?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / drivers / net / wireless / intel / iwlwifi / pcie / tx.c
blob18650dccdb58429cef06b02c957e18e0706bee2b
1 /******************************************************************************
2 *
3 * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4 * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
5 * Copyright(c) 2016 Intel Deutschland GmbH
6 *
7 * Portions of this file are derived from the ipw3945 project, as well
8 * as portions of the ieee80211 subsystem header files.
9 *
10 * This program is free software; you can redistribute it and/or modify it
11 * under the terms of version 2 of the GNU General Public License as
12 * published by the Free Software Foundation.
13 *
14 * This program is distributed in the hope that it will be useful, but WITHOUT
15 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 * more details.
18 *
19 * You should have received a copy of the GNU General Public License along with
20 * this program; if not, write to the Free Software Foundation, Inc.,
21 * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
22 *
23 * The full GNU General Public License is included in this distribution in the
24 * file called LICENSE.
25 *
26 * Contact Information:
27 * Intel Linux Wireless <linuxwifi@intel.com>
28 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
29 *
30 *****************************************************************************/
31 #include <linux/etherdevice.h>
32 #include <linux/ieee80211.h>
33 #include <linux/slab.h>
34 #include <linux/sched.h>
35 #include <linux/pm_runtime.h>
36 #include <net/ip6_checksum.h>
37 #include <net/tso.h>
38
39 #include "iwl-debug.h"
40 #include "iwl-csr.h"
41 #include "iwl-prph.h"
42 #include "iwl-io.h"
43 #include "iwl-scd.h"
44 #include "iwl-op-mode.h"
45 #include "internal.h"
46 /* FIXME: need to abstract out TX command (once we know what it looks like) */
47 #include "dvm/commands.h"
48
49 #define IWL_TX_CRC_SIZE 4
50 #define IWL_TX_DELIMITER_SIZE 4
51
52 /*************** DMA-QUEUE-GENERAL-FUNCTIONS *****
53 * DMA services
54 *
55 * Theory of operation
56 *
57 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
58 * of buffer descriptors, each of which points to one or more data buffers for
59 * the device to read from or fill. Driver and device exchange status of each
60 * queue via "read" and "write" pointers. Driver keeps minimum of 2 empty
61 * entries in each circular buffer, to protect against confusing empty and full
62 * queue states.
63 *
64 * The device reads or writes the data in the queues via the device's several
65 * DMA/FIFO channels. Each queue is mapped to a single DMA channel.
66 *
67 * For Tx queue, there are low mark and high mark limits. If, after queuing
68 * the packet for Tx, free space become < low mark, Tx queue stopped. When
69 * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
70 * Tx queue resumed.
71 *
72 ***************************************************/
73
74 static int iwl_queue_space(const struct iwl_queue *q)
75 {
76 unsigned int max;
77 unsigned int used;
78
79 /*
80 * To avoid ambiguity between empty and completely full queues, there
81 * should always be less than TFD_QUEUE_SIZE_MAX elements in the queue.
82 * If q->n_window is smaller than TFD_QUEUE_SIZE_MAX, there is no need
83 * to reserve any queue entries for this purpose.
84 */
85 if (q->n_window < TFD_QUEUE_SIZE_MAX)
86 max = q->n_window;
87 else
88 max = TFD_QUEUE_SIZE_MAX - 1;
89
90 /*
91 * TFD_QUEUE_SIZE_MAX is a power of 2, so the following is equivalent to
92 * modulo by TFD_QUEUE_SIZE_MAX and is well defined.
93 */
94 used = (q->write_ptr - q->read_ptr) & (TFD_QUEUE_SIZE_MAX - 1);
95
96 if (WARN_ON(used > max))
97 return 0;
98
99 return max - used;
100 }
101
102 /*
103 * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
104 */
105 static int iwl_queue_init(struct iwl_queue *q, int slots_num, u32 id)
106 {
107 q->n_window = slots_num;
108 q->id = id;
109
110 /* slots_num must be power-of-two size, otherwise
111 * get_cmd_index is broken. */
112 if (WARN_ON(!is_power_of_2(slots_num)))
113 return -EINVAL;
114
115 q->low_mark = q->n_window / 4;
116 if (q->low_mark < 4)
117 q->low_mark = 4;
118
119 q->high_mark = q->n_window / 8;
120 if (q->high_mark < 2)
121 q->high_mark = 2;
122
123 q->write_ptr = 0;
124 q->read_ptr = 0;
125
126 return 0;
127 }
128
129 static int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
130 struct iwl_dma_ptr *ptr, size_t size)
131 {
132 if (WARN_ON(ptr->addr))
133 return -EINVAL;
134
135 ptr->addr = dma_alloc_coherent(trans->dev, size,
136 &ptr->dma, GFP_KERNEL);
137 if (!ptr->addr)
138 return -ENOMEM;
139 ptr->size = size;
140 return 0;
141 }
142
143 static void iwl_pcie_free_dma_ptr(struct iwl_trans *trans,
144 struct iwl_dma_ptr *ptr)
145 {
146 if (unlikely(!ptr->addr))
147 return;
148
149 dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
150 memset(ptr, 0, sizeof(*ptr));
151 }
152
153 static void iwl_pcie_txq_stuck_timer(unsigned long data)
154 {
155 struct iwl_txq *txq = (void *)data;
156 struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
157 struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
158
159 spin_lock(&txq->lock);
160 /* check if triggered erroneously */
161 if (txq->q.read_ptr == txq->q.write_ptr) {
162 spin_unlock(&txq->lock);
163 return;
164 }
165 spin_unlock(&txq->lock);
166
167 IWL_ERR(trans, "Queue %d stuck for %u ms.\n", txq->q.id,
168 jiffies_to_msecs(txq->wd_timeout));
169
170 iwl_trans_pcie_log_scd_error(trans, txq);
171
172 iwl_force_nmi(trans);
173 }
174
175 /*
176 * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
177 */
178 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
179 struct iwl_txq *txq, u16 byte_cnt)
180 {
181 struct iwlagn_scd_bc_tbl *scd_bc_tbl;
182 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
183 int write_ptr = txq->q.write_ptr;
184 int txq_id = txq->q.id;
185 u8 sec_ctl = 0;
186 u8 sta_id = 0;
187 u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
188 __le16 bc_ent;
189 struct iwl_tx_cmd *tx_cmd =
190 (void *) txq->entries[txq->q.write_ptr].cmd->payload;
191
192 scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
193
194 sta_id = tx_cmd->sta_id;
195 sec_ctl = tx_cmd->sec_ctl;
196
197 switch (sec_ctl & TX_CMD_SEC_MSK) {
198 case TX_CMD_SEC_CCM:
199 len += IEEE80211_CCMP_MIC_LEN;
200 break;
201 case TX_CMD_SEC_TKIP:
202 len += IEEE80211_TKIP_ICV_LEN;
203 break;
204 case TX_CMD_SEC_WEP:
205 len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
206 break;
207 }
208
209 if (trans_pcie->bc_table_dword)
210 len = DIV_ROUND_UP(len, 4);
211
212 if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
213 return;
214
215 bc_ent = cpu_to_le16(len | (sta_id << 12));
216
217 scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
218
219 if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
220 scd_bc_tbl[txq_id].
221 tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
222 }
223
224 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
225 struct iwl_txq *txq)
226 {
227 struct iwl_trans_pcie *trans_pcie =
228 IWL_TRANS_GET_PCIE_TRANS(trans);
229 struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
230 int txq_id = txq->q.id;
231 int read_ptr = txq->q.read_ptr;
232 u8 sta_id = 0;
233 __le16 bc_ent;
234 struct iwl_tx_cmd *tx_cmd =
235 (void *)txq->entries[txq->q.read_ptr].cmd->payload;
236
237 WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
238
239 if (txq_id != trans_pcie->cmd_queue)
240 sta_id = tx_cmd->sta_id;
241
242 bc_ent = cpu_to_le16(1 | (sta_id << 12));
243 scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
244
245 if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
246 scd_bc_tbl[txq_id].
247 tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
248 }
249
250 /*
251 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
252 */
253 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
254 struct iwl_txq *txq)
255 {
256 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
257 u32 reg = 0;
258 int txq_id = txq->q.id;
259
260 lockdep_assert_held(&txq->lock);
261
262 /*
263 * explicitly wake up the NIC if:
264 * 1. shadow registers aren't enabled
265 * 2. NIC is woken up for CMD regardless of shadow outside this function
266 * 3. there is a chance that the NIC is asleep
267 */
268 if (!trans->cfg->base_params->shadow_reg_enable &&
269 txq_id != trans_pcie->cmd_queue &&
270 test_bit(STATUS_TPOWER_PMI, &trans->status)) {
271 /*
272 * wake up nic if it's powered down ...
273 * uCode will wake up, and interrupt us again, so next
274 * time we'll skip this part.
275 */
276 reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
277
278 if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
279 IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
280 txq_id, reg);
281 iwl_set_bit(trans, CSR_GP_CNTRL,
282 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
283 txq->need_update = true;
284 return;
285 }
286 }
287
288 /*
289 * if not in power-save mode, uCode will never sleep when we're
290 * trying to tx (during RFKILL, we're not trying to tx).
291 */
292 IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->q.write_ptr);
293 if (!txq->block)
294 iwl_write32(trans, HBUS_TARG_WRPTR,
295 txq->q.write_ptr | (txq_id << 8));
296 }
297
298 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
299 {
300 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
301 int i;
302
303 for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
304 struct iwl_txq *txq = &trans_pcie->txq[i];
305
306 spin_lock_bh(&txq->lock);
307 if (trans_pcie->txq[i].need_update) {
308 iwl_pcie_txq_inc_wr_ptr(trans, txq);
309 trans_pcie->txq[i].need_update = false;
310 }
311 spin_unlock_bh(&txq->lock);
312 }
313 }
314
315 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx)
316 {
317 struct iwl_tfd_tb *tb = &tfd->tbs[idx];
318
319 dma_addr_t addr = get_unaligned_le32(&tb->lo);
320 if (sizeof(dma_addr_t) > sizeof(u32))
321 addr |=
322 ((dma_addr_t)(le16_to_cpu(tb->hi_n_len) & 0xF) << 16) << 16;
323
324 return addr;
325 }
326
327 static inline void iwl_pcie_tfd_set_tb(struct iwl_tfd *tfd, u8 idx,
328 dma_addr_t addr, u16 len)
329 {
330 struct iwl_tfd_tb *tb = &tfd->tbs[idx];
331 u16 hi_n_len = len << 4;
332
333 put_unaligned_le32(addr, &tb->lo);
334 if (sizeof(dma_addr_t) > sizeof(u32))
335 hi_n_len |= ((addr >> 16) >> 16) & 0xF;
336
337 tb->hi_n_len = cpu_to_le16(hi_n_len);
338
339 tfd->num_tbs = idx + 1;
340 }
341
342 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_tfd *tfd)
343 {
344 return tfd->num_tbs & 0x1f;
345 }
346
347 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
348 struct iwl_cmd_meta *meta,
349 struct iwl_tfd *tfd)
350 {
351 int i;
352 int num_tbs;
353
354 /* Sanity check on number of chunks */
355 num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
356
357 if (num_tbs >= IWL_NUM_OF_TBS) {
358 IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
359 /* @todo issue fatal error, it is quite serious situation */
360 return;
361 }
362
363 /* first TB is never freed - it's the bidirectional DMA data */
364
365 for (i = 1; i < num_tbs; i++) {
366 if (meta->flags & BIT(i + CMD_TB_BITMAP_POS))
367 dma_unmap_page(trans->dev,
368 iwl_pcie_tfd_tb_get_addr(tfd, i),
369 iwl_pcie_tfd_tb_get_len(tfd, i),
370 DMA_TO_DEVICE);
371 else
372 dma_unmap_single(trans->dev,
373 iwl_pcie_tfd_tb_get_addr(tfd, i),
374 iwl_pcie_tfd_tb_get_len(tfd, i),
375 DMA_TO_DEVICE);
376 }
377 tfd->num_tbs = 0;
378 }
379
380 /*
381 * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
382 * @trans - transport private data
383 * @txq - tx queue
384 * @dma_dir - the direction of the DMA mapping
385 *
386 * Does NOT advance any TFD circular buffer read/write indexes
387 * Does NOT free the TFD itself (which is within circular buffer)
388 */
389 static void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
390 {
391 struct iwl_tfd *tfd_tmp = txq->tfds;
392
393 /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
394 * idx is bounded by n_window
395 */
396 int rd_ptr = txq->q.read_ptr;
397 int idx = get_cmd_index(&txq->q, rd_ptr);
398
399 lockdep_assert_held(&txq->lock);
400
401 /* We have only q->n_window txq->entries, but we use
402 * TFD_QUEUE_SIZE_MAX tfds
403 */
404 iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, &tfd_tmp[rd_ptr]);
405
406 /* free SKB */
407 if (txq->entries) {
408 struct sk_buff *skb;
409
410 skb = txq->entries[idx].skb;
411
412 /* Can be called from irqs-disabled context
413 * If skb is not NULL, it means that the whole queue is being
414 * freed and that the queue is not empty - free the skb
415 */
416 if (skb) {
417 iwl_op_mode_free_skb(trans->op_mode, skb);
418 txq->entries[idx].skb = NULL;
419 }
420 }
421 }
422
423 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
424 dma_addr_t addr, u16 len, bool reset)
425 {
426 struct iwl_queue *q;
427 struct iwl_tfd *tfd, *tfd_tmp;
428 u32 num_tbs;
429
430 q = &txq->q;
431 tfd_tmp = txq->tfds;
432 tfd = &tfd_tmp[q->write_ptr];
433
434 if (reset)
435 memset(tfd, 0, sizeof(*tfd));
436
437 num_tbs = iwl_pcie_tfd_get_num_tbs(tfd);
438
439 /* Each TFD can point to a maximum 20 Tx buffers */
440 if (num_tbs >= IWL_NUM_OF_TBS) {
441 IWL_ERR(trans, "Error can not send more than %d chunks\n",
442 IWL_NUM_OF_TBS);
443 return -EINVAL;
444 }
445
446 if (WARN(addr & ~IWL_TX_DMA_MASK,
447 "Unaligned address = %llx\n", (unsigned long long)addr))
448 return -EINVAL;
449
450 iwl_pcie_tfd_set_tb(tfd, num_tbs, addr, len);
451
452 return num_tbs;
453 }
454
455 static int iwl_pcie_txq_alloc(struct iwl_trans *trans,
456 struct iwl_txq *txq, int slots_num,
457 u32 txq_id)
458 {
459 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
460 size_t tfd_sz = sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX;
461 size_t tb0_buf_sz;
462 int i;
463
464 if (WARN_ON(txq->entries || txq->tfds))
465 return -EINVAL;
466
467 setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer,
468 (unsigned long)txq);
469 txq->trans_pcie = trans_pcie;
470
471 txq->q.n_window = slots_num;
472
473 txq->entries = kcalloc(slots_num,
474 sizeof(struct iwl_pcie_txq_entry),
475 GFP_KERNEL);
476
477 if (!txq->entries)
478 goto error;
479
480 if (txq_id == trans_pcie->cmd_queue)
481 for (i = 0; i < slots_num; i++) {
482 txq->entries[i].cmd =
483 kmalloc(sizeof(struct iwl_device_cmd),
484 GFP_KERNEL);
485 if (!txq->entries[i].cmd)
486 goto error;
487 }
488
489 /* Circular buffer of transmit frame descriptors (TFDs),
490 * shared with device */
491 txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
492 &txq->q.dma_addr, GFP_KERNEL);
493 if (!txq->tfds)
494 goto error;
495
496 BUILD_BUG_ON(IWL_FIRST_TB_SIZE_ALIGN != sizeof(*txq->first_tb_bufs));
497
498 tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
499
500 txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
501 &txq->first_tb_dma,
502 GFP_KERNEL);
503 if (!txq->first_tb_bufs)
504 goto err_free_tfds;
505
506 txq->q.id = txq_id;
507
508 return 0;
509 err_free_tfds:
510 dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->q.dma_addr);
511 error:
512 if (txq->entries && txq_id == trans_pcie->cmd_queue)
513 for (i = 0; i < slots_num; i++)
514 kfree(txq->entries[i].cmd);
515 kfree(txq->entries);
516 txq->entries = NULL;
517
518 return -ENOMEM;
519
520 }
521
522 static int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
523 int slots_num, u32 txq_id)
524 {
525 int ret;
526
527 txq->need_update = false;
528
529 /* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise
530 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
531 BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1));
532
533 /* Initialize queue's high/low-water marks, and head/tail indexes */
534 ret = iwl_queue_init(&txq->q, slots_num, txq_id);
535 if (ret)
536 return ret;
537
538 spin_lock_init(&txq->lock);
539 __skb_queue_head_init(&txq->overflow_q);
540
541 /*
542 * Tell nic where to find circular buffer of Tx Frame Descriptors for
543 * given Tx queue, and enable the DMA channel used for that queue.
544 * Circular buffer (TFD queue in DRAM) physical base address */
545 if (trans->cfg->use_tfh)
546 iwl_write_direct64(trans,
547 FH_MEM_CBBC_QUEUE(trans, txq_id),
548 txq->q.dma_addr);
549 else
550 iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
551 txq->q.dma_addr >> 8);
552
553 return 0;
554 }
555
556 static void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
557 struct sk_buff *skb)
558 {
559 struct page **page_ptr;
560
561 page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
562
563 if (*page_ptr) {
564 __free_page(*page_ptr);
565 *page_ptr = NULL;
566 }
567 }
568
569 static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
570 {
571 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
572
573 lockdep_assert_held(&trans_pcie->reg_lock);
574
575 if (trans_pcie->ref_cmd_in_flight) {
576 trans_pcie->ref_cmd_in_flight = false;
577 IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
578 iwl_trans_unref(trans);
579 }
580
581 if (!trans->cfg->base_params->apmg_wake_up_wa)
582 return;
583 if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
584 return;
585
586 trans_pcie->cmd_hold_nic_awake = false;
587 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
588 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
589 }
590
591 /*
592 * iwl_pcie_txq_unmap - Unmap any remaining DMA mappings and free skb's
593 */
594 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
595 {
596 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
597 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
598 struct iwl_queue *q = &txq->q;
599
600 spin_lock_bh(&txq->lock);
601 while (q->write_ptr != q->read_ptr) {
602 IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
603 txq_id, q->read_ptr);
604
605 if (txq_id != trans_pcie->cmd_queue) {
606 struct sk_buff *skb = txq->entries[q->read_ptr].skb;
607
608 if (WARN_ON_ONCE(!skb))
609 continue;
610
611 iwl_pcie_free_tso_page(trans_pcie, skb);
612 }
613 iwl_pcie_txq_free_tfd(trans, txq);
614 q->read_ptr = iwl_queue_inc_wrap(q->read_ptr);
615
616 if (q->read_ptr == q->write_ptr) {
617 unsigned long flags;
618
619 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
620 if (txq_id != trans_pcie->cmd_queue) {
621 IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
622 q->id);
623 iwl_trans_unref(trans);
624 } else {
625 iwl_pcie_clear_cmd_in_flight(trans);
626 }
627 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
628 }
629 }
630 txq->active = false;
631
632 while (!skb_queue_empty(&txq->overflow_q)) {
633 struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
634
635 iwl_op_mode_free_skb(trans->op_mode, skb);
636 }
637
638 spin_unlock_bh(&txq->lock);
639
640 /* just in case - this queue may have been stopped */
641 iwl_wake_queue(trans, txq);
642 }
643
644 /*
645 * iwl_pcie_txq_free - Deallocate DMA queue.
646 * @txq: Transmit queue to deallocate.
647 *
648 * Empty queue by removing and destroying all BD's.
649 * Free all buffers.
650 * 0-fill, but do not free "txq" descriptor structure.
651 */
652 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
653 {
654 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
655 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
656 struct device *dev = trans->dev;
657 int i;
658
659 if (WARN_ON(!txq))
660 return;
661
662 iwl_pcie_txq_unmap(trans, txq_id);
663
664 /* De-alloc array of command/tx buffers */
665 if (txq_id == trans_pcie->cmd_queue)
666 for (i = 0; i < txq->q.n_window; i++) {
667 kzfree(txq->entries[i].cmd);
668 kzfree(txq->entries[i].free_buf);
669 }
670
671 /* De-alloc circular buffer of TFDs */
672 if (txq->tfds) {
673 dma_free_coherent(dev,
674 sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX,
675 txq->tfds, txq->q.dma_addr);
676 txq->q.dma_addr = 0;
677 txq->tfds = NULL;
678
679 dma_free_coherent(dev,
680 sizeof(*txq->first_tb_bufs) * txq->q.n_window,
681 txq->first_tb_bufs, txq->first_tb_dma);
682 }
683
684 kfree(txq->entries);
685 txq->entries = NULL;
686
687 del_timer_sync(&txq->stuck_timer);
688
689 /* 0-fill queue descriptor structure */
690 memset(txq, 0, sizeof(*txq));
691 }
692
693 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
694 {
695 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
696 int nq = trans->cfg->base_params->num_of_queues;
697 int chan;
698 u32 reg_val;
699 int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
700 SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
701
702 /* make sure all queue are not stopped/used */
703 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
704 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
705
706 trans_pcie->scd_base_addr =
707 iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
708
709 WARN_ON(scd_base_addr != 0 &&
710 scd_base_addr != trans_pcie->scd_base_addr);
711
712 /* reset context data, TX status and translation data */
713 iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
714 SCD_CONTEXT_MEM_LOWER_BOUND,
715 NULL, clear_dwords);
716
717 iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
718 trans_pcie->scd_bc_tbls.dma >> 10);
719
720 /* The chain extension of the SCD doesn't work well. This feature is
721 * enabled by default by the HW, so we need to disable it manually.
722 */
723 if (trans->cfg->base_params->scd_chain_ext_wa)
724 iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
725
726 iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
727 trans_pcie->cmd_fifo,
728 trans_pcie->cmd_q_wdg_timeout);
729
730 /* Activate all Tx DMA/FIFO channels */
731 iwl_scd_activate_fifos(trans);
732
733 /* Enable DMA channel */
734 for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
735 iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
736 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
737 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
738
739 /* Update FH chicken bits */
740 reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
741 iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
742 reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
743
744 /* Enable L1-Active */
745 if (trans->cfg->device_family != IWL_DEVICE_FAMILY_8000)
746 iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
747 APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
748 }
749
750 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
751 {
752 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
753 int txq_id;
754
755 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
756 txq_id++) {
757 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
758 if (trans->cfg->use_tfh)
759 iwl_write_direct64(trans,
760 FH_MEM_CBBC_QUEUE(trans, txq_id),
761 txq->q.dma_addr);
762 else
763 iwl_write_direct32(trans,
764 FH_MEM_CBBC_QUEUE(trans, txq_id),
765 txq->q.dma_addr >> 8);
766 iwl_pcie_txq_unmap(trans, txq_id);
767 txq->q.read_ptr = 0;
768 txq->q.write_ptr = 0;
769 }
770
771 /* Tell NIC where to find the "keep warm" buffer */
772 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
773 trans_pcie->kw.dma >> 4);
774
775 /*
776 * Send 0 as the scd_base_addr since the device may have be reset
777 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
778 * contain garbage.
779 */
780 iwl_pcie_tx_start(trans, 0);
781 }
782
783 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
784 {
785 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
786 unsigned long flags;
787 int ch, ret;
788 u32 mask = 0;
789
790 spin_lock(&trans_pcie->irq_lock);
791
792 if (!iwl_trans_grab_nic_access(trans, &flags))
793 goto out;
794
795 /* Stop each Tx DMA channel */
796 for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
797 iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
798 mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
799 }
800
801 /* Wait for DMA channels to be idle */
802 ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
803 if (ret < 0)
804 IWL_ERR(trans,
805 "Failing on timeout while stopping DMA channel %d [0x%08x]\n",
806 ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
807
808 iwl_trans_release_nic_access(trans, &flags);
809
810 out:
811 spin_unlock(&trans_pcie->irq_lock);
812 }
813
814 /*
815 * iwl_pcie_tx_stop - Stop all Tx DMA channels
816 */
817 int iwl_pcie_tx_stop(struct iwl_trans *trans)
818 {
819 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
820 int txq_id;
821
822 /* Turn off all Tx DMA fifos */
823 iwl_scd_deactivate_fifos(trans);
824
825 /* Turn off all Tx DMA channels */
826 iwl_pcie_tx_stop_fh(trans);
827
828 /*
829 * This function can be called before the op_mode disabled the
830 * queues. This happens when we have an rfkill interrupt.
831 * Since we stop Tx altogether - mark the queues as stopped.
832 */
833 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
834 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
835
836 /* This can happen: start_hw, stop_device */
837 if (!trans_pcie->txq)
838 return 0;
839
840 /* Unmap DMA from host system and free skb's */
841 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
842 txq_id++)
843 iwl_pcie_txq_unmap(trans, txq_id);
844
845 return 0;
846 }
847
848 /*
849 * iwl_trans_tx_free - Free TXQ Context
850 *
851 * Destroy all TX DMA queues and structures
852 */
853 void iwl_pcie_tx_free(struct iwl_trans *trans)
854 {
855 int txq_id;
856 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
857
858 /* Tx queues */
859 if (trans_pcie->txq) {
860 for (txq_id = 0;
861 txq_id < trans->cfg->base_params->num_of_queues; txq_id++)
862 iwl_pcie_txq_free(trans, txq_id);
863 }
864
865 kfree(trans_pcie->txq);
866 trans_pcie->txq = NULL;
867
868 iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
869
870 iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
871 }
872
873 /*
874 * iwl_pcie_tx_alloc - allocate TX context
875 * Allocate all Tx DMA structures and initialize them
876 */
877 static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
878 {
879 int ret;
880 int txq_id, slots_num;
881 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
882
883 u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
884 sizeof(struct iwlagn_scd_bc_tbl);
885
886 /*It is not allowed to alloc twice, so warn when this happens.
887 * We cannot rely on the previous allocation, so free and fail */
888 if (WARN_ON(trans_pcie->txq)) {
889 ret = -EINVAL;
890 goto error;
891 }
892
893 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
894 scd_bc_tbls_size);
895 if (ret) {
896 IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
897 goto error;
898 }
899
900 /* Alloc keep-warm buffer */
901 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
902 if (ret) {
903 IWL_ERR(trans, "Keep Warm allocation failed\n");
904 goto error;
905 }
906
907 trans_pcie->txq = kcalloc(trans->cfg->base_params->num_of_queues,
908 sizeof(struct iwl_txq), GFP_KERNEL);
909 if (!trans_pcie->txq) {
910 IWL_ERR(trans, "Not enough memory for txq\n");
911 ret = -ENOMEM;
912 goto error;
913 }
914
915 /* Alloc and init all Tx queues, including the command queue (#4/#9) */
916 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
917 txq_id++) {
918 slots_num = (txq_id == trans_pcie->cmd_queue) ?
919 TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
920 ret = iwl_pcie_txq_alloc(trans, &trans_pcie->txq[txq_id],
921 slots_num, txq_id);
922 if (ret) {
923 IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
924 goto error;
925 }
926 }
927
928 return 0;
929
930 error:
931 iwl_pcie_tx_free(trans);
932
933 return ret;
934 }
935 int iwl_pcie_tx_init(struct iwl_trans *trans)
936 {
937 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
938 int ret;
939 int txq_id, slots_num;
940 bool alloc = false;
941
942 if (!trans_pcie->txq) {
943 ret = iwl_pcie_tx_alloc(trans);
944 if (ret)
945 goto error;
946 alloc = true;
947 }
948
949 spin_lock(&trans_pcie->irq_lock);
950
951 /* Turn off all Tx DMA fifos */
952 iwl_scd_deactivate_fifos(trans);
953
954 /* Tell NIC where to find the "keep warm" buffer */
955 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
956 trans_pcie->kw.dma >> 4);
957
958 spin_unlock(&trans_pcie->irq_lock);
959
960 /* Alloc and init all Tx queues, including the command queue (#4/#9) */
961 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
962 txq_id++) {
963 slots_num = (txq_id == trans_pcie->cmd_queue) ?
964 TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
965 ret = iwl_pcie_txq_init(trans, &trans_pcie->txq[txq_id],
966 slots_num, txq_id);
967 if (ret) {
968 IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
969 goto error;
970 }
971 }
972
973 if (trans->cfg->use_tfh)
974 iwl_write_direct32(trans, TFH_TRANSFER_MODE,
975 TFH_TRANSFER_MAX_PENDING_REQ |
976 TFH_CHUNK_SIZE_128 |
977 TFH_CHUNK_SPLIT_MODE);
978
979 iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
980 if (trans->cfg->base_params->num_of_queues > 20)
981 iwl_set_bits_prph(trans, SCD_GP_CTRL,
982 SCD_GP_CTRL_ENABLE_31_QUEUES);
983
984 return 0;
985 error:
986 /*Upon error, free only if we allocated something */
987 if (alloc)
988 iwl_pcie_tx_free(trans);
989 return ret;
990 }
991
992 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
993 {
994 lockdep_assert_held(&txq->lock);
995
996 if (!txq->wd_timeout)
997 return;
998
999 /*
1000 * station is asleep and we send data - that must
1001 * be uAPSD or PS-Poll. Don't rearm the timer.
1002 */
1003 if (txq->frozen)
1004 return;
1005
1006 /*
1007 * if empty delete timer, otherwise move timer forward
1008 * since we're making progress on this queue
1009 */
1010 if (txq->q.read_ptr == txq->q.write_ptr)
1011 del_timer(&txq->stuck_timer);
1012 else
1013 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1014 }
1015
1016 /* Frees buffers until index _not_ inclusive */
1017 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1018 struct sk_buff_head *skbs)
1019 {
1020 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1021 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1022 int tfd_num = ssn & (TFD_QUEUE_SIZE_MAX - 1);
1023 struct iwl_queue *q = &txq->q;
1024 int last_to_free;
1025
1026 /* This function is not meant to release cmd queue*/
1027 if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1028 return;
1029
1030 spin_lock_bh(&txq->lock);
1031
1032 if (!txq->active) {
1033 IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1034 txq_id, ssn);
1035 goto out;
1036 }
1037
1038 if (txq->q.read_ptr == tfd_num)
1039 goto out;
1040
1041 IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1042 txq_id, txq->q.read_ptr, tfd_num, ssn);
1043
1044 /*Since we free until index _not_ inclusive, the one before index is
1045 * the last we will free. This one must be used */
1046 last_to_free = iwl_queue_dec_wrap(tfd_num);
1047
1048 if (!iwl_queue_used(q, last_to_free)) {
1049 IWL_ERR(trans,
1050 "%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1051 __func__, txq_id, last_to_free, TFD_QUEUE_SIZE_MAX,
1052 q->write_ptr, q->read_ptr);
1053 goto out;
1054 }
1055
1056 if (WARN_ON(!skb_queue_empty(skbs)))
1057 goto out;
1058
1059 for (;
1060 q->read_ptr != tfd_num;
1061 q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1062 struct sk_buff *skb = txq->entries[txq->q.read_ptr].skb;
1063
1064 if (WARN_ON_ONCE(!skb))
1065 continue;
1066
1067 iwl_pcie_free_tso_page(trans_pcie, skb);
1068
1069 __skb_queue_tail(skbs, skb);
1070
1071 txq->entries[txq->q.read_ptr].skb = NULL;
1072
1073 iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1074
1075 iwl_pcie_txq_free_tfd(trans, txq);
1076 }
1077
1078 iwl_pcie_txq_progress(txq);
1079
1080 if (iwl_queue_space(&txq->q) > txq->q.low_mark &&
1081 test_bit(txq_id, trans_pcie->queue_stopped)) {
1082 struct sk_buff_head overflow_skbs;
1083
1084 __skb_queue_head_init(&overflow_skbs);
1085 skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1086
1087 /*
1088 * This is tricky: we are in reclaim path which is non
1089 * re-entrant, so noone will try to take the access the
1090 * txq data from that path. We stopped tx, so we can't
1091 * have tx as well. Bottom line, we can unlock and re-lock
1092 * later.
1093 */
1094 spin_unlock_bh(&txq->lock);
1095
1096 while (!skb_queue_empty(&overflow_skbs)) {
1097 struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1098 struct iwl_device_cmd *dev_cmd_ptr;
1099
1100 dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1101 trans_pcie->dev_cmd_offs);
1102
1103 /*
1104 * Note that we can very well be overflowing again.
1105 * In that case, iwl_queue_space will be small again
1106 * and we won't wake mac80211's queue.
1107 */
1108 iwl_trans_pcie_tx(trans, skb, dev_cmd_ptr, txq_id);
1109 }
1110 spin_lock_bh(&txq->lock);
1111
1112 if (iwl_queue_space(&txq->q) > txq->q.low_mark)
1113 iwl_wake_queue(trans, txq);
1114 }
1115
1116 if (q->read_ptr == q->write_ptr) {
1117 IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", q->id);
1118 iwl_trans_unref(trans);
1119 }
1120
1121 out:
1122 spin_unlock_bh(&txq->lock);
1123 }
1124
1125 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1126 const struct iwl_host_cmd *cmd)
1127 {
1128 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1129 int ret;
1130
1131 lockdep_assert_held(&trans_pcie->reg_lock);
1132
1133 if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
1134 !trans_pcie->ref_cmd_in_flight) {
1135 trans_pcie->ref_cmd_in_flight = true;
1136 IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
1137 iwl_trans_ref(trans);
1138 }
1139
1140 /*
1141 * wake up the NIC to make sure that the firmware will see the host
1142 * command - we will let the NIC sleep once all the host commands
1143 * returned. This needs to be done only on NICs that have
1144 * apmg_wake_up_wa set.
1145 */
1146 if (trans->cfg->base_params->apmg_wake_up_wa &&
1147 !trans_pcie->cmd_hold_nic_awake) {
1148 __iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1149 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1150
1151 ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1152 CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1153 (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1154 CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1155 15000);
1156 if (ret < 0) {
1157 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1158 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1159 IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1160 return -EIO;
1161 }
1162 trans_pcie->cmd_hold_nic_awake = true;
1163 }
1164
1165 return 0;
1166 }
1167
1168 /*
1169 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1170 *
1171 * When FW advances 'R' index, all entries between old and new 'R' index
1172 * need to be reclaimed. As result, some free space forms. If there is
1173 * enough free space (> low mark), wake the stack that feeds us.
1174 */
1175 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1176 {
1177 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1178 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1179 struct iwl_queue *q = &txq->q;
1180 unsigned long flags;
1181 int nfreed = 0;
1182
1183 lockdep_assert_held(&txq->lock);
1184
1185 if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(q, idx))) {
1186 IWL_ERR(trans,
1187 "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1188 __func__, txq_id, idx, TFD_QUEUE_SIZE_MAX,
1189 q->write_ptr, q->read_ptr);
1190 return;
1191 }
1192
1193 for (idx = iwl_queue_inc_wrap(idx); q->read_ptr != idx;
1194 q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1195
1196 if (nfreed++ > 0) {
1197 IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1198 idx, q->write_ptr, q->read_ptr);
1199 iwl_force_nmi(trans);
1200 }
1201 }
1202
1203 if (q->read_ptr == q->write_ptr) {
1204 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1205 iwl_pcie_clear_cmd_in_flight(trans);
1206 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1207 }
1208
1209 iwl_pcie_txq_progress(txq);
1210 }
1211
1212 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1213 u16 txq_id)
1214 {
1215 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1216 u32 tbl_dw_addr;
1217 u32 tbl_dw;
1218 u16 scd_q2ratid;
1219
1220 scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1221
1222 tbl_dw_addr = trans_pcie->scd_base_addr +
1223 SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1224
1225 tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1226
1227 if (txq_id & 0x1)
1228 tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1229 else
1230 tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1231
1232 iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1233
1234 return 0;
1235 }
1236
1237 /* Receiver address (actually, Rx station's index into station table),
1238 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1239 #define BUILD_RAxTID(sta_id, tid) (((sta_id) << 4) + (tid))
1240
1241 void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1242 const struct iwl_trans_txq_scd_cfg *cfg,
1243 unsigned int wdg_timeout)
1244 {
1245 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1246 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1247 int fifo = -1;
1248
1249 if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1250 WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1251
1252 txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1253
1254 if (cfg) {
1255 fifo = cfg->fifo;
1256
1257 /* Disable the scheduler prior configuring the cmd queue */
1258 if (txq_id == trans_pcie->cmd_queue &&
1259 trans_pcie->scd_set_active)
1260 iwl_scd_enable_set_active(trans, 0);
1261
1262 /* Stop this Tx queue before configuring it */
1263 iwl_scd_txq_set_inactive(trans, txq_id);
1264
1265 /* Set this queue as a chain-building queue unless it is CMD */
1266 if (txq_id != trans_pcie->cmd_queue)
1267 iwl_scd_txq_set_chain(trans, txq_id);
1268
1269 if (cfg->aggregate) {
1270 u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1271
1272 /* Map receiver-address / traffic-ID to this queue */
1273 iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1274
1275 /* enable aggregations for the queue */
1276 iwl_scd_txq_enable_agg(trans, txq_id);
1277 txq->ampdu = true;
1278 } else {
1279 /*
1280 * disable aggregations for the queue, this will also
1281 * make the ra_tid mapping configuration irrelevant
1282 * since it is now a non-AGG queue.
1283 */
1284 iwl_scd_txq_disable_agg(trans, txq_id);
1285
1286 ssn = txq->q.read_ptr;
1287 }
1288 }
1289
1290 /* Place first TFD at index corresponding to start sequence number.
1291 * Assumes that ssn_idx is valid (!= 0xFFF) */
1292 txq->q.read_ptr = (ssn & 0xff);
1293 txq->q.write_ptr = (ssn & 0xff);
1294 iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1295 (ssn & 0xff) | (txq_id << 8));
1296
1297 if (cfg) {
1298 u8 frame_limit = cfg->frame_limit;
1299
1300 iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1301
1302 /* Set up Tx window size and frame limit for this queue */
1303 iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1304 SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1305 iwl_trans_write_mem32(trans,
1306 trans_pcie->scd_base_addr +
1307 SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1308 ((frame_limit << SCD_QUEUE_CTX_REG2_WIN_SIZE_POS) &
1309 SCD_QUEUE_CTX_REG2_WIN_SIZE_MSK) |
1310 ((frame_limit << SCD_QUEUE_CTX_REG2_FRAME_LIMIT_POS) &
1311 SCD_QUEUE_CTX_REG2_FRAME_LIMIT_MSK));
1312
1313 /* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1314 iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1315 (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1316 (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1317 (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1318 SCD_QUEUE_STTS_REG_MSK);
1319
1320 /* enable the scheduler for this queue (only) */
1321 if (txq_id == trans_pcie->cmd_queue &&
1322 trans_pcie->scd_set_active)
1323 iwl_scd_enable_set_active(trans, BIT(txq_id));
1324
1325 IWL_DEBUG_TX_QUEUES(trans,
1326 "Activate queue %d on FIFO %d WrPtr: %d\n",
1327 txq_id, fifo, ssn & 0xff);
1328 } else {
1329 IWL_DEBUG_TX_QUEUES(trans,
1330 "Activate queue %d WrPtr: %d\n",
1331 txq_id, ssn & 0xff);
1332 }
1333
1334 txq->active = true;
1335 }
1336
1337 void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
1338 bool shared_mode)
1339 {
1340 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1341 struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1342
1343 txq->ampdu = !shared_mode;
1344 }
1345
1346 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1347 bool configure_scd)
1348 {
1349 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1350 u32 stts_addr = trans_pcie->scd_base_addr +
1351 SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1352 static const u32 zero_val[4] = {};
1353
1354 trans_pcie->txq[txq_id].frozen_expiry_remainder = 0;
1355 trans_pcie->txq[txq_id].frozen = false;
1356
1357 /*
1358 * Upon HW Rfkill - we stop the device, and then stop the queues
1359 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1360 * allow the op_mode to call txq_disable after it already called
1361 * stop_device.
1362 */
1363 if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1364 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1365 "queue %d not used", txq_id);
1366 return;
1367 }
1368
1369 if (configure_scd) {
1370 iwl_scd_txq_set_inactive(trans, txq_id);
1371
1372 iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1373 ARRAY_SIZE(zero_val));
1374 }
1375
1376 iwl_pcie_txq_unmap(trans, txq_id);
1377 trans_pcie->txq[txq_id].ampdu = false;
1378
1379 IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1380 }
1381
1382 /*************** HOST COMMAND QUEUE FUNCTIONS *****/
1383
1384 /*
1385 * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1386 * @priv: device private data point
1387 * @cmd: a pointer to the ucode command structure
1388 *
1389 * The function returns < 0 values to indicate the operation
1390 * failed. On success, it returns the index (>= 0) of command in the
1391 * command queue.
1392 */
1393 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1394 struct iwl_host_cmd *cmd)
1395 {
1396 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1397 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1398 struct iwl_queue *q = &txq->q;
1399 struct iwl_device_cmd *out_cmd;
1400 struct iwl_cmd_meta *out_meta;
1401 unsigned long flags;
1402 void *dup_buf = NULL;
1403 dma_addr_t phys_addr;
1404 int idx;
1405 u16 copy_size, cmd_size, tb0_size;
1406 bool had_nocopy = false;
1407 u8 group_id = iwl_cmd_groupid(cmd->id);
1408 int i, ret;
1409 u32 cmd_pos;
1410 const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1411 u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1412
1413 if (WARN(!trans_pcie->wide_cmd_header &&
1414 group_id > IWL_ALWAYS_LONG_GROUP,
1415 "unsupported wide command %#x\n", cmd->id))
1416 return -EINVAL;
1417
1418 if (group_id != 0) {
1419 copy_size = sizeof(struct iwl_cmd_header_wide);
1420 cmd_size = sizeof(struct iwl_cmd_header_wide);
1421 } else {
1422 copy_size = sizeof(struct iwl_cmd_header);
1423 cmd_size = sizeof(struct iwl_cmd_header);
1424 }
1425
1426 /* need one for the header if the first is NOCOPY */
1427 BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1428
1429 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1430 cmddata[i] = cmd->data[i];
1431 cmdlen[i] = cmd->len[i];
1432
1433 if (!cmd->len[i])
1434 continue;
1435
1436 /* need at least IWL_FIRST_TB_SIZE copied */
1437 if (copy_size < IWL_FIRST_TB_SIZE) {
1438 int copy = IWL_FIRST_TB_SIZE - copy_size;
1439
1440 if (copy > cmdlen[i])
1441 copy = cmdlen[i];
1442 cmdlen[i] -= copy;
1443 cmddata[i] += copy;
1444 copy_size += copy;
1445 }
1446
1447 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1448 had_nocopy = true;
1449 if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1450 idx = -EINVAL;
1451 goto free_dup_buf;
1452 }
1453 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1454 /*
1455 * This is also a chunk that isn't copied
1456 * to the static buffer so set had_nocopy.
1457 */
1458 had_nocopy = true;
1459
1460 /* only allowed once */
1461 if (WARN_ON(dup_buf)) {
1462 idx = -EINVAL;
1463 goto free_dup_buf;
1464 }
1465
1466 dup_buf = kmemdup(cmddata[i], cmdlen[i],
1467 GFP_ATOMIC);
1468 if (!dup_buf)
1469 return -ENOMEM;
1470 } else {
1471 /* NOCOPY must not be followed by normal! */
1472 if (WARN_ON(had_nocopy)) {
1473 idx = -EINVAL;
1474 goto free_dup_buf;
1475 }
1476 copy_size += cmdlen[i];
1477 }
1478 cmd_size += cmd->len[i];
1479 }
1480
1481 /*
1482 * If any of the command structures end up being larger than
1483 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1484 * allocated into separate TFDs, then we will need to
1485 * increase the size of the buffers.
1486 */
1487 if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1488 "Command %s (%#x) is too large (%d bytes)\n",
1489 iwl_get_cmd_string(trans, cmd->id),
1490 cmd->id, copy_size)) {
1491 idx = -EINVAL;
1492 goto free_dup_buf;
1493 }
1494
1495 spin_lock_bh(&txq->lock);
1496
1497 if (iwl_queue_space(q) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1498 spin_unlock_bh(&txq->lock);
1499
1500 IWL_ERR(trans, "No space in command queue\n");
1501 iwl_op_mode_cmd_queue_full(trans->op_mode);
1502 idx = -ENOSPC;
1503 goto free_dup_buf;
1504 }
1505
1506 idx = get_cmd_index(q, q->write_ptr);
1507 out_cmd = txq->entries[idx].cmd;
1508 out_meta = &txq->entries[idx].meta;
1509
1510 memset(out_meta, 0, sizeof(*out_meta)); /* re-initialize to NULL */
1511 if (cmd->flags & CMD_WANT_SKB)
1512 out_meta->source = cmd;
1513
1514 /* set up the header */
1515 if (group_id != 0) {
1516 out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1517 out_cmd->hdr_wide.group_id = group_id;
1518 out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1519 out_cmd->hdr_wide.length =
1520 cpu_to_le16(cmd_size -
1521 sizeof(struct iwl_cmd_header_wide));
1522 out_cmd->hdr_wide.reserved = 0;
1523 out_cmd->hdr_wide.sequence =
1524 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1525 INDEX_TO_SEQ(q->write_ptr));
1526
1527 cmd_pos = sizeof(struct iwl_cmd_header_wide);
1528 copy_size = sizeof(struct iwl_cmd_header_wide);
1529 } else {
1530 out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1531 out_cmd->hdr.sequence =
1532 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1533 INDEX_TO_SEQ(q->write_ptr));
1534 out_cmd->hdr.group_id = 0;
1535
1536 cmd_pos = sizeof(struct iwl_cmd_header);
1537 copy_size = sizeof(struct iwl_cmd_header);
1538 }
1539
1540 /* and copy the data that needs to be copied */
1541 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1542 int copy;
1543
1544 if (!cmd->len[i])
1545 continue;
1546
1547 /* copy everything if not nocopy/dup */
1548 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1549 IWL_HCMD_DFL_DUP))) {
1550 copy = cmd->len[i];
1551
1552 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1553 cmd_pos += copy;
1554 copy_size += copy;
1555 continue;
1556 }
1557
1558 /*
1559 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1560 * in total (for bi-directional DMA), but copy up to what
1561 * we can fit into the payload for debug dump purposes.
1562 */
1563 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1564
1565 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1566 cmd_pos += copy;
1567
1568 /* However, treat copy_size the proper way, we need it below */
1569 if (copy_size < IWL_FIRST_TB_SIZE) {
1570 copy = IWL_FIRST_TB_SIZE - copy_size;
1571
1572 if (copy > cmd->len[i])
1573 copy = cmd->len[i];
1574 copy_size += copy;
1575 }
1576 }
1577
1578 IWL_DEBUG_HC(trans,
1579 "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1580 iwl_get_cmd_string(trans, cmd->id),
1581 group_id, out_cmd->hdr.cmd,
1582 le16_to_cpu(out_cmd->hdr.sequence),
1583 cmd_size, q->write_ptr, idx, trans_pcie->cmd_queue);
1584
1585 /* start the TFD with the minimum copy bytes */
1586 tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1587 memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1588 iwl_pcie_txq_build_tfd(trans, txq,
1589 iwl_pcie_get_first_tb_dma(txq, idx),
1590 tb0_size, true);
1591
1592 /* map first command fragment, if any remains */
1593 if (copy_size > tb0_size) {
1594 phys_addr = dma_map_single(trans->dev,
1595 ((u8 *)&out_cmd->hdr) + tb0_size,
1596 copy_size - tb0_size,
1597 DMA_TO_DEVICE);
1598 if (dma_mapping_error(trans->dev, phys_addr)) {
1599 iwl_pcie_tfd_unmap(trans, out_meta,
1600 &txq->tfds[q->write_ptr]);
1601 idx = -ENOMEM;
1602 goto out;
1603 }
1604
1605 iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1606 copy_size - tb0_size, false);
1607 }
1608
1609 /* map the remaining (adjusted) nocopy/dup fragments */
1610 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1611 const void *data = cmddata[i];
1612
1613 if (!cmdlen[i])
1614 continue;
1615 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1616 IWL_HCMD_DFL_DUP)))
1617 continue;
1618 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1619 data = dup_buf;
1620 phys_addr = dma_map_single(trans->dev, (void *)data,
1621 cmdlen[i], DMA_TO_DEVICE);
1622 if (dma_mapping_error(trans->dev, phys_addr)) {
1623 iwl_pcie_tfd_unmap(trans, out_meta,
1624 &txq->tfds[q->write_ptr]);
1625 idx = -ENOMEM;
1626 goto out;
1627 }
1628
1629 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1630 }
1631
1632 BUILD_BUG_ON(IWL_NUM_OF_TBS + CMD_TB_BITMAP_POS >
1633 sizeof(out_meta->flags) * BITS_PER_BYTE);
1634 out_meta->flags = cmd->flags;
1635 if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1636 kzfree(txq->entries[idx].free_buf);
1637 txq->entries[idx].free_buf = dup_buf;
1638
1639 trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1640
1641 /* start timer if queue currently empty */
1642 if (q->read_ptr == q->write_ptr && txq->wd_timeout)
1643 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1644
1645 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1646 ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1647 if (ret < 0) {
1648 idx = ret;
1649 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1650 goto out;
1651 }
1652
1653 /* Increment and update queue's write index */
1654 q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
1655 iwl_pcie_txq_inc_wr_ptr(trans, txq);
1656
1657 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1658
1659 out:
1660 spin_unlock_bh(&txq->lock);
1661 free_dup_buf:
1662 if (idx < 0)
1663 kfree(dup_buf);
1664 return idx;
1665 }
1666
1667 /*
1668 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1669 * @rxb: Rx buffer to reclaim
1670 */
1671 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1672 struct iwl_rx_cmd_buffer *rxb)
1673 {
1674 struct iwl_rx_packet *pkt = rxb_addr(rxb);
1675 u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1676 u8 group_id = iwl_cmd_groupid(pkt->hdr.group_id);
1677 u32 cmd_id;
1678 int txq_id = SEQ_TO_QUEUE(sequence);
1679 int index = SEQ_TO_INDEX(sequence);
1680 int cmd_index;
1681 struct iwl_device_cmd *cmd;
1682 struct iwl_cmd_meta *meta;
1683 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1684 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1685
1686 /* If a Tx command is being handled and it isn't in the actual
1687 * command queue then there a command routing bug has been introduced
1688 * in the queue management code. */
1689 if (WARN(txq_id != trans_pcie->cmd_queue,
1690 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1691 txq_id, trans_pcie->cmd_queue, sequence,
1692 trans_pcie->txq[trans_pcie->cmd_queue].q.read_ptr,
1693 trans_pcie->txq[trans_pcie->cmd_queue].q.write_ptr)) {
1694 iwl_print_hex_error(trans, pkt, 32);
1695 return;
1696 }
1697
1698 spin_lock_bh(&txq->lock);
1699
1700 cmd_index = get_cmd_index(&txq->q, index);
1701 cmd = txq->entries[cmd_index].cmd;
1702 meta = &txq->entries[cmd_index].meta;
1703 cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1704
1705 iwl_pcie_tfd_unmap(trans, meta, &txq->tfds[index]);
1706
1707 /* Input error checking is done when commands are added to queue. */
1708 if (meta->flags & CMD_WANT_SKB) {
1709 struct page *p = rxb_steal_page(rxb);
1710
1711 meta->source->resp_pkt = pkt;
1712 meta->source->_rx_page_addr = (unsigned long)page_address(p);
1713 meta->source->_rx_page_order = trans_pcie->rx_page_order;
1714 }
1715
1716 if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1717 iwl_op_mode_async_cb(trans->op_mode, cmd);
1718
1719 iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1720
1721 if (!(meta->flags & CMD_ASYNC)) {
1722 if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1723 IWL_WARN(trans,
1724 "HCMD_ACTIVE already clear for command %s\n",
1725 iwl_get_cmd_string(trans, cmd_id));
1726 }
1727 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1728 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1729 iwl_get_cmd_string(trans, cmd_id));
1730 wake_up(&trans_pcie->wait_command_queue);
1731 }
1732
1733 if (meta->flags & CMD_MAKE_TRANS_IDLE) {
1734 IWL_DEBUG_INFO(trans, "complete %s - mark trans as idle\n",
1735 iwl_get_cmd_string(trans, cmd->hdr.cmd));
1736 set_bit(STATUS_TRANS_IDLE, &trans->status);
1737 wake_up(&trans_pcie->d0i3_waitq);
1738 }
1739
1740 if (meta->flags & CMD_WAKE_UP_TRANS) {
1741 IWL_DEBUG_INFO(trans, "complete %s - clear trans idle flag\n",
1742 iwl_get_cmd_string(trans, cmd->hdr.cmd));
1743 clear_bit(STATUS_TRANS_IDLE, &trans->status);
1744 wake_up(&trans_pcie->d0i3_waitq);
1745 }
1746
1747 meta->flags = 0;
1748
1749 spin_unlock_bh(&txq->lock);
1750 }
1751
1752 #define HOST_COMPLETE_TIMEOUT (2 * HZ)
1753
1754 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1755 struct iwl_host_cmd *cmd)
1756 {
1757 int ret;
1758
1759 /* An asynchronous command can not expect an SKB to be set. */
1760 if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1761 return -EINVAL;
1762
1763 ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1764 if (ret < 0) {
1765 IWL_ERR(trans,
1766 "Error sending %s: enqueue_hcmd failed: %d\n",
1767 iwl_get_cmd_string(trans, cmd->id), ret);
1768 return ret;
1769 }
1770 return 0;
1771 }
1772
1773 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1774 struct iwl_host_cmd *cmd)
1775 {
1776 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1777 int cmd_idx;
1778 int ret;
1779
1780 IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1781 iwl_get_cmd_string(trans, cmd->id));
1782
1783 if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1784 &trans->status),
1785 "Command %s: a command is already active!\n",
1786 iwl_get_cmd_string(trans, cmd->id)))
1787 return -EIO;
1788
1789 IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1790 iwl_get_cmd_string(trans, cmd->id));
1791
1792 if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
1793 ret = wait_event_timeout(trans_pcie->d0i3_waitq,
1794 pm_runtime_active(&trans_pcie->pci_dev->dev),
1795 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
1796 if (!ret) {
1797 IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
1798 return -ETIMEDOUT;
1799 }
1800 }
1801
1802 cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1803 if (cmd_idx < 0) {
1804 ret = cmd_idx;
1805 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1806 IWL_ERR(trans,
1807 "Error sending %s: enqueue_hcmd failed: %d\n",
1808 iwl_get_cmd_string(trans, cmd->id), ret);
1809 return ret;
1810 }
1811
1812 ret = wait_event_timeout(trans_pcie->wait_command_queue,
1813 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1814 &trans->status),
1815 HOST_COMPLETE_TIMEOUT);
1816 if (!ret) {
1817 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1818 struct iwl_queue *q = &txq->q;
1819
1820 IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1821 iwl_get_cmd_string(trans, cmd->id),
1822 jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1823
1824 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1825 q->read_ptr, q->write_ptr);
1826
1827 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1828 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1829 iwl_get_cmd_string(trans, cmd->id));
1830 ret = -ETIMEDOUT;
1831
1832 iwl_force_nmi(trans);
1833 iwl_trans_fw_error(trans);
1834
1835 goto cancel;
1836 }
1837
1838 if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1839 IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1840 iwl_get_cmd_string(trans, cmd->id));
1841 dump_stack();
1842 ret = -EIO;
1843 goto cancel;
1844 }
1845
1846 if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1847 test_bit(STATUS_RFKILL, &trans->status)) {
1848 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1849 ret = -ERFKILL;
1850 goto cancel;
1851 }
1852
1853 if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1854 IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1855 iwl_get_cmd_string(trans, cmd->id));
1856 ret = -EIO;
1857 goto cancel;
1858 }
1859
1860 return 0;
1861
1862 cancel:
1863 if (cmd->flags & CMD_WANT_SKB) {
1864 /*
1865 * Cancel the CMD_WANT_SKB flag for the cmd in the
1866 * TX cmd queue. Otherwise in case the cmd comes
1867 * in later, it will possibly set an invalid
1868 * address (cmd->meta.source).
1869 */
1870 trans_pcie->txq[trans_pcie->cmd_queue].
1871 entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1872 }
1873
1874 if (cmd->resp_pkt) {
1875 iwl_free_resp(cmd);
1876 cmd->resp_pkt = NULL;
1877 }
1878
1879 return ret;
1880 }
1881
1882 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1883 {
1884 if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1885 test_bit(STATUS_RFKILL, &trans->status)) {
1886 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1887 cmd->id);
1888 return -ERFKILL;
1889 }
1890
1891 if (cmd->flags & CMD_ASYNC)
1892 return iwl_pcie_send_hcmd_async(trans, cmd);
1893
1894 /* We still can fail on RFKILL that can be asserted while we wait */
1895 return iwl_pcie_send_hcmd_sync(trans, cmd);
1896 }
1897
1898 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
1899 struct iwl_txq *txq, u8 hdr_len,
1900 struct iwl_cmd_meta *out_meta,
1901 struct iwl_device_cmd *dev_cmd, u16 tb1_len)
1902 {
1903 struct iwl_queue *q = &txq->q;
1904 u16 tb2_len;
1905 int i;
1906
1907 /*
1908 * Set up TFD's third entry to point directly to remainder
1909 * of skb's head, if any
1910 */
1911 tb2_len = skb_headlen(skb) - hdr_len;
1912
1913 if (tb2_len > 0) {
1914 dma_addr_t tb2_phys = dma_map_single(trans->dev,
1915 skb->data + hdr_len,
1916 tb2_len, DMA_TO_DEVICE);
1917 if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) {
1918 iwl_pcie_tfd_unmap(trans, out_meta,
1919 &txq->tfds[q->write_ptr]);
1920 return -EINVAL;
1921 }
1922 iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false);
1923 }
1924
1925 /* set up the remaining entries to point to the data */
1926 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1927 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1928 dma_addr_t tb_phys;
1929 int tb_idx;
1930
1931 if (!skb_frag_size(frag))
1932 continue;
1933
1934 tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
1935 skb_frag_size(frag), DMA_TO_DEVICE);
1936
1937 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
1938 iwl_pcie_tfd_unmap(trans, out_meta,
1939 &txq->tfds[q->write_ptr]);
1940 return -EINVAL;
1941 }
1942 tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
1943 skb_frag_size(frag), false);
1944
1945 out_meta->flags |= BIT(tb_idx + CMD_TB_BITMAP_POS);
1946 }
1947
1948 trace_iwlwifi_dev_tx(trans->dev, skb,
1949 &txq->tfds[txq->q.write_ptr],
1950 sizeof(struct iwl_tfd),
1951 &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
1952 skb->data + hdr_len, tb2_len);
1953 trace_iwlwifi_dev_tx_data(trans->dev, skb,
1954 hdr_len, skb->len - hdr_len);
1955 return 0;
1956 }
1957
1958 #ifdef CONFIG_INET
1959 static struct iwl_tso_hdr_page *
1960 get_page_hdr(struct iwl_trans *trans, size_t len)
1961 {
1962 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1963 struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
1964
1965 if (!p->page)
1966 goto alloc;
1967
1968 /* enough room on this page */
1969 if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
1970 return p;
1971
1972 /* We don't have enough room on this page, get a new one. */
1973 __free_page(p->page);
1974
1975 alloc:
1976 p->page = alloc_page(GFP_ATOMIC);
1977 if (!p->page)
1978 return NULL;
1979 p->pos = page_address(p->page);
1980 return p;
1981 }
1982
1983 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
1984 bool ipv6, unsigned int len)
1985 {
1986 if (ipv6) {
1987 struct ipv6hdr *iphv6 = iph;
1988
1989 tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
1990 len + tcph->doff * 4,
1991 IPPROTO_TCP, 0);
1992 } else {
1993 struct iphdr *iphv4 = iph;
1994
1995 ip_send_check(iphv4);
1996 tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
1997 len + tcph->doff * 4,
1998 IPPROTO_TCP, 0);
1999 }
2000 }
2001
2002 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2003 struct iwl_txq *txq, u8 hdr_len,
2004 struct iwl_cmd_meta *out_meta,
2005 struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2006 {
2007 struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
2008 struct ieee80211_hdr *hdr = (void *)skb->data;
2009 unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
2010 unsigned int mss = skb_shinfo(skb)->gso_size;
2011 struct iwl_queue *q = &txq->q;
2012 u16 length, iv_len, amsdu_pad;
2013 u8 *start_hdr;
2014 struct iwl_tso_hdr_page *hdr_page;
2015 struct page **page_ptr;
2016 int ret;
2017 struct tso_t tso;
2018
2019 /* if the packet is protected, then it must be CCMP or GCMP */
2020 BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
2021 iv_len = ieee80211_has_protected(hdr->frame_control) ?
2022 IEEE80211_CCMP_HDR_LEN : 0;
2023
2024 trace_iwlwifi_dev_tx(trans->dev, skb,
2025 &txq->tfds[txq->q.write_ptr],
2026 sizeof(struct iwl_tfd),
2027 &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
2028 NULL, 0);
2029
2030 ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
2031 snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
2032 total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
2033 amsdu_pad = 0;
2034
2035 /* total amount of header we may need for this A-MSDU */
2036 hdr_room = DIV_ROUND_UP(total_len, mss) *
2037 (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
2038
2039 /* Our device supports 9 segments at most, it will fit in 1 page */
2040 hdr_page = get_page_hdr(trans, hdr_room);
2041 if (!hdr_page)
2042 return -ENOMEM;
2043
2044 get_page(hdr_page->page);
2045 start_hdr = hdr_page->pos;
2046 page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
2047 *page_ptr = hdr_page->page;
2048 memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
2049 hdr_page->pos += iv_len;
2050
2051 /*
2052 * Pull the ieee80211 header + IV to be able to use TSO core,
2053 * we will restore it for the tx_status flow.
2054 */
2055 skb_pull(skb, hdr_len + iv_len);
2056
2057 tso_start(skb, &tso);
2058
2059 while (total_len) {
2060 /* this is the data left for this subframe */
2061 unsigned int data_left =
2062 min_t(unsigned int, mss, total_len);
2063 struct sk_buff *csum_skb = NULL;
2064 unsigned int hdr_tb_len;
2065 dma_addr_t hdr_tb_phys;
2066 struct tcphdr *tcph;
2067 u8 *iph;
2068
2069 total_len -= data_left;
2070
2071 memset(hdr_page->pos, 0, amsdu_pad);
2072 hdr_page->pos += amsdu_pad;
2073 amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2074 data_left)) & 0x3;
2075 ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2076 hdr_page->pos += ETH_ALEN;
2077 ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2078 hdr_page->pos += ETH_ALEN;
2079
2080 length = snap_ip_tcp_hdrlen + data_left;
2081 *((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2082 hdr_page->pos += sizeof(length);
2083
2084 /*
2085 * This will copy the SNAP as well which will be considered
2086 * as MAC header.
2087 */
2088 tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2089 iph = hdr_page->pos + 8;
2090 tcph = (void *)(iph + ip_hdrlen);
2091
2092 /* For testing on current hardware only */
2093 if (trans_pcie->sw_csum_tx) {
2094 csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2095 GFP_ATOMIC);
2096 if (!csum_skb) {
2097 ret = -ENOMEM;
2098 goto out_unmap;
2099 }
2100
2101 iwl_compute_pseudo_hdr_csum(iph, tcph,
2102 skb->protocol ==
2103 htons(ETH_P_IPV6),
2104 data_left);
2105
2106 memcpy(skb_put(csum_skb, tcp_hdrlen(skb)),
2107 tcph, tcp_hdrlen(skb));
2108 skb_set_transport_header(csum_skb, 0);
2109 csum_skb->csum_start =
2110 (unsigned char *)tcp_hdr(csum_skb) -
2111 csum_skb->head;
2112 }
2113
2114 hdr_page->pos += snap_ip_tcp_hdrlen;
2115
2116 hdr_tb_len = hdr_page->pos - start_hdr;
2117 hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2118 hdr_tb_len, DMA_TO_DEVICE);
2119 if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2120 dev_kfree_skb(csum_skb);
2121 ret = -EINVAL;
2122 goto out_unmap;
2123 }
2124 iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2125 hdr_tb_len, false);
2126 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr,
2127 hdr_tb_len);
2128
2129 /* prepare the start_hdr for the next subframe */
2130 start_hdr = hdr_page->pos;
2131
2132 /* put the payload */
2133 while (data_left) {
2134 unsigned int size = min_t(unsigned int, tso.size,
2135 data_left);
2136 dma_addr_t tb_phys;
2137
2138 if (trans_pcie->sw_csum_tx)
2139 memcpy(skb_put(csum_skb, size), tso.data, size);
2140
2141 tb_phys = dma_map_single(trans->dev, tso.data,
2142 size, DMA_TO_DEVICE);
2143 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2144 dev_kfree_skb(csum_skb);
2145 ret = -EINVAL;
2146 goto out_unmap;
2147 }
2148
2149 iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2150 size, false);
2151 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
2152 size);
2153
2154 data_left -= size;
2155 tso_build_data(skb, &tso, size);
2156 }
2157
2158 /* For testing on early hardware only */
2159 if (trans_pcie->sw_csum_tx) {
2160 __wsum csum;
2161
2162 csum = skb_checksum(csum_skb,
2163 skb_checksum_start_offset(csum_skb),
2164 csum_skb->len -
2165 skb_checksum_start_offset(csum_skb),
2166 0);
2167 dev_kfree_skb(csum_skb);
2168 dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2169 hdr_tb_len, DMA_TO_DEVICE);
2170 tcph->check = csum_fold(csum);
2171 dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2172 hdr_tb_len, DMA_TO_DEVICE);
2173 }
2174 }
2175
2176 /* re -add the WiFi header and IV */
2177 skb_push(skb, hdr_len + iv_len);
2178
2179 return 0;
2180
2181 out_unmap:
2182 iwl_pcie_tfd_unmap(trans, out_meta, &txq->tfds[q->write_ptr]);
2183 return ret;
2184 }
2185 #else /* CONFIG_INET */
2186 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2187 struct iwl_txq *txq, u8 hdr_len,
2188 struct iwl_cmd_meta *out_meta,
2189 struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2190 {
2191 /* No A-MSDU without CONFIG_INET */
2192 WARN_ON(1);
2193
2194 return -1;
2195 }
2196 #endif /* CONFIG_INET */
2197
2198 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2199 struct iwl_device_cmd *dev_cmd, int txq_id)
2200 {
2201 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2202 struct ieee80211_hdr *hdr;
2203 struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2204 struct iwl_cmd_meta *out_meta;
2205 struct iwl_txq *txq;
2206 struct iwl_queue *q;
2207 dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2208 void *tb1_addr;
2209 u16 len, tb1_len;
2210 bool wait_write_ptr;
2211 __le16 fc;
2212 u8 hdr_len;
2213 u16 wifi_seq;
2214 bool amsdu;
2215
2216 txq = &trans_pcie->txq[txq_id];
2217 q = &txq->q;
2218
2219 if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2220 "TX on unused queue %d\n", txq_id))
2221 return -EINVAL;
2222
2223 if (unlikely(trans_pcie->sw_csum_tx &&
2224 skb->ip_summed == CHECKSUM_PARTIAL)) {
2225 int offs = skb_checksum_start_offset(skb);
2226 int csum_offs = offs + skb->csum_offset;
2227 __wsum csum;
2228
2229 if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2230 return -1;
2231
2232 csum = skb_checksum(skb, offs, skb->len - offs, 0);
2233 *(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2234
2235 skb->ip_summed = CHECKSUM_UNNECESSARY;
2236 }
2237
2238 if (skb_is_nonlinear(skb) &&
2239 skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS &&
2240 __skb_linearize(skb))
2241 return -ENOMEM;
2242
2243 /* mac80211 always puts the full header into the SKB's head,
2244 * so there's no need to check if it's readable there
2245 */
2246 hdr = (struct ieee80211_hdr *)skb->data;
2247 fc = hdr->frame_control;
2248 hdr_len = ieee80211_hdrlen(fc);
2249
2250 spin_lock(&txq->lock);
2251
2252 if (iwl_queue_space(q) < q->high_mark) {
2253 iwl_stop_queue(trans, txq);
2254
2255 /* don't put the packet on the ring, if there is no room */
2256 if (unlikely(iwl_queue_space(q) < 3)) {
2257 struct iwl_device_cmd **dev_cmd_ptr;
2258
2259 dev_cmd_ptr = (void *)((u8 *)skb->cb +
2260 trans_pcie->dev_cmd_offs);
2261
2262 *dev_cmd_ptr = dev_cmd;
2263 __skb_queue_tail(&txq->overflow_q, skb);
2264
2265 spin_unlock(&txq->lock);
2266 return 0;
2267 }
2268 }
2269
2270 /* In AGG mode, the index in the ring must correspond to the WiFi
2271 * sequence number. This is a HW requirements to help the SCD to parse
2272 * the BA.
2273 * Check here that the packets are in the right place on the ring.
2274 */
2275 wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2276 WARN_ONCE(txq->ampdu &&
2277 (wifi_seq & 0xff) != q->write_ptr,
2278 "Q: %d WiFi Seq %d tfdNum %d",
2279 txq_id, wifi_seq, q->write_ptr);
2280
2281 /* Set up driver data for this TFD */
2282 txq->entries[q->write_ptr].skb = skb;
2283 txq->entries[q->write_ptr].cmd = dev_cmd;
2284
2285 dev_cmd->hdr.sequence =
2286 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2287 INDEX_TO_SEQ(q->write_ptr)));
2288
2289 tb0_phys = iwl_pcie_get_first_tb_dma(txq, q->write_ptr);
2290 scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2291 offsetof(struct iwl_tx_cmd, scratch);
2292
2293 tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2294 tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2295
2296 /* Set up first empty entry in queue's array of Tx/cmd buffers */
2297 out_meta = &txq->entries[q->write_ptr].meta;
2298 out_meta->flags = 0;
2299
2300 /*
2301 * The second TB (tb1) points to the remainder of the TX command
2302 * and the 802.11 header - dword aligned size
2303 * (This calculation modifies the TX command, so do it before the
2304 * setup of the first TB)
2305 */
2306 len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2307 hdr_len - IWL_FIRST_TB_SIZE;
2308 /* do not align A-MSDU to dword as the subframe header aligns it */
2309 amsdu = ieee80211_is_data_qos(fc) &&
2310 (*ieee80211_get_qos_ctl(hdr) &
2311 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
2312 if (trans_pcie->sw_csum_tx || !amsdu) {
2313 tb1_len = ALIGN(len, 4);
2314 /* Tell NIC about any 2-byte padding after MAC header */
2315 if (tb1_len != len)
2316 tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
2317 } else {
2318 tb1_len = len;
2319 }
2320
2321 /* The first TB points to bi-directional DMA data */
2322 memcpy(&txq->first_tb_bufs[q->write_ptr], &dev_cmd->hdr,
2323 IWL_FIRST_TB_SIZE);
2324 iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2325 IWL_FIRST_TB_SIZE, true);
2326
2327 /* there must be data left over for TB1 or this code must be changed */
2328 BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
2329
2330 /* map the data for TB1 */
2331 tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
2332 tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2333 if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2334 goto out_err;
2335 iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2336
2337 if (amsdu) {
2338 if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2339 out_meta, dev_cmd,
2340 tb1_len)))
2341 goto out_err;
2342 } else if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2343 out_meta, dev_cmd, tb1_len))) {
2344 goto out_err;
2345 }
2346
2347 /* Set up entry for this TFD in Tx byte-count array */
2348 iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len));
2349
2350 wait_write_ptr = ieee80211_has_morefrags(fc);
2351
2352 /* start timer if queue currently empty */
2353 if (q->read_ptr == q->write_ptr) {
2354 if (txq->wd_timeout) {
2355 /*
2356 * If the TXQ is active, then set the timer, if not,
2357 * set the timer in remainder so that the timer will
2358 * be armed with the right value when the station will
2359 * wake up.
2360 */
2361 if (!txq->frozen)
2362 mod_timer(&txq->stuck_timer,
2363 jiffies + txq->wd_timeout);
2364 else
2365 txq->frozen_expiry_remainder = txq->wd_timeout;
2366 }
2367 IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", q->id);
2368 iwl_trans_ref(trans);
2369 }
2370
2371 /* Tell device the write index *just past* this latest filled TFD */
2372 q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
2373 if (!wait_write_ptr)
2374 iwl_pcie_txq_inc_wr_ptr(trans, txq);
2375
2376 /*
2377 * At this point the frame is "transmitted" successfully
2378 * and we will get a TX status notification eventually.
2379 */
2380 spin_unlock(&txq->lock);
2381 return 0;
2382 out_err:
2383 spin_unlock(&txq->lock);
2384 return -1;
2385 }
CRIS linux kernel tree