1 /******************************************************************************
2  *
3  * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4  * Copyright(c) 2013 - 2014 Intel Mobile Communications GmbH
5  *
6  * Portions of this file are derived from the ipw3945 project, as well
7  * as portions of the ieee80211 subsystem header files.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of version 2 of the GNU General Public License as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along with
19  * this program; if not, write to the Free Software Foundation, Inc.,
20  * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
21  *
22  * The full GNU General Public License is included in this distribution in the
23  * file called LICENSE.
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  *****************************************************************************/
30 #include <linux/etherdevice.h>
31 #include <linux/ieee80211.h>
32 #include <linux/slab.h>
33 #include <linux/sched.h>
34 #include <net/ip6_checksum.h>
35 #include <net/tso.h>
36 #include <net/ip6_checksum.h>
37 
38 #include "iwl-debug.h"
39 #include "iwl-csr.h"
40 #include "iwl-prph.h"
41 #include "iwl-io.h"
42 #include "iwl-scd.h"
43 #include "iwl-op-mode.h"
44 #include "internal.h"
45 /* FIXME: need to abstract out TX command (once we know what it looks like) */
46 #include "dvm/commands.h"
47 
48 #define IWL_TX_CRC_SIZE 4
49 #define IWL_TX_DELIMITER_SIZE 4
50 
51 /*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
52  * DMA services
53  *
54  * Theory of operation
55  *
56  * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
57  * of buffer descriptors, each of which points to one or more data buffers for
58  * the device to read from or fill.  Driver and device exchange status of each
59  * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
60  * entries in each circular buffer, to protect against confusing empty and full
61  * queue states.
62  *
63  * The device reads or writes the data in the queues via the device's several
64  * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
65  *
66  * For Tx queue, there are low mark and high mark limits. If, after queuing
67  * the packet for Tx, free space become < low mark, Tx queue stopped. When
68  * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
69  * Tx queue resumed.
70  *
71  ***************************************************/
72 static int iwl_queue_space(const struct iwl_queue *q)
73 {
74 	unsigned int max;
75 	unsigned int used;
76 
77 	/*
78 	 * To avoid ambiguity between empty and completely full queues, there
79 	 * should always be less than TFD_QUEUE_SIZE_MAX elements in the queue.
80 	 * If q->n_window is smaller than TFD_QUEUE_SIZE_MAX, there is no need
81 	 * to reserve any queue entries for this purpose.
82 	 */
83 	if (q->n_window < TFD_QUEUE_SIZE_MAX)
84 		max = q->n_window;
85 	else
86 		max = TFD_QUEUE_SIZE_MAX - 1;
87 
88 	/*
89 	 * TFD_QUEUE_SIZE_MAX is a power of 2, so the following is equivalent to
90 	 * modulo by TFD_QUEUE_SIZE_MAX and is well defined.
91 	 */
92 	used = (q->write_ptr - q->read_ptr) & (TFD_QUEUE_SIZE_MAX - 1);
93 
94 	if (WARN_ON(used > max))
95 		return 0;
96 
97 	return max - used;
98 }
99 
100 /*
101  * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
102  */
103 static int iwl_queue_init(struct iwl_queue *q, int slots_num, u32 id)
104 {
105 	q->n_window = slots_num;
106 	q->id = id;
107 
108 	/* slots_num must be power-of-two size, otherwise
109 	 * get_cmd_index is broken. */
110 	if (WARN_ON(!is_power_of_2(slots_num)))
111 		return -EINVAL;
112 
113 	q->low_mark = q->n_window / 4;
114 	if (q->low_mark < 4)
115 		q->low_mark = 4;
116 
117 	q->high_mark = q->n_window / 8;
118 	if (q->high_mark < 2)
119 		q->high_mark = 2;
120 
121 	q->write_ptr = 0;
122 	q->read_ptr = 0;
123 
124 	return 0;
125 }
126 
127 static int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
128 				  struct iwl_dma_ptr *ptr, size_t size)
129 {
130 	if (WARN_ON(ptr->addr))
131 		return -EINVAL;
132 
133 	ptr->addr = dma_alloc_coherent(trans->dev, size,
134 				       &ptr->dma, GFP_KERNEL);
135 	if (!ptr->addr)
136 		return -ENOMEM;
137 	ptr->size = size;
138 	return 0;
139 }
140 
141 static void iwl_pcie_free_dma_ptr(struct iwl_trans *trans,
142 				  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 	u32 scd_sram_addr = trans_pcie->scd_base_addr +
157 				SCD_TX_STTS_QUEUE_OFFSET(txq->q.id);
158 	u8 buf[16];
159 	int i;
160 
161 	spin_lock(&txq->lock);
162 	/* check if triggered erroneously */
163 	if (txq->q.read_ptr == txq->q.write_ptr) {
164 		spin_unlock(&txq->lock);
165 		return;
166 	}
167 	spin_unlock(&txq->lock);
168 
169 	IWL_ERR(trans, "Queue %d stuck for %u ms.\n", txq->q.id,
170 		jiffies_to_msecs(txq->wd_timeout));
171 	IWL_ERR(trans, "Current SW read_ptr %d write_ptr %d\n",
172 		txq->q.read_ptr, txq->q.write_ptr);
173 
174 	iwl_trans_read_mem_bytes(trans, scd_sram_addr, buf, sizeof(buf));
175 
176 	iwl_print_hex_error(trans, buf, sizeof(buf));
177 
178 	for (i = 0; i < FH_TCSR_CHNL_NUM; i++)
179 		IWL_ERR(trans, "FH TRBs(%d) = 0x%08x\n", i,
180 			iwl_read_direct32(trans, FH_TX_TRB_REG(i)));
181 
182 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
183 		u32 status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(i));
184 		u8 fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7;
185 		bool active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE));
186 		u32 tbl_dw =
187 			iwl_trans_read_mem32(trans,
188 					     trans_pcie->scd_base_addr +
189 					     SCD_TRANS_TBL_OFFSET_QUEUE(i));
190 
191 		if (i & 0x1)
192 			tbl_dw = (tbl_dw & 0xFFFF0000) >> 16;
193 		else
194 			tbl_dw = tbl_dw & 0x0000FFFF;
195 
196 		IWL_ERR(trans,
197 			"Q %d is %sactive and mapped to fifo %d ra_tid 0x%04x [%d,%d]\n",
198 			i, active ? "" : "in", fifo, tbl_dw,
199 			iwl_read_prph(trans, SCD_QUEUE_RDPTR(i)) &
200 				(TFD_QUEUE_SIZE_MAX - 1),
201 			iwl_read_prph(trans, SCD_QUEUE_WRPTR(i)));
202 	}
203 
204 	iwl_force_nmi(trans);
205 }
206 
207 /*
208  * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
209  */
210 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
211 					     struct iwl_txq *txq, u16 byte_cnt)
212 {
213 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
214 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
215 	int write_ptr = txq->q.write_ptr;
216 	int txq_id = txq->q.id;
217 	u8 sec_ctl = 0;
218 	u8 sta_id = 0;
219 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
220 	__le16 bc_ent;
221 	struct iwl_tx_cmd *tx_cmd =
222 		(void *) txq->entries[txq->q.write_ptr].cmd->payload;
223 
224 	scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
225 
226 	sta_id = tx_cmd->sta_id;
227 	sec_ctl = tx_cmd->sec_ctl;
228 
229 	switch (sec_ctl & TX_CMD_SEC_MSK) {
230 	case TX_CMD_SEC_CCM:
231 		len += IEEE80211_CCMP_MIC_LEN;
232 		break;
233 	case TX_CMD_SEC_TKIP:
234 		len += IEEE80211_TKIP_ICV_LEN;
235 		break;
236 	case TX_CMD_SEC_WEP:
237 		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
238 		break;
239 	}
240 
241 	if (trans_pcie->bc_table_dword)
242 		len = DIV_ROUND_UP(len, 4);
243 
244 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
245 		return;
246 
247 	bc_ent = cpu_to_le16(len | (sta_id << 12));
248 
249 	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
250 
251 	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
252 		scd_bc_tbl[txq_id].
253 			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
254 }
255 
256 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
257 					    struct iwl_txq *txq)
258 {
259 	struct iwl_trans_pcie *trans_pcie =
260 		IWL_TRANS_GET_PCIE_TRANS(trans);
261 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
262 	int txq_id = txq->q.id;
263 	int read_ptr = txq->q.read_ptr;
264 	u8 sta_id = 0;
265 	__le16 bc_ent;
266 	struct iwl_tx_cmd *tx_cmd =
267 		(void *)txq->entries[txq->q.read_ptr].cmd->payload;
268 
269 	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
270 
271 	if (txq_id != trans_pcie->cmd_queue)
272 		sta_id = tx_cmd->sta_id;
273 
274 	bc_ent = cpu_to_le16(1 | (sta_id << 12));
275 	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
276 
277 	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
278 		scd_bc_tbl[txq_id].
279 			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
280 }
281 
282 /*
283  * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
284  */
285 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
286 				    struct iwl_txq *txq)
287 {
288 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
289 	u32 reg = 0;
290 	int txq_id = txq->q.id;
291 
292 	lockdep_assert_held(&txq->lock);
293 
294 	/*
295 	 * explicitly wake up the NIC if:
296 	 * 1. shadow registers aren't enabled
297 	 * 2. NIC is woken up for CMD regardless of shadow outside this function
298 	 * 3. there is a chance that the NIC is asleep
299 	 */
300 	if (!trans->cfg->base_params->shadow_reg_enable &&
301 	    txq_id != trans_pcie->cmd_queue &&
302 	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
303 		/*
304 		 * wake up nic if it's powered down ...
305 		 * uCode will wake up, and interrupt us again, so next
306 		 * time we'll skip this part.
307 		 */
308 		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
309 
310 		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
311 			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
312 				       txq_id, reg);
313 			iwl_set_bit(trans, CSR_GP_CNTRL,
314 				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
315 			txq->need_update = true;
316 			return;
317 		}
318 	}
319 
320 	/*
321 	 * if not in power-save mode, uCode will never sleep when we're
322 	 * trying to tx (during RFKILL, we're not trying to tx).
323 	 */
324 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->q.write_ptr);
325 	if (!txq->block)
326 		iwl_write32(trans, HBUS_TARG_WRPTR,
327 			    txq->q.write_ptr | (txq_id << 8));
328 }
329 
330 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
331 {
332 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
333 	int i;
334 
335 	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
336 		struct iwl_txq *txq = &trans_pcie->txq[i];
337 
338 		spin_lock_bh(&txq->lock);
339 		if (trans_pcie->txq[i].need_update) {
340 			iwl_pcie_txq_inc_wr_ptr(trans, txq);
341 			trans_pcie->txq[i].need_update = false;
342 		}
343 		spin_unlock_bh(&txq->lock);
344 	}
345 }
346 
347 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx)
348 {
349 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
350 
351 	dma_addr_t addr = get_unaligned_le32(&tb->lo);
352 	if (sizeof(dma_addr_t) > sizeof(u32))
353 		addr |=
354 		((dma_addr_t)(le16_to_cpu(tb->hi_n_len) & 0xF) << 16) << 16;
355 
356 	return addr;
357 }
358 
359 static inline void iwl_pcie_tfd_set_tb(struct iwl_tfd *tfd, u8 idx,
360 				       dma_addr_t addr, u16 len)
361 {
362 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
363 	u16 hi_n_len = len << 4;
364 
365 	put_unaligned_le32(addr, &tb->lo);
366 	if (sizeof(dma_addr_t) > sizeof(u32))
367 		hi_n_len |= ((addr >> 16) >> 16) & 0xF;
368 
369 	tb->hi_n_len = cpu_to_le16(hi_n_len);
370 
371 	tfd->num_tbs = idx + 1;
372 }
373 
374 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_tfd *tfd)
375 {
376 	return tfd->num_tbs & 0x1f;
377 }
378 
379 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
380 			       struct iwl_cmd_meta *meta,
381 			       struct iwl_tfd *tfd)
382 {
383 	int i;
384 	int num_tbs;
385 
386 	/* Sanity check on number of chunks */
387 	num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
388 
389 	if (num_tbs >= IWL_NUM_OF_TBS) {
390 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
391 		/* @todo issue fatal error, it is quite serious situation */
392 		return;
393 	}
394 
395 	/* first TB is never freed - it's the scratchbuf data */
396 
397 	for (i = 1; i < num_tbs; i++) {
398 		if (meta->flags & BIT(i + CMD_TB_BITMAP_POS))
399 			dma_unmap_page(trans->dev,
400 				       iwl_pcie_tfd_tb_get_addr(tfd, i),
401 				       iwl_pcie_tfd_tb_get_len(tfd, i),
402 				       DMA_TO_DEVICE);
403 		else
404 			dma_unmap_single(trans->dev,
405 					 iwl_pcie_tfd_tb_get_addr(tfd, i),
406 					 iwl_pcie_tfd_tb_get_len(tfd, i),
407 					 DMA_TO_DEVICE);
408 	}
409 	tfd->num_tbs = 0;
410 }
411 
412 /*
413  * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
414  * @trans - transport private data
415  * @txq - tx queue
416  * @dma_dir - the direction of the DMA mapping
417  *
418  * Does NOT advance any TFD circular buffer read/write indexes
419  * Does NOT free the TFD itself (which is within circular buffer)
420  */
421 static void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
422 {
423 	struct iwl_tfd *tfd_tmp = txq->tfds;
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->q.read_ptr;
429 	int idx = get_cmd_index(&txq->q, 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, &tfd_tmp[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_queue *q;
459 	struct iwl_tfd *tfd, *tfd_tmp;
460 	u32 num_tbs;
461 
462 	q = &txq->q;
463 	tfd_tmp = txq->tfds;
464 	tfd = &tfd_tmp[q->write_ptr];
465 
466 	if (reset)
467 		memset(tfd, 0, sizeof(*tfd));
468 
469 	num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
470 
471 	/* Each TFD can point to a maximum 20 Tx buffers */
472 	if (num_tbs >= IWL_NUM_OF_TBS) {
473 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
474 			IWL_NUM_OF_TBS);
475 		return -EINVAL;
476 	}
477 
478 	if (WARN(addr & ~IWL_TX_DMA_MASK,
479 		 "Unaligned address = %llx\n", (unsigned long long)addr))
480 		return -EINVAL;
481 
482 	iwl_pcie_tfd_set_tb(tfd, num_tbs, addr, len);
483 
484 	return num_tbs;
485 }
486 
487 static int iwl_pcie_txq_alloc(struct iwl_trans *trans,
488 			       struct iwl_txq *txq, int slots_num,
489 			       u32 txq_id)
490 {
491 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
492 	size_t tfd_sz = sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX;
493 	size_t scratchbuf_sz;
494 	int i;
495 
496 	if (WARN_ON(txq->entries || txq->tfds))
497 		return -EINVAL;
498 
499 	setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer,
500 		    (unsigned long)txq);
501 	txq->trans_pcie = trans_pcie;
502 
503 	txq->q.n_window = slots_num;
504 
505 	txq->entries = kcalloc(slots_num,
506 			       sizeof(struct iwl_pcie_txq_entry),
507 			       GFP_KERNEL);
508 
509 	if (!txq->entries)
510 		goto error;
511 
512 	if (txq_id == trans_pcie->cmd_queue)
513 		for (i = 0; i < slots_num; i++) {
514 			txq->entries[i].cmd =
515 				kmalloc(sizeof(struct iwl_device_cmd),
516 					GFP_KERNEL);
517 			if (!txq->entries[i].cmd)
518 				goto error;
519 		}
520 
521 	/* Circular buffer of transmit frame descriptors (TFDs),
522 	 * shared with device */
523 	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
524 				       &txq->q.dma_addr, GFP_KERNEL);
525 	if (!txq->tfds)
526 		goto error;
527 
528 	BUILD_BUG_ON(IWL_HCMD_SCRATCHBUF_SIZE != sizeof(*txq->scratchbufs));
529 	BUILD_BUG_ON(offsetof(struct iwl_pcie_txq_scratch_buf, scratch) !=
530 			sizeof(struct iwl_cmd_header) +
531 			offsetof(struct iwl_tx_cmd, scratch));
532 
533 	scratchbuf_sz = sizeof(*txq->scratchbufs) * slots_num;
534 
535 	txq->scratchbufs = dma_alloc_coherent(trans->dev, scratchbuf_sz,
536 					      &txq->scratchbufs_dma,
537 					      GFP_KERNEL);
538 	if (!txq->scratchbufs)
539 		goto err_free_tfds;
540 
541 	txq->q.id = txq_id;
542 
543 	return 0;
544 err_free_tfds:
545 	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->q.dma_addr);
546 error:
547 	if (txq->entries && txq_id == trans_pcie->cmd_queue)
548 		for (i = 0; i < slots_num; i++)
549 			kfree(txq->entries[i].cmd);
550 	kfree(txq->entries);
551 	txq->entries = NULL;
552 
553 	return -ENOMEM;
554 
555 }
556 
557 static int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
558 			      int slots_num, u32 txq_id)
559 {
560 	int ret;
561 
562 	txq->need_update = false;
563 
564 	/* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise
565 	 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
566 	BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1));
567 
568 	/* Initialize queue's high/low-water marks, and head/tail indexes */
569 	ret = iwl_queue_init(&txq->q, slots_num, txq_id);
570 	if (ret)
571 		return ret;
572 
573 	spin_lock_init(&txq->lock);
574 
575 	/*
576 	 * Tell nic where to find circular buffer of Tx Frame Descriptors for
577 	 * given Tx queue, and enable the DMA channel used for that queue.
578 	 * Circular buffer (TFD queue in DRAM) physical base address */
579 	iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id),
580 			   txq->q.dma_addr >> 8);
581 
582 	return 0;
583 }
584 
585 static void iwl_pcie_free_tso_page(struct sk_buff *skb)
586 {
587 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
588 
589 	if (info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA]) {
590 		struct page *page =
591 			info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA];
592 
593 		__free_page(page);
594 		info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA] = NULL;
595 	}
596 }
597 
598 /*
599  * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
600  */
601 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
602 {
603 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
604 	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
605 	struct iwl_queue *q = &txq->q;
606 
607 	spin_lock_bh(&txq->lock);
608 	while (q->write_ptr != q->read_ptr) {
609 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
610 				   txq_id, q->read_ptr);
611 
612 		if (txq_id != trans_pcie->cmd_queue) {
613 			struct sk_buff *skb = txq->entries[q->read_ptr].skb;
614 
615 			if (WARN_ON_ONCE(!skb))
616 				continue;
617 
618 			iwl_pcie_free_tso_page(skb);
619 		}
620 		iwl_pcie_txq_free_tfd(trans, txq);
621 		q->read_ptr = iwl_queue_inc_wrap(q->read_ptr);
622 	}
623 	txq->active = false;
624 	spin_unlock_bh(&txq->lock);
625 
626 	/* just in case - this queue may have been stopped */
627 	iwl_wake_queue(trans, txq);
628 }
629 
630 /*
631  * iwl_pcie_txq_free - Deallocate DMA queue.
632  * @txq: Transmit queue to deallocate.
633  *
634  * Empty queue by removing and destroying all BD's.
635  * Free all buffers.
636  * 0-fill, but do not free "txq" descriptor structure.
637  */
638 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
639 {
640 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
641 	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
642 	struct device *dev = trans->dev;
643 	int i;
644 
645 	if (WARN_ON(!txq))
646 		return;
647 
648 	iwl_pcie_txq_unmap(trans, txq_id);
649 
650 	/* De-alloc array of command/tx buffers */
651 	if (txq_id == trans_pcie->cmd_queue)
652 		for (i = 0; i < txq->q.n_window; i++) {
653 			kzfree(txq->entries[i].cmd);
654 			kzfree(txq->entries[i].free_buf);
655 		}
656 
657 	/* De-alloc circular buffer of TFDs */
658 	if (txq->tfds) {
659 		dma_free_coherent(dev,
660 				  sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX,
661 				  txq->tfds, txq->q.dma_addr);
662 		txq->q.dma_addr = 0;
663 		txq->tfds = NULL;
664 
665 		dma_free_coherent(dev,
666 				  sizeof(*txq->scratchbufs) * txq->q.n_window,
667 				  txq->scratchbufs, txq->scratchbufs_dma);
668 	}
669 
670 	kfree(txq->entries);
671 	txq->entries = NULL;
672 
673 	del_timer_sync(&txq->stuck_timer);
674 
675 	/* 0-fill queue descriptor structure */
676 	memset(txq, 0, sizeof(*txq));
677 }
678 
679 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
680 {
681 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
682 	int nq = trans->cfg->base_params->num_of_queues;
683 	int chan;
684 	u32 reg_val;
685 	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
686 				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
687 
688 	/* make sure all queue are not stopped/used */
689 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
690 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
691 
692 	trans_pcie->scd_base_addr =
693 		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
694 
695 	WARN_ON(scd_base_addr != 0 &&
696 		scd_base_addr != trans_pcie->scd_base_addr);
697 
698 	/* reset context data, TX status and translation data */
699 	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
700 				   SCD_CONTEXT_MEM_LOWER_BOUND,
701 			    NULL, clear_dwords);
702 
703 	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
704 		       trans_pcie->scd_bc_tbls.dma >> 10);
705 
706 	/* The chain extension of the SCD doesn't work well. This feature is
707 	 * enabled by default by the HW, so we need to disable it manually.
708 	 */
709 	if (trans->cfg->base_params->scd_chain_ext_wa)
710 		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
711 
712 	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
713 				trans_pcie->cmd_fifo,
714 				trans_pcie->cmd_q_wdg_timeout);
715 
716 	/* Activate all Tx DMA/FIFO channels */
717 	iwl_scd_activate_fifos(trans);
718 
719 	/* Enable DMA channel */
720 	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
721 		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
722 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
723 				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
724 
725 	/* Update FH chicken bits */
726 	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
727 	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
728 			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
729 
730 	/* Enable L1-Active */
731 	if (trans->cfg->device_family != IWL_DEVICE_FAMILY_8000)
732 		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
733 				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
734 }
735 
736 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
737 {
738 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
739 	int txq_id;
740 
741 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
742 	     txq_id++) {
743 		struct iwl_txq *txq = &trans_pcie->txq[txq_id];
744 
745 		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id),
746 				   txq->q.dma_addr >> 8);
747 		iwl_pcie_txq_unmap(trans, txq_id);
748 		txq->q.read_ptr = 0;
749 		txq->q.write_ptr = 0;
750 	}
751 
752 	/* Tell NIC where to find the "keep warm" buffer */
753 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
754 			   trans_pcie->kw.dma >> 4);
755 
756 	/*
757 	 * Send 0 as the scd_base_addr since the device may have be reset
758 	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
759 	 * contain garbage.
760 	 */
761 	iwl_pcie_tx_start(trans, 0);
762 }
763 
764 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
765 {
766 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
767 	unsigned long flags;
768 	int ch, ret;
769 	u32 mask = 0;
770 
771 	spin_lock(&trans_pcie->irq_lock);
772 
773 	if (!iwl_trans_grab_nic_access(trans, &flags))
774 		goto out;
775 
776 	/* Stop each Tx DMA channel */
777 	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
778 		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
779 		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
780 	}
781 
782 	/* Wait for DMA channels to be idle */
783 	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
784 	if (ret < 0)
785 		IWL_ERR(trans,
786 			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
787 			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
788 
789 	iwl_trans_release_nic_access(trans, &flags);
790 
791 out:
792 	spin_unlock(&trans_pcie->irq_lock);
793 }
794 
795 /*
796  * iwl_pcie_tx_stop - Stop all Tx DMA channels
797  */
798 int iwl_pcie_tx_stop(struct iwl_trans *trans)
799 {
800 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
801 	int txq_id;
802 
803 	/* Turn off all Tx DMA fifos */
804 	iwl_scd_deactivate_fifos(trans);
805 
806 	/* Turn off all Tx DMA channels */
807 	iwl_pcie_tx_stop_fh(trans);
808 
809 	/*
810 	 * This function can be called before the op_mode disabled the
811 	 * queues. This happens when we have an rfkill interrupt.
812 	 * Since we stop Tx altogether - mark the queues as stopped.
813 	 */
814 	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
815 	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
816 
817 	/* This can happen: start_hw, stop_device */
818 	if (!trans_pcie->txq)
819 		return 0;
820 
821 	/* Unmap DMA from host system and free skb's */
822 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
823 	     txq_id++)
824 		iwl_pcie_txq_unmap(trans, txq_id);
825 
826 	return 0;
827 }
828 
829 /*
830  * iwl_trans_tx_free - Free TXQ Context
831  *
832  * Destroy all TX DMA queues and structures
833  */
834 void iwl_pcie_tx_free(struct iwl_trans *trans)
835 {
836 	int txq_id;
837 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
838 
839 	/* Tx queues */
840 	if (trans_pcie->txq) {
841 		for (txq_id = 0;
842 		     txq_id < trans->cfg->base_params->num_of_queues; txq_id++)
843 			iwl_pcie_txq_free(trans, txq_id);
844 	}
845 
846 	kfree(trans_pcie->txq);
847 	trans_pcie->txq = NULL;
848 
849 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
850 
851 	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
852 }
853 
854 /*
855  * iwl_pcie_tx_alloc - allocate TX context
856  * Allocate all Tx DMA structures and initialize them
857  */
858 static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
859 {
860 	int ret;
861 	int txq_id, slots_num;
862 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
863 
864 	u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
865 			sizeof(struct iwlagn_scd_bc_tbl);
866 
867 	/*It is not allowed to alloc twice, so warn when this happens.
868 	 * We cannot rely on the previous allocation, so free and fail */
869 	if (WARN_ON(trans_pcie->txq)) {
870 		ret = -EINVAL;
871 		goto error;
872 	}
873 
874 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
875 				   scd_bc_tbls_size);
876 	if (ret) {
877 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
878 		goto error;
879 	}
880 
881 	/* Alloc keep-warm buffer */
882 	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
883 	if (ret) {
884 		IWL_ERR(trans, "Keep Warm allocation failed\n");
885 		goto error;
886 	}
887 
888 	trans_pcie->txq = kcalloc(trans->cfg->base_params->num_of_queues,
889 				  sizeof(struct iwl_txq), GFP_KERNEL);
890 	if (!trans_pcie->txq) {
891 		IWL_ERR(trans, "Not enough memory for txq\n");
892 		ret = -ENOMEM;
893 		goto error;
894 	}
895 
896 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
897 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
898 	     txq_id++) {
899 		slots_num = (txq_id == trans_pcie->cmd_queue) ?
900 					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
901 		ret = iwl_pcie_txq_alloc(trans, &trans_pcie->txq[txq_id],
902 					  slots_num, txq_id);
903 		if (ret) {
904 			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
905 			goto error;
906 		}
907 	}
908 
909 	return 0;
910 
911 error:
912 	iwl_pcie_tx_free(trans);
913 
914 	return ret;
915 }
916 int iwl_pcie_tx_init(struct iwl_trans *trans)
917 {
918 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
919 	int ret;
920 	int txq_id, slots_num;
921 	bool alloc = false;
922 
923 	if (!trans_pcie->txq) {
924 		ret = iwl_pcie_tx_alloc(trans);
925 		if (ret)
926 			goto error;
927 		alloc = true;
928 	}
929 
930 	spin_lock(&trans_pcie->irq_lock);
931 
932 	/* Turn off all Tx DMA fifos */
933 	iwl_scd_deactivate_fifos(trans);
934 
935 	/* Tell NIC where to find the "keep warm" buffer */
936 	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
937 			   trans_pcie->kw.dma >> 4);
938 
939 	spin_unlock(&trans_pcie->irq_lock);
940 
941 	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
942 	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
943 	     txq_id++) {
944 		slots_num = (txq_id == trans_pcie->cmd_queue) ?
945 					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
946 		ret = iwl_pcie_txq_init(trans, &trans_pcie->txq[txq_id],
947 					 slots_num, txq_id);
948 		if (ret) {
949 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
950 			goto error;
951 		}
952 	}
953 
954 	iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
955 	if (trans->cfg->base_params->num_of_queues > 20)
956 		iwl_set_bits_prph(trans, SCD_GP_CTRL,
957 				  SCD_GP_CTRL_ENABLE_31_QUEUES);
958 
959 	return 0;
960 error:
961 	/*Upon error, free only if we allocated something */
962 	if (alloc)
963 		iwl_pcie_tx_free(trans);
964 	return ret;
965 }
966 
967 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
968 {
969 	lockdep_assert_held(&txq->lock);
970 
971 	if (!txq->wd_timeout)
972 		return;
973 
974 	/*
975 	 * station is asleep and we send data - that must
976 	 * be uAPSD or PS-Poll. Don't rearm the timer.
977 	 */
978 	if (txq->frozen)
979 		return;
980 
981 	/*
982 	 * if empty delete timer, otherwise move timer forward
983 	 * since we're making progress on this queue
984 	 */
985 	if (txq->q.read_ptr == txq->q.write_ptr)
986 		del_timer(&txq->stuck_timer);
987 	else
988 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
989 }
990 
991 /* Frees buffers until index _not_ inclusive */
992 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
993 			    struct sk_buff_head *skbs)
994 {
995 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
996 	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
997 	int tfd_num = ssn & (TFD_QUEUE_SIZE_MAX - 1);
998 	struct iwl_queue *q = &txq->q;
999 	int last_to_free;
1000 
1001 	/* This function is not meant to release cmd queue*/
1002 	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1003 		return;
1004 
1005 	spin_lock_bh(&txq->lock);
1006 
1007 	if (!txq->active) {
1008 		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1009 				    txq_id, ssn);
1010 		goto out;
1011 	}
1012 
1013 	if (txq->q.read_ptr == tfd_num)
1014 		goto out;
1015 
1016 	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1017 			   txq_id, txq->q.read_ptr, tfd_num, ssn);
1018 
1019 	/*Since we free until index _not_ inclusive, the one before index is
1020 	 * the last we will free. This one must be used */
1021 	last_to_free = iwl_queue_dec_wrap(tfd_num);
1022 
1023 	if (!iwl_queue_used(q, last_to_free)) {
1024 		IWL_ERR(trans,
1025 			"%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1026 			__func__, txq_id, last_to_free, TFD_QUEUE_SIZE_MAX,
1027 			q->write_ptr, q->read_ptr);
1028 		goto out;
1029 	}
1030 
1031 	if (WARN_ON(!skb_queue_empty(skbs)))
1032 		goto out;
1033 
1034 	for (;
1035 	     q->read_ptr != tfd_num;
1036 	     q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1037 		struct sk_buff *skb = txq->entries[txq->q.read_ptr].skb;
1038 
1039 		if (WARN_ON_ONCE(!skb))
1040 			continue;
1041 
1042 		iwl_pcie_free_tso_page(skb);
1043 
1044 		__skb_queue_tail(skbs, skb);
1045 
1046 		txq->entries[txq->q.read_ptr].skb = NULL;
1047 
1048 		iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1049 
1050 		iwl_pcie_txq_free_tfd(trans, txq);
1051 	}
1052 
1053 	iwl_pcie_txq_progress(txq);
1054 
1055 	if (iwl_queue_space(&txq->q) > txq->q.low_mark)
1056 		iwl_wake_queue(trans, txq);
1057 
1058 	if (q->read_ptr == q->write_ptr) {
1059 		IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", q->id);
1060 		iwl_trans_pcie_unref(trans);
1061 	}
1062 
1063 out:
1064 	spin_unlock_bh(&txq->lock);
1065 }
1066 
1067 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1068 				      const struct iwl_host_cmd *cmd)
1069 {
1070 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1071 	int ret;
1072 
1073 	lockdep_assert_held(&trans_pcie->reg_lock);
1074 
1075 	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
1076 	    !trans_pcie->ref_cmd_in_flight) {
1077 		trans_pcie->ref_cmd_in_flight = true;
1078 		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
1079 		iwl_trans_pcie_ref(trans);
1080 	}
1081 
1082 	/*
1083 	 * wake up the NIC to make sure that the firmware will see the host
1084 	 * command - we will let the NIC sleep once all the host commands
1085 	 * returned. This needs to be done only on NICs that have
1086 	 * apmg_wake_up_wa set.
1087 	 */
1088 	if (trans->cfg->base_params->apmg_wake_up_wa &&
1089 	    !trans_pcie->cmd_hold_nic_awake) {
1090 		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1091 					 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1092 
1093 		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1094 				   CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1095 				   (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1096 				    CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1097 				   15000);
1098 		if (ret < 0) {
1099 			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1100 					CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1101 			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1102 			return -EIO;
1103 		}
1104 		trans_pcie->cmd_hold_nic_awake = true;
1105 	}
1106 
1107 	return 0;
1108 }
1109 
1110 static int iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
1111 {
1112 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1113 
1114 	lockdep_assert_held(&trans_pcie->reg_lock);
1115 
1116 	if (trans_pcie->ref_cmd_in_flight) {
1117 		trans_pcie->ref_cmd_in_flight = false;
1118 		IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
1119 		iwl_trans_pcie_unref(trans);
1120 	}
1121 
1122 	if (trans->cfg->base_params->apmg_wake_up_wa) {
1123 		if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
1124 			return 0;
1125 
1126 		trans_pcie->cmd_hold_nic_awake = false;
1127 		__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1128 					   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1129 	}
1130 	return 0;
1131 }
1132 
1133 /*
1134  * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1135  *
1136  * When FW advances 'R' index, all entries between old and new 'R' index
1137  * need to be reclaimed. As result, some free space forms.  If there is
1138  * enough free space (> low mark), wake the stack that feeds us.
1139  */
1140 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1141 {
1142 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1143 	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1144 	struct iwl_queue *q = &txq->q;
1145 	unsigned long flags;
1146 	int nfreed = 0;
1147 
1148 	lockdep_assert_held(&txq->lock);
1149 
1150 	if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(q, idx))) {
1151 		IWL_ERR(trans,
1152 			"%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1153 			__func__, txq_id, idx, TFD_QUEUE_SIZE_MAX,
1154 			q->write_ptr, q->read_ptr);
1155 		return;
1156 	}
1157 
1158 	for (idx = iwl_queue_inc_wrap(idx); q->read_ptr != idx;
1159 	     q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1160 
1161 		if (nfreed++ > 0) {
1162 			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1163 				idx, q->write_ptr, q->read_ptr);
1164 			iwl_force_nmi(trans);
1165 		}
1166 	}
1167 
1168 	if (q->read_ptr == q->write_ptr) {
1169 		spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1170 		iwl_pcie_clear_cmd_in_flight(trans);
1171 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1172 	}
1173 
1174 	iwl_pcie_txq_progress(txq);
1175 }
1176 
1177 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1178 				 u16 txq_id)
1179 {
1180 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1181 	u32 tbl_dw_addr;
1182 	u32 tbl_dw;
1183 	u16 scd_q2ratid;
1184 
1185 	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1186 
1187 	tbl_dw_addr = trans_pcie->scd_base_addr +
1188 			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1189 
1190 	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1191 
1192 	if (txq_id & 0x1)
1193 		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1194 	else
1195 		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1196 
1197 	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1198 
1199 	return 0;
1200 }
1201 
1202 /* Receiver address (actually, Rx station's index into station table),
1203  * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1204 #define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
1205 
1206 void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1207 			       const struct iwl_trans_txq_scd_cfg *cfg,
1208 			       unsigned int wdg_timeout)
1209 {
1210 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1211 	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1212 	int fifo = -1;
1213 
1214 	if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1215 		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1216 
1217 	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1218 
1219 	if (cfg) {
1220 		fifo = cfg->fifo;
1221 
1222 		/* Disable the scheduler prior configuring the cmd queue */
1223 		if (txq_id == trans_pcie->cmd_queue &&
1224 		    trans_pcie->scd_set_active)
1225 			iwl_scd_enable_set_active(trans, 0);
1226 
1227 		/* Stop this Tx queue before configuring it */
1228 		iwl_scd_txq_set_inactive(trans, txq_id);
1229 
1230 		/* Set this queue as a chain-building queue unless it is CMD */
1231 		if (txq_id != trans_pcie->cmd_queue)
1232 			iwl_scd_txq_set_chain(trans, txq_id);
1233 
1234 		if (cfg->aggregate) {
1235 			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1236 
1237 			/* Map receiver-address / traffic-ID to this queue */
1238 			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1239 
1240 			/* enable aggregations for the queue */
1241 			iwl_scd_txq_enable_agg(trans, txq_id);
1242 			txq->ampdu = true;
1243 		} else {
1244 			/*
1245 			 * disable aggregations for the queue, this will also
1246 			 * make the ra_tid mapping configuration irrelevant
1247 			 * since it is now a non-AGG queue.
1248 			 */
1249 			iwl_scd_txq_disable_agg(trans, txq_id);
1250 
1251 			ssn = txq->q.read_ptr;
1252 		}
1253 	}
1254 
1255 	/* Place first TFD at index corresponding to start sequence number.
1256 	 * Assumes that ssn_idx is valid (!= 0xFFF) */
1257 	txq->q.read_ptr = (ssn & 0xff);
1258 	txq->q.write_ptr = (ssn & 0xff);
1259 	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1260 			   (ssn & 0xff) | (txq_id << 8));
1261 
1262 	if (cfg) {
1263 		u8 frame_limit = cfg->frame_limit;
1264 
1265 		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1266 
1267 		/* Set up Tx window size and frame limit for this queue */
1268 		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1269 				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1270 		iwl_trans_write_mem32(trans,
1271 			trans_pcie->scd_base_addr +
1272 			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1273 			((frame_limit << SCD_QUEUE_CTX_REG2_WIN_SIZE_POS) &
1274 					SCD_QUEUE_CTX_REG2_WIN_SIZE_MSK) |
1275 			((frame_limit << SCD_QUEUE_CTX_REG2_FRAME_LIMIT_POS) &
1276 					SCD_QUEUE_CTX_REG2_FRAME_LIMIT_MSK));
1277 
1278 		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1279 		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1280 			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1281 			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1282 			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1283 			       SCD_QUEUE_STTS_REG_MSK);
1284 
1285 		/* enable the scheduler for this queue (only) */
1286 		if (txq_id == trans_pcie->cmd_queue &&
1287 		    trans_pcie->scd_set_active)
1288 			iwl_scd_enable_set_active(trans, BIT(txq_id));
1289 
1290 		IWL_DEBUG_TX_QUEUES(trans,
1291 				    "Activate queue %d on FIFO %d WrPtr: %d\n",
1292 				    txq_id, fifo, ssn & 0xff);
1293 	} else {
1294 		IWL_DEBUG_TX_QUEUES(trans,
1295 				    "Activate queue %d WrPtr: %d\n",
1296 				    txq_id, ssn & 0xff);
1297 	}
1298 
1299 	txq->active = true;
1300 }
1301 
1302 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1303 				bool configure_scd)
1304 {
1305 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1306 	u32 stts_addr = trans_pcie->scd_base_addr +
1307 			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1308 	static const u32 zero_val[4] = {};
1309 
1310 	trans_pcie->txq[txq_id].frozen_expiry_remainder = 0;
1311 	trans_pcie->txq[txq_id].frozen = false;
1312 
1313 	/*
1314 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1315 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1316 	 * allow the op_mode to call txq_disable after it already called
1317 	 * stop_device.
1318 	 */
1319 	if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1320 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1321 			  "queue %d not used", txq_id);
1322 		return;
1323 	}
1324 
1325 	if (configure_scd) {
1326 		iwl_scd_txq_set_inactive(trans, txq_id);
1327 
1328 		iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1329 				    ARRAY_SIZE(zero_val));
1330 	}
1331 
1332 	iwl_pcie_txq_unmap(trans, txq_id);
1333 	trans_pcie->txq[txq_id].ampdu = false;
1334 
1335 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1336 }
1337 
1338 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1339 
1340 /*
1341  * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1342  * @priv: device private data point
1343  * @cmd: a pointer to the ucode command structure
1344  *
1345  * The function returns < 0 values to indicate the operation
1346  * failed. On success, it returns the index (>= 0) of command in the
1347  * command queue.
1348  */
1349 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1350 				 struct iwl_host_cmd *cmd)
1351 {
1352 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1353 	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1354 	struct iwl_queue *q = &txq->q;
1355 	struct iwl_device_cmd *out_cmd;
1356 	struct iwl_cmd_meta *out_meta;
1357 	unsigned long flags;
1358 	void *dup_buf = NULL;
1359 	dma_addr_t phys_addr;
1360 	int idx;
1361 	u16 copy_size, cmd_size, scratch_size;
1362 	bool had_nocopy = false;
1363 	u8 group_id = iwl_cmd_groupid(cmd->id);
1364 	int i, ret;
1365 	u32 cmd_pos;
1366 	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1367 	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1368 
1369 	if (WARN(!trans_pcie->wide_cmd_header &&
1370 		 group_id > IWL_ALWAYS_LONG_GROUP,
1371 		 "unsupported wide command %#x\n", cmd->id))
1372 		return -EINVAL;
1373 
1374 	if (group_id != 0) {
1375 		copy_size = sizeof(struct iwl_cmd_header_wide);
1376 		cmd_size = sizeof(struct iwl_cmd_header_wide);
1377 	} else {
1378 		copy_size = sizeof(struct iwl_cmd_header);
1379 		cmd_size = sizeof(struct iwl_cmd_header);
1380 	}
1381 
1382 	/* need one for the header if the first is NOCOPY */
1383 	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1384 
1385 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1386 		cmddata[i] = cmd->data[i];
1387 		cmdlen[i] = cmd->len[i];
1388 
1389 		if (!cmd->len[i])
1390 			continue;
1391 
1392 		/* need at least IWL_HCMD_SCRATCHBUF_SIZE copied */
1393 		if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) {
1394 			int copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size;
1395 
1396 			if (copy > cmdlen[i])
1397 				copy = cmdlen[i];
1398 			cmdlen[i] -= copy;
1399 			cmddata[i] += copy;
1400 			copy_size += copy;
1401 		}
1402 
1403 		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1404 			had_nocopy = true;
1405 			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1406 				idx = -EINVAL;
1407 				goto free_dup_buf;
1408 			}
1409 		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1410 			/*
1411 			 * This is also a chunk that isn't copied
1412 			 * to the static buffer so set had_nocopy.
1413 			 */
1414 			had_nocopy = true;
1415 
1416 			/* only allowed once */
1417 			if (WARN_ON(dup_buf)) {
1418 				idx = -EINVAL;
1419 				goto free_dup_buf;
1420 			}
1421 
1422 			dup_buf = kmemdup(cmddata[i], cmdlen[i],
1423 					  GFP_ATOMIC);
1424 			if (!dup_buf)
1425 				return -ENOMEM;
1426 		} else {
1427 			/* NOCOPY must not be followed by normal! */
1428 			if (WARN_ON(had_nocopy)) {
1429 				idx = -EINVAL;
1430 				goto free_dup_buf;
1431 			}
1432 			copy_size += cmdlen[i];
1433 		}
1434 		cmd_size += cmd->len[i];
1435 	}
1436 
1437 	/*
1438 	 * If any of the command structures end up being larger than
1439 	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1440 	 * allocated into separate TFDs, then we will need to
1441 	 * increase the size of the buffers.
1442 	 */
1443 	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1444 		 "Command %s (%#x) is too large (%d bytes)\n",
1445 		 iwl_get_cmd_string(trans, cmd->id),
1446 		 cmd->id, copy_size)) {
1447 		idx = -EINVAL;
1448 		goto free_dup_buf;
1449 	}
1450 
1451 	spin_lock_bh(&txq->lock);
1452 
1453 	if (iwl_queue_space(q) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1454 		spin_unlock_bh(&txq->lock);
1455 
1456 		IWL_ERR(trans, "No space in command queue\n");
1457 		iwl_op_mode_cmd_queue_full(trans->op_mode);
1458 		idx = -ENOSPC;
1459 		goto free_dup_buf;
1460 	}
1461 
1462 	idx = get_cmd_index(q, q->write_ptr);
1463 	out_cmd = txq->entries[idx].cmd;
1464 	out_meta = &txq->entries[idx].meta;
1465 
1466 	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1467 	if (cmd->flags & CMD_WANT_SKB)
1468 		out_meta->source = cmd;
1469 
1470 	/* set up the header */
1471 	if (group_id != 0) {
1472 		out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1473 		out_cmd->hdr_wide.group_id = group_id;
1474 		out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1475 		out_cmd->hdr_wide.length =
1476 			cpu_to_le16(cmd_size -
1477 				    sizeof(struct iwl_cmd_header_wide));
1478 		out_cmd->hdr_wide.reserved = 0;
1479 		out_cmd->hdr_wide.sequence =
1480 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1481 						 INDEX_TO_SEQ(q->write_ptr));
1482 
1483 		cmd_pos = sizeof(struct iwl_cmd_header_wide);
1484 		copy_size = sizeof(struct iwl_cmd_header_wide);
1485 	} else {
1486 		out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1487 		out_cmd->hdr.sequence =
1488 			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1489 						 INDEX_TO_SEQ(q->write_ptr));
1490 		out_cmd->hdr.group_id = 0;
1491 
1492 		cmd_pos = sizeof(struct iwl_cmd_header);
1493 		copy_size = sizeof(struct iwl_cmd_header);
1494 	}
1495 
1496 	/* and copy the data that needs to be copied */
1497 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1498 		int copy;
1499 
1500 		if (!cmd->len[i])
1501 			continue;
1502 
1503 		/* copy everything if not nocopy/dup */
1504 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1505 					   IWL_HCMD_DFL_DUP))) {
1506 			copy = cmd->len[i];
1507 
1508 			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1509 			cmd_pos += copy;
1510 			copy_size += copy;
1511 			continue;
1512 		}
1513 
1514 		/*
1515 		 * Otherwise we need at least IWL_HCMD_SCRATCHBUF_SIZE copied
1516 		 * in total (for the scratchbuf handling), but copy up to what
1517 		 * we can fit into the payload for debug dump purposes.
1518 		 */
1519 		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1520 
1521 		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1522 		cmd_pos += copy;
1523 
1524 		/* However, treat copy_size the proper way, we need it below */
1525 		if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) {
1526 			copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size;
1527 
1528 			if (copy > cmd->len[i])
1529 				copy = cmd->len[i];
1530 			copy_size += copy;
1531 		}
1532 	}
1533 
1534 	IWL_DEBUG_HC(trans,
1535 		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1536 		     iwl_get_cmd_string(trans, cmd->id),
1537 		     group_id, out_cmd->hdr.cmd,
1538 		     le16_to_cpu(out_cmd->hdr.sequence),
1539 		     cmd_size, q->write_ptr, idx, trans_pcie->cmd_queue);
1540 
1541 	/* start the TFD with the scratchbuf */
1542 	scratch_size = min_t(int, copy_size, IWL_HCMD_SCRATCHBUF_SIZE);
1543 	memcpy(&txq->scratchbufs[q->write_ptr], &out_cmd->hdr, scratch_size);
1544 	iwl_pcie_txq_build_tfd(trans, txq,
1545 			       iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr),
1546 			       scratch_size, true);
1547 
1548 	/* map first command fragment, if any remains */
1549 	if (copy_size > scratch_size) {
1550 		phys_addr = dma_map_single(trans->dev,
1551 					   ((u8 *)&out_cmd->hdr) + scratch_size,
1552 					   copy_size - scratch_size,
1553 					   DMA_TO_DEVICE);
1554 		if (dma_mapping_error(trans->dev, phys_addr)) {
1555 			iwl_pcie_tfd_unmap(trans, out_meta,
1556 					   &txq->tfds[q->write_ptr]);
1557 			idx = -ENOMEM;
1558 			goto out;
1559 		}
1560 
1561 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1562 				       copy_size - scratch_size, false);
1563 	}
1564 
1565 	/* map the remaining (adjusted) nocopy/dup fragments */
1566 	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1567 		const void *data = cmddata[i];
1568 
1569 		if (!cmdlen[i])
1570 			continue;
1571 		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1572 					   IWL_HCMD_DFL_DUP)))
1573 			continue;
1574 		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1575 			data = dup_buf;
1576 		phys_addr = dma_map_single(trans->dev, (void *)data,
1577 					   cmdlen[i], DMA_TO_DEVICE);
1578 		if (dma_mapping_error(trans->dev, phys_addr)) {
1579 			iwl_pcie_tfd_unmap(trans, out_meta,
1580 					   &txq->tfds[q->write_ptr]);
1581 			idx = -ENOMEM;
1582 			goto out;
1583 		}
1584 
1585 		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1586 	}
1587 
1588 	BUILD_BUG_ON(IWL_NUM_OF_TBS + CMD_TB_BITMAP_POS >
1589 		     sizeof(out_meta->flags) * BITS_PER_BYTE);
1590 	out_meta->flags = cmd->flags;
1591 	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1592 		kzfree(txq->entries[idx].free_buf);
1593 	txq->entries[idx].free_buf = dup_buf;
1594 
1595 	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1596 
1597 	/* start timer if queue currently empty */
1598 	if (q->read_ptr == q->write_ptr && txq->wd_timeout)
1599 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1600 
1601 	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1602 	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1603 	if (ret < 0) {
1604 		idx = ret;
1605 		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1606 		goto out;
1607 	}
1608 
1609 	/* Increment and update queue's write index */
1610 	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
1611 	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1612 
1613 	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1614 
1615  out:
1616 	spin_unlock_bh(&txq->lock);
1617  free_dup_buf:
1618 	if (idx < 0)
1619 		kfree(dup_buf);
1620 	return idx;
1621 }
1622 
1623 /*
1624  * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1625  * @rxb: Rx buffer to reclaim
1626  */
1627 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1628 			    struct iwl_rx_cmd_buffer *rxb)
1629 {
1630 	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1631 	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1632 	u8 group_id = iwl_cmd_groupid(pkt->hdr.group_id);
1633 	u32 cmd_id;
1634 	int txq_id = SEQ_TO_QUEUE(sequence);
1635 	int index = SEQ_TO_INDEX(sequence);
1636 	int cmd_index;
1637 	struct iwl_device_cmd *cmd;
1638 	struct iwl_cmd_meta *meta;
1639 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1640 	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1641 
1642 	/* If a Tx command is being handled and it isn't in the actual
1643 	 * command queue then there a command routing bug has been introduced
1644 	 * in the queue management code. */
1645 	if (WARN(txq_id != trans_pcie->cmd_queue,
1646 		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1647 		 txq_id, trans_pcie->cmd_queue, sequence,
1648 		 trans_pcie->txq[trans_pcie->cmd_queue].q.read_ptr,
1649 		 trans_pcie->txq[trans_pcie->cmd_queue].q.write_ptr)) {
1650 		iwl_print_hex_error(trans, pkt, 32);
1651 		return;
1652 	}
1653 
1654 	spin_lock_bh(&txq->lock);
1655 
1656 	cmd_index = get_cmd_index(&txq->q, index);
1657 	cmd = txq->entries[cmd_index].cmd;
1658 	meta = &txq->entries[cmd_index].meta;
1659 	cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1660 
1661 	iwl_pcie_tfd_unmap(trans, meta, &txq->tfds[index]);
1662 
1663 	/* Input error checking is done when commands are added to queue. */
1664 	if (meta->flags & CMD_WANT_SKB) {
1665 		struct page *p = rxb_steal_page(rxb);
1666 
1667 		meta->source->resp_pkt = pkt;
1668 		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1669 		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1670 	}
1671 
1672 	if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1673 		iwl_op_mode_async_cb(trans->op_mode, cmd);
1674 
1675 	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1676 
1677 	if (!(meta->flags & CMD_ASYNC)) {
1678 		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1679 			IWL_WARN(trans,
1680 				 "HCMD_ACTIVE already clear for command %s\n",
1681 				 iwl_get_cmd_string(trans, cmd_id));
1682 		}
1683 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1684 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1685 			       iwl_get_cmd_string(trans, cmd_id));
1686 		wake_up(&trans_pcie->wait_command_queue);
1687 	}
1688 
1689 	meta->flags = 0;
1690 
1691 	spin_unlock_bh(&txq->lock);
1692 }
1693 
1694 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1695 
1696 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1697 				    struct iwl_host_cmd *cmd)
1698 {
1699 	int ret;
1700 
1701 	/* An asynchronous command can not expect an SKB to be set. */
1702 	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1703 		return -EINVAL;
1704 
1705 	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1706 	if (ret < 0) {
1707 		IWL_ERR(trans,
1708 			"Error sending %s: enqueue_hcmd failed: %d\n",
1709 			iwl_get_cmd_string(trans, cmd->id), ret);
1710 		return ret;
1711 	}
1712 	return 0;
1713 }
1714 
1715 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1716 				   struct iwl_host_cmd *cmd)
1717 {
1718 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1719 	int cmd_idx;
1720 	int ret;
1721 
1722 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1723 		       iwl_get_cmd_string(trans, cmd->id));
1724 
1725 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1726 				  &trans->status),
1727 		 "Command %s: a command is already active!\n",
1728 		 iwl_get_cmd_string(trans, cmd->id)))
1729 		return -EIO;
1730 
1731 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1732 		       iwl_get_cmd_string(trans, cmd->id));
1733 
1734 	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1735 	if (cmd_idx < 0) {
1736 		ret = cmd_idx;
1737 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1738 		IWL_ERR(trans,
1739 			"Error sending %s: enqueue_hcmd failed: %d\n",
1740 			iwl_get_cmd_string(trans, cmd->id), ret);
1741 		return ret;
1742 	}
1743 
1744 	ret = wait_event_timeout(trans_pcie->wait_command_queue,
1745 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1746 					   &trans->status),
1747 				 HOST_COMPLETE_TIMEOUT);
1748 	if (!ret) {
1749 		struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1750 		struct iwl_queue *q = &txq->q;
1751 
1752 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1753 			iwl_get_cmd_string(trans, cmd->id),
1754 			jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1755 
1756 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1757 			q->read_ptr, q->write_ptr);
1758 
1759 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1760 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1761 			       iwl_get_cmd_string(trans, cmd->id));
1762 		ret = -ETIMEDOUT;
1763 
1764 		iwl_force_nmi(trans);
1765 		iwl_trans_fw_error(trans);
1766 
1767 		goto cancel;
1768 	}
1769 
1770 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1771 		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1772 			iwl_get_cmd_string(trans, cmd->id));
1773 		dump_stack();
1774 		ret = -EIO;
1775 		goto cancel;
1776 	}
1777 
1778 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1779 	    test_bit(STATUS_RFKILL, &trans->status)) {
1780 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1781 		ret = -ERFKILL;
1782 		goto cancel;
1783 	}
1784 
1785 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1786 		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1787 			iwl_get_cmd_string(trans, cmd->id));
1788 		ret = -EIO;
1789 		goto cancel;
1790 	}
1791 
1792 	return 0;
1793 
1794 cancel:
1795 	if (cmd->flags & CMD_WANT_SKB) {
1796 		/*
1797 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1798 		 * TX cmd queue. Otherwise in case the cmd comes
1799 		 * in later, it will possibly set an invalid
1800 		 * address (cmd->meta.source).
1801 		 */
1802 		trans_pcie->txq[trans_pcie->cmd_queue].
1803 			entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1804 	}
1805 
1806 	if (cmd->resp_pkt) {
1807 		iwl_free_resp(cmd);
1808 		cmd->resp_pkt = NULL;
1809 	}
1810 
1811 	return ret;
1812 }
1813 
1814 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1815 {
1816 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1817 	    test_bit(STATUS_RFKILL, &trans->status)) {
1818 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1819 				  cmd->id);
1820 		return -ERFKILL;
1821 	}
1822 
1823 	if (cmd->flags & CMD_ASYNC)
1824 		return iwl_pcie_send_hcmd_async(trans, cmd);
1825 
1826 	/* We still can fail on RFKILL that can be asserted while we wait */
1827 	return iwl_pcie_send_hcmd_sync(trans, cmd);
1828 }
1829 
1830 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
1831 			     struct iwl_txq *txq, u8 hdr_len,
1832 			     struct iwl_cmd_meta *out_meta,
1833 			     struct iwl_device_cmd *dev_cmd, u16 tb1_len)
1834 {
1835 	struct iwl_queue *q = &txq->q;
1836 	u16 tb2_len;
1837 	int i;
1838 
1839 	/*
1840 	 * Set up TFD's third entry to point directly to remainder
1841 	 * of skb's head, if any
1842 	 */
1843 	tb2_len = skb_headlen(skb) - hdr_len;
1844 
1845 	if (tb2_len > 0) {
1846 		dma_addr_t tb2_phys = dma_map_single(trans->dev,
1847 						     skb->data + hdr_len,
1848 						     tb2_len, DMA_TO_DEVICE);
1849 		if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) {
1850 			iwl_pcie_tfd_unmap(trans, out_meta,
1851 					   &txq->tfds[q->write_ptr]);
1852 			return -EINVAL;
1853 		}
1854 		iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false);
1855 	}
1856 
1857 	/* set up the remaining entries to point to the data */
1858 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1859 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1860 		dma_addr_t tb_phys;
1861 		int tb_idx;
1862 
1863 		if (!skb_frag_size(frag))
1864 			continue;
1865 
1866 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
1867 					   skb_frag_size(frag), DMA_TO_DEVICE);
1868 
1869 		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
1870 			iwl_pcie_tfd_unmap(trans, out_meta,
1871 					   &txq->tfds[q->write_ptr]);
1872 			return -EINVAL;
1873 		}
1874 		tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
1875 						skb_frag_size(frag), false);
1876 
1877 		out_meta->flags |= BIT(tb_idx + CMD_TB_BITMAP_POS);
1878 	}
1879 
1880 	trace_iwlwifi_dev_tx(trans->dev, skb,
1881 			     &txq->tfds[txq->q.write_ptr],
1882 			     sizeof(struct iwl_tfd),
1883 			     &dev_cmd->hdr, IWL_HCMD_SCRATCHBUF_SIZE + tb1_len,
1884 			     skb->data + hdr_len, tb2_len);
1885 	trace_iwlwifi_dev_tx_data(trans->dev, skb,
1886 				  hdr_len, skb->len - hdr_len);
1887 	return 0;
1888 }
1889 
1890 #ifdef CONFIG_INET
1891 static struct iwl_tso_hdr_page *
1892 get_page_hdr(struct iwl_trans *trans, size_t len)
1893 {
1894 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1895 	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
1896 
1897 	if (!p->page)
1898 		goto alloc;
1899 
1900 	/* enough room on this page */
1901 	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
1902 		return p;
1903 
1904 	/* We don't have enough room on this page, get a new one. */
1905 	__free_page(p->page);
1906 
1907 alloc:
1908 	p->page = alloc_page(GFP_ATOMIC);
1909 	if (!p->page)
1910 		return NULL;
1911 	p->pos = page_address(p->page);
1912 	return p;
1913 }
1914 
1915 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
1916 					bool ipv6, unsigned int len)
1917 {
1918 	if (ipv6) {
1919 		struct ipv6hdr *iphv6 = iph;
1920 
1921 		tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
1922 					       len + tcph->doff * 4,
1923 					       IPPROTO_TCP, 0);
1924 	} else {
1925 		struct iphdr *iphv4 = iph;
1926 
1927 		ip_send_check(iphv4);
1928 		tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
1929 						 len + tcph->doff * 4,
1930 						 IPPROTO_TCP, 0);
1931 	}
1932 }
1933 
1934 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
1935 				   struct iwl_txq *txq, u8 hdr_len,
1936 				   struct iwl_cmd_meta *out_meta,
1937 				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
1938 {
1939 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
1940 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
1941 	struct ieee80211_hdr *hdr = (void *)skb->data;
1942 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
1943 	unsigned int mss = skb_shinfo(skb)->gso_size;
1944 	struct iwl_queue *q = &txq->q;
1945 	u16 length, iv_len, amsdu_pad;
1946 	u8 *start_hdr;
1947 	struct iwl_tso_hdr_page *hdr_page;
1948 	int ret;
1949 	struct tso_t tso;
1950 
1951 	/* if the packet is protected, then it must be CCMP or GCMP */
1952 	BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
1953 	iv_len = ieee80211_has_protected(hdr->frame_control) ?
1954 		IEEE80211_CCMP_HDR_LEN : 0;
1955 
1956 	trace_iwlwifi_dev_tx(trans->dev, skb,
1957 			     &txq->tfds[txq->q.write_ptr],
1958 			     sizeof(struct iwl_tfd),
1959 			     &dev_cmd->hdr, IWL_HCMD_SCRATCHBUF_SIZE + tb1_len,
1960 			     NULL, 0);
1961 
1962 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
1963 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
1964 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
1965 	amsdu_pad = 0;
1966 
1967 	/* total amount of header we may need for this A-MSDU */
1968 	hdr_room = DIV_ROUND_UP(total_len, mss) *
1969 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
1970 
1971 	/* Our device supports 9 segments at most, it will fit in 1 page */
1972 	hdr_page = get_page_hdr(trans, hdr_room);
1973 	if (!hdr_page)
1974 		return -ENOMEM;
1975 
1976 	get_page(hdr_page->page);
1977 	start_hdr = hdr_page->pos;
1978 	info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA] = hdr_page->page;
1979 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
1980 	hdr_page->pos += iv_len;
1981 
1982 	/*
1983 	 * Pull the ieee80211 header + IV to be able to use TSO core,
1984 	 * we will restore it for the tx_status flow.
1985 	 */
1986 	skb_pull(skb, hdr_len + iv_len);
1987 
1988 	tso_start(skb, &tso);
1989 
1990 	while (total_len) {
1991 		/* this is the data left for this subframe */
1992 		unsigned int data_left =
1993 			min_t(unsigned int, mss, total_len);
1994 		struct sk_buff *csum_skb = NULL;
1995 		unsigned int hdr_tb_len;
1996 		dma_addr_t hdr_tb_phys;
1997 		struct tcphdr *tcph;
1998 		u8 *iph;
1999 
2000 		total_len -= data_left;
2001 
2002 		memset(hdr_page->pos, 0, amsdu_pad);
2003 		hdr_page->pos += amsdu_pad;
2004 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2005 				  data_left)) & 0x3;
2006 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2007 		hdr_page->pos += ETH_ALEN;
2008 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2009 		hdr_page->pos += ETH_ALEN;
2010 
2011 		length = snap_ip_tcp_hdrlen + data_left;
2012 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2013 		hdr_page->pos += sizeof(length);
2014 
2015 		/*
2016 		 * This will copy the SNAP as well which will be considered
2017 		 * as MAC header.
2018 		 */
2019 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2020 		iph = hdr_page->pos + 8;
2021 		tcph = (void *)(iph + ip_hdrlen);
2022 
2023 		/* For testing on current hardware only */
2024 		if (trans_pcie->sw_csum_tx) {
2025 			csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2026 					     GFP_ATOMIC);
2027 			if (!csum_skb) {
2028 				ret = -ENOMEM;
2029 				goto out_unmap;
2030 			}
2031 
2032 			iwl_compute_pseudo_hdr_csum(iph, tcph,
2033 						    skb->protocol ==
2034 							htons(ETH_P_IPV6),
2035 						    data_left);
2036 
2037 			memcpy(skb_put(csum_skb, tcp_hdrlen(skb)),
2038 			       tcph, tcp_hdrlen(skb));
2039 			skb_set_transport_header(csum_skb, 0);
2040 			csum_skb->csum_start =
2041 				(unsigned char *)tcp_hdr(csum_skb) -
2042 						 csum_skb->head;
2043 		}
2044 
2045 		hdr_page->pos += snap_ip_tcp_hdrlen;
2046 
2047 		hdr_tb_len = hdr_page->pos - start_hdr;
2048 		hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2049 					     hdr_tb_len, DMA_TO_DEVICE);
2050 		if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2051 			dev_kfree_skb(csum_skb);
2052 			ret = -EINVAL;
2053 			goto out_unmap;
2054 		}
2055 		iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2056 				       hdr_tb_len, false);
2057 		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr,
2058 					       hdr_tb_len);
2059 
2060 		/* prepare the start_hdr for the next subframe */
2061 		start_hdr = hdr_page->pos;
2062 
2063 		/* put the payload */
2064 		while (data_left) {
2065 			unsigned int size = min_t(unsigned int, tso.size,
2066 						  data_left);
2067 			dma_addr_t tb_phys;
2068 
2069 			if (trans_pcie->sw_csum_tx)
2070 				memcpy(skb_put(csum_skb, size), tso.data, size);
2071 
2072 			tb_phys = dma_map_single(trans->dev, tso.data,
2073 						 size, DMA_TO_DEVICE);
2074 			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2075 				dev_kfree_skb(csum_skb);
2076 				ret = -EINVAL;
2077 				goto out_unmap;
2078 			}
2079 
2080 			iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2081 					       size, false);
2082 			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
2083 						       size);
2084 
2085 			data_left -= size;
2086 			tso_build_data(skb, &tso, size);
2087 		}
2088 
2089 		/* For testing on early hardware only */
2090 		if (trans_pcie->sw_csum_tx) {
2091 			__wsum csum;
2092 
2093 			csum = skb_checksum(csum_skb,
2094 					    skb_checksum_start_offset(csum_skb),
2095 					    csum_skb->len -
2096 					    skb_checksum_start_offset(csum_skb),
2097 					    0);
2098 			dev_kfree_skb(csum_skb);
2099 			dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2100 						hdr_tb_len, DMA_TO_DEVICE);
2101 			tcph->check = csum_fold(csum);
2102 			dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2103 						   hdr_tb_len, DMA_TO_DEVICE);
2104 		}
2105 	}
2106 
2107 	/* re -add the WiFi header and IV */
2108 	skb_push(skb, hdr_len + iv_len);
2109 
2110 	return 0;
2111 
2112 out_unmap:
2113 	iwl_pcie_tfd_unmap(trans, out_meta, &txq->tfds[q->write_ptr]);
2114 	return ret;
2115 }
2116 #else /* CONFIG_INET */
2117 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2118 				   struct iwl_txq *txq, u8 hdr_len,
2119 				   struct iwl_cmd_meta *out_meta,
2120 				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2121 {
2122 	/* No A-MSDU without CONFIG_INET */
2123 	WARN_ON(1);
2124 
2125 	return -1;
2126 }
2127 #endif /* CONFIG_INET */
2128 
2129 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2130 		      struct iwl_device_cmd *dev_cmd, int txq_id)
2131 {
2132 	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2133 	struct ieee80211_hdr *hdr;
2134 	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2135 	struct iwl_cmd_meta *out_meta;
2136 	struct iwl_txq *txq;
2137 	struct iwl_queue *q;
2138 	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2139 	void *tb1_addr;
2140 	u16 len, tb1_len;
2141 	bool wait_write_ptr;
2142 	__le16 fc;
2143 	u8 hdr_len;
2144 	u16 wifi_seq;
2145 
2146 	txq = &trans_pcie->txq[txq_id];
2147 	q = &txq->q;
2148 
2149 	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2150 		      "TX on unused queue %d\n", txq_id))
2151 		return -EINVAL;
2152 
2153 	if (unlikely(trans_pcie->sw_csum_tx &&
2154 		     skb->ip_summed == CHECKSUM_PARTIAL)) {
2155 		int offs = skb_checksum_start_offset(skb);
2156 		int csum_offs = offs + skb->csum_offset;
2157 		__wsum csum;
2158 
2159 		if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2160 			return -1;
2161 
2162 		csum = skb_checksum(skb, offs, skb->len - offs, 0);
2163 		*(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2164 	}
2165 
2166 	if (skb_is_nonlinear(skb) &&
2167 	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS &&
2168 	    __skb_linearize(skb))
2169 		return -ENOMEM;
2170 
2171 	/* mac80211 always puts the full header into the SKB's head,
2172 	 * so there's no need to check if it's readable there
2173 	 */
2174 	hdr = (struct ieee80211_hdr *)skb->data;
2175 	fc = hdr->frame_control;
2176 	hdr_len = ieee80211_hdrlen(fc);
2177 
2178 	spin_lock(&txq->lock);
2179 
2180 	/* In AGG mode, the index in the ring must correspond to the WiFi
2181 	 * sequence number. This is a HW requirements to help the SCD to parse
2182 	 * the BA.
2183 	 * Check here that the packets are in the right place on the ring.
2184 	 */
2185 	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2186 	WARN_ONCE(txq->ampdu &&
2187 		  (wifi_seq & 0xff) != q->write_ptr,
2188 		  "Q: %d WiFi Seq %d tfdNum %d",
2189 		  txq_id, wifi_seq, q->write_ptr);
2190 
2191 	/* Set up driver data for this TFD */
2192 	txq->entries[q->write_ptr].skb = skb;
2193 	txq->entries[q->write_ptr].cmd = dev_cmd;
2194 
2195 	dev_cmd->hdr.sequence =
2196 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2197 			    INDEX_TO_SEQ(q->write_ptr)));
2198 
2199 	tb0_phys = iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr);
2200 	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2201 		       offsetof(struct iwl_tx_cmd, scratch);
2202 
2203 	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2204 	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2205 
2206 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
2207 	out_meta = &txq->entries[q->write_ptr].meta;
2208 	out_meta->flags = 0;
2209 
2210 	/*
2211 	 * The second TB (tb1) points to the remainder of the TX command
2212 	 * and the 802.11 header - dword aligned size
2213 	 * (This calculation modifies the TX command, so do it before the
2214 	 * setup of the first TB)
2215 	 */
2216 	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2217 	      hdr_len - IWL_HCMD_SCRATCHBUF_SIZE;
2218 	tb1_len = ALIGN(len, 4);
2219 
2220 	/* Tell NIC about any 2-byte padding after MAC header */
2221 	if (tb1_len != len)
2222 		tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
2223 
2224 	/* The first TB points to the scratchbuf data - min_copy bytes */
2225 	memcpy(&txq->scratchbufs[q->write_ptr], &dev_cmd->hdr,
2226 	       IWL_HCMD_SCRATCHBUF_SIZE);
2227 	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2228 			       IWL_HCMD_SCRATCHBUF_SIZE, true);
2229 
2230 	/* there must be data left over for TB1 or this code must be changed */
2231 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_HCMD_SCRATCHBUF_SIZE);
2232 
2233 	/* map the data for TB1 */
2234 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_HCMD_SCRATCHBUF_SIZE;
2235 	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2236 	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2237 		goto out_err;
2238 	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2239 
2240 	if (ieee80211_is_data_qos(fc) &&
2241 	    (*ieee80211_get_qos_ctl(hdr) & IEEE80211_QOS_CTL_A_MSDU_PRESENT)) {
2242 		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2243 						     out_meta, dev_cmd,
2244 						     tb1_len)))
2245 			goto out_err;
2246 	} else if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2247 				       out_meta, dev_cmd, tb1_len))) {
2248 		goto out_err;
2249 	}
2250 
2251 	/* Set up entry for this TFD in Tx byte-count array */
2252 	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len));
2253 
2254 	wait_write_ptr = ieee80211_has_morefrags(fc);
2255 
2256 	/* start timer if queue currently empty */
2257 	if (q->read_ptr == q->write_ptr) {
2258 		if (txq->wd_timeout) {
2259 			/*
2260 			 * If the TXQ is active, then set the timer, if not,
2261 			 * set the timer in remainder so that the timer will
2262 			 * be armed with the right value when the station will
2263 			 * wake up.
2264 			 */
2265 			if (!txq->frozen)
2266 				mod_timer(&txq->stuck_timer,
2267 					  jiffies + txq->wd_timeout);
2268 			else
2269 				txq->frozen_expiry_remainder = txq->wd_timeout;
2270 		}
2271 		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", q->id);
2272 		iwl_trans_pcie_ref(trans);
2273 	}
2274 
2275 	/* Tell device the write index *just past* this latest filled TFD */
2276 	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
2277 	if (!wait_write_ptr)
2278 		iwl_pcie_txq_inc_wr_ptr(trans, txq);
2279 
2280 	/*
2281 	 * At this point the frame is "transmitted" successfully
2282 	 * and we will get a TX status notification eventually.
2283 	 */
2284 	if (iwl_queue_space(q) < q->high_mark) {
2285 		if (wait_write_ptr)
2286 			iwl_pcie_txq_inc_wr_ptr(trans, txq);
2287 		else
2288 			iwl_stop_queue(trans, txq);
2289 	}
2290 	spin_unlock(&txq->lock);
2291 	return 0;
2292 out_err:
2293 	spin_unlock(&txq->lock);
2294 	return -1;
2295 }
2296