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 <net/ip6_checksum.h>
69 #include <net/tso.h>
70 
71 #include "iwl-debug.h"
72 #include "iwl-csr.h"
73 #include "iwl-prph.h"
74 #include "iwl-io.h"
75 #include "iwl-scd.h"
76 #include "iwl-op-mode.h"
77 #include "internal.h"
78 #include "fw/api/tx.h"
79 
80 #define IWL_TX_CRC_SIZE 4
81 #define IWL_TX_DELIMITER_SIZE 4
82 
83 /*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
84  * DMA services
85  *
86  * Theory of operation
87  *
88  * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
89  * of buffer descriptors, each of which points to one or more data buffers for
90  * the device to read from or fill.  Driver and device exchange status of each
91  * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
92  * entries in each circular buffer, to protect against confusing empty and full
93  * queue states.
94  *
95  * The device reads or writes the data in the queues via the device's several
96  * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
97  *
98  * For Tx queue, there are low mark and high mark limits. If, after queuing
99  * the packet for Tx, free space become < low mark, Tx queue stopped. When
100  * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
101  * Tx queue resumed.
102  *
103  ***************************************************/
104 
105 int iwl_queue_space(struct iwl_trans *trans, const struct iwl_txq *q)
106 {
107 	unsigned int max;
108 	unsigned int used;
109 
110 	/*
111 	 * To avoid ambiguity between empty and completely full queues, there
112 	 * should always be less than max_tfd_queue_size elements in the queue.
113 	 * If q->n_window is smaller than max_tfd_queue_size, there is no need
114 	 * to reserve any queue entries for this purpose.
115 	 */
116 	if (q->n_window < trans->trans_cfg->base_params->max_tfd_queue_size)
117 		max = q->n_window;
118 	else
119 		max = trans->trans_cfg->base_params->max_tfd_queue_size - 1;
120 
121 	/*
122 	 * max_tfd_queue_size is a power of 2, so the following is equivalent to
123 	 * modulo by max_tfd_queue_size and is well defined.
124 	 */
125 	used = (q->write_ptr - q->read_ptr) &
126 		(trans->trans_cfg->base_params->max_tfd_queue_size - 1);
127 
128 	if (WARN_ON(used > max))
129 		return 0;
130 
131 	return max - used;
132 }
133 
134 /*
135  * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
136  */
137 static int iwl_queue_init(struct iwl_txq *q, int slots_num)
138 {
139 	q->n_window = slots_num;
140 
141 	/* slots_num must be power-of-two size, otherwise
142 	 * iwl_pcie_get_cmd_index is broken. */
143 	if (WARN_ON(!is_power_of_2(slots_num)))
144 		return -EINVAL;
145 
146 	q->low_mark = q->n_window / 4;
147 	if (q->low_mark < 4)
148 		q->low_mark = 4;
149 
150 	q->high_mark = q->n_window / 8;
151 	if (q->high_mark < 2)
152 		q->high_mark = 2;
153 
154 	q->write_ptr = 0;
155 	q->read_ptr = 0;
156 
157 	return 0;
158 }
159 
160 int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
161 			   struct iwl_dma_ptr *ptr, size_t size)
162 {
163 	if (WARN_ON(ptr->addr))
164 		return -EINVAL;
165 
166 	ptr->addr = dma_alloc_coherent(trans->dev, size,
167 				       &ptr->dma, GFP_KERNEL);
168 	if (!ptr->addr)
169 		return -ENOMEM;
170 	ptr->size = size;
171 	return 0;
172 }
173 
174 void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr)
175 {
176 	if (unlikely(!ptr->addr))
177 		return;
178 
179 	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
180 	memset(ptr, 0, sizeof(*ptr));
181 }
182 
183 static void iwl_pcie_txq_stuck_timer(struct timer_list *t)
184 {
185 	struct iwl_txq *txq = from_timer(txq, t, stuck_timer);
186 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
187 	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
188 
189 	spin_lock(&txq->lock);
190 	/* check if triggered erroneously */
191 	if (txq->read_ptr == txq->write_ptr) {
192 		spin_unlock(&txq->lock);
193 		return;
194 	}
195 	spin_unlock(&txq->lock);
196 
197 	iwl_trans_pcie_log_scd_error(trans, txq);
198 
199 	iwl_force_nmi(trans);
200 }
201 
202 /*
203  * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
204  */
205 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
206 					     struct iwl_txq *txq, u16 byte_cnt,
207 					     int num_tbs)
208 {
209 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
210 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
211 	int write_ptr = txq->write_ptr;
212 	int txq_id = txq->id;
213 	u8 sec_ctl = 0;
214 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
215 	__le16 bc_ent;
216 	struct iwl_device_tx_cmd *dev_cmd = txq->entries[txq->write_ptr].cmd;
217 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
218 	u8 sta_id = tx_cmd->sta_id;
219 
220 	scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
221 
222 	sec_ctl = tx_cmd->sec_ctl;
223 
224 	switch (sec_ctl & TX_CMD_SEC_MSK) {
225 	case TX_CMD_SEC_CCM:
226 		len += IEEE80211_CCMP_MIC_LEN;
227 		break;
228 	case TX_CMD_SEC_TKIP:
229 		len += IEEE80211_TKIP_ICV_LEN;
230 		break;
231 	case TX_CMD_SEC_WEP:
232 		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
233 		break;
234 	}
235 	if (trans_pcie->bc_table_dword)
236 		len = DIV_ROUND_UP(len, 4);
237 
238 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
239 		return;
240 
241 	bc_ent = cpu_to_le16(len | (sta_id << 12));
242 
243 	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
244 
245 	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
246 		scd_bc_tbl[txq_id].
247 			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
248 }
249 
250 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
251 					    struct iwl_txq *txq)
252 {
253 	struct iwl_trans_pcie *trans_pcie =
254 		IWL_TRANS_GET_PCIE_TRANS(trans);
255 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
256 	int txq_id = txq->id;
257 	int read_ptr = txq->read_ptr;
258 	u8 sta_id = 0;
259 	__le16 bc_ent;
260 	struct iwl_device_tx_cmd *dev_cmd = txq->entries[read_ptr].cmd;
261 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
262 
263 	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
264 
265 	if (txq_id != trans_pcie->cmd_queue)
266 		sta_id = tx_cmd->sta_id;
267 
268 	bc_ent = cpu_to_le16(1 | (sta_id << 12));
269 
270 	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
271 
272 	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
273 		scd_bc_tbl[txq_id].
274 			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
275 }
276 
277 /*
278  * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
279  */
280 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
281 				    struct iwl_txq *txq)
282 {
283 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
284 	u32 reg = 0;
285 	int txq_id = txq->id;
286 
287 	lockdep_assert_held(&txq->lock);
288 
289 	/*
290 	 * explicitly wake up the NIC if:
291 	 * 1. shadow registers aren't enabled
292 	 * 2. NIC is woken up for CMD regardless of shadow outside this function
293 	 * 3. there is a chance that the NIC is asleep
294 	 */
295 	if (!trans->trans_cfg->base_params->shadow_reg_enable &&
296 	    txq_id != trans_pcie->cmd_queue &&
297 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
298 		/*
299 		 * wake up nic if it's powered down ...
300 		 * uCode will wake up, and interrupt us again, so next
301 		 * time we'll skip this part.
302 		 */
303 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
304 
305 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
306 			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
307 				       txq_id, reg);
308 			iwl_set_bit(trans, CSR_GP_CNTRL,
309 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
310 			txq->need_update = true;
311 			return;
312 		}
313 	}
314 
315 	/*
316 	 * if not in power-save mode, uCode will never sleep when we're
317 	 * trying to tx (during RFKILL, we're not trying to tx).
318 	 */
319 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr);
320 	if (!txq->block)
321 		iwl_write32(trans, HBUS_TARG_WRPTR,
322 			    txq->write_ptr | (txq_id << 8));
323 }
324 
325 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
326 {
327 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
328 	int i;
329 
330 	for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) {
331 		struct iwl_txq *txq = trans_pcie->txq[i];
332 
333 		if (!test_bit(i, trans_pcie->queue_used))
334 			continue;
335 
336 		spin_lock_bh(&txq->lock);
337 		if (txq->need_update) {
338 			iwl_pcie_txq_inc_wr_ptr(trans, txq);
339 			txq->need_update = false;
340 		}
341 		spin_unlock_bh(&txq->lock);
342 	}
343 }
344 
345 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_trans *trans,
346 						  void *_tfd, u8 idx)
347 {
348 
349 	if (trans->trans_cfg->use_tfh) {
350 		struct iwl_tfh_tfd *tfd = _tfd;
351 		struct iwl_tfh_tb *tb = &tfd->tbs[idx];
352 
353 		return (dma_addr_t)(le64_to_cpu(tb->addr));
354 	} else {
355 		struct iwl_tfd *tfd = _tfd;
356 		struct iwl_tfd_tb *tb = &tfd->tbs[idx];
357 		dma_addr_t addr = get_unaligned_le32(&tb->lo);
358 		dma_addr_t hi_len;
359 
360 		if (sizeof(dma_addr_t) <= sizeof(u32))
361 			return addr;
362 
363 		hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
364 
365 		/*
366 		 * shift by 16 twice to avoid warnings on 32-bit
367 		 * (where this code never runs anyway due to the
368 		 * if statement above)
369 		 */
370 		return addr | ((hi_len << 16) << 16);
371 	}
372 }
373 
374 static inline void iwl_pcie_tfd_set_tb(struct iwl_trans *trans, void *tfd,
375 				       u8 idx, dma_addr_t addr, u16 len)
376 {
377 	struct iwl_tfd *tfd_fh = (void *)tfd;
378 	struct iwl_tfd_tb *tb = &tfd_fh->tbs[idx];
379 
380 	u16 hi_n_len = len << 4;
381 
382 	put_unaligned_le32(addr, &tb->lo);
383 	hi_n_len |= iwl_get_dma_hi_addr(addr);
384 
385 	tb->hi_n_len = cpu_to_le16(hi_n_len);
386 
387 	tfd_fh->num_tbs = idx + 1;
388 }
389 
390 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_trans *trans, void *_tfd)
391 {
392 	if (trans->trans_cfg->use_tfh) {
393 		struct iwl_tfh_tfd *tfd = _tfd;
394 
395 		return le16_to_cpu(tfd->num_tbs) & 0x1f;
396 	} else {
397 		struct iwl_tfd *tfd = _tfd;
398 
399 		return tfd->num_tbs & 0x1f;
400 	}
401 }
402 
403 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
404 			       struct iwl_cmd_meta *meta,
405 			       struct iwl_txq *txq, int index)
406 {
407 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
408 	int i, num_tbs;
409 	void *tfd = iwl_pcie_get_tfd(trans, txq, index);
410 
411 	/* Sanity check on number of chunks */
412 	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
413 
414 	if (num_tbs > trans_pcie->max_tbs) {
415 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
416 		/* @todo issue fatal error, it is quite serious situation */
417 		return;
418 	}
419 
420 	/* first TB is never freed - it's the bidirectional DMA data */
421 
422 	for (i = 1; i < num_tbs; i++) {
423 		if (meta->tbs & BIT(i))
424 			dma_unmap_page(trans->dev,
425 				       iwl_pcie_tfd_tb_get_addr(trans, tfd, i),
426 				       iwl_pcie_tfd_tb_get_len(trans, tfd, i),
427 				       DMA_TO_DEVICE);
428 		else
429 			dma_unmap_single(trans->dev,
430 					 iwl_pcie_tfd_tb_get_addr(trans, tfd,
431 								  i),
432 					 iwl_pcie_tfd_tb_get_len(trans, tfd,
433 								 i),
434 					 DMA_TO_DEVICE);
435 	}
436 
437 	meta->tbs = 0;
438 
439 	if (trans->trans_cfg->use_tfh) {
440 		struct iwl_tfh_tfd *tfd_fh = (void *)tfd;
441 
442 		tfd_fh->num_tbs = 0;
443 	} else {
444 		struct iwl_tfd *tfd_fh = (void *)tfd;
445 
446 		tfd_fh->num_tbs = 0;
447 	}
448 
449 }
450 
451 /*
452  * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
453  * @trans - transport private data
454  * @txq - tx queue
455  * @dma_dir - the direction of the DMA mapping
456  *
457  * Does NOT advance any TFD circular buffer read/write indexes
458  * Does NOT free the TFD itself (which is within circular buffer)
459  */
460 void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
461 {
462 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
463 	 * idx is bounded by n_window
464 	 */
465 	int rd_ptr = txq->read_ptr;
466 	int idx = iwl_pcie_get_cmd_index(txq, rd_ptr);
467 
468 	lockdep_assert_held(&txq->lock);
469 
470 	/* We have only q->n_window txq->entries, but we use
471 	 * TFD_QUEUE_SIZE_MAX tfds
472 	 */
473 	iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, txq, rd_ptr);
474 
475 	/* free SKB */
476 	if (txq->entries) {
477 		struct sk_buff *skb;
478 
479 		skb = txq->entries[idx].skb;
480 
481 		/* Can be called from irqs-disabled context
482 		 * If skb is not NULL, it means that the whole queue is being
483 		 * freed and that the queue is not empty - free the skb
484 		 */
485 		if (skb) {
486 			iwl_op_mode_free_skb(trans->op_mode, skb);
487 			txq->entries[idx].skb = NULL;
488 		}
489 	}
490 }
491 
492 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
493 				  dma_addr_t addr, u16 len, bool reset)
494 {
495 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
496 	void *tfd;
497 	u32 num_tbs;
498 
499 	tfd = txq->tfds + trans_pcie->tfd_size * txq->write_ptr;
500 
501 	if (reset)
502 		memset(tfd, 0, trans_pcie->tfd_size);
503 
504 	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
505 
506 	/* Each TFD can point to a maximum max_tbs Tx buffers */
507 	if (num_tbs >= trans_pcie->max_tbs) {
508 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
509 			trans_pcie->max_tbs);
510 		return -EINVAL;
511 	}
512 
513 	if (WARN(addr & ~IWL_TX_DMA_MASK,
514 		 "Unaligned address = %llx\n", (unsigned long long)addr))
515 		return -EINVAL;
516 
517 	iwl_pcie_tfd_set_tb(trans, tfd, num_tbs, addr, len);
518 
519 	return num_tbs;
520 }
521 
522 int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq,
523 		       int slots_num, bool cmd_queue)
524 {
525 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
526 	size_t tfd_sz = trans_pcie->tfd_size *
527 		trans->trans_cfg->base_params->max_tfd_queue_size;
528 	size_t tb0_buf_sz;
529 	int i;
530 
531 	if (WARN_ON(txq->entries || txq->tfds))
532 		return -EINVAL;
533 
534 	if (trans->trans_cfg->use_tfh)
535 		tfd_sz = trans_pcie->tfd_size * slots_num;
536 
537 	timer_setup(&txq->stuck_timer, iwl_pcie_txq_stuck_timer, 0);
538 	txq->trans_pcie = trans_pcie;
539 
540 	txq->n_window = slots_num;
541 
542 	txq->entries = kcalloc(slots_num,
543 			       sizeof(struct iwl_pcie_txq_entry),
544 			       GFP_KERNEL);
545 
546 	if (!txq->entries)
547 		goto error;
548 
549 	if (cmd_queue)
550 		for (i = 0; i < slots_num; i++) {
551 			txq->entries[i].cmd =
552 				kmalloc(sizeof(struct iwl_device_cmd),
553 					GFP_KERNEL);
554 			if (!txq->entries[i].cmd)
555 				goto error;
556 		}
557 
558 	/* Circular buffer of transmit frame descriptors (TFDs),
559 	 * shared with device */
560 	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
561 				       &txq->dma_addr, GFP_KERNEL);
562 	if (!txq->tfds)
563 		goto error;
564 
565 	BUILD_BUG_ON(IWL_FIRST_TB_SIZE_ALIGN != sizeof(*txq->first_tb_bufs));
566 
567 	tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
568 
569 	txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
570 					      &txq->first_tb_dma,
571 					      GFP_KERNEL);
572 	if (!txq->first_tb_bufs)
573 		goto err_free_tfds;
574 
575 	return 0;
576 err_free_tfds:
577 	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
578 error:
579 	if (txq->entries && cmd_queue)
580 		for (i = 0; i < slots_num; i++)
581 			kfree(txq->entries[i].cmd);
582 	kfree(txq->entries);
583 	txq->entries = NULL;
584 
585 	return -ENOMEM;
586 
587 }
588 
589 int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
590 		      int slots_num, bool cmd_queue)
591 {
592 	int ret;
593 	u32 tfd_queue_max_size =
594 		trans->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 	struct page *next;
628 
629 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
630 	next = *page_ptr;
631 	*page_ptr = NULL;
632 
633 	while (next) {
634 		struct page *tmp = next;
635 
636 		next = *(void **)(page_address(next) + PAGE_SIZE -
637 				  sizeof(void *));
638 		__free_page(tmp);
639 	}
640 }
641 
642 static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
643 {
644 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
645 
646 	lockdep_assert_held(&trans_pcie->reg_lock);
647 
648 	if (!trans->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 				   CSR_GP_CNTRL_REG_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_pcie_clear_cmd_in_flight(trans);
688 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
689 		}
690 	}
691 
692 	while (!skb_queue_empty(&txq->overflow_q)) {
693 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
694 
695 		iwl_op_mode_free_skb(trans->op_mode, skb);
696 	}
697 
698 	spin_unlock_bh(&txq->lock);
699 
700 	/* just in case - this queue may have been stopped */
701 	iwl_wake_queue(trans, txq);
702 }
703 
704 /*
705  * iwl_pcie_txq_free - Deallocate DMA queue.
706  * @txq: Transmit queue to deallocate.
707  *
708  * Empty queue by removing and destroying all BD's.
709  * Free all buffers.
710  * 0-fill, but do not free "txq" descriptor structure.
711  */
712 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
713 {
714 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
715 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
716 	struct device *dev = trans->dev;
717 	int i;
718 
719 	if (WARN_ON(!txq))
720 		return;
721 
722 	iwl_pcie_txq_unmap(trans, txq_id);
723 
724 	/* De-alloc array of command/tx buffers */
725 	if (txq_id == trans_pcie->cmd_queue)
726 		for (i = 0; i < txq->n_window; i++) {
727 			kzfree(txq->entries[i].cmd);
728 			kzfree(txq->entries[i].free_buf);
729 		}
730 
731 	/* De-alloc circular buffer of TFDs */
732 	if (txq->tfds) {
733 		dma_free_coherent(dev,
734 				  trans_pcie->tfd_size *
735 				  trans->trans_cfg->base_params->max_tfd_queue_size,
736 				  txq->tfds, txq->dma_addr);
737 		txq->dma_addr = 0;
738 		txq->tfds = NULL;
739 
740 		dma_free_coherent(dev,
741 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
742 				  txq->first_tb_bufs, txq->first_tb_dma);
743 	}
744 
745 	kfree(txq->entries);
746 	txq->entries = NULL;
747 
748 	del_timer_sync(&txq->stuck_timer);
749 
750 	/* 0-fill queue descriptor structure */
751 	memset(txq, 0, sizeof(*txq));
752 }
753 
754 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
755 {
756 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
757 	int nq = trans->trans_cfg->base_params->num_of_queues;
758 	int chan;
759 	u32 reg_val;
760 	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
761 				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
762 
763 	/* make sure all queue are not stopped/used */
764 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
765 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
766 
767 	trans_pcie->scd_base_addr =
768 		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
769 
770 	WARN_ON(scd_base_addr != 0 &&
771 		scd_base_addr != trans_pcie->scd_base_addr);
772 
773 	/* reset context data, TX status and translation data */
774 	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
775 				   SCD_CONTEXT_MEM_LOWER_BOUND,
776 			    NULL, clear_dwords);
777 
778 	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
779 		       trans_pcie->scd_bc_tbls.dma >> 10);
780 
781 	/* The chain extension of the SCD doesn't work well. This feature is
782 	 * enabled by default by the HW, so we need to disable it manually.
783 	 */
784 	if (trans->trans_cfg->base_params->scd_chain_ext_wa)
785 		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
786 
787 	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
788 				trans_pcie->cmd_fifo,
789 				trans_pcie->cmd_q_wdg_timeout);
790 
791 	/* Activate all Tx DMA/FIFO channels */
792 	iwl_scd_activate_fifos(trans);
793 
794 	/* Enable DMA channel */
795 	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
796 		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
797 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
798 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
799 
800 	/* Update FH chicken bits */
801 	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
802 	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
803 			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
804 
805 	/* Enable L1-Active */
806 	if (trans->trans_cfg->device_family < IWL_DEVICE_FAMILY_8000)
807 		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
808 				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
809 }
810 
811 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
812 {
813 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
814 	int txq_id;
815 
816 	/*
817 	 * we should never get here in gen2 trans mode return early to avoid
818 	 * having invalid accesses
819 	 */
820 	if (WARN_ON_ONCE(trans->trans_cfg->gen2))
821 		return;
822 
823 	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
824 	     txq_id++) {
825 		struct iwl_txq *txq = trans_pcie->txq[txq_id];
826 		if (trans->trans_cfg->use_tfh)
827 			iwl_write_direct64(trans,
828 					   FH_MEM_CBBC_QUEUE(trans, txq_id),
829 					   txq->dma_addr);
830 		else
831 			iwl_write_direct32(trans,
832 					   FH_MEM_CBBC_QUEUE(trans, txq_id),
833 					   txq->dma_addr >> 8);
834 		iwl_pcie_txq_unmap(trans, txq_id);
835 		txq->read_ptr = 0;
836 		txq->write_ptr = 0;
837 	}
838 
839 	/* Tell NIC where to find the "keep warm" buffer */
840 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
841 			   trans_pcie->kw.dma >> 4);
842 
843 	/*
844 	 * Send 0 as the scd_base_addr since the device may have be reset
845 	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
846 	 * contain garbage.
847 	 */
848 	iwl_pcie_tx_start(trans, 0);
849 }
850 
851 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
852 {
853 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
854 	unsigned long flags;
855 	int ch, ret;
856 	u32 mask = 0;
857 
858 	spin_lock(&trans_pcie->irq_lock);
859 
860 	if (!iwl_trans_grab_nic_access(trans, &flags))
861 		goto out;
862 
863 	/* Stop each Tx DMA channel */
864 	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
865 		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
866 		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
867 	}
868 
869 	/* Wait for DMA channels to be idle */
870 	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
871 	if (ret < 0)
872 		IWL_ERR(trans,
873 			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
874 			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
875 
876 	iwl_trans_release_nic_access(trans, &flags);
877 
878 out:
879 	spin_unlock(&trans_pcie->irq_lock);
880 }
881 
882 /*
883  * iwl_pcie_tx_stop - Stop all Tx DMA channels
884  */
885 int iwl_pcie_tx_stop(struct iwl_trans *trans)
886 {
887 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
888 	int txq_id;
889 
890 	/* Turn off all Tx DMA fifos */
891 	iwl_scd_deactivate_fifos(trans);
892 
893 	/* Turn off all Tx DMA channels */
894 	iwl_pcie_tx_stop_fh(trans);
895 
896 	/*
897 	 * This function can be called before the op_mode disabled the
898 	 * queues. This happens when we have an rfkill interrupt.
899 	 * Since we stop Tx altogether - mark the queues as stopped.
900 	 */
901 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
902 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
903 
904 	/* This can happen: start_hw, stop_device */
905 	if (!trans_pcie->txq_memory)
906 		return 0;
907 
908 	/* Unmap DMA from host system and free skb's */
909 	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
910 	     txq_id++)
911 		iwl_pcie_txq_unmap(trans, txq_id);
912 
913 	return 0;
914 }
915 
916 /*
917  * iwl_trans_tx_free - Free TXQ Context
918  *
919  * Destroy all TX DMA queues and structures
920  */
921 void iwl_pcie_tx_free(struct iwl_trans *trans)
922 {
923 	int txq_id;
924 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
925 
926 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
927 
928 	/* Tx queues */
929 	if (trans_pcie->txq_memory) {
930 		for (txq_id = 0;
931 		     txq_id < trans->trans_cfg->base_params->num_of_queues;
932 		     txq_id++) {
933 			iwl_pcie_txq_free(trans, txq_id);
934 			trans_pcie->txq[txq_id] = NULL;
935 		}
936 	}
937 
938 	kfree(trans_pcie->txq_memory);
939 	trans_pcie->txq_memory = NULL;
940 
941 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
942 
943 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
944 }
945 
946 /*
947  * iwl_pcie_tx_alloc - allocate TX context
948  * Allocate all Tx DMA structures and initialize them
949  */
950 static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
951 {
952 	int ret;
953 	int txq_id, slots_num;
954 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
955 	u16 bc_tbls_size = trans->trans_cfg->base_params->num_of_queues;
956 
957 	bc_tbls_size *= (trans->trans_cfg->device_family >=
958 			 IWL_DEVICE_FAMILY_AX210) ?
959 		sizeof(struct iwl_gen3_bc_tbl) :
960 		sizeof(struct iwlagn_scd_bc_tbl);
961 
962 	/*It is not allowed to alloc twice, so warn when this happens.
963 	 * We cannot rely on the previous allocation, so free and fail */
964 	if (WARN_ON(trans_pcie->txq_memory)) {
965 		ret = -EINVAL;
966 		goto error;
967 	}
968 
969 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
970 				     bc_tbls_size);
971 	if (ret) {
972 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
973 		goto error;
974 	}
975 
976 	/* Alloc keep-warm buffer */
977 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
978 	if (ret) {
979 		IWL_ERR(trans, "Keep Warm allocation failed\n");
980 		goto error;
981 	}
982 
983 	trans_pcie->txq_memory =
984 		kcalloc(trans->trans_cfg->base_params->num_of_queues,
985 			sizeof(struct iwl_txq), GFP_KERNEL);
986 	if (!trans_pcie->txq_memory) {
987 		IWL_ERR(trans, "Not enough memory for txq\n");
988 		ret = -ENOMEM;
989 		goto error;
990 	}
991 
992 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
993 	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
994 	     txq_id++) {
995 		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
996 
997 		if (cmd_queue)
998 			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
999 					  trans->cfg->min_txq_size);
1000 		else
1001 			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
1002 					  trans->cfg->min_256_ba_txq_size);
1003 		trans_pcie->txq[txq_id] = &trans_pcie->txq_memory[txq_id];
1004 		ret = iwl_pcie_txq_alloc(trans, trans_pcie->txq[txq_id],
1005 					 slots_num, cmd_queue);
1006 		if (ret) {
1007 			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1008 			goto error;
1009 		}
1010 		trans_pcie->txq[txq_id]->id = txq_id;
1011 	}
1012 
1013 	return 0;
1014 
1015 error:
1016 	iwl_pcie_tx_free(trans);
1017 
1018 	return ret;
1019 }
1020 
1021 int iwl_pcie_tx_init(struct iwl_trans *trans)
1022 {
1023 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1024 	int ret;
1025 	int txq_id, slots_num;
1026 	bool alloc = false;
1027 
1028 	if (!trans_pcie->txq_memory) {
1029 		ret = iwl_pcie_tx_alloc(trans);
1030 		if (ret)
1031 			goto error;
1032 		alloc = true;
1033 	}
1034 
1035 	spin_lock(&trans_pcie->irq_lock);
1036 
1037 	/* Turn off all Tx DMA fifos */
1038 	iwl_scd_deactivate_fifos(trans);
1039 
1040 	/* Tell NIC where to find the "keep warm" buffer */
1041 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
1042 			   trans_pcie->kw.dma >> 4);
1043 
1044 	spin_unlock(&trans_pcie->irq_lock);
1045 
1046 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
1047 	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
1048 	     txq_id++) {
1049 		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
1050 
1051 		if (cmd_queue)
1052 			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
1053 					  trans->cfg->min_txq_size);
1054 		else
1055 			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
1056 					  trans->cfg->min_256_ba_txq_size);
1057 		ret = iwl_pcie_txq_init(trans, trans_pcie->txq[txq_id],
1058 					slots_num, cmd_queue);
1059 		if (ret) {
1060 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1061 			goto error;
1062 		}
1063 
1064 		/*
1065 		 * Tell nic where to find circular buffer of TFDs for a
1066 		 * given Tx queue, and enable the DMA channel used for that
1067 		 * queue.
1068 		 * Circular buffer (TFD queue in DRAM) physical base address
1069 		 */
1070 		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
1071 				   trans_pcie->txq[txq_id]->dma_addr >> 8);
1072 	}
1073 
1074 	iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
1075 	if (trans->trans_cfg->base_params->num_of_queues > 20)
1076 		iwl_set_bits_prph(trans, SCD_GP_CTRL,
1077 				  SCD_GP_CTRL_ENABLE_31_QUEUES);
1078 
1079 	return 0;
1080 error:
1081 	/*Upon error, free only if we allocated something */
1082 	if (alloc)
1083 		iwl_pcie_tx_free(trans);
1084 	return ret;
1085 }
1086 
1087 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
1088 {
1089 	lockdep_assert_held(&txq->lock);
1090 
1091 	if (!txq->wd_timeout)
1092 		return;
1093 
1094 	/*
1095 	 * station is asleep and we send data - that must
1096 	 * be uAPSD or PS-Poll. Don't rearm the timer.
1097 	 */
1098 	if (txq->frozen)
1099 		return;
1100 
1101 	/*
1102 	 * if empty delete timer, otherwise move timer forward
1103 	 * since we're making progress on this queue
1104 	 */
1105 	if (txq->read_ptr == txq->write_ptr)
1106 		del_timer(&txq->stuck_timer);
1107 	else
1108 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1109 }
1110 
1111 /* Frees buffers until index _not_ inclusive */
1112 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1113 			    struct sk_buff_head *skbs)
1114 {
1115 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1116 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1117 	int tfd_num = iwl_pcie_get_cmd_index(txq, ssn);
1118 	int read_ptr = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1119 	int last_to_free;
1120 
1121 	/* This function is not meant to release cmd queue*/
1122 	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1123 		return;
1124 
1125 	spin_lock_bh(&txq->lock);
1126 
1127 	if (!test_bit(txq_id, trans_pcie->queue_used)) {
1128 		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1129 				    txq_id, ssn);
1130 		goto out;
1131 	}
1132 
1133 	if (read_ptr == tfd_num)
1134 		goto out;
1135 
1136 	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1137 			   txq_id, txq->read_ptr, tfd_num, ssn);
1138 
1139 	/*Since we free until index _not_ inclusive, the one before index is
1140 	 * the last we will free. This one must be used */
1141 	last_to_free = iwl_queue_dec_wrap(trans, tfd_num);
1142 
1143 	if (!iwl_queue_used(txq, last_to_free)) {
1144 		IWL_ERR(trans,
1145 			"%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1146 			__func__, txq_id, last_to_free,
1147 			trans->trans_cfg->base_params->max_tfd_queue_size,
1148 			txq->write_ptr, txq->read_ptr);
1149 		goto out;
1150 	}
1151 
1152 	if (WARN_ON(!skb_queue_empty(skbs)))
1153 		goto out;
1154 
1155 	for (;
1156 	     read_ptr != tfd_num;
1157 	     txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr),
1158 	     read_ptr = iwl_pcie_get_cmd_index(txq, txq->read_ptr)) {
1159 		struct sk_buff *skb = txq->entries[read_ptr].skb;
1160 
1161 		if (WARN_ON_ONCE(!skb))
1162 			continue;
1163 
1164 		iwl_pcie_free_tso_page(trans_pcie, skb);
1165 
1166 		__skb_queue_tail(skbs, skb);
1167 
1168 		txq->entries[read_ptr].skb = NULL;
1169 
1170 		if (!trans->trans_cfg->use_tfh)
1171 			iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1172 
1173 		iwl_pcie_txq_free_tfd(trans, txq);
1174 	}
1175 
1176 	iwl_pcie_txq_progress(txq);
1177 
1178 	if (iwl_queue_space(trans, txq) > txq->low_mark &&
1179 	    test_bit(txq_id, trans_pcie->queue_stopped)) {
1180 		struct sk_buff_head overflow_skbs;
1181 
1182 		__skb_queue_head_init(&overflow_skbs);
1183 		skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1184 
1185 		/*
1186 		 * We are going to transmit from the overflow queue.
1187 		 * Remember this state so that wait_for_txq_empty will know we
1188 		 * are adding more packets to the TFD queue. It cannot rely on
1189 		 * the state of &txq->overflow_q, as we just emptied it, but
1190 		 * haven't TXed the content yet.
1191 		 */
1192 		txq->overflow_tx = true;
1193 
1194 		/*
1195 		 * This is tricky: we are in reclaim path which is non
1196 		 * re-entrant, so noone will try to take the access the
1197 		 * txq data from that path. We stopped tx, so we can't
1198 		 * have tx as well. Bottom line, we can unlock and re-lock
1199 		 * later.
1200 		 */
1201 		spin_unlock_bh(&txq->lock);
1202 
1203 		while (!skb_queue_empty(&overflow_skbs)) {
1204 			struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1205 			struct iwl_device_tx_cmd *dev_cmd_ptr;
1206 
1207 			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1208 						 trans_pcie->dev_cmd_offs);
1209 
1210 			/*
1211 			 * Note that we can very well be overflowing again.
1212 			 * In that case, iwl_queue_space will be small again
1213 			 * and we won't wake mac80211's queue.
1214 			 */
1215 			iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id);
1216 		}
1217 
1218 		if (iwl_queue_space(trans, txq) > txq->low_mark)
1219 			iwl_wake_queue(trans, txq);
1220 
1221 		spin_lock_bh(&txq->lock);
1222 		txq->overflow_tx = false;
1223 	}
1224 
1225 out:
1226 	spin_unlock_bh(&txq->lock);
1227 }
1228 
1229 /* Set wr_ptr of specific device and txq  */
1230 void iwl_trans_pcie_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr)
1231 {
1232 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1233 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1234 
1235 	spin_lock_bh(&txq->lock);
1236 
1237 	txq->write_ptr = ptr;
1238 	txq->read_ptr = txq->write_ptr;
1239 
1240 	spin_unlock_bh(&txq->lock);
1241 }
1242 
1243 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1244 				      const struct iwl_host_cmd *cmd)
1245 {
1246 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1247 	int ret;
1248 
1249 	lockdep_assert_held(&trans_pcie->reg_lock);
1250 
1251 	/* Make sure the NIC is still alive in the bus */
1252 	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1253 		return -ENODEV;
1254 
1255 	/*
1256 	 * wake up the NIC to make sure that the firmware will see the host
1257 	 * command - we will let the NIC sleep once all the host commands
1258 	 * returned. This needs to be done only on NICs that have
1259 	 * apmg_wake_up_wa set.
1260 	 */
1261 	if (trans->trans_cfg->base_params->apmg_wake_up_wa &&
1262 	    !trans_pcie->cmd_hold_nic_awake) {
1263 		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1264 					 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1265 
1266 		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1267 				   CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1268 				   (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1269 				    CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1270 				   15000);
1271 		if (ret < 0) {
1272 			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1273 					CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1274 			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1275 			return -EIO;
1276 		}
1277 		trans_pcie->cmd_hold_nic_awake = true;
1278 	}
1279 
1280 	return 0;
1281 }
1282 
1283 /*
1284  * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1285  *
1286  * When FW advances 'R' index, all entries between old and new 'R' index
1287  * need to be reclaimed. As result, some free space forms.  If there is
1288  * enough free space (> low mark), wake the stack that feeds us.
1289  */
1290 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1291 {
1292 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1293 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1294 	unsigned long flags;
1295 	int nfreed = 0;
1296 	u16 r;
1297 
1298 	lockdep_assert_held(&txq->lock);
1299 
1300 	idx = iwl_pcie_get_cmd_index(txq, idx);
1301 	r = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1302 
1303 	if (idx >= trans->trans_cfg->base_params->max_tfd_queue_size ||
1304 	    (!iwl_queue_used(txq, idx))) {
1305 		WARN_ONCE(test_bit(txq_id, trans_pcie->queue_used),
1306 			  "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1307 			  __func__, txq_id, idx,
1308 			  trans->trans_cfg->base_params->max_tfd_queue_size,
1309 			  txq->write_ptr, txq->read_ptr);
1310 		return;
1311 	}
1312 
1313 	for (idx = iwl_queue_inc_wrap(trans, idx); r != idx;
1314 	     r = iwl_queue_inc_wrap(trans, r)) {
1315 		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
1316 
1317 		if (nfreed++ > 0) {
1318 			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1319 				idx, txq->write_ptr, r);
1320 			iwl_force_nmi(trans);
1321 		}
1322 	}
1323 
1324 	if (txq->read_ptr == txq->write_ptr) {
1325 		spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1326 		iwl_pcie_clear_cmd_in_flight(trans);
1327 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1328 	}
1329 
1330 	iwl_pcie_txq_progress(txq);
1331 }
1332 
1333 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1334 				 u16 txq_id)
1335 {
1336 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1337 	u32 tbl_dw_addr;
1338 	u32 tbl_dw;
1339 	u16 scd_q2ratid;
1340 
1341 	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1342 
1343 	tbl_dw_addr = trans_pcie->scd_base_addr +
1344 			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1345 
1346 	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1347 
1348 	if (txq_id & 0x1)
1349 		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1350 	else
1351 		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1352 
1353 	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1354 
1355 	return 0;
1356 }
1357 
1358 /* Receiver address (actually, Rx station's index into station table),
1359  * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1360 #define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
1361 
1362 bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1363 			       const struct iwl_trans_txq_scd_cfg *cfg,
1364 			       unsigned int wdg_timeout)
1365 {
1366 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1367 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1368 	int fifo = -1;
1369 	bool scd_bug = false;
1370 
1371 	if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1372 		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1373 
1374 	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1375 
1376 	if (cfg) {
1377 		fifo = cfg->fifo;
1378 
1379 		/* Disable the scheduler prior configuring the cmd queue */
1380 		if (txq_id == trans_pcie->cmd_queue &&
1381 		    trans_pcie->scd_set_active)
1382 			iwl_scd_enable_set_active(trans, 0);
1383 
1384 		/* Stop this Tx queue before configuring it */
1385 		iwl_scd_txq_set_inactive(trans, txq_id);
1386 
1387 		/* Set this queue as a chain-building queue unless it is CMD */
1388 		if (txq_id != trans_pcie->cmd_queue)
1389 			iwl_scd_txq_set_chain(trans, txq_id);
1390 
1391 		if (cfg->aggregate) {
1392 			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1393 
1394 			/* Map receiver-address / traffic-ID to this queue */
1395 			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1396 
1397 			/* enable aggregations for the queue */
1398 			iwl_scd_txq_enable_agg(trans, txq_id);
1399 			txq->ampdu = true;
1400 		} else {
1401 			/*
1402 			 * disable aggregations for the queue, this will also
1403 			 * make the ra_tid mapping configuration irrelevant
1404 			 * since it is now a non-AGG queue.
1405 			 */
1406 			iwl_scd_txq_disable_agg(trans, txq_id);
1407 
1408 			ssn = txq->read_ptr;
1409 		}
1410 	} else {
1411 		/*
1412 		 * If we need to move the SCD write pointer by steps of
1413 		 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let
1414 		 * the op_mode know by returning true later.
1415 		 * Do this only in case cfg is NULL since this trick can
1416 		 * be done only if we have DQA enabled which is true for mvm
1417 		 * only. And mvm never sets a cfg pointer.
1418 		 * This is really ugly, but this is the easiest way out for
1419 		 * this sad hardware issue.
1420 		 * This bug has been fixed on devices 9000 and up.
1421 		 */
1422 		scd_bug = !trans->trans_cfg->mq_rx_supported &&
1423 			!((ssn - txq->write_ptr) & 0x3f) &&
1424 			(ssn != txq->write_ptr);
1425 		if (scd_bug)
1426 			ssn++;
1427 	}
1428 
1429 	/* Place first TFD at index corresponding to start sequence number.
1430 	 * Assumes that ssn_idx is valid (!= 0xFFF) */
1431 	txq->read_ptr = (ssn & 0xff);
1432 	txq->write_ptr = (ssn & 0xff);
1433 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1434 			   (ssn & 0xff) | (txq_id << 8));
1435 
1436 	if (cfg) {
1437 		u8 frame_limit = cfg->frame_limit;
1438 
1439 		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1440 
1441 		/* Set up Tx window size and frame limit for this queue */
1442 		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1443 				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1444 		iwl_trans_write_mem32(trans,
1445 			trans_pcie->scd_base_addr +
1446 			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1447 			SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) |
1448 			SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit));
1449 
1450 		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1451 		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1452 			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1453 			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1454 			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1455 			       SCD_QUEUE_STTS_REG_MSK);
1456 
1457 		/* enable the scheduler for this queue (only) */
1458 		if (txq_id == trans_pcie->cmd_queue &&
1459 		    trans_pcie->scd_set_active)
1460 			iwl_scd_enable_set_active(trans, BIT(txq_id));
1461 
1462 		IWL_DEBUG_TX_QUEUES(trans,
1463 				    "Activate queue %d on FIFO %d WrPtr: %d\n",
1464 				    txq_id, fifo, ssn & 0xff);
1465 	} else {
1466 		IWL_DEBUG_TX_QUEUES(trans,
1467 				    "Activate queue %d WrPtr: %d\n",
1468 				    txq_id, ssn & 0xff);
1469 	}
1470 
1471 	return scd_bug;
1472 }
1473 
1474 void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
1475 					bool shared_mode)
1476 {
1477 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1478 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1479 
1480 	txq->ampdu = !shared_mode;
1481 }
1482 
1483 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1484 				bool configure_scd)
1485 {
1486 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1487 	u32 stts_addr = trans_pcie->scd_base_addr +
1488 			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1489 	static const u32 zero_val[4] = {};
1490 
1491 	trans_pcie->txq[txq_id]->frozen_expiry_remainder = 0;
1492 	trans_pcie->txq[txq_id]->frozen = false;
1493 
1494 	/*
1495 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1496 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1497 	 * allow the op_mode to call txq_disable after it already called
1498 	 * stop_device.
1499 	 */
1500 	if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1501 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1502 			  "queue %d not used", txq_id);
1503 		return;
1504 	}
1505 
1506 	if (configure_scd) {
1507 		iwl_scd_txq_set_inactive(trans, txq_id);
1508 
1509 		iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1510 				    ARRAY_SIZE(zero_val));
1511 	}
1512 
1513 	iwl_pcie_txq_unmap(trans, txq_id);
1514 	trans_pcie->txq[txq_id]->ampdu = false;
1515 
1516 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1517 }
1518 
1519 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1520 
1521 /*
1522  * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1523  * @priv: device private data point
1524  * @cmd: a pointer to the ucode command structure
1525  *
1526  * The function returns < 0 values to indicate the operation
1527  * failed. On success, it returns the index (>= 0) of command in the
1528  * command queue.
1529  */
1530 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1531 				 struct iwl_host_cmd *cmd)
1532 {
1533 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1534 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1535 	struct iwl_device_cmd *out_cmd;
1536 	struct iwl_cmd_meta *out_meta;
1537 	unsigned long flags;
1538 	void *dup_buf = NULL;
1539 	dma_addr_t phys_addr;
1540 	int idx;
1541 	u16 copy_size, cmd_size, tb0_size;
1542 	bool had_nocopy = false;
1543 	u8 group_id = iwl_cmd_groupid(cmd->id);
1544 	int i, ret;
1545 	u32 cmd_pos;
1546 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1547 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1548 
1549 	if (WARN(!trans->wide_cmd_header &&
1550 		 group_id > IWL_ALWAYS_LONG_GROUP,
1551 		 "unsupported wide command %#x\n", cmd->id))
1552 		return -EINVAL;
1553 
1554 	if (group_id != 0) {
1555 		copy_size = sizeof(struct iwl_cmd_header_wide);
1556 		cmd_size = sizeof(struct iwl_cmd_header_wide);
1557 	} else {
1558 		copy_size = sizeof(struct iwl_cmd_header);
1559 		cmd_size = sizeof(struct iwl_cmd_header);
1560 	}
1561 
1562 	/* need one for the header if the first is NOCOPY */
1563 	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1564 
1565 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1566 		cmddata[i] = cmd->data[i];
1567 		cmdlen[i] = cmd->len[i];
1568 
1569 		if (!cmd->len[i])
1570 			continue;
1571 
1572 		/* need at least IWL_FIRST_TB_SIZE copied */
1573 		if (copy_size < IWL_FIRST_TB_SIZE) {
1574 			int copy = IWL_FIRST_TB_SIZE - copy_size;
1575 
1576 			if (copy > cmdlen[i])
1577 				copy = cmdlen[i];
1578 			cmdlen[i] -= copy;
1579 			cmddata[i] += copy;
1580 			copy_size += copy;
1581 		}
1582 
1583 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1584 			had_nocopy = true;
1585 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1586 				idx = -EINVAL;
1587 				goto free_dup_buf;
1588 			}
1589 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1590 			/*
1591 			 * This is also a chunk that isn't copied
1592 			 * to the static buffer so set had_nocopy.
1593 			 */
1594 			had_nocopy = true;
1595 
1596 			/* only allowed once */
1597 			if (WARN_ON(dup_buf)) {
1598 				idx = -EINVAL;
1599 				goto free_dup_buf;
1600 			}
1601 
1602 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
1603 					  GFP_ATOMIC);
1604 			if (!dup_buf)
1605 				return -ENOMEM;
1606 		} else {
1607 			/* NOCOPY must not be followed by normal! */
1608 			if (WARN_ON(had_nocopy)) {
1609 				idx = -EINVAL;
1610 				goto free_dup_buf;
1611 			}
1612 			copy_size += cmdlen[i];
1613 		}
1614 		cmd_size += cmd->len[i];
1615 	}
1616 
1617 	/*
1618 	 * If any of the command structures end up being larger than
1619 	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1620 	 * allocated into separate TFDs, then we will need to
1621 	 * increase the size of the buffers.
1622 	 */
1623 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1624 		 "Command %s (%#x) is too large (%d bytes)\n",
1625 		 iwl_get_cmd_string(trans, cmd->id),
1626 		 cmd->id, copy_size)) {
1627 		idx = -EINVAL;
1628 		goto free_dup_buf;
1629 	}
1630 
1631 	spin_lock_bh(&txq->lock);
1632 
1633 	if (iwl_queue_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1634 		spin_unlock_bh(&txq->lock);
1635 
1636 		IWL_ERR(trans, "No space in command queue\n");
1637 		iwl_op_mode_cmd_queue_full(trans->op_mode);
1638 		idx = -ENOSPC;
1639 		goto free_dup_buf;
1640 	}
1641 
1642 	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
1643 	out_cmd = txq->entries[idx].cmd;
1644 	out_meta = &txq->entries[idx].meta;
1645 
1646 	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1647 	if (cmd->flags & CMD_WANT_SKB)
1648 		out_meta->source = cmd;
1649 
1650 	/* set up the header */
1651 	if (group_id != 0) {
1652 		out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1653 		out_cmd->hdr_wide.group_id = group_id;
1654 		out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1655 		out_cmd->hdr_wide.length =
1656 			cpu_to_le16(cmd_size -
1657 				    sizeof(struct iwl_cmd_header_wide));
1658 		out_cmd->hdr_wide.reserved = 0;
1659 		out_cmd->hdr_wide.sequence =
1660 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1661 						 INDEX_TO_SEQ(txq->write_ptr));
1662 
1663 		cmd_pos = sizeof(struct iwl_cmd_header_wide);
1664 		copy_size = sizeof(struct iwl_cmd_header_wide);
1665 	} else {
1666 		out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1667 		out_cmd->hdr.sequence =
1668 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1669 						 INDEX_TO_SEQ(txq->write_ptr));
1670 		out_cmd->hdr.group_id = 0;
1671 
1672 		cmd_pos = sizeof(struct iwl_cmd_header);
1673 		copy_size = sizeof(struct iwl_cmd_header);
1674 	}
1675 
1676 	/* and copy the data that needs to be copied */
1677 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1678 		int copy;
1679 
1680 		if (!cmd->len[i])
1681 			continue;
1682 
1683 		/* copy everything if not nocopy/dup */
1684 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1685 					   IWL_HCMD_DFL_DUP))) {
1686 			copy = cmd->len[i];
1687 
1688 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1689 			cmd_pos += copy;
1690 			copy_size += copy;
1691 			continue;
1692 		}
1693 
1694 		/*
1695 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1696 		 * in total (for bi-directional DMA), but copy up to what
1697 		 * we can fit into the payload for debug dump purposes.
1698 		 */
1699 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1700 
1701 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1702 		cmd_pos += copy;
1703 
1704 		/* However, treat copy_size the proper way, we need it below */
1705 		if (copy_size < IWL_FIRST_TB_SIZE) {
1706 			copy = IWL_FIRST_TB_SIZE - copy_size;
1707 
1708 			if (copy > cmd->len[i])
1709 				copy = cmd->len[i];
1710 			copy_size += copy;
1711 		}
1712 	}
1713 
1714 	IWL_DEBUG_HC(trans,
1715 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1716 		     iwl_get_cmd_string(trans, cmd->id),
1717 		     group_id, out_cmd->hdr.cmd,
1718 		     le16_to_cpu(out_cmd->hdr.sequence),
1719 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
1720 
1721 	/* start the TFD with the minimum copy bytes */
1722 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1723 	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1724 	iwl_pcie_txq_build_tfd(trans, txq,
1725 			       iwl_pcie_get_first_tb_dma(txq, idx),
1726 			       tb0_size, true);
1727 
1728 	/* map first command fragment, if any remains */
1729 	if (copy_size > tb0_size) {
1730 		phys_addr = dma_map_single(trans->dev,
1731 					   ((u8 *)&out_cmd->hdr) + tb0_size,
1732 					   copy_size - tb0_size,
1733 					   DMA_TO_DEVICE);
1734 		if (dma_mapping_error(trans->dev, phys_addr)) {
1735 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1736 					   txq->write_ptr);
1737 			idx = -ENOMEM;
1738 			goto out;
1739 		}
1740 
1741 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1742 				       copy_size - tb0_size, false);
1743 	}
1744 
1745 	/* map the remaining (adjusted) nocopy/dup fragments */
1746 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1747 		const void *data = cmddata[i];
1748 
1749 		if (!cmdlen[i])
1750 			continue;
1751 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1752 					   IWL_HCMD_DFL_DUP)))
1753 			continue;
1754 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1755 			data = dup_buf;
1756 		phys_addr = dma_map_single(trans->dev, (void *)data,
1757 					   cmdlen[i], DMA_TO_DEVICE);
1758 		if (dma_mapping_error(trans->dev, phys_addr)) {
1759 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1760 					   txq->write_ptr);
1761 			idx = -ENOMEM;
1762 			goto out;
1763 		}
1764 
1765 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1766 	}
1767 
1768 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
1769 	out_meta->flags = cmd->flags;
1770 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1771 		kzfree(txq->entries[idx].free_buf);
1772 	txq->entries[idx].free_buf = dup_buf;
1773 
1774 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1775 
1776 	/* start timer if queue currently empty */
1777 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
1778 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1779 
1780 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1781 	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1782 	if (ret < 0) {
1783 		idx = ret;
1784 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1785 		goto out;
1786 	}
1787 
1788 	/* Increment and update queue's write index */
1789 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
1790 	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1791 
1792 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1793 
1794  out:
1795 	spin_unlock_bh(&txq->lock);
1796  free_dup_buf:
1797 	if (idx < 0)
1798 		kfree(dup_buf);
1799 	return idx;
1800 }
1801 
1802 /*
1803  * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1804  * @rxb: Rx buffer to reclaim
1805  */
1806 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1807 			    struct iwl_rx_cmd_buffer *rxb)
1808 {
1809 	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1810 	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1811 	u8 group_id;
1812 	u32 cmd_id;
1813 	int txq_id = SEQ_TO_QUEUE(sequence);
1814 	int index = SEQ_TO_INDEX(sequence);
1815 	int cmd_index;
1816 	struct iwl_device_cmd *cmd;
1817 	struct iwl_cmd_meta *meta;
1818 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1819 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1820 
1821 	/* If a Tx command is being handled and it isn't in the actual
1822 	 * command queue then there a command routing bug has been introduced
1823 	 * in the queue management code. */
1824 	if (WARN(txq_id != trans_pcie->cmd_queue,
1825 		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1826 		 txq_id, trans_pcie->cmd_queue, sequence, txq->read_ptr,
1827 		 txq->write_ptr)) {
1828 		iwl_print_hex_error(trans, pkt, 32);
1829 		return;
1830 	}
1831 
1832 	spin_lock_bh(&txq->lock);
1833 
1834 	cmd_index = iwl_pcie_get_cmd_index(txq, index);
1835 	cmd = txq->entries[cmd_index].cmd;
1836 	meta = &txq->entries[cmd_index].meta;
1837 	group_id = cmd->hdr.group_id;
1838 	cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1839 
1840 	iwl_pcie_tfd_unmap(trans, meta, txq, index);
1841 
1842 	/* Input error checking is done when commands are added to queue. */
1843 	if (meta->flags & CMD_WANT_SKB) {
1844 		struct page *p = rxb_steal_page(rxb);
1845 
1846 		meta->source->resp_pkt = pkt;
1847 		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1848 		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1849 	}
1850 
1851 	if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1852 		iwl_op_mode_async_cb(trans->op_mode, cmd);
1853 
1854 	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1855 
1856 	if (!(meta->flags & CMD_ASYNC)) {
1857 		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1858 			IWL_WARN(trans,
1859 				 "HCMD_ACTIVE already clear for command %s\n",
1860 				 iwl_get_cmd_string(trans, cmd_id));
1861 		}
1862 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1863 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1864 			       iwl_get_cmd_string(trans, cmd_id));
1865 		wake_up(&trans_pcie->wait_command_queue);
1866 	}
1867 
1868 	meta->flags = 0;
1869 
1870 	spin_unlock_bh(&txq->lock);
1871 }
1872 
1873 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1874 
1875 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1876 				    struct iwl_host_cmd *cmd)
1877 {
1878 	int ret;
1879 
1880 	/* An asynchronous command can not expect an SKB to be set. */
1881 	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1882 		return -EINVAL;
1883 
1884 	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1885 	if (ret < 0) {
1886 		IWL_ERR(trans,
1887 			"Error sending %s: enqueue_hcmd failed: %d\n",
1888 			iwl_get_cmd_string(trans, cmd->id), ret);
1889 		return ret;
1890 	}
1891 	return 0;
1892 }
1893 
1894 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1895 				   struct iwl_host_cmd *cmd)
1896 {
1897 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1898 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1899 	int cmd_idx;
1900 	int ret;
1901 
1902 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1903 		       iwl_get_cmd_string(trans, cmd->id));
1904 
1905 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1906 				  &trans->status),
1907 		 "Command %s: a command is already active!\n",
1908 		 iwl_get_cmd_string(trans, cmd->id)))
1909 		return -EIO;
1910 
1911 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1912 		       iwl_get_cmd_string(trans, cmd->id));
1913 
1914 	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1915 	if (cmd_idx < 0) {
1916 		ret = cmd_idx;
1917 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1918 		IWL_ERR(trans,
1919 			"Error sending %s: enqueue_hcmd failed: %d\n",
1920 			iwl_get_cmd_string(trans, cmd->id), ret);
1921 		return ret;
1922 	}
1923 
1924 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
1925 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1926 					   &trans->status),
1927 				 HOST_COMPLETE_TIMEOUT);
1928 	if (!ret) {
1929 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1930 			iwl_get_cmd_string(trans, cmd->id),
1931 			jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1932 
1933 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1934 			txq->read_ptr, txq->write_ptr);
1935 
1936 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1937 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1938 			       iwl_get_cmd_string(trans, cmd->id));
1939 		ret = -ETIMEDOUT;
1940 
1941 		iwl_trans_pcie_sync_nmi(trans);
1942 		goto cancel;
1943 	}
1944 
1945 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1946 		iwl_trans_pcie_dump_regs(trans);
1947 		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1948 			iwl_get_cmd_string(trans, cmd->id));
1949 		dump_stack();
1950 		ret = -EIO;
1951 		goto cancel;
1952 	}
1953 
1954 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1955 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1956 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1957 		ret = -ERFKILL;
1958 		goto cancel;
1959 	}
1960 
1961 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1962 		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1963 			iwl_get_cmd_string(trans, cmd->id));
1964 		ret = -EIO;
1965 		goto cancel;
1966 	}
1967 
1968 	return 0;
1969 
1970 cancel:
1971 	if (cmd->flags & CMD_WANT_SKB) {
1972 		/*
1973 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1974 		 * TX cmd queue. Otherwise in case the cmd comes
1975 		 * in later, it will possibly set an invalid
1976 		 * address (cmd->meta.source).
1977 		 */
1978 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1979 	}
1980 
1981 	if (cmd->resp_pkt) {
1982 		iwl_free_resp(cmd);
1983 		cmd->resp_pkt = NULL;
1984 	}
1985 
1986 	return ret;
1987 }
1988 
1989 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1990 {
1991 	/* Make sure the NIC is still alive in the bus */
1992 	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1993 		return -ENODEV;
1994 
1995 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1996 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1997 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1998 				  cmd->id);
1999 		return -ERFKILL;
2000 	}
2001 
2002 	if (cmd->flags & CMD_ASYNC)
2003 		return iwl_pcie_send_hcmd_async(trans, cmd);
2004 
2005 	/* We still can fail on RFKILL that can be asserted while we wait */
2006 	return iwl_pcie_send_hcmd_sync(trans, cmd);
2007 }
2008 
2009 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
2010 			     struct iwl_txq *txq, u8 hdr_len,
2011 			     struct iwl_cmd_meta *out_meta)
2012 {
2013 	u16 head_tb_len;
2014 	int i;
2015 
2016 	/*
2017 	 * Set up TFD's third entry to point directly to remainder
2018 	 * of skb's head, if any
2019 	 */
2020 	head_tb_len = skb_headlen(skb) - hdr_len;
2021 
2022 	if (head_tb_len > 0) {
2023 		dma_addr_t tb_phys = dma_map_single(trans->dev,
2024 						    skb->data + hdr_len,
2025 						    head_tb_len, DMA_TO_DEVICE);
2026 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
2027 			return -EINVAL;
2028 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb->data + hdr_len,
2029 					tb_phys, head_tb_len);
2030 		iwl_pcie_txq_build_tfd(trans, txq, tb_phys, head_tb_len, false);
2031 	}
2032 
2033 	/* set up the remaining entries to point to the data */
2034 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2035 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2036 		dma_addr_t tb_phys;
2037 		int tb_idx;
2038 
2039 		if (!skb_frag_size(frag))
2040 			continue;
2041 
2042 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
2043 					   skb_frag_size(frag), DMA_TO_DEVICE);
2044 
2045 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
2046 			return -EINVAL;
2047 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb_frag_address(frag),
2048 					tb_phys, skb_frag_size(frag));
2049 		tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2050 						skb_frag_size(frag), false);
2051 		if (tb_idx < 0)
2052 			return tb_idx;
2053 
2054 		out_meta->tbs |= BIT(tb_idx);
2055 	}
2056 
2057 	return 0;
2058 }
2059 
2060 #ifdef CONFIG_INET
2061 struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len,
2062 				      struct sk_buff *skb)
2063 {
2064 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2065 	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
2066 	struct page **page_ptr;
2067 
2068 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
2069 
2070 	if (WARN_ON(*page_ptr))
2071 		return NULL;
2072 
2073 	if (!p->page)
2074 		goto alloc;
2075 
2076 	/*
2077 	 * Check if there's enough room on this page
2078 	 *
2079 	 * Note that we put a page chaining pointer *last* in the
2080 	 * page - we need it somewhere, and if it's there then we
2081 	 * avoid DMA mapping the last bits of the page which may
2082 	 * trigger the 32-bit boundary hardware bug.
2083 	 *
2084 	 * (see also get_workaround_page() in tx-gen2.c)
2085 	 */
2086 	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE -
2087 			   sizeof(void *))
2088 		goto out;
2089 
2090 	/* We don't have enough room on this page, get a new one. */
2091 	__free_page(p->page);
2092 
2093 alloc:
2094 	p->page = alloc_page(GFP_ATOMIC);
2095 	if (!p->page)
2096 		return NULL;
2097 	p->pos = page_address(p->page);
2098 	/* set the chaining pointer to NULL */
2099 	*(void **)(page_address(p->page) + PAGE_SIZE - sizeof(void *)) = NULL;
2100 out:
2101 	*page_ptr = p->page;
2102 	get_page(p->page);
2103 	return p;
2104 }
2105 
2106 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
2107 					bool ipv6, unsigned int len)
2108 {
2109 	if (ipv6) {
2110 		struct ipv6hdr *iphv6 = iph;
2111 
2112 		tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
2113 					       len + tcph->doff * 4,
2114 					       IPPROTO_TCP, 0);
2115 	} else {
2116 		struct iphdr *iphv4 = iph;
2117 
2118 		ip_send_check(iphv4);
2119 		tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
2120 						 len + tcph->doff * 4,
2121 						 IPPROTO_TCP, 0);
2122 	}
2123 }
2124 
2125 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2126 				   struct iwl_txq *txq, u8 hdr_len,
2127 				   struct iwl_cmd_meta *out_meta,
2128 				   struct iwl_device_tx_cmd *dev_cmd,
2129 				   u16 tb1_len)
2130 {
2131 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
2132 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
2133 	struct ieee80211_hdr *hdr = (void *)skb->data;
2134 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
2135 	unsigned int mss = skb_shinfo(skb)->gso_size;
2136 	u16 length, iv_len, amsdu_pad;
2137 	u8 *start_hdr;
2138 	struct iwl_tso_hdr_page *hdr_page;
2139 	struct tso_t tso;
2140 
2141 	/* if the packet is protected, then it must be CCMP or GCMP */
2142 	BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
2143 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
2144 		IEEE80211_CCMP_HDR_LEN : 0;
2145 
2146 	trace_iwlwifi_dev_tx(trans->dev, skb,
2147 			     iwl_pcie_get_tfd(trans, txq, txq->write_ptr),
2148 			     trans_pcie->tfd_size,
2149 			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0);
2150 
2151 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
2152 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
2153 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
2154 	amsdu_pad = 0;
2155 
2156 	/* total amount of header we may need for this A-MSDU */
2157 	hdr_room = DIV_ROUND_UP(total_len, mss) *
2158 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
2159 
2160 	/* Our device supports 9 segments at most, it will fit in 1 page */
2161 	hdr_page = get_page_hdr(trans, hdr_room, skb);
2162 	if (!hdr_page)
2163 		return -ENOMEM;
2164 
2165 	start_hdr = hdr_page->pos;
2166 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
2167 	hdr_page->pos += iv_len;
2168 
2169 	/*
2170 	 * Pull the ieee80211 header + IV to be able to use TSO core,
2171 	 * we will restore it for the tx_status flow.
2172 	 */
2173 	skb_pull(skb, hdr_len + iv_len);
2174 
2175 	/*
2176 	 * Remove the length of all the headers that we don't actually
2177 	 * have in the MPDU by themselves, but that we duplicate into
2178 	 * all the different MSDUs inside the A-MSDU.
2179 	 */
2180 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
2181 
2182 	tso_start(skb, &tso);
2183 
2184 	while (total_len) {
2185 		/* this is the data left for this subframe */
2186 		unsigned int data_left =
2187 			min_t(unsigned int, mss, total_len);
2188 		struct sk_buff *csum_skb = NULL;
2189 		unsigned int hdr_tb_len;
2190 		dma_addr_t hdr_tb_phys;
2191 		struct tcphdr *tcph;
2192 		u8 *iph, *subf_hdrs_start = hdr_page->pos;
2193 
2194 		total_len -= data_left;
2195 
2196 		memset(hdr_page->pos, 0, amsdu_pad);
2197 		hdr_page->pos += amsdu_pad;
2198 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2199 				  data_left)) & 0x3;
2200 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2201 		hdr_page->pos += ETH_ALEN;
2202 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2203 		hdr_page->pos += ETH_ALEN;
2204 
2205 		length = snap_ip_tcp_hdrlen + data_left;
2206 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2207 		hdr_page->pos += sizeof(length);
2208 
2209 		/*
2210 		 * This will copy the SNAP as well which will be considered
2211 		 * as MAC header.
2212 		 */
2213 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2214 		iph = hdr_page->pos + 8;
2215 		tcph = (void *)(iph + ip_hdrlen);
2216 
2217 		/* For testing on current hardware only */
2218 		if (trans_pcie->sw_csum_tx) {
2219 			csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2220 					     GFP_ATOMIC);
2221 			if (!csum_skb)
2222 				return -ENOMEM;
2223 
2224 			iwl_compute_pseudo_hdr_csum(iph, tcph,
2225 						    skb->protocol ==
2226 							htons(ETH_P_IPV6),
2227 						    data_left);
2228 
2229 			skb_put_data(csum_skb, tcph, tcp_hdrlen(skb));
2230 			skb_reset_transport_header(csum_skb);
2231 			csum_skb->csum_start =
2232 				(unsigned char *)tcp_hdr(csum_skb) -
2233 						 csum_skb->head;
2234 		}
2235 
2236 		hdr_page->pos += snap_ip_tcp_hdrlen;
2237 
2238 		hdr_tb_len = hdr_page->pos - start_hdr;
2239 		hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2240 					     hdr_tb_len, DMA_TO_DEVICE);
2241 		if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2242 			dev_kfree_skb(csum_skb);
2243 			return -EINVAL;
2244 		}
2245 		iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2246 				       hdr_tb_len, false);
2247 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
2248 					hdr_tb_phys, hdr_tb_len);
2249 		/* add this subframe's headers' length to the tx_cmd */
2250 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
2251 
2252 		/* prepare the start_hdr for the next subframe */
2253 		start_hdr = hdr_page->pos;
2254 
2255 		/* put the payload */
2256 		while (data_left) {
2257 			unsigned int size = min_t(unsigned int, tso.size,
2258 						  data_left);
2259 			dma_addr_t tb_phys;
2260 
2261 			if (trans_pcie->sw_csum_tx)
2262 				skb_put_data(csum_skb, tso.data, size);
2263 
2264 			tb_phys = dma_map_single(trans->dev, tso.data,
2265 						 size, DMA_TO_DEVICE);
2266 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2267 				dev_kfree_skb(csum_skb);
2268 				return -EINVAL;
2269 			}
2270 
2271 			iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2272 					       size, false);
2273 			trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data,
2274 						tb_phys, size);
2275 
2276 			data_left -= size;
2277 			tso_build_data(skb, &tso, size);
2278 		}
2279 
2280 		/* For testing on early hardware only */
2281 		if (trans_pcie->sw_csum_tx) {
2282 			__wsum csum;
2283 
2284 			csum = skb_checksum(csum_skb,
2285 					    skb_checksum_start_offset(csum_skb),
2286 					    csum_skb->len -
2287 					    skb_checksum_start_offset(csum_skb),
2288 					    0);
2289 			dev_kfree_skb(csum_skb);
2290 			dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2291 						hdr_tb_len, DMA_TO_DEVICE);
2292 			tcph->check = csum_fold(csum);
2293 			dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2294 						   hdr_tb_len, DMA_TO_DEVICE);
2295 		}
2296 	}
2297 
2298 	/* re -add the WiFi header and IV */
2299 	skb_push(skb, hdr_len + iv_len);
2300 
2301 	return 0;
2302 }
2303 #else /* CONFIG_INET */
2304 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2305 				   struct iwl_txq *txq, u8 hdr_len,
2306 				   struct iwl_cmd_meta *out_meta,
2307 				   struct iwl_device_tx_cmd *dev_cmd,
2308 				   u16 tb1_len)
2309 {
2310 	/* No A-MSDU without CONFIG_INET */
2311 	WARN_ON(1);
2312 
2313 	return -1;
2314 }
2315 #endif /* CONFIG_INET */
2316 
2317 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2318 		      struct iwl_device_tx_cmd *dev_cmd, int txq_id)
2319 {
2320 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2321 	struct ieee80211_hdr *hdr;
2322 	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2323 	struct iwl_cmd_meta *out_meta;
2324 	struct iwl_txq *txq;
2325 	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2326 	void *tb1_addr;
2327 	void *tfd;
2328 	u16 len, tb1_len;
2329 	bool wait_write_ptr;
2330 	__le16 fc;
2331 	u8 hdr_len;
2332 	u16 wifi_seq;
2333 	bool amsdu;
2334 
2335 	txq = trans_pcie->txq[txq_id];
2336 
2337 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2338 		      "TX on unused queue %d\n", txq_id))
2339 		return -EINVAL;
2340 
2341 	if (unlikely(trans_pcie->sw_csum_tx &&
2342 		     skb->ip_summed == CHECKSUM_PARTIAL)) {
2343 		int offs = skb_checksum_start_offset(skb);
2344 		int csum_offs = offs + skb->csum_offset;
2345 		__wsum csum;
2346 
2347 		if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2348 			return -1;
2349 
2350 		csum = skb_checksum(skb, offs, skb->len - offs, 0);
2351 		*(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2352 
2353 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2354 	}
2355 
2356 	if (skb_is_nonlinear(skb) &&
2357 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
2358 	    __skb_linearize(skb))
2359 		return -ENOMEM;
2360 
2361 	/* mac80211 always puts the full header into the SKB's head,
2362 	 * so there's no need to check if it's readable there
2363 	 */
2364 	hdr = (struct ieee80211_hdr *)skb->data;
2365 	fc = hdr->frame_control;
2366 	hdr_len = ieee80211_hdrlen(fc);
2367 
2368 	spin_lock(&txq->lock);
2369 
2370 	if (iwl_queue_space(trans, txq) < txq->high_mark) {
2371 		iwl_stop_queue(trans, txq);
2372 
2373 		/* don't put the packet on the ring, if there is no room */
2374 		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
2375 			struct iwl_device_tx_cmd **dev_cmd_ptr;
2376 
2377 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
2378 					       trans_pcie->dev_cmd_offs);
2379 
2380 			*dev_cmd_ptr = dev_cmd;
2381 			__skb_queue_tail(&txq->overflow_q, skb);
2382 
2383 			spin_unlock(&txq->lock);
2384 			return 0;
2385 		}
2386 	}
2387 
2388 	/* In AGG mode, the index in the ring must correspond to the WiFi
2389 	 * sequence number. This is a HW requirements to help the SCD to parse
2390 	 * the BA.
2391 	 * Check here that the packets are in the right place on the ring.
2392 	 */
2393 	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2394 	WARN_ONCE(txq->ampdu &&
2395 		  (wifi_seq & 0xff) != txq->write_ptr,
2396 		  "Q: %d WiFi Seq %d tfdNum %d",
2397 		  txq_id, wifi_seq, txq->write_ptr);
2398 
2399 	/* Set up driver data for this TFD */
2400 	txq->entries[txq->write_ptr].skb = skb;
2401 	txq->entries[txq->write_ptr].cmd = dev_cmd;
2402 
2403 	dev_cmd->hdr.sequence =
2404 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2405 			    INDEX_TO_SEQ(txq->write_ptr)));
2406 
2407 	tb0_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
2408 	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2409 		       offsetof(struct iwl_tx_cmd, scratch);
2410 
2411 	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2412 	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2413 
2414 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
2415 	out_meta = &txq->entries[txq->write_ptr].meta;
2416 	out_meta->flags = 0;
2417 
2418 	/*
2419 	 * The second TB (tb1) points to the remainder of the TX command
2420 	 * and the 802.11 header - dword aligned size
2421 	 * (This calculation modifies the TX command, so do it before the
2422 	 * setup of the first TB)
2423 	 */
2424 	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2425 	      hdr_len - IWL_FIRST_TB_SIZE;
2426 	/* do not align A-MSDU to dword as the subframe header aligns it */
2427 	amsdu = ieee80211_is_data_qos(fc) &&
2428 		(*ieee80211_get_qos_ctl(hdr) &
2429 		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
2430 	if (trans_pcie->sw_csum_tx || !amsdu) {
2431 		tb1_len = ALIGN(len, 4);
2432 		/* Tell NIC about any 2-byte padding after MAC header */
2433 		if (tb1_len != len)
2434 			tx_cmd->tx_flags |= cpu_to_le32(TX_CMD_FLG_MH_PAD);
2435 	} else {
2436 		tb1_len = len;
2437 	}
2438 
2439 	/*
2440 	 * The first TB points to bi-directional DMA data, we'll
2441 	 * memcpy the data into it later.
2442 	 */
2443 	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2444 			       IWL_FIRST_TB_SIZE, true);
2445 
2446 	/* there must be data left over for TB1 or this code must be changed */
2447 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
2448 
2449 	/* map the data for TB1 */
2450 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
2451 	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2452 	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2453 		goto out_err;
2454 	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2455 
2456 	trace_iwlwifi_dev_tx(trans->dev, skb,
2457 			     iwl_pcie_get_tfd(trans, txq,
2458 					      txq->write_ptr),
2459 			     trans_pcie->tfd_size,
2460 			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
2461 			     hdr_len);
2462 
2463 	/*
2464 	 * If gso_size wasn't set, don't give the frame "amsdu treatment"
2465 	 * (adding subframes, etc.).
2466 	 * This can happen in some testing flows when the amsdu was already
2467 	 * pre-built, and we just need to send the resulting skb.
2468 	 */
2469 	if (amsdu && skb_shinfo(skb)->gso_size) {
2470 		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2471 						     out_meta, dev_cmd,
2472 						     tb1_len)))
2473 			goto out_err;
2474 	} else {
2475 		struct sk_buff *frag;
2476 
2477 		if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2478 					       out_meta)))
2479 			goto out_err;
2480 
2481 		skb_walk_frags(skb, frag) {
2482 			if (unlikely(iwl_fill_data_tbs(trans, frag, txq, 0,
2483 						       out_meta)))
2484 				goto out_err;
2485 		}
2486 	}
2487 
2488 	/* building the A-MSDU might have changed this data, so memcpy it now */
2489 	memcpy(&txq->first_tb_bufs[txq->write_ptr], dev_cmd, IWL_FIRST_TB_SIZE);
2490 
2491 	tfd = iwl_pcie_get_tfd(trans, txq, txq->write_ptr);
2492 	/* Set up entry for this TFD in Tx byte-count array */
2493 	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len),
2494 					 iwl_pcie_tfd_get_num_tbs(trans, tfd));
2495 
2496 	wait_write_ptr = ieee80211_has_morefrags(fc);
2497 
2498 	/* start timer if queue currently empty */
2499 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) {
2500 		/*
2501 		 * If the TXQ is active, then set the timer, if not,
2502 		 * set the timer in remainder so that the timer will
2503 		 * be armed with the right value when the station will
2504 		 * wake up.
2505 		 */
2506 		if (!txq->frozen)
2507 			mod_timer(&txq->stuck_timer,
2508 				  jiffies + txq->wd_timeout);
2509 		else
2510 			txq->frozen_expiry_remainder = txq->wd_timeout;
2511 	}
2512 
2513 	/* Tell device the write index *just past* this latest filled TFD */
2514 	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
2515 	if (!wait_write_ptr)
2516 		iwl_pcie_txq_inc_wr_ptr(trans, txq);
2517 
2518 	/*
2519 	 * At this point the frame is "transmitted" successfully
2520 	 * and we will get a TX status notification eventually.
2521 	 */
2522 	spin_unlock(&txq->lock);
2523 	return 0;
2524 out_err:
2525 	iwl_pcie_tfd_unmap(trans, out_meta, txq, txq->write_ptr);
2526 	spin_unlock(&txq->lock);
2527 	return -1;
2528 }
2529