1 /******************************************************************************
2  *
3  * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4  * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
5  * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
6  *
7  * Portions of this file are derived from the ipw3945 project, as well
8  * as portions of the ieee80211 subsystem header files.
9  *
10  * This program is free software; you can redistribute it and/or modify it
11  * under the terms of version 2 of the GNU General Public License as
12  * published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful, but WITHOUT
15  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
17  * more details.
18  *
19  * You should have received a copy of the GNU General Public License along with
20  * this program; if not, write to the Free Software Foundation, Inc.,
21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
22  *
23  * The full GNU General Public License is included in this distribution in the
24  * file called LICENSE.
25  *
26  * Contact Information:
27  *  Intel Linux Wireless <linuxwifi@intel.com>
28  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
29  *
30  *****************************************************************************/
31 #include <linux/etherdevice.h>
32 #include <linux/ieee80211.h>
33 #include <linux/slab.h>
34 #include <linux/sched.h>
35 #include <linux/pm_runtime.h>
36 #include <net/ip6_checksum.h>
37 #include <net/tso.h>
38 
39 #include "iwl-debug.h"
40 #include "iwl-csr.h"
41 #include "iwl-prph.h"
42 #include "iwl-io.h"
43 #include "iwl-scd.h"
44 #include "iwl-op-mode.h"
45 #include "internal.h"
46 /* FIXME: need to abstract out TX command (once we know what it looks like) */
47 #include "dvm/commands.h"
48 
49 #define IWL_TX_CRC_SIZE 4
50 #define IWL_TX_DELIMITER_SIZE 4
51 
52 /*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
53  * DMA services
54  *
55  * Theory of operation
56  *
57  * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
58  * of buffer descriptors, each of which points to one or more data buffers for
59  * the device to read from or fill.  Driver and device exchange status of each
60  * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
61  * entries in each circular buffer, to protect against confusing empty and full
62  * queue states.
63  *
64  * The device reads or writes the data in the queues via the device's several
65  * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
66  *
67  * For Tx queue, there are low mark and high mark limits. If, after queuing
68  * the packet for Tx, free space become < low mark, Tx queue stopped. When
69  * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
70  * Tx queue resumed.
71  *
72  ***************************************************/
73 
74 int iwl_queue_space(const struct iwl_txq *q)
75 {
76 	unsigned int max;
77 	unsigned int used;
78 
79 	/*
80 	 * To avoid ambiguity between empty and completely full queues, there
81 	 * should always be less than TFD_QUEUE_SIZE_MAX elements in the queue.
82 	 * If q->n_window is smaller than TFD_QUEUE_SIZE_MAX, there is no need
83 	 * to reserve any queue entries for this purpose.
84 	 */
85 	if (q->n_window < TFD_QUEUE_SIZE_MAX)
86 		max = q->n_window;
87 	else
88 		max = TFD_QUEUE_SIZE_MAX - 1;
89 
90 	/*
91 	 * TFD_QUEUE_SIZE_MAX is a power of 2, so the following is equivalent to
92 	 * modulo by TFD_QUEUE_SIZE_MAX and is well defined.
93 	 */
94 	used = (q->write_ptr - q->read_ptr) & (TFD_QUEUE_SIZE_MAX - 1);
95 
96 	if (WARN_ON(used > max))
97 		return 0;
98 
99 	return max - used;
100 }
101 
102 /*
103  * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
104  */
105 static int iwl_queue_init(struct iwl_txq *q, int slots_num)
106 {
107 	q->n_window = slots_num;
108 
109 	/* slots_num must be power-of-two size, otherwise
110 	 * get_cmd_index is broken. */
111 	if (WARN_ON(!is_power_of_2(slots_num)))
112 		return -EINVAL;
113 
114 	q->low_mark = q->n_window / 4;
115 	if (q->low_mark < 4)
116 		q->low_mark = 4;
117 
118 	q->high_mark = q->n_window / 8;
119 	if (q->high_mark < 2)
120 		q->high_mark = 2;
121 
122 	q->write_ptr = 0;
123 	q->read_ptr = 0;
124 
125 	return 0;
126 }
127 
128 int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
129 			   struct iwl_dma_ptr *ptr, size_t size)
130 {
131 	if (WARN_ON(ptr->addr))
132 		return -EINVAL;
133 
134 	ptr->addr = dma_alloc_coherent(trans->dev, size,
135 				       &ptr->dma, GFP_KERNEL);
136 	if (!ptr->addr)
137 		return -ENOMEM;
138 	ptr->size = size;
139 	return 0;
140 }
141 
142 void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr)
143 {
144 	if (unlikely(!ptr->addr))
145 		return;
146 
147 	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
148 	memset(ptr, 0, sizeof(*ptr));
149 }
150 
151 static void iwl_pcie_txq_stuck_timer(unsigned long data)
152 {
153 	struct iwl_txq *txq = (void *)data;
154 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
155 	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
156 
157 	spin_lock(&txq->lock);
158 	/* check if triggered erroneously */
159 	if (txq->read_ptr == txq->write_ptr) {
160 		spin_unlock(&txq->lock);
161 		return;
162 	}
163 	spin_unlock(&txq->lock);
164 
165 	iwl_trans_pcie_log_scd_error(trans, txq);
166 
167 	iwl_force_nmi(trans);
168 }
169 
170 /*
171  * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
172  */
173 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
174 					     struct iwl_txq *txq, u16 byte_cnt,
175 					     int num_tbs)
176 {
177 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
178 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
179 	int write_ptr = txq->write_ptr;
180 	int txq_id = txq->id;
181 	u8 sec_ctl = 0;
182 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
183 	__le16 bc_ent;
184 	struct iwl_tx_cmd *tx_cmd =
185 		(void *)txq->entries[txq->write_ptr].cmd->payload;
186 	u8 sta_id = tx_cmd->sta_id;
187 
188 	scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
189 
190 	sec_ctl = tx_cmd->sec_ctl;
191 
192 	switch (sec_ctl & TX_CMD_SEC_MSK) {
193 	case TX_CMD_SEC_CCM:
194 		len += IEEE80211_CCMP_MIC_LEN;
195 		break;
196 	case TX_CMD_SEC_TKIP:
197 		len += IEEE80211_TKIP_ICV_LEN;
198 		break;
199 	case TX_CMD_SEC_WEP:
200 		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
201 		break;
202 	}
203 	if (trans_pcie->bc_table_dword)
204 		len = DIV_ROUND_UP(len, 4);
205 
206 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
207 		return;
208 
209 	bc_ent = cpu_to_le16(len | (sta_id << 12));
210 
211 	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
212 
213 	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
214 		scd_bc_tbl[txq_id].
215 			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
216 }
217 
218 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
219 					    struct iwl_txq *txq)
220 {
221 	struct iwl_trans_pcie *trans_pcie =
222 		IWL_TRANS_GET_PCIE_TRANS(trans);
223 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
224 	int txq_id = txq->id;
225 	int read_ptr = txq->read_ptr;
226 	u8 sta_id = 0;
227 	__le16 bc_ent;
228 	struct iwl_tx_cmd *tx_cmd =
229 		(void *)txq->entries[read_ptr].cmd->payload;
230 
231 	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
232 
233 	if (txq_id != trans_pcie->cmd_queue)
234 		sta_id = tx_cmd->sta_id;
235 
236 	bc_ent = cpu_to_le16(1 | (sta_id << 12));
237 
238 	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
239 
240 	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
241 		scd_bc_tbl[txq_id].
242 			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
243 }
244 
245 /*
246  * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
247  */
248 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
249 				    struct iwl_txq *txq)
250 {
251 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
252 	u32 reg = 0;
253 	int txq_id = txq->id;
254 
255 	lockdep_assert_held(&txq->lock);
256 
257 	/*
258 	 * explicitly wake up the NIC if:
259 	 * 1. shadow registers aren't enabled
260 	 * 2. NIC is woken up for CMD regardless of shadow outside this function
261 	 * 3. there is a chance that the NIC is asleep
262 	 */
263 	if (!trans->cfg->base_params->shadow_reg_enable &&
264 	    txq_id != trans_pcie->cmd_queue &&
265 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
266 		/*
267 		 * wake up nic if it's powered down ...
268 		 * uCode will wake up, and interrupt us again, so next
269 		 * time we'll skip this part.
270 		 */
271 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
272 
273 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
274 			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
275 				       txq_id, reg);
276 			iwl_set_bit(trans, CSR_GP_CNTRL,
277 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
278 			txq->need_update = true;
279 			return;
280 		}
281 	}
282 
283 	/*
284 	 * if not in power-save mode, uCode will never sleep when we're
285 	 * trying to tx (during RFKILL, we're not trying to tx).
286 	 */
287 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr);
288 	if (!txq->block)
289 		iwl_write32(trans, HBUS_TARG_WRPTR,
290 			    txq->write_ptr | (txq_id << 8));
291 }
292 
293 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
294 {
295 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
296 	int i;
297 
298 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
299 		struct iwl_txq *txq = trans_pcie->txq[i];
300 
301 		spin_lock_bh(&txq->lock);
302 		if (txq->need_update) {
303 			iwl_pcie_txq_inc_wr_ptr(trans, txq);
304 			txq->need_update = false;
305 		}
306 		spin_unlock_bh(&txq->lock);
307 	}
308 }
309 
310 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_trans *trans,
311 						  void *_tfd, u8 idx)
312 {
313 
314 	if (trans->cfg->use_tfh) {
315 		struct iwl_tfh_tfd *tfd = _tfd;
316 		struct iwl_tfh_tb *tb = &tfd->tbs[idx];
317 
318 		return (dma_addr_t)(le64_to_cpu(tb->addr));
319 	} else {
320 		struct iwl_tfd *tfd = _tfd;
321 		struct iwl_tfd_tb *tb = &tfd->tbs[idx];
322 		dma_addr_t addr = get_unaligned_le32(&tb->lo);
323 		dma_addr_t hi_len;
324 
325 		if (sizeof(dma_addr_t) <= sizeof(u32))
326 			return addr;
327 
328 		hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
329 
330 		/*
331 		 * shift by 16 twice to avoid warnings on 32-bit
332 		 * (where this code never runs anyway due to the
333 		 * if statement above)
334 		 */
335 		return addr | ((hi_len << 16) << 16);
336 	}
337 }
338 
339 static inline void iwl_pcie_tfd_set_tb(struct iwl_trans *trans, void *tfd,
340 				       u8 idx, dma_addr_t addr, u16 len)
341 {
342 	struct iwl_tfd *tfd_fh = (void *)tfd;
343 	struct iwl_tfd_tb *tb = &tfd_fh->tbs[idx];
344 
345 	u16 hi_n_len = len << 4;
346 
347 	put_unaligned_le32(addr, &tb->lo);
348 	hi_n_len |= iwl_get_dma_hi_addr(addr);
349 
350 	tb->hi_n_len = cpu_to_le16(hi_n_len);
351 
352 	tfd_fh->num_tbs = idx + 1;
353 }
354 
355 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_trans *trans, void *_tfd)
356 {
357 	if (trans->cfg->use_tfh) {
358 		struct iwl_tfh_tfd *tfd = _tfd;
359 
360 		return le16_to_cpu(tfd->num_tbs) & 0x1f;
361 	} else {
362 		struct iwl_tfd *tfd = _tfd;
363 
364 		return tfd->num_tbs & 0x1f;
365 	}
366 }
367 
368 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
369 			       struct iwl_cmd_meta *meta,
370 			       struct iwl_txq *txq, int index)
371 {
372 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
373 	int i, num_tbs;
374 	void *tfd = iwl_pcie_get_tfd(trans_pcie, txq, index);
375 
376 	/* Sanity check on number of chunks */
377 	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
378 
379 	if (num_tbs >= trans_pcie->max_tbs) {
380 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
381 		/* @todo issue fatal error, it is quite serious situation */
382 		return;
383 	}
384 
385 	/* first TB is never freed - it's the bidirectional DMA data */
386 
387 	for (i = 1; i < num_tbs; i++) {
388 		if (meta->tbs & BIT(i))
389 			dma_unmap_page(trans->dev,
390 				       iwl_pcie_tfd_tb_get_addr(trans, tfd, i),
391 				       iwl_pcie_tfd_tb_get_len(trans, tfd, i),
392 				       DMA_TO_DEVICE);
393 		else
394 			dma_unmap_single(trans->dev,
395 					 iwl_pcie_tfd_tb_get_addr(trans, tfd,
396 								  i),
397 					 iwl_pcie_tfd_tb_get_len(trans, tfd,
398 								 i),
399 					 DMA_TO_DEVICE);
400 	}
401 
402 	if (trans->cfg->use_tfh) {
403 		struct iwl_tfh_tfd *tfd_fh = (void *)tfd;
404 
405 		tfd_fh->num_tbs = 0;
406 	} else {
407 		struct iwl_tfd *tfd_fh = (void *)tfd;
408 
409 		tfd_fh->num_tbs = 0;
410 	}
411 
412 }
413 
414 /*
415  * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
416  * @trans - transport private data
417  * @txq - tx queue
418  * @dma_dir - the direction of the DMA mapping
419  *
420  * Does NOT advance any TFD circular buffer read/write indexes
421  * Does NOT free the TFD itself (which is within circular buffer)
422  */
423 void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
424 {
425 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
426 	 * idx is bounded by n_window
427 	 */
428 	int rd_ptr = txq->read_ptr;
429 	int idx = get_cmd_index(txq, rd_ptr);
430 
431 	lockdep_assert_held(&txq->lock);
432 
433 	/* We have only q->n_window txq->entries, but we use
434 	 * TFD_QUEUE_SIZE_MAX tfds
435 	 */
436 	iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, txq, rd_ptr);
437 
438 	/* free SKB */
439 	if (txq->entries) {
440 		struct sk_buff *skb;
441 
442 		skb = txq->entries[idx].skb;
443 
444 		/* Can be called from irqs-disabled context
445 		 * If skb is not NULL, it means that the whole queue is being
446 		 * freed and that the queue is not empty - free the skb
447 		 */
448 		if (skb) {
449 			iwl_op_mode_free_skb(trans->op_mode, skb);
450 			txq->entries[idx].skb = NULL;
451 		}
452 	}
453 }
454 
455 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
456 				  dma_addr_t addr, u16 len, bool reset)
457 {
458 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
459 	void *tfd;
460 	u32 num_tbs;
461 
462 	tfd = txq->tfds + trans_pcie->tfd_size * txq->write_ptr;
463 
464 	if (reset)
465 		memset(tfd, 0, trans_pcie->tfd_size);
466 
467 	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
468 
469 	/* Each TFD can point to a maximum max_tbs Tx buffers */
470 	if (num_tbs >= trans_pcie->max_tbs) {
471 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
472 			trans_pcie->max_tbs);
473 		return -EINVAL;
474 	}
475 
476 	if (WARN(addr & ~IWL_TX_DMA_MASK,
477 		 "Unaligned address = %llx\n", (unsigned long long)addr))
478 		return -EINVAL;
479 
480 	iwl_pcie_tfd_set_tb(trans, tfd, num_tbs, addr, len);
481 
482 	return num_tbs;
483 }
484 
485 int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq,
486 		       int slots_num, bool cmd_queue)
487 {
488 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
489 	size_t tfd_sz = trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX;
490 	size_t tb0_buf_sz;
491 	int i;
492 
493 	if (WARN_ON(txq->entries || txq->tfds))
494 		return -EINVAL;
495 
496 	setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer,
497 		    (unsigned long)txq);
498 	txq->trans_pcie = trans_pcie;
499 
500 	txq->n_window = slots_num;
501 
502 	txq->entries = kcalloc(slots_num,
503 			       sizeof(struct iwl_pcie_txq_entry),
504 			       GFP_KERNEL);
505 
506 	if (!txq->entries)
507 		goto error;
508 
509 	if (cmd_queue)
510 		for (i = 0; i < slots_num; i++) {
511 			txq->entries[i].cmd =
512 				kmalloc(sizeof(struct iwl_device_cmd),
513 					GFP_KERNEL);
514 			if (!txq->entries[i].cmd)
515 				goto error;
516 		}
517 
518 	/* Circular buffer of transmit frame descriptors (TFDs),
519 	 * shared with device */
520 	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
521 				       &txq->dma_addr, GFP_KERNEL);
522 	if (!txq->tfds)
523 		goto error;
524 
525 	BUILD_BUG_ON(IWL_FIRST_TB_SIZE_ALIGN != sizeof(*txq->first_tb_bufs));
526 
527 	tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
528 
529 	txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
530 					      &txq->first_tb_dma,
531 					      GFP_KERNEL);
532 	if (!txq->first_tb_bufs)
533 		goto err_free_tfds;
534 
535 	return 0;
536 err_free_tfds:
537 	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
538 error:
539 	if (txq->entries && cmd_queue)
540 		for (i = 0; i < slots_num; i++)
541 			kfree(txq->entries[i].cmd);
542 	kfree(txq->entries);
543 	txq->entries = NULL;
544 
545 	return -ENOMEM;
546 
547 }
548 
549 int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
550 		      int slots_num, bool cmd_queue)
551 {
552 	int ret;
553 
554 	txq->need_update = false;
555 
556 	/* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise
557 	 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
558 	BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1));
559 
560 	/* Initialize queue's high/low-water marks, and head/tail indexes */
561 	ret = iwl_queue_init(txq, slots_num);
562 	if (ret)
563 		return ret;
564 
565 	spin_lock_init(&txq->lock);
566 
567 	if (cmd_queue) {
568 		static struct lock_class_key iwl_pcie_cmd_queue_lock_class;
569 
570 		lockdep_set_class(&txq->lock, &iwl_pcie_cmd_queue_lock_class);
571 	}
572 
573 	__skb_queue_head_init(&txq->overflow_q);
574 
575 	return 0;
576 }
577 
578 static void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
579 				   struct sk_buff *skb)
580 {
581 	struct page **page_ptr;
582 
583 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
584 
585 	if (*page_ptr) {
586 		__free_page(*page_ptr);
587 		*page_ptr = NULL;
588 	}
589 }
590 
591 static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
592 {
593 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
594 
595 	lockdep_assert_held(&trans_pcie->reg_lock);
596 
597 	if (trans_pcie->ref_cmd_in_flight) {
598 		trans_pcie->ref_cmd_in_flight = false;
599 		IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
600 		iwl_trans_unref(trans);
601 	}
602 
603 	if (!trans->cfg->base_params->apmg_wake_up_wa)
604 		return;
605 	if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
606 		return;
607 
608 	trans_pcie->cmd_hold_nic_awake = false;
609 	__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
610 				   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
611 }
612 
613 /*
614  * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
615  */
616 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
617 {
618 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
619 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
620 
621 	spin_lock_bh(&txq->lock);
622 	while (txq->write_ptr != txq->read_ptr) {
623 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
624 				   txq_id, txq->read_ptr);
625 
626 		if (txq_id != trans_pcie->cmd_queue) {
627 			struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
628 
629 			if (WARN_ON_ONCE(!skb))
630 				continue;
631 
632 			iwl_pcie_free_tso_page(trans_pcie, skb);
633 		}
634 		iwl_pcie_txq_free_tfd(trans, txq);
635 		txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr);
636 
637 		if (txq->read_ptr == txq->write_ptr) {
638 			unsigned long flags;
639 
640 			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
641 			if (txq_id != trans_pcie->cmd_queue) {
642 				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
643 					      txq->id);
644 				iwl_trans_unref(trans);
645 			} else {
646 				iwl_pcie_clear_cmd_in_flight(trans);
647 			}
648 			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
649 		}
650 	}
651 
652 	while (!skb_queue_empty(&txq->overflow_q)) {
653 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
654 
655 		iwl_op_mode_free_skb(trans->op_mode, skb);
656 	}
657 
658 	spin_unlock_bh(&txq->lock);
659 
660 	/* just in case - this queue may have been stopped */
661 	iwl_wake_queue(trans, txq);
662 }
663 
664 /*
665  * iwl_pcie_txq_free - Deallocate DMA queue.
666  * @txq: Transmit queue to deallocate.
667  *
668  * Empty queue by removing and destroying all BD's.
669  * Free all buffers.
670  * 0-fill, but do not free "txq" descriptor structure.
671  */
672 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
673 {
674 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
675 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
676 	struct device *dev = trans->dev;
677 	int i;
678 
679 	if (WARN_ON(!txq))
680 		return;
681 
682 	iwl_pcie_txq_unmap(trans, txq_id);
683 
684 	/* De-alloc array of command/tx buffers */
685 	if (txq_id == trans_pcie->cmd_queue)
686 		for (i = 0; i < txq->n_window; i++) {
687 			kzfree(txq->entries[i].cmd);
688 			kzfree(txq->entries[i].free_buf);
689 		}
690 
691 	/* De-alloc circular buffer of TFDs */
692 	if (txq->tfds) {
693 		dma_free_coherent(dev,
694 				  trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX,
695 				  txq->tfds, txq->dma_addr);
696 		txq->dma_addr = 0;
697 		txq->tfds = NULL;
698 
699 		dma_free_coherent(dev,
700 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
701 				  txq->first_tb_bufs, txq->first_tb_dma);
702 	}
703 
704 	kfree(txq->entries);
705 	txq->entries = NULL;
706 
707 	del_timer_sync(&txq->stuck_timer);
708 
709 	/* 0-fill queue descriptor structure */
710 	memset(txq, 0, sizeof(*txq));
711 }
712 
713 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
714 {
715 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
716 	int nq = trans->cfg->base_params->num_of_queues;
717 	int chan;
718 	u32 reg_val;
719 	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
720 				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
721 
722 	/* make sure all queue are not stopped/used */
723 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
724 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
725 
726 	trans_pcie->scd_base_addr =
727 		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
728 
729 	WARN_ON(scd_base_addr != 0 &&
730 		scd_base_addr != trans_pcie->scd_base_addr);
731 
732 	/* reset context data, TX status and translation data */
733 	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
734 				   SCD_CONTEXT_MEM_LOWER_BOUND,
735 			    NULL, clear_dwords);
736 
737 	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
738 		       trans_pcie->scd_bc_tbls.dma >> 10);
739 
740 	/* The chain extension of the SCD doesn't work well. This feature is
741 	 * enabled by default by the HW, so we need to disable it manually.
742 	 */
743 	if (trans->cfg->base_params->scd_chain_ext_wa)
744 		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
745 
746 	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
747 				trans_pcie->cmd_fifo,
748 				trans_pcie->cmd_q_wdg_timeout);
749 
750 	/* Activate all Tx DMA/FIFO channels */
751 	iwl_scd_activate_fifos(trans);
752 
753 	/* Enable DMA channel */
754 	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
755 		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
756 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
757 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
758 
759 	/* Update FH chicken bits */
760 	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
761 	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
762 			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
763 
764 	/* Enable L1-Active */
765 	if (trans->cfg->device_family < IWL_DEVICE_FAMILY_8000)
766 		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
767 				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
768 }
769 
770 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
771 {
772 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
773 	int txq_id;
774 
775 	/*
776 	 * we should never get here in gen2 trans mode return early to avoid
777 	 * having invalid accesses
778 	 */
779 	if (WARN_ON_ONCE(trans->cfg->gen2))
780 		return;
781 
782 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
783 	     txq_id++) {
784 		struct iwl_txq *txq = trans_pcie->txq[txq_id];
785 		if (trans->cfg->use_tfh)
786 			iwl_write_direct64(trans,
787 					   FH_MEM_CBBC_QUEUE(trans, txq_id),
788 					   txq->dma_addr);
789 		else
790 			iwl_write_direct32(trans,
791 					   FH_MEM_CBBC_QUEUE(trans, txq_id),
792 					   txq->dma_addr >> 8);
793 		iwl_pcie_txq_unmap(trans, txq_id);
794 		txq->read_ptr = 0;
795 		txq->write_ptr = 0;
796 	}
797 
798 	/* Tell NIC where to find the "keep warm" buffer */
799 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
800 			   trans_pcie->kw.dma >> 4);
801 
802 	/*
803 	 * Send 0 as the scd_base_addr since the device may have be reset
804 	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
805 	 * contain garbage.
806 	 */
807 	iwl_pcie_tx_start(trans, 0);
808 }
809 
810 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
811 {
812 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
813 	unsigned long flags;
814 	int ch, ret;
815 	u32 mask = 0;
816 
817 	spin_lock(&trans_pcie->irq_lock);
818 
819 	if (!iwl_trans_grab_nic_access(trans, &flags))
820 		goto out;
821 
822 	/* Stop each Tx DMA channel */
823 	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
824 		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
825 		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
826 	}
827 
828 	/* Wait for DMA channels to be idle */
829 	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
830 	if (ret < 0)
831 		IWL_ERR(trans,
832 			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
833 			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
834 
835 	iwl_trans_release_nic_access(trans, &flags);
836 
837 out:
838 	spin_unlock(&trans_pcie->irq_lock);
839 }
840 
841 /*
842  * iwl_pcie_tx_stop - Stop all Tx DMA channels
843  */
844 int iwl_pcie_tx_stop(struct iwl_trans *trans)
845 {
846 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
847 	int txq_id;
848 
849 	/* Turn off all Tx DMA fifos */
850 	iwl_scd_deactivate_fifos(trans);
851 
852 	/* Turn off all Tx DMA channels */
853 	iwl_pcie_tx_stop_fh(trans);
854 
855 	/*
856 	 * This function can be called before the op_mode disabled the
857 	 * queues. This happens when we have an rfkill interrupt.
858 	 * Since we stop Tx altogether - mark the queues as stopped.
859 	 */
860 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
861 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
862 
863 	/* This can happen: start_hw, stop_device */
864 	if (!trans_pcie->txq_memory)
865 		return 0;
866 
867 	/* Unmap DMA from host system and free skb's */
868 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
869 	     txq_id++)
870 		iwl_pcie_txq_unmap(trans, txq_id);
871 
872 	return 0;
873 }
874 
875 /*
876  * iwl_trans_tx_free - Free TXQ Context
877  *
878  * Destroy all TX DMA queues and structures
879  */
880 void iwl_pcie_tx_free(struct iwl_trans *trans)
881 {
882 	int txq_id;
883 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
884 
885 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
886 
887 	/* Tx queues */
888 	if (trans_pcie->txq_memory) {
889 		for (txq_id = 0;
890 		     txq_id < trans->cfg->base_params->num_of_queues;
891 		     txq_id++) {
892 			iwl_pcie_txq_free(trans, txq_id);
893 			trans_pcie->txq[txq_id] = NULL;
894 		}
895 	}
896 
897 	kfree(trans_pcie->txq_memory);
898 	trans_pcie->txq_memory = NULL;
899 
900 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
901 
902 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
903 }
904 
905 /*
906  * iwl_pcie_tx_alloc - allocate TX context
907  * Allocate all Tx DMA structures and initialize them
908  */
909 static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
910 {
911 	int ret;
912 	int txq_id, slots_num;
913 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
914 
915 	u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
916 			sizeof(struct iwlagn_scd_bc_tbl);
917 
918 	/*It is not allowed to alloc twice, so warn when this happens.
919 	 * We cannot rely on the previous allocation, so free and fail */
920 	if (WARN_ON(trans_pcie->txq_memory)) {
921 		ret = -EINVAL;
922 		goto error;
923 	}
924 
925 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
926 				   scd_bc_tbls_size);
927 	if (ret) {
928 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
929 		goto error;
930 	}
931 
932 	/* Alloc keep-warm buffer */
933 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
934 	if (ret) {
935 		IWL_ERR(trans, "Keep Warm allocation failed\n");
936 		goto error;
937 	}
938 
939 	trans_pcie->txq_memory = kcalloc(trans->cfg->base_params->num_of_queues,
940 					 sizeof(struct iwl_txq), GFP_KERNEL);
941 	if (!trans_pcie->txq_memory) {
942 		IWL_ERR(trans, "Not enough memory for txq\n");
943 		ret = -ENOMEM;
944 		goto error;
945 	}
946 
947 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
948 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
949 	     txq_id++) {
950 		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
951 
952 		slots_num = cmd_queue ? TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
953 		trans_pcie->txq[txq_id] = &trans_pcie->txq_memory[txq_id];
954 		ret = iwl_pcie_txq_alloc(trans, trans_pcie->txq[txq_id],
955 					 slots_num, cmd_queue);
956 		if (ret) {
957 			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
958 			goto error;
959 		}
960 		trans_pcie->txq[txq_id]->id = txq_id;
961 	}
962 
963 	return 0;
964 
965 error:
966 	iwl_pcie_tx_free(trans);
967 
968 	return ret;
969 }
970 
971 int iwl_pcie_tx_init(struct iwl_trans *trans)
972 {
973 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
974 	int ret;
975 	int txq_id, slots_num;
976 	bool alloc = false;
977 
978 	if (!trans_pcie->txq_memory) {
979 		ret = iwl_pcie_tx_alloc(trans);
980 		if (ret)
981 			goto error;
982 		alloc = true;
983 	}
984 
985 	spin_lock(&trans_pcie->irq_lock);
986 
987 	/* Turn off all Tx DMA fifos */
988 	iwl_scd_deactivate_fifos(trans);
989 
990 	/* Tell NIC where to find the "keep warm" buffer */
991 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
992 			   trans_pcie->kw.dma >> 4);
993 
994 	spin_unlock(&trans_pcie->irq_lock);
995 
996 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
997 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
998 	     txq_id++) {
999 		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
1000 
1001 		slots_num = cmd_queue ? TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
1002 		ret = iwl_pcie_txq_init(trans, trans_pcie->txq[txq_id],
1003 					slots_num, cmd_queue);
1004 		if (ret) {
1005 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1006 			goto error;
1007 		}
1008 
1009 		/*
1010 		 * Tell nic where to find circular buffer of TFDs for a
1011 		 * given Tx queue, and enable the DMA channel used for that
1012 		 * queue.
1013 		 * Circular buffer (TFD queue in DRAM) physical base address
1014 		 */
1015 		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
1016 				   trans_pcie->txq[txq_id]->dma_addr >> 8);
1017 	}
1018 
1019 	iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
1020 	if (trans->cfg->base_params->num_of_queues > 20)
1021 		iwl_set_bits_prph(trans, SCD_GP_CTRL,
1022 				  SCD_GP_CTRL_ENABLE_31_QUEUES);
1023 
1024 	return 0;
1025 error:
1026 	/*Upon error, free only if we allocated something */
1027 	if (alloc)
1028 		iwl_pcie_tx_free(trans);
1029 	return ret;
1030 }
1031 
1032 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
1033 {
1034 	lockdep_assert_held(&txq->lock);
1035 
1036 	if (!txq->wd_timeout)
1037 		return;
1038 
1039 	/*
1040 	 * station is asleep and we send data - that must
1041 	 * be uAPSD or PS-Poll. Don't rearm the timer.
1042 	 */
1043 	if (txq->frozen)
1044 		return;
1045 
1046 	/*
1047 	 * if empty delete timer, otherwise move timer forward
1048 	 * since we're making progress on this queue
1049 	 */
1050 	if (txq->read_ptr == txq->write_ptr)
1051 		del_timer(&txq->stuck_timer);
1052 	else
1053 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1054 }
1055 
1056 /* Frees buffers until index _not_ inclusive */
1057 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1058 			    struct sk_buff_head *skbs)
1059 {
1060 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1061 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1062 	int tfd_num = ssn & (TFD_QUEUE_SIZE_MAX - 1);
1063 	int last_to_free;
1064 
1065 	/* This function is not meant to release cmd queue*/
1066 	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1067 		return;
1068 
1069 	spin_lock_bh(&txq->lock);
1070 
1071 	if (!test_bit(txq_id, trans_pcie->queue_used)) {
1072 		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1073 				    txq_id, ssn);
1074 		goto out;
1075 	}
1076 
1077 	if (txq->read_ptr == tfd_num)
1078 		goto out;
1079 
1080 	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1081 			   txq_id, txq->read_ptr, tfd_num, ssn);
1082 
1083 	/*Since we free until index _not_ inclusive, the one before index is
1084 	 * the last we will free. This one must be used */
1085 	last_to_free = iwl_queue_dec_wrap(tfd_num);
1086 
1087 	if (!iwl_queue_used(txq, last_to_free)) {
1088 		IWL_ERR(trans,
1089 			"%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1090 			__func__, txq_id, last_to_free, TFD_QUEUE_SIZE_MAX,
1091 			txq->write_ptr, txq->read_ptr);
1092 		goto out;
1093 	}
1094 
1095 	if (WARN_ON(!skb_queue_empty(skbs)))
1096 		goto out;
1097 
1098 	for (;
1099 	     txq->read_ptr != tfd_num;
1100 	     txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr)) {
1101 		struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
1102 
1103 		if (WARN_ON_ONCE(!skb))
1104 			continue;
1105 
1106 		iwl_pcie_free_tso_page(trans_pcie, skb);
1107 
1108 		__skb_queue_tail(skbs, skb);
1109 
1110 		txq->entries[txq->read_ptr].skb = NULL;
1111 
1112 		if (!trans->cfg->use_tfh)
1113 			iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1114 
1115 		iwl_pcie_txq_free_tfd(trans, txq);
1116 	}
1117 
1118 	iwl_pcie_txq_progress(txq);
1119 
1120 	if (iwl_queue_space(txq) > txq->low_mark &&
1121 	    test_bit(txq_id, trans_pcie->queue_stopped)) {
1122 		struct sk_buff_head overflow_skbs;
1123 
1124 		__skb_queue_head_init(&overflow_skbs);
1125 		skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1126 
1127 		/*
1128 		 * This is tricky: we are in reclaim path which is non
1129 		 * re-entrant, so noone will try to take the access the
1130 		 * txq data from that path. We stopped tx, so we can't
1131 		 * have tx as well. Bottom line, we can unlock and re-lock
1132 		 * later.
1133 		 */
1134 		spin_unlock_bh(&txq->lock);
1135 
1136 		while (!skb_queue_empty(&overflow_skbs)) {
1137 			struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1138 			struct iwl_device_cmd *dev_cmd_ptr;
1139 
1140 			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1141 						 trans_pcie->dev_cmd_offs);
1142 
1143 			/*
1144 			 * Note that we can very well be overflowing again.
1145 			 * In that case, iwl_queue_space will be small again
1146 			 * and we won't wake mac80211's queue.
1147 			 */
1148 			iwl_trans_pcie_tx(trans, skb, dev_cmd_ptr, txq_id);
1149 		}
1150 		spin_lock_bh(&txq->lock);
1151 
1152 		if (iwl_queue_space(txq) > txq->low_mark)
1153 			iwl_wake_queue(trans, txq);
1154 	}
1155 
1156 	if (txq->read_ptr == txq->write_ptr) {
1157 		IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", txq->id);
1158 		iwl_trans_unref(trans);
1159 	}
1160 
1161 out:
1162 	spin_unlock_bh(&txq->lock);
1163 }
1164 
1165 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1166 				      const struct iwl_host_cmd *cmd)
1167 {
1168 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1169 	int ret;
1170 
1171 	lockdep_assert_held(&trans_pcie->reg_lock);
1172 
1173 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
1174 	    !trans_pcie->ref_cmd_in_flight) {
1175 		trans_pcie->ref_cmd_in_flight = true;
1176 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
1177 		iwl_trans_ref(trans);
1178 	}
1179 
1180 	/*
1181 	 * wake up the NIC to make sure that the firmware will see the host
1182 	 * command - we will let the NIC sleep once all the host commands
1183 	 * returned. This needs to be done only on NICs that have
1184 	 * apmg_wake_up_wa set.
1185 	 */
1186 	if (trans->cfg->base_params->apmg_wake_up_wa &&
1187 	    !trans_pcie->cmd_hold_nic_awake) {
1188 		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1189 					 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1190 
1191 		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1192 				   CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1193 				   (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1194 				    CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1195 				   15000);
1196 		if (ret < 0) {
1197 			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1198 					CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1199 			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1200 			return -EIO;
1201 		}
1202 		trans_pcie->cmd_hold_nic_awake = true;
1203 	}
1204 
1205 	return 0;
1206 }
1207 
1208 /*
1209  * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1210  *
1211  * When FW advances 'R' index, all entries between old and new 'R' index
1212  * need to be reclaimed. As result, some free space forms.  If there is
1213  * enough free space (> low mark), wake the stack that feeds us.
1214  */
1215 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1216 {
1217 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1218 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1219 	unsigned long flags;
1220 	int nfreed = 0;
1221 
1222 	lockdep_assert_held(&txq->lock);
1223 
1224 	if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(txq, idx))) {
1225 		IWL_ERR(trans,
1226 			"%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1227 			__func__, txq_id, idx, TFD_QUEUE_SIZE_MAX,
1228 			txq->write_ptr, txq->read_ptr);
1229 		return;
1230 	}
1231 
1232 	for (idx = iwl_queue_inc_wrap(idx); txq->read_ptr != idx;
1233 	     txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr)) {
1234 
1235 		if (nfreed++ > 0) {
1236 			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1237 				idx, txq->write_ptr, txq->read_ptr);
1238 			iwl_force_nmi(trans);
1239 		}
1240 	}
1241 
1242 	if (txq->read_ptr == txq->write_ptr) {
1243 		spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1244 		iwl_pcie_clear_cmd_in_flight(trans);
1245 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1246 	}
1247 
1248 	iwl_pcie_txq_progress(txq);
1249 }
1250 
1251 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1252 				 u16 txq_id)
1253 {
1254 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1255 	u32 tbl_dw_addr;
1256 	u32 tbl_dw;
1257 	u16 scd_q2ratid;
1258 
1259 	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1260 
1261 	tbl_dw_addr = trans_pcie->scd_base_addr +
1262 			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1263 
1264 	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1265 
1266 	if (txq_id & 0x1)
1267 		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1268 	else
1269 		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1270 
1271 	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1272 
1273 	return 0;
1274 }
1275 
1276 /* Receiver address (actually, Rx station's index into station table),
1277  * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1278 #define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
1279 
1280 bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1281 			       const struct iwl_trans_txq_scd_cfg *cfg,
1282 			       unsigned int wdg_timeout)
1283 {
1284 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1285 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1286 	int fifo = -1;
1287 	bool scd_bug = false;
1288 
1289 	if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1290 		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1291 
1292 	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1293 
1294 	if (cfg) {
1295 		fifo = cfg->fifo;
1296 
1297 		/* Disable the scheduler prior configuring the cmd queue */
1298 		if (txq_id == trans_pcie->cmd_queue &&
1299 		    trans_pcie->scd_set_active)
1300 			iwl_scd_enable_set_active(trans, 0);
1301 
1302 		/* Stop this Tx queue before configuring it */
1303 		iwl_scd_txq_set_inactive(trans, txq_id);
1304 
1305 		/* Set this queue as a chain-building queue unless it is CMD */
1306 		if (txq_id != trans_pcie->cmd_queue)
1307 			iwl_scd_txq_set_chain(trans, txq_id);
1308 
1309 		if (cfg->aggregate) {
1310 			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1311 
1312 			/* Map receiver-address / traffic-ID to this queue */
1313 			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1314 
1315 			/* enable aggregations for the queue */
1316 			iwl_scd_txq_enable_agg(trans, txq_id);
1317 			txq->ampdu = true;
1318 		} else {
1319 			/*
1320 			 * disable aggregations for the queue, this will also
1321 			 * make the ra_tid mapping configuration irrelevant
1322 			 * since it is now a non-AGG queue.
1323 			 */
1324 			iwl_scd_txq_disable_agg(trans, txq_id);
1325 
1326 			ssn = txq->read_ptr;
1327 		}
1328 	} else {
1329 		/*
1330 		 * If we need to move the SCD write pointer by steps of
1331 		 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let
1332 		 * the op_mode know by returning true later.
1333 		 * Do this only in case cfg is NULL since this trick can
1334 		 * be done only if we have DQA enabled which is true for mvm
1335 		 * only. And mvm never sets a cfg pointer.
1336 		 * This is really ugly, but this is the easiest way out for
1337 		 * this sad hardware issue.
1338 		 * This bug has been fixed on devices 9000 and up.
1339 		 */
1340 		scd_bug = !trans->cfg->mq_rx_supported &&
1341 			!((ssn - txq->write_ptr) & 0x3f) &&
1342 			(ssn != txq->write_ptr);
1343 		if (scd_bug)
1344 			ssn++;
1345 	}
1346 
1347 	/* Place first TFD at index corresponding to start sequence number.
1348 	 * Assumes that ssn_idx is valid (!= 0xFFF) */
1349 	txq->read_ptr = (ssn & 0xff);
1350 	txq->write_ptr = (ssn & 0xff);
1351 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1352 			   (ssn & 0xff) | (txq_id << 8));
1353 
1354 	if (cfg) {
1355 		u8 frame_limit = cfg->frame_limit;
1356 
1357 		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1358 
1359 		/* Set up Tx window size and frame limit for this queue */
1360 		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1361 				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1362 		iwl_trans_write_mem32(trans,
1363 			trans_pcie->scd_base_addr +
1364 			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1365 			SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) |
1366 			SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit));
1367 
1368 		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1369 		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1370 			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1371 			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1372 			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1373 			       SCD_QUEUE_STTS_REG_MSK);
1374 
1375 		/* enable the scheduler for this queue (only) */
1376 		if (txq_id == trans_pcie->cmd_queue &&
1377 		    trans_pcie->scd_set_active)
1378 			iwl_scd_enable_set_active(trans, BIT(txq_id));
1379 
1380 		IWL_DEBUG_TX_QUEUES(trans,
1381 				    "Activate queue %d on FIFO %d WrPtr: %d\n",
1382 				    txq_id, fifo, ssn & 0xff);
1383 	} else {
1384 		IWL_DEBUG_TX_QUEUES(trans,
1385 				    "Activate queue %d WrPtr: %d\n",
1386 				    txq_id, ssn & 0xff);
1387 	}
1388 
1389 	return scd_bug;
1390 }
1391 
1392 void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
1393 					bool shared_mode)
1394 {
1395 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1396 	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1397 
1398 	txq->ampdu = !shared_mode;
1399 }
1400 
1401 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1402 				bool configure_scd)
1403 {
1404 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1405 	u32 stts_addr = trans_pcie->scd_base_addr +
1406 			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1407 	static const u32 zero_val[4] = {};
1408 
1409 	trans_pcie->txq[txq_id]->frozen_expiry_remainder = 0;
1410 	trans_pcie->txq[txq_id]->frozen = false;
1411 
1412 	/*
1413 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1414 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1415 	 * allow the op_mode to call txq_disable after it already called
1416 	 * stop_device.
1417 	 */
1418 	if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1419 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1420 			  "queue %d not used", txq_id);
1421 		return;
1422 	}
1423 
1424 	if (configure_scd) {
1425 		iwl_scd_txq_set_inactive(trans, txq_id);
1426 
1427 		iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1428 				    ARRAY_SIZE(zero_val));
1429 	}
1430 
1431 	iwl_pcie_txq_unmap(trans, txq_id);
1432 	trans_pcie->txq[txq_id]->ampdu = false;
1433 
1434 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1435 }
1436 
1437 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1438 
1439 /*
1440  * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1441  * @priv: device private data point
1442  * @cmd: a pointer to the ucode command structure
1443  *
1444  * The function returns < 0 values to indicate the operation
1445  * failed. On success, it returns the index (>= 0) of command in the
1446  * command queue.
1447  */
1448 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1449 				 struct iwl_host_cmd *cmd)
1450 {
1451 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1452 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1453 	struct iwl_device_cmd *out_cmd;
1454 	struct iwl_cmd_meta *out_meta;
1455 	unsigned long flags;
1456 	void *dup_buf = NULL;
1457 	dma_addr_t phys_addr;
1458 	int idx;
1459 	u16 copy_size, cmd_size, tb0_size;
1460 	bool had_nocopy = false;
1461 	u8 group_id = iwl_cmd_groupid(cmd->id);
1462 	int i, ret;
1463 	u32 cmd_pos;
1464 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1465 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1466 
1467 	if (WARN(!trans->wide_cmd_header &&
1468 		 group_id > IWL_ALWAYS_LONG_GROUP,
1469 		 "unsupported wide command %#x\n", cmd->id))
1470 		return -EINVAL;
1471 
1472 	if (group_id != 0) {
1473 		copy_size = sizeof(struct iwl_cmd_header_wide);
1474 		cmd_size = sizeof(struct iwl_cmd_header_wide);
1475 	} else {
1476 		copy_size = sizeof(struct iwl_cmd_header);
1477 		cmd_size = sizeof(struct iwl_cmd_header);
1478 	}
1479 
1480 	/* need one for the header if the first is NOCOPY */
1481 	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1482 
1483 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1484 		cmddata[i] = cmd->data[i];
1485 		cmdlen[i] = cmd->len[i];
1486 
1487 		if (!cmd->len[i])
1488 			continue;
1489 
1490 		/* need at least IWL_FIRST_TB_SIZE copied */
1491 		if (copy_size < IWL_FIRST_TB_SIZE) {
1492 			int copy = IWL_FIRST_TB_SIZE - copy_size;
1493 
1494 			if (copy > cmdlen[i])
1495 				copy = cmdlen[i];
1496 			cmdlen[i] -= copy;
1497 			cmddata[i] += copy;
1498 			copy_size += copy;
1499 		}
1500 
1501 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1502 			had_nocopy = true;
1503 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1504 				idx = -EINVAL;
1505 				goto free_dup_buf;
1506 			}
1507 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1508 			/*
1509 			 * This is also a chunk that isn't copied
1510 			 * to the static buffer so set had_nocopy.
1511 			 */
1512 			had_nocopy = true;
1513 
1514 			/* only allowed once */
1515 			if (WARN_ON(dup_buf)) {
1516 				idx = -EINVAL;
1517 				goto free_dup_buf;
1518 			}
1519 
1520 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
1521 					  GFP_ATOMIC);
1522 			if (!dup_buf)
1523 				return -ENOMEM;
1524 		} else {
1525 			/* NOCOPY must not be followed by normal! */
1526 			if (WARN_ON(had_nocopy)) {
1527 				idx = -EINVAL;
1528 				goto free_dup_buf;
1529 			}
1530 			copy_size += cmdlen[i];
1531 		}
1532 		cmd_size += cmd->len[i];
1533 	}
1534 
1535 	/*
1536 	 * If any of the command structures end up being larger than
1537 	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1538 	 * allocated into separate TFDs, then we will need to
1539 	 * increase the size of the buffers.
1540 	 */
1541 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1542 		 "Command %s (%#x) is too large (%d bytes)\n",
1543 		 iwl_get_cmd_string(trans, cmd->id),
1544 		 cmd->id, copy_size)) {
1545 		idx = -EINVAL;
1546 		goto free_dup_buf;
1547 	}
1548 
1549 	spin_lock_bh(&txq->lock);
1550 
1551 	if (iwl_queue_space(txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1552 		spin_unlock_bh(&txq->lock);
1553 
1554 		IWL_ERR(trans, "No space in command queue\n");
1555 		iwl_op_mode_cmd_queue_full(trans->op_mode);
1556 		idx = -ENOSPC;
1557 		goto free_dup_buf;
1558 	}
1559 
1560 	idx = get_cmd_index(txq, txq->write_ptr);
1561 	out_cmd = txq->entries[idx].cmd;
1562 	out_meta = &txq->entries[idx].meta;
1563 
1564 	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1565 	if (cmd->flags & CMD_WANT_SKB)
1566 		out_meta->source = cmd;
1567 
1568 	/* set up the header */
1569 	if (group_id != 0) {
1570 		out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1571 		out_cmd->hdr_wide.group_id = group_id;
1572 		out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1573 		out_cmd->hdr_wide.length =
1574 			cpu_to_le16(cmd_size -
1575 				    sizeof(struct iwl_cmd_header_wide));
1576 		out_cmd->hdr_wide.reserved = 0;
1577 		out_cmd->hdr_wide.sequence =
1578 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1579 						 INDEX_TO_SEQ(txq->write_ptr));
1580 
1581 		cmd_pos = sizeof(struct iwl_cmd_header_wide);
1582 		copy_size = sizeof(struct iwl_cmd_header_wide);
1583 	} else {
1584 		out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1585 		out_cmd->hdr.sequence =
1586 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1587 						 INDEX_TO_SEQ(txq->write_ptr));
1588 		out_cmd->hdr.group_id = 0;
1589 
1590 		cmd_pos = sizeof(struct iwl_cmd_header);
1591 		copy_size = sizeof(struct iwl_cmd_header);
1592 	}
1593 
1594 	/* and copy the data that needs to be copied */
1595 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1596 		int copy;
1597 
1598 		if (!cmd->len[i])
1599 			continue;
1600 
1601 		/* copy everything if not nocopy/dup */
1602 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1603 					   IWL_HCMD_DFL_DUP))) {
1604 			copy = cmd->len[i];
1605 
1606 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1607 			cmd_pos += copy;
1608 			copy_size += copy;
1609 			continue;
1610 		}
1611 
1612 		/*
1613 		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1614 		 * in total (for bi-directional DMA), but copy up to what
1615 		 * we can fit into the payload for debug dump purposes.
1616 		 */
1617 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1618 
1619 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1620 		cmd_pos += copy;
1621 
1622 		/* However, treat copy_size the proper way, we need it below */
1623 		if (copy_size < IWL_FIRST_TB_SIZE) {
1624 			copy = IWL_FIRST_TB_SIZE - copy_size;
1625 
1626 			if (copy > cmd->len[i])
1627 				copy = cmd->len[i];
1628 			copy_size += copy;
1629 		}
1630 	}
1631 
1632 	IWL_DEBUG_HC(trans,
1633 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1634 		     iwl_get_cmd_string(trans, cmd->id),
1635 		     group_id, out_cmd->hdr.cmd,
1636 		     le16_to_cpu(out_cmd->hdr.sequence),
1637 		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
1638 
1639 	/* start the TFD with the minimum copy bytes */
1640 	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1641 	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1642 	iwl_pcie_txq_build_tfd(trans, txq,
1643 			       iwl_pcie_get_first_tb_dma(txq, idx),
1644 			       tb0_size, true);
1645 
1646 	/* map first command fragment, if any remains */
1647 	if (copy_size > tb0_size) {
1648 		phys_addr = dma_map_single(trans->dev,
1649 					   ((u8 *)&out_cmd->hdr) + tb0_size,
1650 					   copy_size - tb0_size,
1651 					   DMA_TO_DEVICE);
1652 		if (dma_mapping_error(trans->dev, phys_addr)) {
1653 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1654 					   txq->write_ptr);
1655 			idx = -ENOMEM;
1656 			goto out;
1657 		}
1658 
1659 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1660 				       copy_size - tb0_size, false);
1661 	}
1662 
1663 	/* map the remaining (adjusted) nocopy/dup fragments */
1664 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1665 		const void *data = cmddata[i];
1666 
1667 		if (!cmdlen[i])
1668 			continue;
1669 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1670 					   IWL_HCMD_DFL_DUP)))
1671 			continue;
1672 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1673 			data = dup_buf;
1674 		phys_addr = dma_map_single(trans->dev, (void *)data,
1675 					   cmdlen[i], DMA_TO_DEVICE);
1676 		if (dma_mapping_error(trans->dev, phys_addr)) {
1677 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1678 					   txq->write_ptr);
1679 			idx = -ENOMEM;
1680 			goto out;
1681 		}
1682 
1683 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1684 	}
1685 
1686 	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
1687 	out_meta->flags = cmd->flags;
1688 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1689 		kzfree(txq->entries[idx].free_buf);
1690 	txq->entries[idx].free_buf = dup_buf;
1691 
1692 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1693 
1694 	/* start timer if queue currently empty */
1695 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
1696 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1697 
1698 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1699 	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1700 	if (ret < 0) {
1701 		idx = ret;
1702 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1703 		goto out;
1704 	}
1705 
1706 	/* Increment and update queue's write index */
1707 	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
1708 	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1709 
1710 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1711 
1712  out:
1713 	spin_unlock_bh(&txq->lock);
1714  free_dup_buf:
1715 	if (idx < 0)
1716 		kfree(dup_buf);
1717 	return idx;
1718 }
1719 
1720 /*
1721  * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1722  * @rxb: Rx buffer to reclaim
1723  */
1724 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1725 			    struct iwl_rx_cmd_buffer *rxb)
1726 {
1727 	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1728 	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1729 	u8 group_id;
1730 	u32 cmd_id;
1731 	int txq_id = SEQ_TO_QUEUE(sequence);
1732 	int index = SEQ_TO_INDEX(sequence);
1733 	int cmd_index;
1734 	struct iwl_device_cmd *cmd;
1735 	struct iwl_cmd_meta *meta;
1736 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1737 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1738 
1739 	/* If a Tx command is being handled and it isn't in the actual
1740 	 * command queue then there a command routing bug has been introduced
1741 	 * in the queue management code. */
1742 	if (WARN(txq_id != trans_pcie->cmd_queue,
1743 		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1744 		 txq_id, trans_pcie->cmd_queue, sequence, txq->read_ptr,
1745 		 txq->write_ptr)) {
1746 		iwl_print_hex_error(trans, pkt, 32);
1747 		return;
1748 	}
1749 
1750 	spin_lock_bh(&txq->lock);
1751 
1752 	cmd_index = get_cmd_index(txq, index);
1753 	cmd = txq->entries[cmd_index].cmd;
1754 	meta = &txq->entries[cmd_index].meta;
1755 	group_id = cmd->hdr.group_id;
1756 	cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1757 
1758 	iwl_pcie_tfd_unmap(trans, meta, txq, index);
1759 
1760 	/* Input error checking is done when commands are added to queue. */
1761 	if (meta->flags & CMD_WANT_SKB) {
1762 		struct page *p = rxb_steal_page(rxb);
1763 
1764 		meta->source->resp_pkt = pkt;
1765 		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1766 		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1767 	}
1768 
1769 	if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1770 		iwl_op_mode_async_cb(trans->op_mode, cmd);
1771 
1772 	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1773 
1774 	if (!(meta->flags & CMD_ASYNC)) {
1775 		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1776 			IWL_WARN(trans,
1777 				 "HCMD_ACTIVE already clear for command %s\n",
1778 				 iwl_get_cmd_string(trans, cmd_id));
1779 		}
1780 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1781 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1782 			       iwl_get_cmd_string(trans, cmd_id));
1783 		wake_up(&trans_pcie->wait_command_queue);
1784 	}
1785 
1786 	if (meta->flags & CMD_MAKE_TRANS_IDLE) {
1787 		IWL_DEBUG_INFO(trans, "complete %s - mark trans as idle\n",
1788 			       iwl_get_cmd_string(trans, cmd->hdr.cmd));
1789 		set_bit(STATUS_TRANS_IDLE, &trans->status);
1790 		wake_up(&trans_pcie->d0i3_waitq);
1791 	}
1792 
1793 	if (meta->flags & CMD_WAKE_UP_TRANS) {
1794 		IWL_DEBUG_INFO(trans, "complete %s - clear trans idle flag\n",
1795 			       iwl_get_cmd_string(trans, cmd->hdr.cmd));
1796 		clear_bit(STATUS_TRANS_IDLE, &trans->status);
1797 		wake_up(&trans_pcie->d0i3_waitq);
1798 	}
1799 
1800 	meta->flags = 0;
1801 
1802 	spin_unlock_bh(&txq->lock);
1803 }
1804 
1805 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1806 
1807 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1808 				    struct iwl_host_cmd *cmd)
1809 {
1810 	int ret;
1811 
1812 	/* An asynchronous command can not expect an SKB to be set. */
1813 	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1814 		return -EINVAL;
1815 
1816 	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1817 	if (ret < 0) {
1818 		IWL_ERR(trans,
1819 			"Error sending %s: enqueue_hcmd failed: %d\n",
1820 			iwl_get_cmd_string(trans, cmd->id), ret);
1821 		return ret;
1822 	}
1823 	return 0;
1824 }
1825 
1826 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1827 				   struct iwl_host_cmd *cmd)
1828 {
1829 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1830 	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1831 	int cmd_idx;
1832 	int ret;
1833 
1834 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1835 		       iwl_get_cmd_string(trans, cmd->id));
1836 
1837 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1838 				  &trans->status),
1839 		 "Command %s: a command is already active!\n",
1840 		 iwl_get_cmd_string(trans, cmd->id)))
1841 		return -EIO;
1842 
1843 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1844 		       iwl_get_cmd_string(trans, cmd->id));
1845 
1846 	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
1847 		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
1848 				 pm_runtime_active(&trans_pcie->pci_dev->dev),
1849 				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
1850 		if (!ret) {
1851 			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
1852 			return -ETIMEDOUT;
1853 		}
1854 	}
1855 
1856 	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1857 	if (cmd_idx < 0) {
1858 		ret = cmd_idx;
1859 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1860 		IWL_ERR(trans,
1861 			"Error sending %s: enqueue_hcmd failed: %d\n",
1862 			iwl_get_cmd_string(trans, cmd->id), ret);
1863 		return ret;
1864 	}
1865 
1866 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
1867 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1868 					   &trans->status),
1869 				 HOST_COMPLETE_TIMEOUT);
1870 	if (!ret) {
1871 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1872 			iwl_get_cmd_string(trans, cmd->id),
1873 			jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1874 
1875 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1876 			txq->read_ptr, txq->write_ptr);
1877 
1878 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1879 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1880 			       iwl_get_cmd_string(trans, cmd->id));
1881 		ret = -ETIMEDOUT;
1882 
1883 		iwl_force_nmi(trans);
1884 		iwl_trans_fw_error(trans);
1885 
1886 		goto cancel;
1887 	}
1888 
1889 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1890 		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1891 			iwl_get_cmd_string(trans, cmd->id));
1892 		dump_stack();
1893 		ret = -EIO;
1894 		goto cancel;
1895 	}
1896 
1897 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1898 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1899 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1900 		ret = -ERFKILL;
1901 		goto cancel;
1902 	}
1903 
1904 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1905 		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1906 			iwl_get_cmd_string(trans, cmd->id));
1907 		ret = -EIO;
1908 		goto cancel;
1909 	}
1910 
1911 	return 0;
1912 
1913 cancel:
1914 	if (cmd->flags & CMD_WANT_SKB) {
1915 		/*
1916 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1917 		 * TX cmd queue. Otherwise in case the cmd comes
1918 		 * in later, it will possibly set an invalid
1919 		 * address (cmd->meta.source).
1920 		 */
1921 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1922 	}
1923 
1924 	if (cmd->resp_pkt) {
1925 		iwl_free_resp(cmd);
1926 		cmd->resp_pkt = NULL;
1927 	}
1928 
1929 	return ret;
1930 }
1931 
1932 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1933 {
1934 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1935 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1936 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1937 				  cmd->id);
1938 		return -ERFKILL;
1939 	}
1940 
1941 	if (cmd->flags & CMD_ASYNC)
1942 		return iwl_pcie_send_hcmd_async(trans, cmd);
1943 
1944 	/* We still can fail on RFKILL that can be asserted while we wait */
1945 	return iwl_pcie_send_hcmd_sync(trans, cmd);
1946 }
1947 
1948 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
1949 			     struct iwl_txq *txq, u8 hdr_len,
1950 			     struct iwl_cmd_meta *out_meta,
1951 			     struct iwl_device_cmd *dev_cmd, u16 tb1_len)
1952 {
1953 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1954 	u16 tb2_len;
1955 	int i;
1956 
1957 	/*
1958 	 * Set up TFD's third entry to point directly to remainder
1959 	 * of skb's head, if any
1960 	 */
1961 	tb2_len = skb_headlen(skb) - hdr_len;
1962 
1963 	if (tb2_len > 0) {
1964 		dma_addr_t tb2_phys = dma_map_single(trans->dev,
1965 						     skb->data + hdr_len,
1966 						     tb2_len, DMA_TO_DEVICE);
1967 		if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) {
1968 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1969 					   txq->write_ptr);
1970 			return -EINVAL;
1971 		}
1972 		iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false);
1973 	}
1974 
1975 	/* set up the remaining entries to point to the data */
1976 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1977 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1978 		dma_addr_t tb_phys;
1979 		int tb_idx;
1980 
1981 		if (!skb_frag_size(frag))
1982 			continue;
1983 
1984 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
1985 					   skb_frag_size(frag), DMA_TO_DEVICE);
1986 
1987 		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
1988 			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1989 					   txq->write_ptr);
1990 			return -EINVAL;
1991 		}
1992 		tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
1993 						skb_frag_size(frag), false);
1994 
1995 		out_meta->tbs |= BIT(tb_idx);
1996 	}
1997 
1998 	trace_iwlwifi_dev_tx(trans->dev, skb,
1999 			     iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr),
2000 			     trans_pcie->tfd_size,
2001 			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
2002 			     hdr_len);
2003 	trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
2004 	return 0;
2005 }
2006 
2007 #ifdef CONFIG_INET
2008 struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len)
2009 {
2010 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2011 	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
2012 
2013 	if (!p->page)
2014 		goto alloc;
2015 
2016 	/* enough room on this page */
2017 	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
2018 		return p;
2019 
2020 	/* We don't have enough room on this page, get a new one. */
2021 	__free_page(p->page);
2022 
2023 alloc:
2024 	p->page = alloc_page(GFP_ATOMIC);
2025 	if (!p->page)
2026 		return NULL;
2027 	p->pos = page_address(p->page);
2028 	return p;
2029 }
2030 
2031 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
2032 					bool ipv6, unsigned int len)
2033 {
2034 	if (ipv6) {
2035 		struct ipv6hdr *iphv6 = iph;
2036 
2037 		tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
2038 					       len + tcph->doff * 4,
2039 					       IPPROTO_TCP, 0);
2040 	} else {
2041 		struct iphdr *iphv4 = iph;
2042 
2043 		ip_send_check(iphv4);
2044 		tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
2045 						 len + tcph->doff * 4,
2046 						 IPPROTO_TCP, 0);
2047 	}
2048 }
2049 
2050 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2051 				   struct iwl_txq *txq, u8 hdr_len,
2052 				   struct iwl_cmd_meta *out_meta,
2053 				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2054 {
2055 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
2056 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
2057 	struct ieee80211_hdr *hdr = (void *)skb->data;
2058 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
2059 	unsigned int mss = skb_shinfo(skb)->gso_size;
2060 	u16 length, iv_len, amsdu_pad;
2061 	u8 *start_hdr;
2062 	struct iwl_tso_hdr_page *hdr_page;
2063 	struct page **page_ptr;
2064 	int ret;
2065 	struct tso_t tso;
2066 
2067 	/* if the packet is protected, then it must be CCMP or GCMP */
2068 	BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
2069 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
2070 		IEEE80211_CCMP_HDR_LEN : 0;
2071 
2072 	trace_iwlwifi_dev_tx(trans->dev, skb,
2073 			     iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr),
2074 			     trans_pcie->tfd_size,
2075 			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0);
2076 
2077 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
2078 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
2079 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
2080 	amsdu_pad = 0;
2081 
2082 	/* total amount of header we may need for this A-MSDU */
2083 	hdr_room = DIV_ROUND_UP(total_len, mss) *
2084 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
2085 
2086 	/* Our device supports 9 segments at most, it will fit in 1 page */
2087 	hdr_page = get_page_hdr(trans, hdr_room);
2088 	if (!hdr_page)
2089 		return -ENOMEM;
2090 
2091 	get_page(hdr_page->page);
2092 	start_hdr = hdr_page->pos;
2093 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
2094 	*page_ptr = hdr_page->page;
2095 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
2096 	hdr_page->pos += iv_len;
2097 
2098 	/*
2099 	 * Pull the ieee80211 header + IV to be able to use TSO core,
2100 	 * we will restore it for the tx_status flow.
2101 	 */
2102 	skb_pull(skb, hdr_len + iv_len);
2103 
2104 	/*
2105 	 * Remove the length of all the headers that we don't actually
2106 	 * have in the MPDU by themselves, but that we duplicate into
2107 	 * all the different MSDUs inside the A-MSDU.
2108 	 */
2109 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
2110 
2111 	tso_start(skb, &tso);
2112 
2113 	while (total_len) {
2114 		/* this is the data left for this subframe */
2115 		unsigned int data_left =
2116 			min_t(unsigned int, mss, total_len);
2117 		struct sk_buff *csum_skb = NULL;
2118 		unsigned int hdr_tb_len;
2119 		dma_addr_t hdr_tb_phys;
2120 		struct tcphdr *tcph;
2121 		u8 *iph, *subf_hdrs_start = hdr_page->pos;
2122 
2123 		total_len -= data_left;
2124 
2125 		memset(hdr_page->pos, 0, amsdu_pad);
2126 		hdr_page->pos += amsdu_pad;
2127 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2128 				  data_left)) & 0x3;
2129 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2130 		hdr_page->pos += ETH_ALEN;
2131 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2132 		hdr_page->pos += ETH_ALEN;
2133 
2134 		length = snap_ip_tcp_hdrlen + data_left;
2135 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2136 		hdr_page->pos += sizeof(length);
2137 
2138 		/*
2139 		 * This will copy the SNAP as well which will be considered
2140 		 * as MAC header.
2141 		 */
2142 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2143 		iph = hdr_page->pos + 8;
2144 		tcph = (void *)(iph + ip_hdrlen);
2145 
2146 		/* For testing on current hardware only */
2147 		if (trans_pcie->sw_csum_tx) {
2148 			csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2149 					     GFP_ATOMIC);
2150 			if (!csum_skb) {
2151 				ret = -ENOMEM;
2152 				goto out_unmap;
2153 			}
2154 
2155 			iwl_compute_pseudo_hdr_csum(iph, tcph,
2156 						    skb->protocol ==
2157 							htons(ETH_P_IPV6),
2158 						    data_left);
2159 
2160 			skb_put_data(csum_skb, tcph, tcp_hdrlen(skb));
2161 			skb_reset_transport_header(csum_skb);
2162 			csum_skb->csum_start =
2163 				(unsigned char *)tcp_hdr(csum_skb) -
2164 						 csum_skb->head;
2165 		}
2166 
2167 		hdr_page->pos += snap_ip_tcp_hdrlen;
2168 
2169 		hdr_tb_len = hdr_page->pos - start_hdr;
2170 		hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2171 					     hdr_tb_len, DMA_TO_DEVICE);
2172 		if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2173 			dev_kfree_skb(csum_skb);
2174 			ret = -EINVAL;
2175 			goto out_unmap;
2176 		}
2177 		iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2178 				       hdr_tb_len, false);
2179 		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr,
2180 					       hdr_tb_len);
2181 		/* add this subframe's headers' length to the tx_cmd */
2182 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
2183 
2184 		/* prepare the start_hdr for the next subframe */
2185 		start_hdr = hdr_page->pos;
2186 
2187 		/* put the payload */
2188 		while (data_left) {
2189 			unsigned int size = min_t(unsigned int, tso.size,
2190 						  data_left);
2191 			dma_addr_t tb_phys;
2192 
2193 			if (trans_pcie->sw_csum_tx)
2194 				skb_put_data(csum_skb, tso.data, size);
2195 
2196 			tb_phys = dma_map_single(trans->dev, tso.data,
2197 						 size, DMA_TO_DEVICE);
2198 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2199 				dev_kfree_skb(csum_skb);
2200 				ret = -EINVAL;
2201 				goto out_unmap;
2202 			}
2203 
2204 			iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2205 					       size, false);
2206 			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
2207 						       size);
2208 
2209 			data_left -= size;
2210 			tso_build_data(skb, &tso, size);
2211 		}
2212 
2213 		/* For testing on early hardware only */
2214 		if (trans_pcie->sw_csum_tx) {
2215 			__wsum csum;
2216 
2217 			csum = skb_checksum(csum_skb,
2218 					    skb_checksum_start_offset(csum_skb),
2219 					    csum_skb->len -
2220 					    skb_checksum_start_offset(csum_skb),
2221 					    0);
2222 			dev_kfree_skb(csum_skb);
2223 			dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2224 						hdr_tb_len, DMA_TO_DEVICE);
2225 			tcph->check = csum_fold(csum);
2226 			dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2227 						   hdr_tb_len, DMA_TO_DEVICE);
2228 		}
2229 	}
2230 
2231 	/* re -add the WiFi header and IV */
2232 	skb_push(skb, hdr_len + iv_len);
2233 
2234 	return 0;
2235 
2236 out_unmap:
2237 	iwl_pcie_tfd_unmap(trans, out_meta, txq, txq->write_ptr);
2238 	return ret;
2239 }
2240 #else /* CONFIG_INET */
2241 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2242 				   struct iwl_txq *txq, u8 hdr_len,
2243 				   struct iwl_cmd_meta *out_meta,
2244 				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2245 {
2246 	/* No A-MSDU without CONFIG_INET */
2247 	WARN_ON(1);
2248 
2249 	return -1;
2250 }
2251 #endif /* CONFIG_INET */
2252 
2253 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2254 		      struct iwl_device_cmd *dev_cmd, int txq_id)
2255 {
2256 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2257 	struct ieee80211_hdr *hdr;
2258 	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2259 	struct iwl_cmd_meta *out_meta;
2260 	struct iwl_txq *txq;
2261 	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2262 	void *tb1_addr;
2263 	void *tfd;
2264 	u16 len, tb1_len;
2265 	bool wait_write_ptr;
2266 	__le16 fc;
2267 	u8 hdr_len;
2268 	u16 wifi_seq;
2269 	bool amsdu;
2270 
2271 	txq = trans_pcie->txq[txq_id];
2272 
2273 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2274 		      "TX on unused queue %d\n", txq_id))
2275 		return -EINVAL;
2276 
2277 	if (unlikely(trans_pcie->sw_csum_tx &&
2278 		     skb->ip_summed == CHECKSUM_PARTIAL)) {
2279 		int offs = skb_checksum_start_offset(skb);
2280 		int csum_offs = offs + skb->csum_offset;
2281 		__wsum csum;
2282 
2283 		if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2284 			return -1;
2285 
2286 		csum = skb_checksum(skb, offs, skb->len - offs, 0);
2287 		*(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2288 
2289 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2290 	}
2291 
2292 	if (skb_is_nonlinear(skb) &&
2293 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
2294 	    __skb_linearize(skb))
2295 		return -ENOMEM;
2296 
2297 	/* mac80211 always puts the full header into the SKB's head,
2298 	 * so there's no need to check if it's readable there
2299 	 */
2300 	hdr = (struct ieee80211_hdr *)skb->data;
2301 	fc = hdr->frame_control;
2302 	hdr_len = ieee80211_hdrlen(fc);
2303 
2304 	spin_lock(&txq->lock);
2305 
2306 	if (iwl_queue_space(txq) < txq->high_mark) {
2307 		iwl_stop_queue(trans, txq);
2308 
2309 		/* don't put the packet on the ring, if there is no room */
2310 		if (unlikely(iwl_queue_space(txq) < 3)) {
2311 			struct iwl_device_cmd **dev_cmd_ptr;
2312 
2313 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
2314 					       trans_pcie->dev_cmd_offs);
2315 
2316 			*dev_cmd_ptr = dev_cmd;
2317 			__skb_queue_tail(&txq->overflow_q, skb);
2318 
2319 			spin_unlock(&txq->lock);
2320 			return 0;
2321 		}
2322 	}
2323 
2324 	/* In AGG mode, the index in the ring must correspond to the WiFi
2325 	 * sequence number. This is a HW requirements to help the SCD to parse
2326 	 * the BA.
2327 	 * Check here that the packets are in the right place on the ring.
2328 	 */
2329 	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2330 	WARN_ONCE(txq->ampdu &&
2331 		  (wifi_seq & 0xff) != txq->write_ptr,
2332 		  "Q: %d WiFi Seq %d tfdNum %d",
2333 		  txq_id, wifi_seq, txq->write_ptr);
2334 
2335 	/* Set up driver data for this TFD */
2336 	txq->entries[txq->write_ptr].skb = skb;
2337 	txq->entries[txq->write_ptr].cmd = dev_cmd;
2338 
2339 	dev_cmd->hdr.sequence =
2340 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2341 			    INDEX_TO_SEQ(txq->write_ptr)));
2342 
2343 	tb0_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
2344 	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2345 		       offsetof(struct iwl_tx_cmd, scratch);
2346 
2347 	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2348 	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2349 
2350 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
2351 	out_meta = &txq->entries[txq->write_ptr].meta;
2352 	out_meta->flags = 0;
2353 
2354 	/*
2355 	 * The second TB (tb1) points to the remainder of the TX command
2356 	 * and the 802.11 header - dword aligned size
2357 	 * (This calculation modifies the TX command, so do it before the
2358 	 * setup of the first TB)
2359 	 */
2360 	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2361 	      hdr_len - IWL_FIRST_TB_SIZE;
2362 	/* do not align A-MSDU to dword as the subframe header aligns it */
2363 	amsdu = ieee80211_is_data_qos(fc) &&
2364 		(*ieee80211_get_qos_ctl(hdr) &
2365 		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
2366 	if (trans_pcie->sw_csum_tx || !amsdu) {
2367 		tb1_len = ALIGN(len, 4);
2368 		/* Tell NIC about any 2-byte padding after MAC header */
2369 		if (tb1_len != len)
2370 			tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
2371 	} else {
2372 		tb1_len = len;
2373 	}
2374 
2375 	/*
2376 	 * The first TB points to bi-directional DMA data, we'll
2377 	 * memcpy the data into it later.
2378 	 */
2379 	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2380 			       IWL_FIRST_TB_SIZE, true);
2381 
2382 	/* there must be data left over for TB1 or this code must be changed */
2383 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
2384 
2385 	/* map the data for TB1 */
2386 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
2387 	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2388 	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2389 		goto out_err;
2390 	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2391 
2392 	if (amsdu) {
2393 		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2394 						     out_meta, dev_cmd,
2395 						     tb1_len)))
2396 			goto out_err;
2397 	} else if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2398 				       out_meta, dev_cmd, tb1_len))) {
2399 		goto out_err;
2400 	}
2401 
2402 	/* building the A-MSDU might have changed this data, so memcpy it now */
2403 	memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
2404 	       IWL_FIRST_TB_SIZE);
2405 
2406 	tfd = iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr);
2407 	/* Set up entry for this TFD in Tx byte-count array */
2408 	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len),
2409 					 iwl_pcie_tfd_get_num_tbs(trans, tfd));
2410 
2411 	wait_write_ptr = ieee80211_has_morefrags(fc);
2412 
2413 	/* start timer if queue currently empty */
2414 	if (txq->read_ptr == txq->write_ptr) {
2415 		if (txq->wd_timeout) {
2416 			/*
2417 			 * If the TXQ is active, then set the timer, if not,
2418 			 * set the timer in remainder so that the timer will
2419 			 * be armed with the right value when the station will
2420 			 * wake up.
2421 			 */
2422 			if (!txq->frozen)
2423 				mod_timer(&txq->stuck_timer,
2424 					  jiffies + txq->wd_timeout);
2425 			else
2426 				txq->frozen_expiry_remainder = txq->wd_timeout;
2427 		}
2428 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
2429 		iwl_trans_ref(trans);
2430 	}
2431 
2432 	/* Tell device the write index *just past* this latest filled TFD */
2433 	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
2434 	if (!wait_write_ptr)
2435 		iwl_pcie_txq_inc_wr_ptr(trans, txq);
2436 
2437 	/*
2438 	 * At this point the frame is "transmitted" successfully
2439 	 * and we will get a TX status notification eventually.
2440 	 */
2441 	spin_unlock(&txq->lock);
2442 	return 0;
2443 out_err:
2444 	spin_unlock(&txq->lock);
2445 	return -1;
2446 }
2447