 
946 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
947 
948 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
949 }
950 
951 /*
952  * iwl_pcie_tx_alloc - allocate TX context
953  * Allocate all Tx DMA structures and initialize them
954  */
955 static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
956 {
957 	int ret;
958 	int txq_id, slots_num;
959 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
960 	u16 bc_tbls_size = trans->cfg->base_params->num_of_queues;
961 
962 	bc_tbls_size *= (trans->cfg->device_family >= IWL_DEVICE_FAMILY_22560) ?
963 		sizeof(struct iwl_gen3_bc_tbl) :
964 		sizeof(struct iwlagn_scd_bc_tbl);
965 
966 	/*It is not allowed to alloc twice, so warn when this happens.
967 	 * We cannot rely on the previous allocation, so free and fail */
968 	if (WARN_ON(trans_pcie->txq_memory)) {
969 		ret = -EINVAL;
970 		goto error;
971 	}
972 
973 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
974 				     bc_tbls_size);
975 	if (ret) {
976 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
977 		goto error;
978 	}
979 
980 	/* Alloc keep-warm buffer */
981 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
982 	if (ret) {
983 		IWL_ERR(trans, "Keep Warm allocation failed\n");
984 		goto error;
985 	}
986 
987 	trans_pcie->txq_memory = kcalloc(trans->cfg->base_params->num_of_queues,
988 					 sizeof(struct iwl_txq), GFP_KERNEL);
989 	if (!trans_pcie->txq_memory) {
990 		IWL_ERR(trans, "Not enough memory for txq\n");
991 		ret = -ENOMEM;
992 		goto error;
993 	}
994 
995 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
996 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
997 	     txq_id++) {
998 		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
999 
1000 		if (cmd_queue)
1001 			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
1002 					  trans->cfg->min_txq_size);
1003 		else
1004 			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
1005 					  trans->cfg->min_256_ba_txq_size);
1006 		trans_pcie->txq[txq_id] = &trans_pcie->txq_memory[txq_id];
1007 		ret = iwl_pcie_txq_alloc(trans, trans_pcie->txq[txq_id],
1008 					 slots_num, cmd_queue);
1009 		if (ret) {
1010 			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1011 			goto error;
1012 		}
1013 		trans_pcie->txq[txq_id]->id = txq_id;
1014 	}
1015 
1016 	return 0;
1017 
1018 error:
1019 	iwl_pcie_tx_free(trans);
1020 
1021 	return ret;
1022 }
1023 
1024 int iwl_pcie_tx_init(struct iwl_trans *trans)
1025 {
1026 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1027 	int ret;
1028 	int txq_id, slots_num;
1029 	bool alloc = false;
1030 
1031 	if (!trans_pcie->txq_memory) {
1032 		ret = iwl_pcie_tx_alloc(trans);
1033 		if (ret)
1034 			goto error;
1035 		alloc = true;
1036 	}
1037 
1038 	spin_lock(&trans_pcie->irq_lock);
1039 
1040 	/* Turn off all Tx DMA fifos */
1041 	iwl_scd_deactivate_fifos(trans);
1042 
1043 	/* Tell NIC where to find the "keep warm" buffer */
1044 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
1045 			   trans_pcie->kw.dma >> 4);
1046 
1047 	spin_unlock(&trans_pcie->irq_lock);
1048 
1049 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
1050 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
1051 	     txq_id++) {
1052 		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
1053 
1054 		if (cmd_queue)
1055 			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
1056 					  trans->cfg->min_txq_size);
1057 		else
1058 			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
1059 					  trans->cfg->min_256_ba_txq_size);
1060 		ret = iwl_pcie_txq_init(trans, trans_pcie->txq[txq_id],
1061 					slots_num, cmd_queue);
1062 		if (ret) {
1063 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1064 			goto error;
1065 		}
1066 
1067 		/*
1068 		 * Tell nic where to find circular buffer of TFDs for a
1069 		 * given Tx queue, and enable the DMA channel used for that
1070 		 * queue.
1071 		 * Circular buffer (TFD queue in DRAM) physical base address
1072 		 */
1073 		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
1074 				   trans_pcie->txq[txq_id]->dma_addr >> 8);
1075 	}
1076 
1077 	iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
1078 	if (trans->cfg->base_params->num_of_queues > 20)
1079 		iwl_set_bits_prph(trans, SCD_GP_CTRL,
1080 				  SCD_GP_CTRL_ENABLE_31_QUEUES);
1081 
1082 	return 0;
1083 error:
1084 	/*Upon error, free only if we allocated something */
1085 	if (alloc)
1086 		iwl_pcie_tx_free(trans);
1087 	return ret;
1088 }
1089 
1090 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
1091 {
1092 	lockdep_assert_held(&txq->lock);
1093 
1094 	if (!txq->wd_timeout)
1095 		return;
1096 
1097 	/*
1098 	 * station is asleep and we send data - that must
1099 	 * be uAPSD or PS-Poll. Don't rearm the timer.
1100 	 */
1101 	if (txq->frozen)
1102 		return;
1103 
1104 	/*
1105 	 * if empty delete timer, otherwise move timer forward
1106 	 * since we're making progress on this queue
1107 	 */
1108 	if (txq->read_ptr == txq->write_ptr)
1109 		del_timer(&txq->stuck_timer);
1110 	else
1111 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1112 }
1113 
1114 /* Frees buffers until index _not_ inclusive */
1115 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1116 			    struct sk_buff_head *skbs)
1117 {
1118 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1119 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1120 	int tfd_num = iwl_pcie_get_cmd_index(txq, ssn);
1121 	int read_ptr = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1122 	int last_to_free;
1123 
1124 	/* This function is not meant to release cmd queue*/
1125 	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1126 		return;
1127 
1128 	spin_lock_bh(&txq->lock);
1129 
1130 	if (!test_bit(txq_id, trans_pcie->queue_used)) {
1131 		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1132 				    txq_id, ssn);
1133 		goto out;
1134 	}
1135 
1136 	if (read_ptr == tfd_num)
1137 		goto out;
1138 
1139 	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1140 			   txq_id, txq->read_ptr, tfd_num, ssn);
1141 
1142 	/*Since we free until index _not_ inclusive, the one before index is
1143 	 * the last we will free. This one must be used */
1144 	last_to_free = iwl_queue_dec_wrap(trans, tfd_num);
1145 
1146 	if (!iwl_queue_used(txq, last_to_free)) {
1147 		IWL_ERR(trans,
1148 			"%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1149 			__func__, txq_id, last_to_free,
1150 			trans->cfg->base_params->max_tfd_queue_size,
1151 			txq->write_ptr, txq->read_ptr);
1152 		goto out;
1153 	}
1154 
1155 	if (WARN_ON(!skb_queue_empty(skbs)))
1156 		goto out;
1157 
1158 	for (;
1159 	     read_ptr != tfd_num;
1160 	     txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr),
1161 	     read_ptr = iwl_pcie_get_cmd_index(txq, txq->read_ptr)) {
1162 		struct sk_buff *skb = txq->entries[read_ptr].skb;
1163 
1164 		if (WARN_ON_ONCE(!skb))
1165 			continue;
1166 
1167 		iwl_pcie_free_tso_page(trans_pcie, skb);
1168 
1169 		__skb_queue_tail(skbs, skb);
1170 
1171 		txq->entries[read_ptr].skb = NULL;
1172 
1173 		if (!trans->cfg->use_tfh)
1174 			iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1175 
1176 		iwl_pcie_txq_free_tfd(trans, txq);
1177 	}
1178 
1179 	iwl_pcie_txq_progress(txq);
1180 
1181 	if (iwl_queue_space(trans, txq) > txq->low_mark &&
1182 	    test_bit(txq_id, trans_pcie->queue_stopped)) {
1183 		struct sk_buff_head overflow_skbs;
1184 
1185 		__skb_queue_head_init(&overflow_skbs);
1186 		skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1187 
1188 		/*
1189 		 * We are going to transmit from the overflow queue.
1190 		 * Remember this state so that wait_for_txq_empty will know we
1191 		 * are adding more packets to the TFD queue. It cannot rely on
1192 		 * the state of &txq->overflow_q, as we just emptied it, but
1193 		 * haven't TXed the content yet.
1194 		 */
1195 		txq->overflow_tx = true;
1196 
1197 		/*
1198 		 * This is tricky: we are in reclaim path which is non
1199 		 * re-entrant, so noone will try to take the access the
1200 		 * txq data from that path. We stopped tx, so we can't
1201 		 * have tx as well. Bottom line, we can unlock and re-lock
1202 		 * later.
1203 		 */
1204 		spin_unlock_bh(&txq->lock);
1205 
1206 		while (!skb_queue_empty(&overflow_skbs)) {
1207 			struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1208 			struct iwl_device_cmd *dev_cmd_ptr;
1209 
1210 			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1211 						 trans_pcie->dev_cmd_offs);
1212 
1213 			/*
1214 			 * Note that we can very well be overflowing again.
1215 			 * In that case, iwl_queue_space will be small again
1216 			 * and we won't wake mac80211's queue.
1217 			 */
1218 			iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id);
1219 		}
1220 
1221 		if (iwl_queue_space(trans, txq) > txq->low_mark)
1222 			iwl_wake_queue(trans, txq);
1223 
1224 		spin_lock_bh(&txq->lock);
1225 		txq->overflow_tx = false;
1226 	}
1227 
1228 	if (txq->read_ptr == txq->write_ptr) {
1229 		IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", txq->id);
1230 		iwl_trans_unref(trans);
1231 	}
1232 
1233 out:
1234 	spin_unlock_bh(&txq->lock);
1235 }
1236 
1237 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1238 				      const struct iwl_host_cmd *cmd)
1239 {
1240 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1241 	const struct iwl_cfg *cfg = trans->cfg;
1242 	int ret;
1243 
1244 	lockdep_assert_held(&trans_pcie->reg_lock);
1245 
1246 	/* Make sure the NIC is still alive in the bus */
1247 	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1248 		return -ENODEV;
1249 
1250 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
1251 	    !trans_pcie->ref_cmd_in_flight) {
1252 		trans_pcie->ref_cmd_in_flight = true;
1253 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
1254 		iwl_trans_ref(trans);
1255 	}
1256 
1257 	/*
1258 	 * wake up the NIC to make sure that the firmware will see the host
1259 	 * command - we will let the NIC sleep once all the host commands
1260 	 * returned. This needs to be done only on NICs that have
1261 	 * apmg_wake_up_wa set.
1262 	 */
1263 	if (cfg->base_params->apmg_wake_up_wa &&
1264 	    !trans_pcie->cmd_hold_nic_awake) {
1265 		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1266 					 BIT(cfg->csr->flag_mac_access_req));
1267 
1268 		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1269 				   BIT(cfg->csr->flag_val_mac_access_en),
1270 				   (BIT(cfg->csr->flag_mac_clock_ready) |
1271 				    CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1272 				   15000);
1273 		if (ret < 0) {
1274 			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1275 					BIT(cfg->csr->flag_mac_access_req));
1276 			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1277 			return -EIO;
1278 		}
1279 		trans_pcie->cmd_hold_nic_awake = true;
1280 	}
1281 
1282 	return 0;
1283 }
1284 
1285 /*
1286  * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1287  *
1288  * When FW advances 'R' index, all entries between old and new 'R' index
1289  * need to be reclaimed. As result, some free space forms.  If there is
1290  * enough free space (> low mark), wake the stack that feeds us.
1291  */
1292 void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1293 {
1294 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1295 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1296 	unsigned long flags;
1297 	int nfreed = 0;
1298 	u16 r;
1299 
1300 	lockdep_assert_held(&txq->lock);
1301 
1302 	idx = iwl_pcie_get_cmd_index(txq, idx);
1303 	r = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1304 
1305 	if (idx >= trans->cfg->base_params->max_tfd_queue_size ||
1306 	    (!iwl_queue_used(txq, idx))) {
1307 		WARN_ONCE(test_bit(txq_id, trans_pcie->queue_used),
1308 			  "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1309 			  __func__, txq_id, idx,
1310 			  trans->cfg->base_params->max_tfd_queue_size,
1311 			  txq->write_ptr, txq->read_ptr);
1312 		return;
1313 	}
1314 
1315 	for (idx = iwl_queue_inc_wrap(trans, idx); r != idx;
1316 	     r = iwl_queue_inc_wrap(trans, r)) {
1317 		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
1318 
1319 		if (nfreed++ > 0) {
1320 			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1321 				idx, txq->write_ptr, r);
1322 			iwl_force_nmi(trans);
1323 		}
1324 	}
1325 
1326 	if (txq->read_ptr == txq->write_ptr) {
1327 		spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1328 		iwl_pcie_clear_cmd_in_flight(trans);
1329 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1330 	}
1331 
1332 	iwl_pcie_txq_progress(txq);
1333 }
1334 
1335 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1336 				 u16 txq_id)
1337 {
1338 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1339 	u32 tbl_dw_addr;
1340 	u32 tbl_dw;
1341 	u16 scd_q2ratid;
1342 
1343 	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1344 
1345 	tbl_dw_addr = trans_pcie->scd_base_addr +
1346 			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1347 
1348 	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1349 
1350 	if (txq_id & 0x1)
1351 		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1352 	else
1353 		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1354 
1355 	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1356 
1357 	return 0;
1358 }
1359 
1360 /* Receiver address (actually, Rx station's index into station table),
1361  * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1362 #define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
1363 
1364 bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1365 			       const struct iwl_trans_txq_scd_cfg *cfg,
1366 			       unsigned int wdg_timeout)
1367 {
1368 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1369 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1370 	int fifo = -1;
1371 	bool scd_bug = false;
1372 
1373 	if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1374 		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1375 
1376 	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1377 
1378 	if (cfg) {
1379 		fifo = cfg->fifo;
1380 
1381 		/* Disable the scheduler prior configuring the cmd queue */
1382 		if (txq_id == trans_pcie->cmd_queue &&
1383 		    trans_pcie->scd_set_active)
1384 			iwl_scd_enable_set_active(trans, 0);
1385 
1386 		/* Stop this Tx queue before configuring it */
1387 		iwl_scd_txq_set_inactive(trans, txq_id);
1388 
1389 		/* Set this queue as a chain-building queue unless it is CMD */
1390 		if (txq_id != trans_pcie->cmd_queue)
1391 			iwl_scd_txq_set_chain(trans, txq_id);
1392 
1393 		if (cfg->aggregate) {
1394 			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1395 
1396 			/* Map receiver-address / traffic-ID to this queue */
1397 			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1398 
1399 			/* enable aggregations for the queue */
1400 			iwl_scd_txq_enable_agg(trans, txq_id);
1401 			txq->ampdu = true;
1402 		} else {
1403 			/*
1404 			 * disable aggregations for the queue, this will also
1405 			 * make the ra_tid mapping configuration irrelevant
1406 			 * since it is now a non-AGG queue.
1407 			 */
1408 			iwl_scd_txq_disable_agg(trans, txq_id);
1409 
1410 			ssn = txq->read_ptr;
1411 		}
1412 	} else {
1413 		/*
1414 		 * If we need to move the SCD write pointer by steps of
1415 		 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let
1416 		 * the op_mode know by returning true later.
1417 		 * Do this only in case cfg is NULL since this trick can
1418 		 * be done only if we have DQA enabled which is true for mvm
1419 		 * only. And mvm never sets a cfg pointer.
1420 		 * This is really ugly, but this is the easiest way out for
1421 		 * this sad hardware issue.
1422 		 * This bug has been fixed on devices 9000 and up.
1423 		 */
1424 		scd_bug = !trans->cfg->mq_rx_supported &&
1425 			!((ssn - txq->write_ptr) & 0x3f) &&
1426 			(ssn != txq->write_ptr);
1427 		if (scd_bug)
1428 			ssn++;
1429 	}
1430 
1431 	/* Place first TFD at index corresponding to start sequence number.
1432 	 * Assumes that ssn_idx is valid (!= 0xFFF) */
1433 	txq->read_ptr = (ssn & 0xff);
1434 	txq->write_ptr = (ssn & 0xff);
1435 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1436 			   (ssn & 0xff) | (txq_id << 8));
1437 
1438 	if (cfg) {
1439 		u8 frame_limit = cfg->frame_limit;
1440 
1441 		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1442 
1443 		/* Set up Tx window size and frame limit for this queue */
1444 		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1445 				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1446 		iwl_trans_write_mem32(trans,
1447 			trans_pcie->scd_base_addr +
1448 			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1449 			SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) |
1450 			SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit));
1451 
1452 		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1453 		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1454 			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1455 			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1456 			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1457 			       SCD_QUEUE_STTS_REG_MSK);
1458 
1459 		/* enable the scheduler for this queue (only) */
1460 		if (txq_id == trans_pcie->cmd_queue &&
1461 		    trans_pcie->scd_set_active)
1462 			iwl_scd_enable_set_active(trans, BIT(txq_id));
1463 
1464 		IWL_DEBUG_TX_QUEUES(trans,
1465 				    "Activate queue %d on FIFO %d WrPtr: %d\n",
1466 				    txq_id, fifo, ssn & 0xff);
1467 	} else {
1468 		IWL_DEBUG_TX_QUEUES(trans,
1469 				    "Activate queue %d WrPtr: %d\n",
1470 				    txq_id, ssn & 0xff);
1471 	}
1472 
1473 	return scd_bug;
1474 }
1475 
1476 void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
1477 					bool shared_mode)
1478 {
1479 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1480 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1481 
1482 	txq->ampdu = !shared_mode;
1483 }
1484 
1485 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1486 				bool configure_scd)
1487 {
1488 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1489 	u32 stts_addr = trans_pcie->scd_base_addr +
1490 			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1491 	static const u32 zero_val[4] = {};
1492 
1493 	trans_pcie->txq[txq_id]->frozen_expiry_remainder = 0;
1494 	trans_pcie->txq[txq_id]->frozen = false;
1495 
1496 	/*
1497 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1498 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1499 	 * allow the op_mode to call txq_disable after it already called
1500 	 * stop_device.
1501 	 */
1502 	if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1503 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1504 			  "queue %d not used", txq_id);
1505 		return;
1506 	}
1507 
1508 	if (configure_scd) {
1509 		iwl_scd_txq_set_inactive(trans, txq_id);
1510 
1511 		iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1512 				    ARRAY_SIZE(zero_val));
1513 	}
1514 
1515 	iwl_pcie_txq_unmap(trans, txq_id);
1516 	trans_pcie->txq[txq_id]->ampdu = false;
1517 
1518 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1519 }
1520 
1521 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1522 
1523 /*
1524  * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1525  * @priv: device private data point
1526  * @cmd: a pointer to the ucode command structure
1527  *
1528  * The function returns < 0 values to indicate the operation
1529  * failed. On success, it returns the index (>= 0) of command in the
1530  * command queue.
1531  */
1532 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1533 				 struct iwl_host_cmd *cmd)
1534 {
1535 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1536 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1537 	struct iwl_device_cmd *out_cmd;
1538 	struct iwl_cmd_meta *out_meta;
1539 	unsigned long flags;
1540 	void *dup_buf = NULL;
1541 	dma_addr_t phys_addr;
1542 	int idx;
1543 	u16 copy_size, cmd_size, tb0_size;
1544 	bool had_nocopy = false;
1545 	u8 group_id = iwl_cmd_groupid(cmd->id);
1546 	int i, ret;
1547 	u32 cmd_pos;
1548 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1549 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1550 
1551 	if (WARN(!trans->wide_cmd_header &&
1552 		 group_id > IWL_ALWAYS_LONG_GROUP,
1553 		 "unsupported wide command %#x\n", cmd->id))
1554 		return -EINVAL;
1555 
1556 	if (group_id != 0) {
1557 		copy_size = sizeof(struct iwl_cmd_header_wide);
1558 		cmd_size = sizeof(struct iwl_cmd_header_wide);
1559 	} else {
1560 		copy_size = sizeof(struct iwl_cmd_header);
1561 		cmd_size = sizeof(struct iwl_cmd_header);
1562 	}
1563 
1564 	/* need one for the header if the first is NOCOPY */
1565 	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1566 
1567 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1568 		cmddata[i] = cmd->data[i];
1569 		cmdlen[i] = cmd->len[i];
1570 
1571 		if (!cmd->len[i])
1572 			continue;
1573 
1574 		/* need at least IWL_FIRST_TB_SIZE copied */
1575 		if (copy_size < IWL_FIRST_TB_SIZE) {
1576 			int copy = IWL_FIRST_TB_SIZE - copy_size;
1577 
1578 			if (copy > cmdlen[i])
1579 				copy = cmdlen[i];
1580 			cmdlen[i] -= copy;
1581 			cmddata[i] += copy;
1582 			copy_size += copy;
1583 		}
1584 
1585 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1586 			had_nocopy = true;
1587 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1588 				idx = -EINVAL;
1589 				goto free_dup_buf;
1590 			}
1591 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1592 			/*
1593 			 * This is also a chunk that isn't copied
1594 			 * to the static buffer so set had_nocopy.
1595 			 */
1596 			had_nocopy = true;
1597 
1598 			/* only allowed once */
1599 			if (WARN_ON(dup_buf)) {
1600 				idx = -EINVAL;
1601 				goto free_dup_buf;
1602 			}
1603 
1604 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
1605 					  GFP_ATOMIC);
1606 			if (!dup_buf)
1607 				return -ENOMEM;
1608 		} else {
1609 			/* NOCOPY must not be followed by normal! */
1610 			if (WARN_ON(had_nocopy)) {
1611 				idx = -EINVAL;
1612 				goto free_dup_buf;
1613 			}
1614 			copy_size += cmdlen[i];
1615 		}
1616 		cmd_size += cmd->len[i];
1617 	}
1618 
1619 	/*
1620 	 * If any of the command structures end up being larger than
1621 	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1622 	 * allocated into separate TFDs, then we will need to
1623 	 * increase the size of the buffers.
1624 	 */
1625 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1626 		 "Command %s (%#x) is too large (%d bytes)\n",
1627 		 iwl_get_cmd_string(trans, cmd->id),
1628 		 cmd->id, copy_size)) {
1629 		idx = -EINVAL;
1630 		goto free_dup_buf;
1631 	}
1632 
1633 	spin_lock_bh(&txq->lock);
1634 
1635 	if (iwl_queue_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1636 		spin_unlock_bh(&txq->lock);
1637 
1638 		IWL_ERR(trans, "No space in command queue\n");
1639 		iwl_op_mode_cmd_queue_full(trans->op_mode);
1640 		idx = -ENOSPC;
1641 		goto free_dup_buf;
1642 	}
1643 
1644 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
1645 	out_cmd = txq->entries[idx].cmd;
1646 	out_meta = &txq->entries[idx].meta;
1647 
1648 	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1649 	if (cmd->flags & CMD_WANT_SKB)
1650 		out_meta->source = cmd;
1651 
1652 	/* set up the header */
1653 	if (group_id != 0) {
1654 		out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1655 		out_cmd->hdr_wide.group_id = group_id;
1656 		out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1657 		out_cmd->hdr_wide.length =
1658 			cpu_to_le16(cmd_size -
1659 				    sizeof(struct iwl_cmd_header_wide));
1660 		out_cmd->hdr_wide.reserved = 0;
1661 		out_cmd->hdr_wide.sequence =
1662 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1663 						 INDEX_TO_SEQ(txq->write_ptr));
1664 
1665 		cmd_pos = sizeof(struct iwl_cmd_header_wide);
1666 		copy_size = sizeof(struct iwl_cmd_header_wide);
1667 	} else {
1668 		out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1669 		out_cmd->hdr.sequence =
1670 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1671 						 INDEX_TO_SEQ(txq->write_ptr));
1672 		out_cmd->hdr.group_id = 0;
1673 
1674 		cmd_pos = sizeof(struct iwl_cmd_header);
1675 		copy_size = sizeof(struct iwl_cmd_header);
1676 	}
1677 
1678 	/* and copy the data that needs to be copied */
1679 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1680 		int copy;
1681 
1682 		if (!cmd->len[i])
1683 			continue;
1684 
1685 		/* copy everything if not nocopy/dup */
1686 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1687 					   IWL_HCMD_DFL_DUP))) {
1688 			copy = cmd->len[i];
1689 
1690 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1691 			cmd_pos += copy;
1692 			copy_size += copy;
1693 			continue;
1694 		}
1695 
1696 		/*
1697 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1698 		 * in total (for bi-directional DMA), but copy up to what
1699 		 * we can fit into the payload for debug dump purposes.
1700 		 */
1701 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1702 
1703 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1704 		cmd_pos += copy;
1705 
1706 		/* However, treat copy_size the proper way, we need it below */
1707 		if (copy_size < IWL_FIRST_TB_SIZE) {
1708 			copy = IWL_FIRST_TB_SIZE - copy_size;
1709 
1710 			if (copy > cmd->len[i])
1711 				copy = cmd->len[i];
1712 			copy_size += copy;
1713 		}
1714 	}
1715 
1716 	IWL_DEBUG_HC(trans,
1717 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1718 		     iwl_get_cmd_string(trans, cmd->id),
1719 		     group_id, out_cmd->hdr.cmd,
1720 		     le16_to_cpu(out_cmd->hdr.sequence),
1721 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
1722 
1723 	/* start the TFD with the minimum copy bytes */
1724 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1725 	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1726 	iwl_pcie_txq_build_tfd(trans, txq,
1727 			       iwl_pcie_get_first_tb_dma(txq, idx),
1728 			       tb0_size, true);
1729 
1730 	/* map first command fragment, if any remains */
1731 	if (copy_size > tb0_size) {
1732 		phys_addr = dma_map_single(trans->dev,
1733 					   ((u8 *)&out_cmd->hdr) + tb0_size,
1734 					   copy_size - tb0_size,
1735 					   DMA_TO_DEVICE);
1736 		if (dma_mapping_error(trans->dev, phys_addr)) {
1737 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1738 					   txq->write_ptr);
1739 			idx = -ENOMEM;
1740 			goto out;
1741 		}
1742 
1743 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1744 				       copy_size - tb0_size, false);
1745 	}
1746 
1747 	/* map the remaining (adjusted) nocopy/dup fragments */
1748 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1749 		const void *data = cmddata[i];
1750 
1751 		if (!cmdlen[i])
1752 			continue;
1753 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1754 					   IWL_HCMD_DFL_DUP)))
1755 			continue;
1756 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1757 			data = dup_buf;
1758 		phys_addr = dma_map_single(trans->dev, (void *)data,
1759 					   cmdlen[i], DMA_TO_DEVICE);
1760 		if (dma_mapping_error(trans->dev, phys_addr)) {
1761 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1762 					   txq->write_ptr);
1763 			idx = -ENOMEM;
1764 			goto out;
1765 		}
1766 
1767 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1768 	}
1769 
1770 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
1771 	out_meta->flags = cmd->flags;
1772 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1773 		kzfree(txq->entries[idx].free_buf);
1774 	txq->entries[idx].free_buf = dup_buf;
1775 
1776 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1777 
1778 	/* start timer if queue currently empty */
1779 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
1780 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1781 
1782 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1783 	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1784 	if (ret < 0) {
1785 		idx = ret;
1786 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1787 		goto out;
1788 	}
1789 
1790 	/* Increment and update queue's write index */
1791 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
1792 	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1793 
1794 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1795 
1796  out:
1797 	spin_unlock_bh(&txq->lock);
1798  free_dup_buf:
1799 	if (idx < 0)
1800 		kfree(dup_buf);
1801 	return idx;
1802 }
1803 
1804 /*
1805  * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1806  * @rxb: Rx buffer to reclaim
1807  */
1808 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1809 			    struct iwl_rx_cmd_buffer *rxb)
1810 {
1811 	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1812 	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1813 	u8 group_id;
1814 	u32 cmd_id;
1815 	int txq_id = SEQ_TO_QUEUE(sequence);
1816 	int index = SEQ_TO_INDEX(sequence);
1817 	int cmd_index;
1818 	struct iwl_device_cmd *cmd;
1819 	struct iwl_cmd_meta *meta;
1820 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1821 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1822 
1823 	/* If a Tx command is being handled and it isn't in the actual
1824 	 * command queue then there a command routing bug has been introduced
1825 	 * in the queue management code. */
1826 	if (WARN(txq_id != trans_pcie->cmd_queue,
1827 		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1828 		 txq_id, trans_pcie->cmd_queue, sequence, txq->read_ptr,
1829 		 txq->write_ptr)) {
1830 		iwl_print_hex_error(trans, pkt, 32);
1831 		return;
1832 	}
1833 
1834 	spin_lock_bh(&txq->lock);
1835 
1836 	cmd_index = iwl_pcie_get_cmd_index(txq, index);
1837 	cmd = txq->entries[cmd_index].cmd;
1838 	meta = &txq->entries[cmd_index].meta;
1839 	group_id = cmd->hdr.group_id;
1840 	cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1841 
1842 	iwl_pcie_tfd_unmap(trans, meta, txq, index);
1843 
1844 	/* Input error checking is done when commands are added to queue. */
1845 	if (meta->flags & CMD_WANT_SKB) {
1846 		struct page *p = rxb_steal_page(rxb);
1847 
1848 		meta->source->resp_pkt = pkt;
1849 		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1850 		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1851 	}
1852 
1853 	if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1854 		iwl_op_mode_async_cb(trans->op_mode, cmd);
1855 
1856 	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1857 
1858 	if (!(meta->flags & CMD_ASYNC)) {
1859 		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1860 			IWL_WARN(trans,
1861 				 "HCMD_ACTIVE already clear for command %s\n",
1862 				 iwl_get_cmd_string(trans, cmd_id));
1863 		}
1864 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1865 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1866 			       iwl_get_cmd_string(trans, cmd_id));
1867 		wake_up(&trans_pcie->wait_command_queue);
1868 	}
1869 
1870 	if (meta->flags & CMD_MAKE_TRANS_IDLE) {
1871 		IWL_DEBUG_INFO(trans, "complete %s - mark trans as idle\n",
1872 			       iwl_get_cmd_string(trans, cmd->hdr.cmd));
1873 		set_bit(STATUS_TRANS_IDLE, &trans->status);
1874 		wake_up(&trans_pcie->d0i3_waitq);
1875 	}
1876 
1877 	if (meta->flags & CMD_WAKE_UP_TRANS) {
1878 		IWL_DEBUG_INFO(trans, "complete %s - clear trans idle flag\n",
1879 			       iwl_get_cmd_string(trans, cmd->hdr.cmd));
1880 		clear_bit(STATUS_TRANS_IDLE, &trans->status);
1881 		wake_up(&trans_pcie->d0i3_waitq);
1882 	}
1883 
1884 	meta->flags = 0;
1885 
1886 	spin_unlock_bh(&txq->lock);
1887 }
1888 
1889 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1890 
1891 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1892 				    struct iwl_host_cmd *cmd)
1893 {
1894 	int ret;
1895 
1896 	/* An asynchronous command can not expect an SKB to be set. */
1897 	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1898 		return -EINVAL;
1899 
1900 	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1901 	if (ret < 0) {
1902 		IWL_ERR(trans,
1903 			"Error sending %s: enqueue_hcmd failed: %d\n",
1904 			iwl_get_cmd_string(trans, cmd->id), ret);
1905 		return ret;
1906 	}
1907 	return 0;
1908 }
1909 
1910 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1911 				   struct iwl_host_cmd *cmd)
1912 {
1913 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1914 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1915 	int cmd_idx;
1916 	int ret;
1917 
1918 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1919 		       iwl_get_cmd_string(trans, cmd->id));
1920 
1921 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1922 				  &trans->status),
1923 		 "Command %s: a command is already active!\n",
1924 		 iwl_get_cmd_string(trans, cmd->id)))
1925 		return -EIO;
1926 
1927 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1928 		       iwl_get_cmd_string(trans, cmd->id));
1929 
1930 	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
1931 		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
1932 				 pm_runtime_active(&trans_pcie->pci_dev->dev),
1933 				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
1934 		if (!ret) {
1935 			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
1936 			return -ETIMEDOUT;
1937 		}
1938 	}
1939 
1940 	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1941 	if (cmd_idx < 0) {
1942 		ret = cmd_idx;
1943 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1944 		IWL_ERR(trans,
1945 			"Error sending %s: enqueue_hcmd failed: %d\n",
1946 			iwl_get_cmd_string(trans, cmd->id), ret);
1947 		return ret;
1948 	}
1949 
1950 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
1951 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1952 					   &trans->status),
1953 				 HOST_COMPLETE_TIMEOUT);
1954 	if (!ret) {
1955 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1956 			iwl_get_cmd_string(trans, cmd->id),
1957 			jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1958 
1959 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1960 			txq->read_ptr, txq->write_ptr);
1961 
1962 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1963 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1964 			       iwl_get_cmd_string(trans, cmd->id));
1965 		ret = -ETIMEDOUT;
1966 
1967 		iwl_trans_pcie_sync_nmi(trans);
1968 		goto cancel;
1969 	}
1970 
1971 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1972 		iwl_trans_pcie_dump_regs(trans);
1973 		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1974 			iwl_get_cmd_string(trans, cmd->id));
1975 		dump_stack();
1976 		ret = -EIO;
1977 		goto cancel;
1978 	}
1979 
1980 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1981 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1982 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1983 		ret = -ERFKILL;
1984 		goto cancel;
1985 	}
1986 
1987 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1988 		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1989 			iwl_get_cmd_string(trans, cmd->id));
1990 		ret = -EIO;
1991 		goto cancel;
1992 	}
1993 
1994 	return 0;
1995 
1996 cancel:
1997 	if (cmd->flags & CMD_WANT_SKB) {
1998 		/*
1999 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
2000 		 * TX cmd queue. Otherwise in case the cmd comes
2001 		 * in later, it will possibly set an invalid
2002 		 * address (cmd->meta.source).
2003 		 */
2004 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
2005 	}
2006 
2007 	if (cmd->resp_pkt) {
2008 		iwl_free_resp(cmd);
2009 		cmd->resp_pkt = NULL;
2010 	}
2011 
2012 	return ret;
2013 }
2014 
2015 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
2016 {
2017 	/* Make sure the NIC is still alive in the bus */
2018 	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
2019 		return -ENODEV;
2020 
2021 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
2022 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
2023 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
2024 				  cmd->id);
2025 		return -ERFKILL;
2026 	}
2027 
2028 	if (cmd->flags & CMD_ASYNC)
2029 		return iwl_pcie_send_hcmd_async(trans, cmd);
2030 
2031 	/* We still can fail on RFKILL that can be asserted while we wait */
2032 	return iwl_pcie_send_hcmd_sync(trans, cmd);
2033 }
2034 
2035 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
2036 			     struct iwl_txq *txq, u8 hdr_len,
2037 			     struct iwl_cmd_meta *out_meta)
2038 {
2039 	u16 head_tb_len;
2040 	int i;
2041 
2042 	/*
2043 	 * Set up TFD's third entry to point directly to remainder
2044 	 * of skb's head, if any
2045 	 */
2046 	head_tb_len = skb_headlen(skb) - hdr_len;
2047 
2048 	if (head_tb_len > 0) {
2049 		dma_addr_t tb_phys = dma_map_single(trans->dev,
2050 						    skb->data + hdr_len,
2051 						    head_tb_len, DMA_TO_DEVICE);
2052 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
2053 			return -EINVAL;
2054 		trace_iwlwifi_dev_tx_tb(trans->dev, skb,
2055 					skb->data + hdr_len,
2056 					head_tb_len);
2057 		iwl_pcie_txq_build_tfd(trans, txq, tb_phys, head_tb_len, false);
2058 	}
2059 
2060 	/* set up the remaining entries to point to the data */
2061 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2062 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2063 		dma_addr_t tb_phys;
2064 		int tb_idx;
2065 
2066 		if (!skb_frag_size(frag))
2067 			continue;
2068 
2069 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
2070 					   skb_frag_size(frag), DMA_TO_DEVICE);
2071 
2072 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
2073 			return -EINVAL;
2074 		trace_iwlwifi_dev_tx_tb(trans->dev, skb,
2075 					skb_frag_address(frag),
2076 					skb_frag_size(frag));
2077 		tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2078 						skb_frag_size(frag), false);
2079 		if (tb_idx < 0)
2080 			return tb_idx;
2081 
2082 		out_meta->tbs |= BIT(tb_idx);
2083 	}
2084 
2085 	return 0;
2086 }
2087 
2088 #ifdef CONFIG_INET
2089 struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len)
2090 {
2091 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2092 	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
2093 
2094 	if (!p->page)
2095 		goto alloc;
2096 
2097 	/* enough room on this page */
2098 	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
2099 		return p;
2100 
2101 	/* We don't have enough room on this page, get a new one. */
2102 	__free_page(p->page);
2103 
2104 alloc:
2105 	p->page = alloc_page(GFP_ATOMIC);
2106 	if (!p->page)
2107 		return NULL;
2108 	p->pos = page_address(p->page);
2109 	return p;
2110 }
2111 
2112 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
2113 					bool ipv6, unsigned int len)
2114 {
2115 	if (ipv6) {
2116 		struct ipv6hdr *iphv6 = iph;
2117 
2118 		tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
2119 					       len + tcph->doff * 4,
2120 					       IPPROTO_TCP, 0);
2121 	} else {
2122 		struct iphdr *iphv4 = iph;
2123 
2124 		ip_send_check(iphv4);
2125 		tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
2126 						 len + tcph->doff * 4,
2127 						 IPPROTO_TCP, 0);
2128 	}
2129 }
2130 
2131 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2132 				   struct iwl_txq *txq, u8 hdr_len,
2133 				   struct iwl_cmd_meta *out_meta,
2134 				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2135 {
2136 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
2137 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
2138 	struct ieee80211_hdr *hdr = (void *)skb->data;
2139 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
2140 	unsigned int mss = skb_shinfo(skb)->gso_size;
2141 	u16 length, iv_len, amsdu_pad;
2142 	u8 *start_hdr;
2143 	struct iwl_tso_hdr_page *hdr_page;
2144 	struct page **page_ptr;
2145 	struct tso_t tso;
2146 
2147 	/* if the packet is protected, then it must be CCMP or GCMP */
2148 	BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
2149 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
2150 		IEEE80211_CCMP_HDR_LEN : 0;
2151 
2152 	trace_iwlwifi_dev_tx(trans->dev, skb,
2153 			     iwl_pcie_get_tfd(trans, txq, txq->write_ptr),
2154 			     trans_pcie->tfd_size,
2155 			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0);
2156 
2157 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
2158 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
2159 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
2160 	amsdu_pad = 0;
2161 
2162 	/* total amount of header we may need for this A-MSDU */
2163 	hdr_room = DIV_ROUND_UP(total_len, mss) *
2164 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
2165 
2166 	/* Our device supports 9 segments at most, it will fit in 1 page */
2167 	hdr_page = get_page_hdr(trans, hdr_room);
2168 	if (!hdr_page)
2169 		return -ENOMEM;
2170 
2171 	get_page(hdr_page->page);
2172 	start_hdr = hdr_page->pos;
2173 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
2174 	*page_ptr = hdr_page->page;
2175 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
2176 	hdr_page->pos += iv_len;
2177 
2178 	/*
2179 	 * Pull the ieee80211 header + IV to be able to use TSO core,
2180 	 * we will restore it for the tx_status flow.
2181 	 */
2182 	skb_pull(skb, hdr_len + iv_len);
2183 
2184 	/*
2185 	 * Remove the length of all the headers that we don't actually
2186 	 * have in the MPDU by themselves, but that we duplicate into
2187 	 * all the different MSDUs inside the A-MSDU.
2188 	 */
2189 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
2190 
2191 	tso_start(skb, &tso);
2192 
2193 	while (total_len) {
2194 		/* this is the data left for this subframe */
2195 		unsigned int data_left =
2196 			min_t(unsigned int, mss, total_len);
2197 		struct sk_buff *csum_skb = NULL;
2198 		unsigned int hdr_tb_len;
2199 		dma_addr_t hdr_tb_phys;
2200 		struct tcphdr *tcph;
2201 		u8 *iph, *subf_hdrs_start = hdr_page->pos;
2202 
2203 		total_len -= data_left;
2204 
2205 		memset(hdr_page->pos, 0, amsdu_pad);
2206 		hdr_page->pos += amsdu_pad;
2207 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2208 				  data_left)) & 0x3;
2209 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2210 		hdr_page->pos += ETH_ALEN;
2211 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2212 		hdr_page->pos += ETH_ALEN;
2213 
2214 		length = snap_ip_tcp_hdrlen + data_left;
2215 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2216 		hdr_page->pos += sizeof(length);
2217 
2218 		/*
2219 		 * This will copy the SNAP as well which will be considered
2220 		 * as MAC header.
2221 		 */
2222 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2223 		iph = hdr_page->pos + 8;
2224 		tcph = (void *)(iph + ip_hdrlen);
2225 
2226 		/* For testing on current hardware only */
2227 		if (trans_pcie->sw_csum_tx) {
2228 			csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2229 					     GFP_ATOMIC);
2230 			if (!csum_skb)
2231 				return -ENOMEM;
2232 
2233 			iwl_compute_pseudo_hdr_csum(iph, tcph,
2234 						    skb->protocol ==
2235 							htons(ETH_P_IPV6),
2236 						    data_left);
2237 
2238 			skb_put_data(csum_skb, tcph, tcp_hdrlen(skb));
2239 			skb_reset_transport_header(csum_skb);
2240 			csum_skb->csum_start =
2241 				(unsigned char *)tcp_hdr(csum_skb) -
2242 						 csum_skb->head;
2243 		}
2244 
2245 		hdr_page->pos += snap_ip_tcp_hdrlen;
2246 
2247 		hdr_tb_len = hdr_page->pos - start_hdr;
2248 		hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2249 					     hdr_tb_len, DMA_TO_DEVICE);
2250 		if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2251 			dev_kfree_skb(csum_skb);
2252 			return -EINVAL;
2253 		}
2254 		iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2255 				       hdr_tb_len, false);
2256 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
2257 					hdr_tb_len);
2258 		/* add this subframe's headers' length to the tx_cmd */
2259 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
2260 
2261 		/* prepare the start_hdr for the next subframe */
2262 		start_hdr = hdr_page->pos;
2263 
2264 		/* put the payload */
2265 		while (data_left) {
2266 			unsigned int size = min_t(unsigned int, tso.size,
2267 						  data_left);
2268 			dma_addr_t tb_phys;
2269 
2270 			if (trans_pcie->sw_csum_tx)
2271 				skb_put_data(csum_skb, tso.data, size);
2272 
2273 			tb_phys = dma_map_single(trans->dev, tso.data,
2274 						 size, DMA_TO_DEVICE);
2275 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2276 				dev_kfree_skb(csum_skb);
2277 				return -EINVAL;
2278 			}
2279 
2280 			iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2281 					       size, false);
2282 			trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data,
2283 						size);
2284 
2285 			data_left -= size;
2286 			tso_build_data(skb, &tso, size);
2287 		}
2288 
2289 		/* For testing on early hardware only */
2290 		if (trans_pcie->sw_csum_tx) {
2291 			__wsum csum;
2292 
2293 			csum = skb_checksum(csum_skb,
2294 					    skb_checksum_start_offset(csum_skb),
2295 					    csum_skb->len -
2296 					    skb_checksum_start_offset(csum_skb),
2297 					    0);
2298 			dev_kfree_skb(csum_skb);
2299 			dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2300 						hdr_tb_len, DMA_TO_DEVICE);
2301 			tcph->check = csum_fold(csum);
2302 			dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2303 						   hdr_tb_len, DMA_TO_DEVICE);
2304 		}
2305 	}
2306 
2307 	/* re -add the WiFi header and IV */
2308 	skb_push(skb, hdr_len + iv_len);
2309 
2310 	return 0;
2311 }
2312 #else /* CONFIG_INET */
2313 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2314 				   struct iwl_txq *txq, u8 hdr_len,
2315 				   struct iwl_cmd_meta *out_meta,
2316 				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2317 {
2318 	/* No A-MSDU without CONFIG_INET */
2319 	WARN_ON(1);
2320 
2321 	return -1;
2322 }
2323 #endif /* CONFIG_INET */
2324 
2325 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2326 		      struct iwl_device_cmd *dev_cmd, int txq_id)
2327 {
2328 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2329 	struct ieee80211_hdr *hdr;
2330 	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2331 	struct iwl_cmd_meta *out_meta;
2332 	struct iwl_txq *txq;
2333 	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2334 	void *tb1_addr;
2335 	void *tfd;
2336 	u16 len, tb1_len;
2337 	bool wait_write_ptr;
2338 	__le16 fc;
2339 	u8 hdr_len;
2340 	u16 wifi_seq;
2341 	bool amsdu;
2342 
2343 	txq = trans_pcie->txq[txq_id];
2344 
2345 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2346 		      "TX on unused queue %d\n", txq_id))
2347 		return -EINVAL;
2348 
2349 	if (unlikely(trans_pcie->sw_csum_tx &&
2350 		     skb->ip_summed == CHECKSUM_PARTIAL)) {
2351 		int offs = skb_checksum_start_offset(skb);
2352 		int csum_offs = offs + skb->csum_offset;
2353 		__wsum csum;
2354 
2355 		if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2356 			return -1;
2357 
2358 		csum = skb_checksum(skb, offs, skb->len - offs, 0);
2359 		*(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2360 
2361 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2362 	}
2363 
2364 	if (skb_is_nonlinear(skb) &&
2365 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
2366 	    __skb_linearize(skb))
2367 		return -ENOMEM;
2368 
2369 	/* mac80211 always puts the full header into the SKB's head,
2370 	 * so there's no need to check if it's readable there
2371 	 */
2372 	hdr = (struct ieee80211_hdr *)skb->data;
2373 	fc = hdr->frame_control;
2374 	hdr_len = ieee80211_hdrlen(fc);
2375 
2376 	spin_lock(&txq->lock);
2377 
2378 	if (iwl_queue_space(trans, txq) < txq->high_mark) {
2379 		iwl_stop_queue(trans, txq);
2380 
2381 		/* don't put the packet on the ring, if there is no room */
2382 		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
2383 			struct iwl_device_cmd **dev_cmd_ptr;
2384 
2385 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
2386 					       trans_pcie->dev_cmd_offs);
2387 
2388 			*dev_cmd_ptr = dev_cmd;
2389 			__skb_queue_tail(&txq->overflow_q, skb);
2390 
2391 			spin_unlock(&txq->lock);
2392 			return 0;
2393 		}
2394 	}
2395 
2396 	/* In AGG mode, the index in the ring must correspond to the WiFi
2397 	 * sequence number. This is a HW requirements to help the SCD to parse
2398 	 * the BA.
2399 	 * Check here that the packets are in the right place on the ring.
2400 	 */
2401 	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2402 	WARN_ONCE(txq->ampdu &&
2403 		  (wifi_seq & 0xff) != txq->write_ptr,
2404 		  "Q: %d WiFi Seq %d tfdNum %d",
2405 		  txq_id, wifi_seq, txq->write_ptr);
2406 
2407 	/* Set up driver data for this TFD */
2408 	txq->entries[txq->write_ptr].skb = skb;
2409 	txq->entries[txq->write_ptr].cmd = dev_cmd;
2410 
2411 	dev_cmd->hdr.sequence =
2412 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2413 			    INDEX_TO_SEQ(txq->write_ptr)));
2414 
2415 	tb0_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
2416 	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2417 		       offsetof(struct iwl_tx_cmd, scratch);
2418 
2419 	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2420 	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2421 
2422 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
2423 	out_meta = &txq->entries[txq->write_ptr].meta;
2424 	out_meta->flags = 0;
2425 
2426 	/*
2427 	 * The second TB (tb1) points to the remainder of the TX command
2428 	 * and the 802.11 header - dword aligned size
2429 	 * (This calculation modifies the TX command, so do it before the
2430 	 * setup of the first TB)
2431 	 */
2432 	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2433 	      hdr_len - IWL_FIRST_TB_SIZE;
2434 	/* do not align A-MSDU to dword as the subframe header aligns it */
2435 	amsdu = ieee80211_is_data_qos(fc) &&
2436 		(*ieee80211_get_qos_ctl(hdr) &
2437 		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
2438 	if (trans_pcie->sw_csum_tx || !amsdu) {
2439 		tb1_len = ALIGN(len, 4);
2440 		/* Tell NIC about any 2-byte padding after MAC header */
2441 		if (tb1_len != len)
2442 			tx_cmd->tx_flags |= cpu_to_le32(TX_CMD_FLG_MH_PAD);
2443 	} else {
2444 		tb1_len = len;
2445 	}
2446 
2447 	/*
2448 	 * The first TB points to bi-directional DMA data, we'll
2449 	 * memcpy the data into it later.
2450 	 */
2451 	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2452 			       IWL_FIRST_TB_SIZE, true);
2453 
2454 	/* there must be data left over for TB1 or this code must be changed */
2455 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
2456 
2457 	/* map the data for TB1 */
2458 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
2459 	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2460 	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2461 		goto out_err;
2462 	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2463 
2464 	trace_iwlwifi_dev_tx(trans->dev, skb,
2465 			     iwl_pcie_get_tfd(trans, txq,
2466 					      txq->write_ptr),
2467 			     trans_pcie->tfd_size,
2468 			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
2469 			     hdr_len);
2470 
2471 	/*
2472 	 * If gso_size wasn't set, don't give the frame "amsdu treatment"
2473 	 * (adding subframes, etc.).
2474 	 * This can happen in some testing flows when the amsdu was already
2475 	 * pre-built, and we just need to send the resulting skb.
2476 	 */
2477 	if (amsdu && skb_shinfo(skb)->gso_size) {
2478 		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2479 						     out_meta, dev_cmd,
2480 						     tb1_len)))
2481 			goto out_err;
2482 	} else {
2483 		struct sk_buff *frag;
2484 
2485 		if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2486 					       out_meta)))
2487 			goto out_err;
2488 
2489 		skb_walk_frags(skb, frag) {
2490 			if (unlikely(iwl_fill_data_tbs(trans, frag, txq, 0,
2491 						       out_meta)))
2492 				goto out_err;
2493 		}
2494 	}
2495 
2496 	/* building the A-MSDU might have changed this data, so memcpy it now */
2497 	memcpy(&txq->first_tb_bufs[txq->write_ptr], dev_cmd, IWL_FIRST_TB_SIZE);
2498 
2499 	tfd = iwl_pcie_get_tfd(trans, txq, txq->write_ptr);
2500 	/* Set up entry for this TFD in Tx byte-count array */
2501 	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len),
2502 					 iwl_pcie_tfd_get_num_tbs(trans, tfd));
2503 
2504 	wait_write_ptr = ieee80211_has_morefrags(fc);
2505 
2506 	/* start timer if queue currently empty */
2507 	if (txq->read_ptr == txq->write_ptr) {
2508 		if (txq->wd_timeout) {
2509 			/*
2510 			 * If the TXQ is active, then set the timer, if not,
2511 			 * set the timer in remainder so that the timer will
2512 			 * be armed with the right value when the station will
2513 			 * wake up.
2514 			 */
2515 			if (!txq->frozen)
2516 				mod_timer(&txq->stuck_timer,
2517 					  jiffies + txq->wd_timeout);
2518 			else
2519 				txq->frozen_expiry_remainder = txq->wd_timeout;
2520 		}
2521 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
2522 		iwl_trans_ref(trans);
2523 	}
2524 
2525 	/* Tell device the write index *just past* this latest filled TFD */
2526 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
2527 	if (!wait_write_ptr)
2528 		iwl_pcie_txq_inc_wr_ptr(trans, txq);
2529 
2530 	/*
2531 	 * At this point the frame is "transmitted" successfully
2532 	 * and we will get a TX status notification eventually.
2533 	 */
2534 	spin_unlock(&txq->lock);
2535 	return 0;
2536 out_err:
2537 	iwl_pcie_tfd_unmap(trans, out_meta, txq, txq->write_ptr);
2538 	spin_unlock(&txq->lock);
2539 	return -1;
2540 }
2541