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 <ilw@linux.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/slab.h>
32#include <linux/sched.h>
33
34#include "iwl-debug.h"
35#include "iwl-csr.h"
36#include "iwl-prph.h"
37#include "iwl-io.h"
38#include "iwl-scd.h"
39#include "iwl-op-mode.h"
40#include "internal.h"
41/* FIXME: need to abstract out TX command (once we know what it looks like) */
42#include "dvm/commands.h"
43
44#define IWL_TX_CRC_SIZE 4
45#define IWL_TX_DELIMITER_SIZE 4
46
47/*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
48 * DMA services
49 *
50 * Theory of operation
51 *
52 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
53 * of buffer descriptors, each of which points to one or more data buffers for
54 * the device to read from or fill.  Driver and device exchange status of each
55 * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
56 * entries in each circular buffer, to protect against confusing empty and full
57 * queue states.
58 *
59 * The device reads or writes the data in the queues via the device's several
60 * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
61 *
62 * For Tx queue, there are low mark and high mark limits. If, after queuing
63 * the packet for Tx, free space become < low mark, Tx queue stopped. When
64 * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
65 * Tx queue resumed.
66 *
67 ***************************************************/
68static int iwl_queue_space(const struct iwl_queue *q)
69{
70	unsigned int max;
71	unsigned int used;
72
73	/*
74	 * To avoid ambiguity between empty and completely full queues, there
75	 * should always be less than TFD_QUEUE_SIZE_MAX elements in the queue.
76	 * If q->n_window is smaller than TFD_QUEUE_SIZE_MAX, there is no need
77	 * to reserve any queue entries for this purpose.
78	 */
79	if (q->n_window < TFD_QUEUE_SIZE_MAX)
80		max = q->n_window;
81	else
82		max = TFD_QUEUE_SIZE_MAX - 1;
83
84	/*
85	 * TFD_QUEUE_SIZE_MAX is a power of 2, so the following is equivalent to
86	 * modulo by TFD_QUEUE_SIZE_MAX and is well defined.
87	 */
88	used = (q->write_ptr - q->read_ptr) & (TFD_QUEUE_SIZE_MAX - 1);
89
90	if (WARN_ON(used > max))
91		return 0;
92
93	return max - used;
94}
95
96/*
97 * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
98 */
99static int iwl_queue_init(struct iwl_queue *q, int slots_num, u32 id)
100{
101	q->n_window = slots_num;
102	q->id = id;
103
104	/* slots_num must be power-of-two size, otherwise
105	 * get_cmd_index is broken. */
106	if (WARN_ON(!is_power_of_2(slots_num)))
107		return -EINVAL;
108
109	q->low_mark = q->n_window / 4;
110	if (q->low_mark < 4)
111		q->low_mark = 4;
112
113	q->high_mark = q->n_window / 8;
114	if (q->high_mark < 2)
115		q->high_mark = 2;
116
117	q->write_ptr = 0;
118	q->read_ptr = 0;
119
120	return 0;
121}
122
123static int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
124				  struct iwl_dma_ptr *ptr, size_t size)
125{
126	if (WARN_ON(ptr->addr))
127		return -EINVAL;
128
129	ptr->addr = dma_alloc_coherent(trans->dev, size,
130				       &ptr->dma, GFP_KERNEL);
131	if (!ptr->addr)
132		return -ENOMEM;
133	ptr->size = size;
134	return 0;
135}
136
137static void iwl_pcie_free_dma_ptr(struct iwl_trans *trans,
138				  struct iwl_dma_ptr *ptr)
139{
140	if (unlikely(!ptr->addr))
141		return;
142
143	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
144	memset(ptr, 0, sizeof(*ptr));
145}
146
147static void iwl_pcie_txq_stuck_timer(unsigned long data)
148{
149	struct iwl_txq *txq = (void *)data;
150	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
151	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
152	u32 scd_sram_addr = trans_pcie->scd_base_addr +
153				SCD_TX_STTS_QUEUE_OFFSET(txq->q.id);
154	u8 buf[16];
155	int i;
156
157	spin_lock(&txq->lock);
158	/* check if triggered erroneously */
159	if (txq->q.read_ptr == txq->q.write_ptr) {
160		spin_unlock(&txq->lock);
161		return;
162	}
163	spin_unlock(&txq->lock);
164
165	IWL_ERR(trans, "Queue %d stuck for %u ms.\n", txq->q.id,
166		jiffies_to_msecs(txq->wd_timeout));
167	IWL_ERR(trans, "Current SW read_ptr %d write_ptr %d\n",
168		txq->q.read_ptr, txq->q.write_ptr);
169
170	iwl_trans_read_mem_bytes(trans, scd_sram_addr, buf, sizeof(buf));
171
172	iwl_print_hex_error(trans, buf, sizeof(buf));
173
174	for (i = 0; i < FH_TCSR_CHNL_NUM; i++)
175		IWL_ERR(trans, "FH TRBs(%d) = 0x%08x\n", i,
176			iwl_read_direct32(trans, FH_TX_TRB_REG(i)));
177
178	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
179		u32 status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(i));
180		u8 fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7;
181		bool active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE));
182		u32 tbl_dw =
183			iwl_trans_read_mem32(trans,
184					     trans_pcie->scd_base_addr +
185					     SCD_TRANS_TBL_OFFSET_QUEUE(i));
186
187		if (i & 0x1)
188			tbl_dw = (tbl_dw & 0xFFFF0000) >> 16;
189		else
190			tbl_dw = tbl_dw & 0x0000FFFF;
191
192		IWL_ERR(trans,
193			"Q %d is %sactive and mapped to fifo %d ra_tid 0x%04x [%d,%d]\n",
194			i, active ? "" : "in", fifo, tbl_dw,
195			iwl_read_prph(trans, SCD_QUEUE_RDPTR(i)) &
196				(TFD_QUEUE_SIZE_MAX - 1),
197			iwl_read_prph(trans, SCD_QUEUE_WRPTR(i)));
198	}
199
200	iwl_force_nmi(trans);
201}
202
203/*
204 * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
205 */
206static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
207					     struct iwl_txq *txq, u16 byte_cnt)
208{
209	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
210	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
211	int write_ptr = txq->q.write_ptr;
212	int txq_id = txq->q.id;
213	u8 sec_ctl = 0;
214	u8 sta_id = 0;
215	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
216	__le16 bc_ent;
217	struct iwl_tx_cmd *tx_cmd =
218		(void *) txq->entries[txq->q.write_ptr].cmd->payload;
219
220	scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
221
222	WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX);
223
224	sta_id = tx_cmd->sta_id;
225	sec_ctl = tx_cmd->sec_ctl;
226
227	switch (sec_ctl & TX_CMD_SEC_MSK) {
228	case TX_CMD_SEC_CCM:
229		len += IEEE80211_CCMP_MIC_LEN;
230		break;
231	case TX_CMD_SEC_TKIP:
232		len += IEEE80211_TKIP_ICV_LEN;
233		break;
234	case TX_CMD_SEC_WEP:
235		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
236		break;
237	}
238
239	if (trans_pcie->bc_table_dword)
240		len = DIV_ROUND_UP(len, 4);
241
242	bc_ent = cpu_to_le16(len | (sta_id << 12));
243
244	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
245
246	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
247		scd_bc_tbl[txq_id].
248			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
249}
250
251static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
252					    struct iwl_txq *txq)
253{
254	struct iwl_trans_pcie *trans_pcie =
255		IWL_TRANS_GET_PCIE_TRANS(trans);
256	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
257	int txq_id = txq->q.id;
258	int read_ptr = txq->q.read_ptr;
259	u8 sta_id = 0;
260	__le16 bc_ent;
261	struct iwl_tx_cmd *tx_cmd =
262		(void *)txq->entries[txq->q.read_ptr].cmd->payload;
263
264	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
265
266	if (txq_id != trans_pcie->cmd_queue)
267		sta_id = tx_cmd->sta_id;
268
269	bc_ent = cpu_to_le16(1 | (sta_id << 12));
270	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
271
272	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
273		scd_bc_tbl[txq_id].
274			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
275}
276
277/*
278 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
279 */
280static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
281				    struct iwl_txq *txq)
282{
283	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
284	u32 reg = 0;
285	int txq_id = txq->q.id;
286
287	lockdep_assert_held(&txq->lock);
288
289	/*
290	 * explicitly wake up the NIC if:
291	 * 1. shadow registers aren't enabled
292	 * 2. NIC is woken up for CMD regardless of shadow outside this function
293	 * 3. there is a chance that the NIC is asleep
294	 */
295	if (!trans->cfg->base_params->shadow_reg_enable &&
296	    txq_id != trans_pcie->cmd_queue &&
297	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
298		/*
299		 * wake up nic if it's powered down ...
300		 * uCode will wake up, and interrupt us again, so next
301		 * time we'll skip this part.
302		 */
303		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
304
305		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
306			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
307				       txq_id, reg);
308			iwl_set_bit(trans, CSR_GP_CNTRL,
309				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
310			txq->need_update = true;
311			return;
312		}
313	}
314
315	/*
316	 * if not in power-save mode, uCode will never sleep when we're
317	 * trying to tx (during RFKILL, we're not trying to tx).
318	 */
319	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->q.write_ptr);
320	iwl_write32(trans, HBUS_TARG_WRPTR, txq->q.write_ptr | (txq_id << 8));
321}
322
323void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
324{
325	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
326	int i;
327
328	for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
329		struct iwl_txq *txq = &trans_pcie->txq[i];
330
331		spin_lock_bh(&txq->lock);
332		if (trans_pcie->txq[i].need_update) {
333			iwl_pcie_txq_inc_wr_ptr(trans, txq);
334			trans_pcie->txq[i].need_update = false;
335		}
336		spin_unlock_bh(&txq->lock);
337	}
338}
339
340static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx)
341{
342	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
343
344	dma_addr_t addr = get_unaligned_le32(&tb->lo);
345	if (sizeof(dma_addr_t) > sizeof(u32))
346		addr |=
347		((dma_addr_t)(le16_to_cpu(tb->hi_n_len) & 0xF) << 16) << 16;
348
349	return addr;
350}
351
352static inline void iwl_pcie_tfd_set_tb(struct iwl_tfd *tfd, u8 idx,
353				       dma_addr_t addr, u16 len)
354{
355	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
356	u16 hi_n_len = len << 4;
357
358	put_unaligned_le32(addr, &tb->lo);
359	if (sizeof(dma_addr_t) > sizeof(u32))
360		hi_n_len |= ((addr >> 16) >> 16) & 0xF;
361
362	tb->hi_n_len = cpu_to_le16(hi_n_len);
363
364	tfd->num_tbs = idx + 1;
365}
366
367static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_tfd *tfd)
368{
369	return tfd->num_tbs & 0x1f;
370}
371
372static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
373			       struct iwl_cmd_meta *meta,
374			       struct iwl_tfd *tfd)
375{
376	int i;
377	int num_tbs;
378
379	/* Sanity check on number of chunks */
380	num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
381
382	if (num_tbs >= IWL_NUM_OF_TBS) {
383		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
384		/* @todo issue fatal error, it is quite serious situation */
385		return;
386	}
387
388	/* first TB is never freed - it's the scratchbuf data */
389
390	for (i = 1; i < num_tbs; i++)
391		dma_unmap_single(trans->dev, iwl_pcie_tfd_tb_get_addr(tfd, i),
392				 iwl_pcie_tfd_tb_get_len(tfd, i),
393				 DMA_TO_DEVICE);
394
395	tfd->num_tbs = 0;
396}
397
398/*
399 * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
400 * @trans - transport private data
401 * @txq - tx queue
402 * @dma_dir - the direction of the DMA mapping
403 *
404 * Does NOT advance any TFD circular buffer read/write indexes
405 * Does NOT free the TFD itself (which is within circular buffer)
406 */
407static void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
408{
409	struct iwl_tfd *tfd_tmp = txq->tfds;
410
411	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
412	 * idx is bounded by n_window
413	 */
414	int rd_ptr = txq->q.read_ptr;
415	int idx = get_cmd_index(&txq->q, rd_ptr);
416
417	lockdep_assert_held(&txq->lock);
418
419	/* We have only q->n_window txq->entries, but we use
420	 * TFD_QUEUE_SIZE_MAX tfds
421	 */
422	iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, &tfd_tmp[rd_ptr]);
423
424	/* free SKB */
425	if (txq->entries) {
426		struct sk_buff *skb;
427
428		skb = txq->entries[idx].skb;
429
430		/* Can be called from irqs-disabled context
431		 * If skb is not NULL, it means that the whole queue is being
432		 * freed and that the queue is not empty - free the skb
433		 */
434		if (skb) {
435			iwl_op_mode_free_skb(trans->op_mode, skb);
436			txq->entries[idx].skb = NULL;
437		}
438	}
439}
440
441static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
442				  dma_addr_t addr, u16 len, bool reset)
443{
444	struct iwl_queue *q;
445	struct iwl_tfd *tfd, *tfd_tmp;
446	u32 num_tbs;
447
448	q = &txq->q;
449	tfd_tmp = txq->tfds;
450	tfd = &tfd_tmp[q->write_ptr];
451
452	if (reset)
453		memset(tfd, 0, sizeof(*tfd));
454
455	num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
456
457	/* Each TFD can point to a maximum 20 Tx buffers */
458	if (num_tbs >= IWL_NUM_OF_TBS) {
459		IWL_ERR(trans, "Error can not send more than %d chunks\n",
460			IWL_NUM_OF_TBS);
461		return -EINVAL;
462	}
463
464	if (WARN(addr & ~IWL_TX_DMA_MASK,
465		 "Unaligned address = %llx\n", (unsigned long long)addr))
466		return -EINVAL;
467
468	iwl_pcie_tfd_set_tb(tfd, num_tbs, addr, len);
469
470	return 0;
471}
472
473static int iwl_pcie_txq_alloc(struct iwl_trans *trans,
474			       struct iwl_txq *txq, int slots_num,
475			       u32 txq_id)
476{
477	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
478	size_t tfd_sz = sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX;
479	size_t scratchbuf_sz;
480	int i;
481
482	if (WARN_ON(txq->entries || txq->tfds))
483		return -EINVAL;
484
485	setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer,
486		    (unsigned long)txq);
487	txq->trans_pcie = trans_pcie;
488
489	txq->q.n_window = slots_num;
490
491	txq->entries = kcalloc(slots_num,
492			       sizeof(struct iwl_pcie_txq_entry),
493			       GFP_KERNEL);
494
495	if (!txq->entries)
496		goto error;
497
498	if (txq_id == trans_pcie->cmd_queue)
499		for (i = 0; i < slots_num; i++) {
500			txq->entries[i].cmd =
501				kmalloc(sizeof(struct iwl_device_cmd),
502					GFP_KERNEL);
503			if (!txq->entries[i].cmd)
504				goto error;
505		}
506
507	/* Circular buffer of transmit frame descriptors (TFDs),
508	 * shared with device */
509	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
510				       &txq->q.dma_addr, GFP_KERNEL);
511	if (!txq->tfds)
512		goto error;
513
514	BUILD_BUG_ON(IWL_HCMD_SCRATCHBUF_SIZE != sizeof(*txq->scratchbufs));
515	BUILD_BUG_ON(offsetof(struct iwl_pcie_txq_scratch_buf, scratch) !=
516			sizeof(struct iwl_cmd_header) +
517			offsetof(struct iwl_tx_cmd, scratch));
518
519	scratchbuf_sz = sizeof(*txq->scratchbufs) * slots_num;
520
521	txq->scratchbufs = dma_alloc_coherent(trans->dev, scratchbuf_sz,
522					      &txq->scratchbufs_dma,
523					      GFP_KERNEL);
524	if (!txq->scratchbufs)
525		goto err_free_tfds;
526
527	txq->q.id = txq_id;
528
529	return 0;
530err_free_tfds:
531	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->q.dma_addr);
532error:
533	if (txq->entries && txq_id == trans_pcie->cmd_queue)
534		for (i = 0; i < slots_num; i++)
535			kfree(txq->entries[i].cmd);
536	kfree(txq->entries);
537	txq->entries = NULL;
538
539	return -ENOMEM;
540
541}
542
543static int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
544			      int slots_num, u32 txq_id)
545{
546	int ret;
547
548	txq->need_update = false;
549
550	/* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise
551	 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
552	BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1));
553
554	/* Initialize queue's high/low-water marks, and head/tail indexes */
555	ret = iwl_queue_init(&txq->q, slots_num, txq_id);
556	if (ret)
557		return ret;
558
559	spin_lock_init(&txq->lock);
560
561	/*
562	 * Tell nic where to find circular buffer of Tx Frame Descriptors for
563	 * given Tx queue, and enable the DMA channel used for that queue.
564	 * Circular buffer (TFD queue in DRAM) physical base address */
565	iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id),
566			   txq->q.dma_addr >> 8);
567
568	return 0;
569}
570
571/*
572 * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
573 */
574static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
575{
576	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
577	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
578	struct iwl_queue *q = &txq->q;
579
580	spin_lock_bh(&txq->lock);
581	while (q->write_ptr != q->read_ptr) {
582		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
583				   txq_id, q->read_ptr);
584		iwl_pcie_txq_free_tfd(trans, txq);
585		q->read_ptr = iwl_queue_inc_wrap(q->read_ptr);
586	}
587	txq->active = false;
588	spin_unlock_bh(&txq->lock);
589
590	/* just in case - this queue may have been stopped */
591	iwl_wake_queue(trans, txq);
592}
593
594/*
595 * iwl_pcie_txq_free - Deallocate DMA queue.
596 * @txq: Transmit queue to deallocate.
597 *
598 * Empty queue by removing and destroying all BD's.
599 * Free all buffers.
600 * 0-fill, but do not free "txq" descriptor structure.
601 */
602static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
603{
604	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
605	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
606	struct device *dev = trans->dev;
607	int i;
608
609	if (WARN_ON(!txq))
610		return;
611
612	iwl_pcie_txq_unmap(trans, txq_id);
613
614	/* De-alloc array of command/tx buffers */
615	if (txq_id == trans_pcie->cmd_queue)
616		for (i = 0; i < txq->q.n_window; i++) {
617			kzfree(txq->entries[i].cmd);
618			kzfree(txq->entries[i].free_buf);
619		}
620
621	/* De-alloc circular buffer of TFDs */
622	if (txq->tfds) {
623		dma_free_coherent(dev,
624				  sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX,
625				  txq->tfds, txq->q.dma_addr);
626		txq->q.dma_addr = 0;
627		txq->tfds = NULL;
628
629		dma_free_coherent(dev,
630				  sizeof(*txq->scratchbufs) * txq->q.n_window,
631				  txq->scratchbufs, txq->scratchbufs_dma);
632	}
633
634	kfree(txq->entries);
635	txq->entries = NULL;
636
637	del_timer_sync(&txq->stuck_timer);
638
639	/* 0-fill queue descriptor structure */
640	memset(txq, 0, sizeof(*txq));
641}
642
643void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
644{
645	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
646	int nq = trans->cfg->base_params->num_of_queues;
647	int chan;
648	u32 reg_val;
649	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
650				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
651
652	/* make sure all queue are not stopped/used */
653	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
654	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
655
656	trans_pcie->scd_base_addr =
657		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
658
659	WARN_ON(scd_base_addr != 0 &&
660		scd_base_addr != trans_pcie->scd_base_addr);
661
662	/* reset context data, TX status and translation data */
663	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
664				   SCD_CONTEXT_MEM_LOWER_BOUND,
665			    NULL, clear_dwords);
666
667	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
668		       trans_pcie->scd_bc_tbls.dma >> 10);
669
670	/* The chain extension of the SCD doesn't work well. This feature is
671	 * enabled by default by the HW, so we need to disable it manually.
672	 */
673	if (trans->cfg->base_params->scd_chain_ext_wa)
674		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
675
676	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
677				trans_pcie->cmd_fifo,
678				trans_pcie->cmd_q_wdg_timeout);
679
680	/* Activate all Tx DMA/FIFO channels */
681	iwl_scd_activate_fifos(trans);
682
683	/* Enable DMA channel */
684	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
685		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
686				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
687				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
688
689	/* Update FH chicken bits */
690	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
691	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
692			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
693
694	/* Enable L1-Active */
695	if (trans->cfg->device_family != IWL_DEVICE_FAMILY_8000)
696		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
697				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
698}
699
700void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
701{
702	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
703	int txq_id;
704
705	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
706	     txq_id++) {
707		struct iwl_txq *txq = &trans_pcie->txq[txq_id];
708
709		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id),
710				   txq->q.dma_addr >> 8);
711		iwl_pcie_txq_unmap(trans, txq_id);
712		txq->q.read_ptr = 0;
713		txq->q.write_ptr = 0;
714	}
715
716	/* Tell NIC where to find the "keep warm" buffer */
717	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
718			   trans_pcie->kw.dma >> 4);
719
720	/*
721	 * Send 0 as the scd_base_addr since the device may have be reset
722	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
723	 * contain garbage.
724	 */
725	iwl_pcie_tx_start(trans, 0);
726}
727
728static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
729{
730	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
731	unsigned long flags;
732	int ch, ret;
733	u32 mask = 0;
734
735	spin_lock(&trans_pcie->irq_lock);
736
737	if (!iwl_trans_grab_nic_access(trans, false, &flags))
738		goto out;
739
740	/* Stop each Tx DMA channel */
741	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
742		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
743		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
744	}
745
746	/* Wait for DMA channels to be idle */
747	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
748	if (ret < 0)
749		IWL_ERR(trans,
750			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
751			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
752
753	iwl_trans_release_nic_access(trans, &flags);
754
755out:
756	spin_unlock(&trans_pcie->irq_lock);
757}
758
759/*
760 * iwl_pcie_tx_stop - Stop all Tx DMA channels
761 */
762int iwl_pcie_tx_stop(struct iwl_trans *trans)
763{
764	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
765	int txq_id;
766
767	/* Turn off all Tx DMA fifos */
768	iwl_scd_deactivate_fifos(trans);
769
770	/* Turn off all Tx DMA channels */
771	iwl_pcie_tx_stop_fh(trans);
772
773	/*
774	 * This function can be called before the op_mode disabled the
775	 * queues. This happens when we have an rfkill interrupt.
776	 * Since we stop Tx altogether - mark the queues as stopped.
777	 */
778	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
779	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
780
781	/* This can happen: start_hw, stop_device */
782	if (!trans_pcie->txq)
783		return 0;
784
785	/* Unmap DMA from host system and free skb's */
786	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
787	     txq_id++)
788		iwl_pcie_txq_unmap(trans, txq_id);
789
790	return 0;
791}
792
793/*
794 * iwl_trans_tx_free - Free TXQ Context
795 *
796 * Destroy all TX DMA queues and structures
797 */
798void iwl_pcie_tx_free(struct iwl_trans *trans)
799{
800	int txq_id;
801	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
802
803	/* Tx queues */
804	if (trans_pcie->txq) {
805		for (txq_id = 0;
806		     txq_id < trans->cfg->base_params->num_of_queues; txq_id++)
807			iwl_pcie_txq_free(trans, txq_id);
808	}
809
810	kfree(trans_pcie->txq);
811	trans_pcie->txq = NULL;
812
813	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
814
815	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
816}
817
818/*
819 * iwl_pcie_tx_alloc - allocate TX context
820 * Allocate all Tx DMA structures and initialize them
821 */
822static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
823{
824	int ret;
825	int txq_id, slots_num;
826	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
827
828	u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
829			sizeof(struct iwlagn_scd_bc_tbl);
830
831	/*It is not allowed to alloc twice, so warn when this happens.
832	 * We cannot rely on the previous allocation, so free and fail */
833	if (WARN_ON(trans_pcie->txq)) {
834		ret = -EINVAL;
835		goto error;
836	}
837
838	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
839				   scd_bc_tbls_size);
840	if (ret) {
841		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
842		goto error;
843	}
844
845	/* Alloc keep-warm buffer */
846	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
847	if (ret) {
848		IWL_ERR(trans, "Keep Warm allocation failed\n");
849		goto error;
850	}
851
852	trans_pcie->txq = kcalloc(trans->cfg->base_params->num_of_queues,
853				  sizeof(struct iwl_txq), GFP_KERNEL);
854	if (!trans_pcie->txq) {
855		IWL_ERR(trans, "Not enough memory for txq\n");
856		ret = -ENOMEM;
857		goto error;
858	}
859
860	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
861	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
862	     txq_id++) {
863		slots_num = (txq_id == trans_pcie->cmd_queue) ?
864					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
865		ret = iwl_pcie_txq_alloc(trans, &trans_pcie->txq[txq_id],
866					  slots_num, txq_id);
867		if (ret) {
868			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
869			goto error;
870		}
871	}
872
873	return 0;
874
875error:
876	iwl_pcie_tx_free(trans);
877
878	return ret;
879}
880int iwl_pcie_tx_init(struct iwl_trans *trans)
881{
882	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
883	int ret;
884	int txq_id, slots_num;
885	bool alloc = false;
886
887	if (!trans_pcie->txq) {
888		ret = iwl_pcie_tx_alloc(trans);
889		if (ret)
890			goto error;
891		alloc = true;
892	}
893
894	spin_lock(&trans_pcie->irq_lock);
895
896	/* Turn off all Tx DMA fifos */
897	iwl_scd_deactivate_fifos(trans);
898
899	/* Tell NIC where to find the "keep warm" buffer */
900	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
901			   trans_pcie->kw.dma >> 4);
902
903	spin_unlock(&trans_pcie->irq_lock);
904
905	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
906	for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
907	     txq_id++) {
908		slots_num = (txq_id == trans_pcie->cmd_queue) ?
909					TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
910		ret = iwl_pcie_txq_init(trans, &trans_pcie->txq[txq_id],
911					 slots_num, txq_id);
912		if (ret) {
913			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
914			goto error;
915		}
916	}
917
918	if (trans->cfg->base_params->num_of_queues > 20)
919		iwl_set_bits_prph(trans, SCD_GP_CTRL,
920				  SCD_GP_CTRL_ENABLE_31_QUEUES);
921
922	return 0;
923error:
924	/*Upon error, free only if we allocated something */
925	if (alloc)
926		iwl_pcie_tx_free(trans);
927	return ret;
928}
929
930static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
931{
932	lockdep_assert_held(&txq->lock);
933
934	if (!txq->wd_timeout)
935		return;
936
937	/*
938	 * station is asleep and we send data - that must
939	 * be uAPSD or PS-Poll. Don't rearm the timer.
940	 */
941	if (txq->frozen)
942		return;
943
944	/*
945	 * if empty delete timer, otherwise move timer forward
946	 * since we're making progress on this queue
947	 */
948	if (txq->q.read_ptr == txq->q.write_ptr)
949		del_timer(&txq->stuck_timer);
950	else
951		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
952}
953
954/* Frees buffers until index _not_ inclusive */
955void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
956			    struct sk_buff_head *skbs)
957{
958	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
959	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
960	int tfd_num = ssn & (TFD_QUEUE_SIZE_MAX - 1);
961	struct iwl_queue *q = &txq->q;
962	int last_to_free;
963
964	/* This function is not meant to release cmd queue*/
965	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
966		return;
967
968	spin_lock_bh(&txq->lock);
969
970	if (!txq->active) {
971		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
972				    txq_id, ssn);
973		goto out;
974	}
975
976	if (txq->q.read_ptr == tfd_num)
977		goto out;
978
979	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
980			   txq_id, txq->q.read_ptr, tfd_num, ssn);
981
982	/*Since we free until index _not_ inclusive, the one before index is
983	 * the last we will free. This one must be used */
984	last_to_free = iwl_queue_dec_wrap(tfd_num);
985
986	if (!iwl_queue_used(q, last_to_free)) {
987		IWL_ERR(trans,
988			"%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
989			__func__, txq_id, last_to_free, TFD_QUEUE_SIZE_MAX,
990			q->write_ptr, q->read_ptr);
991		goto out;
992	}
993
994	if (WARN_ON(!skb_queue_empty(skbs)))
995		goto out;
996
997	for (;
998	     q->read_ptr != tfd_num;
999	     q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1000
1001		if (WARN_ON_ONCE(txq->entries[txq->q.read_ptr].skb == NULL))
1002			continue;
1003
1004		__skb_queue_tail(skbs, txq->entries[txq->q.read_ptr].skb);
1005
1006		txq->entries[txq->q.read_ptr].skb = NULL;
1007
1008		iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1009
1010		iwl_pcie_txq_free_tfd(trans, txq);
1011	}
1012
1013	iwl_pcie_txq_progress(txq);
1014
1015	if (iwl_queue_space(&txq->q) > txq->q.low_mark)
1016		iwl_wake_queue(trans, txq);
1017
1018	if (q->read_ptr == q->write_ptr) {
1019		IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", q->id);
1020		iwl_trans_pcie_unref(trans);
1021	}
1022
1023out:
1024	spin_unlock_bh(&txq->lock);
1025}
1026
1027static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1028				      const struct iwl_host_cmd *cmd)
1029{
1030	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1031	int ret;
1032
1033	lockdep_assert_held(&trans_pcie->reg_lock);
1034
1035	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
1036	    !trans_pcie->ref_cmd_in_flight) {
1037		trans_pcie->ref_cmd_in_flight = true;
1038		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
1039		iwl_trans_pcie_ref(trans);
1040	}
1041
1042	/*
1043	 * wake up the NIC to make sure that the firmware will see the host
1044	 * command - we will let the NIC sleep once all the host commands
1045	 * returned. This needs to be done only on NICs that have
1046	 * apmg_wake_up_wa set.
1047	 */
1048	if (trans->cfg->base_params->apmg_wake_up_wa &&
1049	    !trans_pcie->cmd_hold_nic_awake) {
1050		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1051					 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1052		if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000)
1053			udelay(2);
1054
1055		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1056				   CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1057				   (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1058				    CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1059				   15000);
1060		if (ret < 0) {
1061			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1062					CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1063			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1064			return -EIO;
1065		}
1066		trans_pcie->cmd_hold_nic_awake = true;
1067	}
1068
1069	return 0;
1070}
1071
1072static int iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
1073{
1074	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1075
1076	lockdep_assert_held(&trans_pcie->reg_lock);
1077
1078	if (trans_pcie->ref_cmd_in_flight) {
1079		trans_pcie->ref_cmd_in_flight = false;
1080		IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
1081		iwl_trans_pcie_unref(trans);
1082	}
1083
1084	if (trans->cfg->base_params->apmg_wake_up_wa) {
1085		if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
1086			return 0;
1087
1088		trans_pcie->cmd_hold_nic_awake = false;
1089		__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1090					   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1091	}
1092	return 0;
1093}
1094
1095/*
1096 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1097 *
1098 * When FW advances 'R' index, all entries between old and new 'R' index
1099 * need to be reclaimed. As result, some free space forms.  If there is
1100 * enough free space (> low mark), wake the stack that feeds us.
1101 */
1102static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1103{
1104	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1105	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1106	struct iwl_queue *q = &txq->q;
1107	unsigned long flags;
1108	int nfreed = 0;
1109
1110	lockdep_assert_held(&txq->lock);
1111
1112	if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(q, idx))) {
1113		IWL_ERR(trans,
1114			"%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1115			__func__, txq_id, idx, TFD_QUEUE_SIZE_MAX,
1116			q->write_ptr, q->read_ptr);
1117		return;
1118	}
1119
1120	for (idx = iwl_queue_inc_wrap(idx); q->read_ptr != idx;
1121	     q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1122
1123		if (nfreed++ > 0) {
1124			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1125				idx, q->write_ptr, q->read_ptr);
1126			iwl_force_nmi(trans);
1127		}
1128	}
1129
1130	if (q->read_ptr == q->write_ptr) {
1131		spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1132		iwl_pcie_clear_cmd_in_flight(trans);
1133		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1134	}
1135
1136	iwl_pcie_txq_progress(txq);
1137}
1138
1139static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1140				 u16 txq_id)
1141{
1142	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1143	u32 tbl_dw_addr;
1144	u32 tbl_dw;
1145	u16 scd_q2ratid;
1146
1147	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1148
1149	tbl_dw_addr = trans_pcie->scd_base_addr +
1150			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1151
1152	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1153
1154	if (txq_id & 0x1)
1155		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1156	else
1157		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1158
1159	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1160
1161	return 0;
1162}
1163
1164/* Receiver address (actually, Rx station's index into station table),
1165 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1166#define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
1167
1168void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1169			       const struct iwl_trans_txq_scd_cfg *cfg,
1170			       unsigned int wdg_timeout)
1171{
1172	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1173	struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1174	int fifo = -1;
1175
1176	if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1177		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1178
1179	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1180
1181	if (cfg) {
1182		fifo = cfg->fifo;
1183
1184		/* Disable the scheduler prior configuring the cmd queue */
1185		if (txq_id == trans_pcie->cmd_queue &&
1186		    trans_pcie->scd_set_active)
1187			iwl_scd_enable_set_active(trans, 0);
1188
1189		/* Stop this Tx queue before configuring it */
1190		iwl_scd_txq_set_inactive(trans, txq_id);
1191
1192		/* Set this queue as a chain-building queue unless it is CMD */
1193		if (txq_id != trans_pcie->cmd_queue)
1194			iwl_scd_txq_set_chain(trans, txq_id);
1195
1196		if (cfg->aggregate) {
1197			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1198
1199			/* Map receiver-address / traffic-ID to this queue */
1200			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1201
1202			/* enable aggregations for the queue */
1203			iwl_scd_txq_enable_agg(trans, txq_id);
1204			txq->ampdu = true;
1205		} else {
1206			/*
1207			 * disable aggregations for the queue, this will also
1208			 * make the ra_tid mapping configuration irrelevant
1209			 * since it is now a non-AGG queue.
1210			 */
1211			iwl_scd_txq_disable_agg(trans, txq_id);
1212
1213			ssn = txq->q.read_ptr;
1214		}
1215	}
1216
1217	/* Place first TFD at index corresponding to start sequence number.
1218	 * Assumes that ssn_idx is valid (!= 0xFFF) */
1219	txq->q.read_ptr = (ssn & 0xff);
1220	txq->q.write_ptr = (ssn & 0xff);
1221	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1222			   (ssn & 0xff) | (txq_id << 8));
1223
1224	if (cfg) {
1225		u8 frame_limit = cfg->frame_limit;
1226
1227		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1228
1229		/* Set up Tx window size and frame limit for this queue */
1230		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1231				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1232		iwl_trans_write_mem32(trans,
1233			trans_pcie->scd_base_addr +
1234			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1235			((frame_limit << SCD_QUEUE_CTX_REG2_WIN_SIZE_POS) &
1236					SCD_QUEUE_CTX_REG2_WIN_SIZE_MSK) |
1237			((frame_limit << SCD_QUEUE_CTX_REG2_FRAME_LIMIT_POS) &
1238					SCD_QUEUE_CTX_REG2_FRAME_LIMIT_MSK));
1239
1240		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1241		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1242			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1243			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1244			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1245			       SCD_QUEUE_STTS_REG_MSK);
1246
1247		/* enable the scheduler for this queue (only) */
1248		if (txq_id == trans_pcie->cmd_queue &&
1249		    trans_pcie->scd_set_active)
1250			iwl_scd_enable_set_active(trans, BIT(txq_id));
1251
1252		IWL_DEBUG_TX_QUEUES(trans,
1253				    "Activate queue %d on FIFO %d WrPtr: %d\n",
1254				    txq_id, fifo, ssn & 0xff);
1255	} else {
1256		IWL_DEBUG_TX_QUEUES(trans,
1257				    "Activate queue %d WrPtr: %d\n",
1258				    txq_id, ssn & 0xff);
1259	}
1260
1261	txq->active = true;
1262}
1263
1264void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1265				bool configure_scd)
1266{
1267	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1268	u32 stts_addr = trans_pcie->scd_base_addr +
1269			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1270	static const u32 zero_val[4] = {};
1271
1272	trans_pcie->txq[txq_id].frozen_expiry_remainder = 0;
1273	trans_pcie->txq[txq_id].frozen = false;
1274
1275	/*
1276	 * Upon HW Rfkill - we stop the device, and then stop the queues
1277	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1278	 * allow the op_mode to call txq_disable after it already called
1279	 * stop_device.
1280	 */
1281	if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1282		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1283			  "queue %d not used", txq_id);
1284		return;
1285	}
1286
1287	if (configure_scd) {
1288		iwl_scd_txq_set_inactive(trans, txq_id);
1289
1290		iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1291				    ARRAY_SIZE(zero_val));
1292	}
1293
1294	iwl_pcie_txq_unmap(trans, txq_id);
1295	trans_pcie->txq[txq_id].ampdu = false;
1296
1297	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1298}
1299
1300/*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1301
1302/*
1303 * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1304 * @priv: device private data point
1305 * @cmd: a pointer to the ucode command structure
1306 *
1307 * The function returns < 0 values to indicate the operation
1308 * failed. On success, it returns the index (>= 0) of command in the
1309 * command queue.
1310 */
1311static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1312				 struct iwl_host_cmd *cmd)
1313{
1314	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1315	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1316	struct iwl_queue *q = &txq->q;
1317	struct iwl_device_cmd *out_cmd;
1318	struct iwl_cmd_meta *out_meta;
1319	unsigned long flags;
1320	void *dup_buf = NULL;
1321	dma_addr_t phys_addr;
1322	int idx;
1323	u16 copy_size, cmd_size, scratch_size;
1324	bool had_nocopy = false;
1325	int i, ret;
1326	u32 cmd_pos;
1327	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1328	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1329
1330	copy_size = sizeof(out_cmd->hdr);
1331	cmd_size = sizeof(out_cmd->hdr);
1332
1333	/* need one for the header if the first is NOCOPY */
1334	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1335
1336	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1337		cmddata[i] = cmd->data[i];
1338		cmdlen[i] = cmd->len[i];
1339
1340		if (!cmd->len[i])
1341			continue;
1342
1343		/* need at least IWL_HCMD_SCRATCHBUF_SIZE copied */
1344		if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) {
1345			int copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size;
1346
1347			if (copy > cmdlen[i])
1348				copy = cmdlen[i];
1349			cmdlen[i] -= copy;
1350			cmddata[i] += copy;
1351			copy_size += copy;
1352		}
1353
1354		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1355			had_nocopy = true;
1356			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1357				idx = -EINVAL;
1358				goto free_dup_buf;
1359			}
1360		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1361			/*
1362			 * This is also a chunk that isn't copied
1363			 * to the static buffer so set had_nocopy.
1364			 */
1365			had_nocopy = true;
1366
1367			/* only allowed once */
1368			if (WARN_ON(dup_buf)) {
1369				idx = -EINVAL;
1370				goto free_dup_buf;
1371			}
1372
1373			dup_buf = kmemdup(cmddata[i], cmdlen[i],
1374					  GFP_ATOMIC);
1375			if (!dup_buf)
1376				return -ENOMEM;
1377		} else {
1378			/* NOCOPY must not be followed by normal! */
1379			if (WARN_ON(had_nocopy)) {
1380				idx = -EINVAL;
1381				goto free_dup_buf;
1382			}
1383			copy_size += cmdlen[i];
1384		}
1385		cmd_size += cmd->len[i];
1386	}
1387
1388	/*
1389	 * If any of the command structures end up being larger than
1390	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1391	 * allocated into separate TFDs, then we will need to
1392	 * increase the size of the buffers.
1393	 */
1394	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1395		 "Command %s (%#x) is too large (%d bytes)\n",
1396		 get_cmd_string(trans_pcie, cmd->id), cmd->id, copy_size)) {
1397		idx = -EINVAL;
1398		goto free_dup_buf;
1399	}
1400
1401	spin_lock_bh(&txq->lock);
1402
1403	if (iwl_queue_space(q) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1404		spin_unlock_bh(&txq->lock);
1405
1406		IWL_ERR(trans, "No space in command queue\n");
1407		iwl_op_mode_cmd_queue_full(trans->op_mode);
1408		idx = -ENOSPC;
1409		goto free_dup_buf;
1410	}
1411
1412	idx = get_cmd_index(q, q->write_ptr);
1413	out_cmd = txq->entries[idx].cmd;
1414	out_meta = &txq->entries[idx].meta;
1415
1416	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1417	if (cmd->flags & CMD_WANT_SKB)
1418		out_meta->source = cmd;
1419
1420	/* set up the header */
1421
1422	out_cmd->hdr.cmd = cmd->id;
1423	out_cmd->hdr.flags = 0;
1424	out_cmd->hdr.sequence =
1425		cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1426					 INDEX_TO_SEQ(q->write_ptr));
1427
1428	/* and copy the data that needs to be copied */
1429	cmd_pos = offsetof(struct iwl_device_cmd, payload);
1430	copy_size = sizeof(out_cmd->hdr);
1431	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1432		int copy;
1433
1434		if (!cmd->len[i])
1435			continue;
1436
1437		/* copy everything if not nocopy/dup */
1438		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1439					   IWL_HCMD_DFL_DUP))) {
1440			copy = cmd->len[i];
1441
1442			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1443			cmd_pos += copy;
1444			copy_size += copy;
1445			continue;
1446		}
1447
1448		/*
1449		 * Otherwise we need at least IWL_HCMD_SCRATCHBUF_SIZE copied
1450		 * in total (for the scratchbuf handling), but copy up to what
1451		 * we can fit into the payload for debug dump purposes.
1452		 */
1453		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1454
1455		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1456		cmd_pos += copy;
1457
1458		/* However, treat copy_size the proper way, we need it below */
1459		if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) {
1460			copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size;
1461
1462			if (copy > cmd->len[i])
1463				copy = cmd->len[i];
1464			copy_size += copy;
1465		}
1466	}
1467
1468	IWL_DEBUG_HC(trans,
1469		     "Sending command %s (#%x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1470		     get_cmd_string(trans_pcie, out_cmd->hdr.cmd),
1471		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
1472		     cmd_size, q->write_ptr, idx, trans_pcie->cmd_queue);
1473
1474	/* start the TFD with the scratchbuf */
1475	scratch_size = min_t(int, copy_size, IWL_HCMD_SCRATCHBUF_SIZE);
1476	memcpy(&txq->scratchbufs[q->write_ptr], &out_cmd->hdr, scratch_size);
1477	iwl_pcie_txq_build_tfd(trans, txq,
1478			       iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr),
1479			       scratch_size, true);
1480
1481	/* map first command fragment, if any remains */
1482	if (copy_size > scratch_size) {
1483		phys_addr = dma_map_single(trans->dev,
1484					   ((u8 *)&out_cmd->hdr) + scratch_size,
1485					   copy_size - scratch_size,
1486					   DMA_TO_DEVICE);
1487		if (dma_mapping_error(trans->dev, phys_addr)) {
1488			iwl_pcie_tfd_unmap(trans, out_meta,
1489					   &txq->tfds[q->write_ptr]);
1490			idx = -ENOMEM;
1491			goto out;
1492		}
1493
1494		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1495				       copy_size - scratch_size, false);
1496	}
1497
1498	/* map the remaining (adjusted) nocopy/dup fragments */
1499	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1500		const void *data = cmddata[i];
1501
1502		if (!cmdlen[i])
1503			continue;
1504		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1505					   IWL_HCMD_DFL_DUP)))
1506			continue;
1507		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1508			data = dup_buf;
1509		phys_addr = dma_map_single(trans->dev, (void *)data,
1510					   cmdlen[i], DMA_TO_DEVICE);
1511		if (dma_mapping_error(trans->dev, phys_addr)) {
1512			iwl_pcie_tfd_unmap(trans, out_meta,
1513					   &txq->tfds[q->write_ptr]);
1514			idx = -ENOMEM;
1515			goto out;
1516		}
1517
1518		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1519	}
1520
1521	out_meta->flags = cmd->flags;
1522	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1523		kzfree(txq->entries[idx].free_buf);
1524	txq->entries[idx].free_buf = dup_buf;
1525
1526	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr);
1527
1528	/* start timer if queue currently empty */
1529	if (q->read_ptr == q->write_ptr && txq->wd_timeout)
1530		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1531
1532	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1533	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1534	if (ret < 0) {
1535		idx = ret;
1536		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1537		goto out;
1538	}
1539
1540	/* Increment and update queue's write index */
1541	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
1542	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1543
1544	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1545
1546 out:
1547	spin_unlock_bh(&txq->lock);
1548 free_dup_buf:
1549	if (idx < 0)
1550		kfree(dup_buf);
1551	return idx;
1552}
1553
1554/*
1555 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1556 * @rxb: Rx buffer to reclaim
1557 * @handler_status: return value of the handler of the command
1558 *	(put in setup_rx_handlers)
1559 *
1560 * If an Rx buffer has an async callback associated with it the callback
1561 * will be executed.  The attached skb (if present) will only be freed
1562 * if the callback returns 1
1563 */
1564void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1565			    struct iwl_rx_cmd_buffer *rxb, int handler_status)
1566{
1567	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1568	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1569	int txq_id = SEQ_TO_QUEUE(sequence);
1570	int index = SEQ_TO_INDEX(sequence);
1571	int cmd_index;
1572	struct iwl_device_cmd *cmd;
1573	struct iwl_cmd_meta *meta;
1574	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1575	struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1576
1577	/* If a Tx command is being handled and it isn't in the actual
1578	 * command queue then there a command routing bug has been introduced
1579	 * in the queue management code. */
1580	if (WARN(txq_id != trans_pcie->cmd_queue,
1581		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1582		 txq_id, trans_pcie->cmd_queue, sequence,
1583		 trans_pcie->txq[trans_pcie->cmd_queue].q.read_ptr,
1584		 trans_pcie->txq[trans_pcie->cmd_queue].q.write_ptr)) {
1585		iwl_print_hex_error(trans, pkt, 32);
1586		return;
1587	}
1588
1589	spin_lock_bh(&txq->lock);
1590
1591	cmd_index = get_cmd_index(&txq->q, index);
1592	cmd = txq->entries[cmd_index].cmd;
1593	meta = &txq->entries[cmd_index].meta;
1594
1595	iwl_pcie_tfd_unmap(trans, meta, &txq->tfds[index]);
1596
1597	/* Input error checking is done when commands are added to queue. */
1598	if (meta->flags & CMD_WANT_SKB) {
1599		struct page *p = rxb_steal_page(rxb);
1600
1601		meta->source->resp_pkt = pkt;
1602		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1603		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1604		meta->source->handler_status = handler_status;
1605	}
1606
1607	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1608
1609	if (!(meta->flags & CMD_ASYNC)) {
1610		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1611			IWL_WARN(trans,
1612				 "HCMD_ACTIVE already clear for command %s\n",
1613				 get_cmd_string(trans_pcie, cmd->hdr.cmd));
1614		}
1615		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1616		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1617			       get_cmd_string(trans_pcie, cmd->hdr.cmd));
1618		wake_up(&trans_pcie->wait_command_queue);
1619	}
1620
1621	meta->flags = 0;
1622
1623	spin_unlock_bh(&txq->lock);
1624}
1625
1626#define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1627
1628static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1629				    struct iwl_host_cmd *cmd)
1630{
1631	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1632	int ret;
1633
1634	/* An asynchronous command can not expect an SKB to be set. */
1635	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1636		return -EINVAL;
1637
1638	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1639	if (ret < 0) {
1640		IWL_ERR(trans,
1641			"Error sending %s: enqueue_hcmd failed: %d\n",
1642			get_cmd_string(trans_pcie, cmd->id), ret);
1643		return ret;
1644	}
1645	return 0;
1646}
1647
1648static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1649				   struct iwl_host_cmd *cmd)
1650{
1651	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1652	int cmd_idx;
1653	int ret;
1654
1655	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1656		       get_cmd_string(trans_pcie, cmd->id));
1657
1658	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1659				  &trans->status),
1660		 "Command %s: a command is already active!\n",
1661		 get_cmd_string(trans_pcie, cmd->id)))
1662		return -EIO;
1663
1664	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1665		       get_cmd_string(trans_pcie, cmd->id));
1666
1667	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1668	if (cmd_idx < 0) {
1669		ret = cmd_idx;
1670		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1671		IWL_ERR(trans,
1672			"Error sending %s: enqueue_hcmd failed: %d\n",
1673			get_cmd_string(trans_pcie, cmd->id), ret);
1674		return ret;
1675	}
1676
1677	ret = wait_event_timeout(trans_pcie->wait_command_queue,
1678				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1679					   &trans->status),
1680				 HOST_COMPLETE_TIMEOUT);
1681	if (!ret) {
1682		struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1683		struct iwl_queue *q = &txq->q;
1684
1685		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1686			get_cmd_string(trans_pcie, cmd->id),
1687			jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1688
1689		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1690			q->read_ptr, q->write_ptr);
1691
1692		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1693		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1694			       get_cmd_string(trans_pcie, cmd->id));
1695		ret = -ETIMEDOUT;
1696
1697		iwl_force_nmi(trans);
1698		iwl_trans_fw_error(trans);
1699
1700		goto cancel;
1701	}
1702
1703	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1704		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1705			get_cmd_string(trans_pcie, cmd->id));
1706		dump_stack();
1707		ret = -EIO;
1708		goto cancel;
1709	}
1710
1711	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1712	    test_bit(STATUS_RFKILL, &trans->status)) {
1713		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1714		ret = -ERFKILL;
1715		goto cancel;
1716	}
1717
1718	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1719		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1720			get_cmd_string(trans_pcie, cmd->id));
1721		ret = -EIO;
1722		goto cancel;
1723	}
1724
1725	return 0;
1726
1727cancel:
1728	if (cmd->flags & CMD_WANT_SKB) {
1729		/*
1730		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1731		 * TX cmd queue. Otherwise in case the cmd comes
1732		 * in later, it will possibly set an invalid
1733		 * address (cmd->meta.source).
1734		 */
1735		trans_pcie->txq[trans_pcie->cmd_queue].
1736			entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1737	}
1738
1739	if (cmd->resp_pkt) {
1740		iwl_free_resp(cmd);
1741		cmd->resp_pkt = NULL;
1742	}
1743
1744	return ret;
1745}
1746
1747int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1748{
1749	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1750	    test_bit(STATUS_RFKILL, &trans->status)) {
1751		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1752				  cmd->id);
1753		return -ERFKILL;
1754	}
1755
1756	if (cmd->flags & CMD_ASYNC)
1757		return iwl_pcie_send_hcmd_async(trans, cmd);
1758
1759	/* We still can fail on RFKILL that can be asserted while we wait */
1760	return iwl_pcie_send_hcmd_sync(trans, cmd);
1761}
1762
1763int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
1764		      struct iwl_device_cmd *dev_cmd, int txq_id)
1765{
1766	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1767	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1768	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
1769	struct iwl_cmd_meta *out_meta;
1770	struct iwl_txq *txq;
1771	struct iwl_queue *q;
1772	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
1773	void *tb1_addr;
1774	u16 len, tb1_len, tb2_len;
1775	bool wait_write_ptr;
1776	__le16 fc = hdr->frame_control;
1777	u8 hdr_len = ieee80211_hdrlen(fc);
1778	u16 wifi_seq;
1779
1780	txq = &trans_pcie->txq[txq_id];
1781	q = &txq->q;
1782
1783	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
1784		      "TX on unused queue %d\n", txq_id))
1785		return -EINVAL;
1786
1787	spin_lock(&txq->lock);
1788
1789	/* In AGG mode, the index in the ring must correspond to the WiFi
1790	 * sequence number. This is a HW requirements to help the SCD to parse
1791	 * the BA.
1792	 * Check here that the packets are in the right place on the ring.
1793	 */
1794	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
1795	WARN_ONCE(txq->ampdu &&
1796		  (wifi_seq & 0xff) != q->write_ptr,
1797		  "Q: %d WiFi Seq %d tfdNum %d",
1798		  txq_id, wifi_seq, q->write_ptr);
1799
1800	/* Set up driver data for this TFD */
1801	txq->entries[q->write_ptr].skb = skb;
1802	txq->entries[q->write_ptr].cmd = dev_cmd;
1803
1804	dev_cmd->hdr.sequence =
1805		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
1806			    INDEX_TO_SEQ(q->write_ptr)));
1807
1808	tb0_phys = iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr);
1809	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
1810		       offsetof(struct iwl_tx_cmd, scratch);
1811
1812	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
1813	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
1814
1815	/* Set up first empty entry in queue's array of Tx/cmd buffers */
1816	out_meta = &txq->entries[q->write_ptr].meta;
1817
1818	/*
1819	 * The second TB (tb1) points to the remainder of the TX command
1820	 * and the 802.11 header - dword aligned size
1821	 * (This calculation modifies the TX command, so do it before the
1822	 * setup of the first TB)
1823	 */
1824	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
1825	      hdr_len - IWL_HCMD_SCRATCHBUF_SIZE;
1826	tb1_len = ALIGN(len, 4);
1827
1828	/* Tell NIC about any 2-byte padding after MAC header */
1829	if (tb1_len != len)
1830		tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
1831
1832	/* The first TB points to the scratchbuf data - min_copy bytes */
1833	memcpy(&txq->scratchbufs[q->write_ptr], &dev_cmd->hdr,
1834	       IWL_HCMD_SCRATCHBUF_SIZE);
1835	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
1836			       IWL_HCMD_SCRATCHBUF_SIZE, true);
1837
1838	/* there must be data left over for TB1 or this code must be changed */
1839	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_HCMD_SCRATCHBUF_SIZE);
1840
1841	/* map the data for TB1 */
1842	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_HCMD_SCRATCHBUF_SIZE;
1843	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
1844	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
1845		goto out_err;
1846	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
1847
1848	/*
1849	 * Set up TFD's third entry to point directly to remainder
1850	 * of skb, if any (802.11 null frames have no payload).
1851	 */
1852	tb2_len = skb->len - hdr_len;
1853	if (tb2_len > 0) {
1854		dma_addr_t tb2_phys = dma_map_single(trans->dev,
1855						     skb->data + hdr_len,
1856						     tb2_len, DMA_TO_DEVICE);
1857		if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) {
1858			iwl_pcie_tfd_unmap(trans, out_meta,
1859					   &txq->tfds[q->write_ptr]);
1860			goto out_err;
1861		}
1862		iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false);
1863	}
1864
1865	/* Set up entry for this TFD in Tx byte-count array */
1866	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len));
1867
1868	trace_iwlwifi_dev_tx(trans->dev, skb,
1869			     &txq->tfds[txq->q.write_ptr],
1870			     sizeof(struct iwl_tfd),
1871			     &dev_cmd->hdr, IWL_HCMD_SCRATCHBUF_SIZE + tb1_len,
1872			     skb->data + hdr_len, tb2_len);
1873	trace_iwlwifi_dev_tx_data(trans->dev, skb,
1874				  skb->data + hdr_len, tb2_len);
1875
1876	wait_write_ptr = ieee80211_has_morefrags(fc);
1877
1878	/* start timer if queue currently empty */
1879	if (q->read_ptr == q->write_ptr) {
1880		if (txq->wd_timeout)
1881			mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1882		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", q->id);
1883		iwl_trans_pcie_ref(trans);
1884	}
1885
1886	/* Tell device the write index *just past* this latest filled TFD */
1887	q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
1888	if (!wait_write_ptr)
1889		iwl_pcie_txq_inc_wr_ptr(trans, txq);
1890
1891	/*
1892	 * At this point the frame is "transmitted" successfully
1893	 * and we will get a TX status notification eventually.
1894	 */
1895	if (iwl_queue_space(q) < q->high_mark) {
1896		if (wait_write_ptr)
1897			iwl_pcie_txq_inc_wr_ptr(trans, txq);
1898		else
1899			iwl_stop_queue(trans, txq);
1900	}
1901	spin_unlock(&txq->lock);
1902	return 0;
1903out_err:
1904	spin_unlock(&txq->lock);
1905	return -1;
1906}
1907