Merge remote-tracking branch 'asoc/fix/rt5645' into asoc-linus
[sfrench/cifs-2.6.git] / drivers / net / wireless / iwlwifi / pcie / tx.c
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  ***************************************************/
68 static 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  */
99 static 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
123 static 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
137 static 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
147 static 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  */
206 static 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
251 static 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  */
280 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
281                                     struct iwl_txq *txq)
282 {
283         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
284         u32 reg = 0;
285         int txq_id = txq->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
323 void 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
340 static 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
352 static 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
367 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_tfd *tfd)
368 {
369         return tfd->num_tbs & 0x1f;
370 }
371
372 static 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  */
407 static 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
441 static 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
473 static 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;
530 err_free_tfds:
531         dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->q.dma_addr);
532 error:
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
543 static 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  */
574 static 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  */
602 static 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
643 void 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
700 void 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
728 static 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
755 out:
756         spin_unlock(&trans_pcie->irq_lock);
757 }
758
759 /*
760  * iwl_pcie_tx_stop - Stop all Tx DMA channels
761  */
762 int 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  */
798 void 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  */
822 static 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
875 error:
876         iwl_pcie_tx_free(trans);
877
878         return ret;
879 }
880 int 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;
923 error:
924         /*Upon error, free only if we allocated something */
925         if (alloc)
926                 iwl_pcie_tx_free(trans);
927         return ret;
928 }
929
930 static 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 */
955 void 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
1023 out:
1024         spin_unlock_bh(&txq->lock);
1025 }
1026
1027 static 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         if (trans_pcie->cmd_in_flight)
1043                 return 0;
1044
1045         trans_pcie->cmd_in_flight = true;
1046
1047         /*
1048          * wake up the NIC to make sure that the firmware will see the host
1049          * command - we will let the NIC sleep once all the host commands
1050          * returned. This needs to be done only on NICs that have
1051          * apmg_wake_up_wa set.
1052          */
1053         if (trans->cfg->base_params->apmg_wake_up_wa) {
1054                 __iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1055                                          CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1056                 if (trans->cfg->device_family == IWL_DEVICE_FAMILY_8000)
1057                         udelay(2);
1058
1059                 ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1060                                    CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1061                                    (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1062                                     CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1063                                    15000);
1064                 if (ret < 0) {
1065                         __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1066                                         CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1067                         trans_pcie->cmd_in_flight = false;
1068                         IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1069                         return -EIO;
1070                 }
1071         }
1072
1073         return 0;
1074 }
1075
1076 static int iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
1077 {
1078         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1079
1080         lockdep_assert_held(&trans_pcie->reg_lock);
1081
1082         if (trans_pcie->ref_cmd_in_flight) {
1083                 trans_pcie->ref_cmd_in_flight = false;
1084                 IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
1085                 iwl_trans_pcie_unref(trans);
1086         }
1087
1088         if (WARN_ON(!trans_pcie->cmd_in_flight))
1089                 return 0;
1090
1091         trans_pcie->cmd_in_flight = false;
1092
1093         if (trans->cfg->base_params->apmg_wake_up_wa)
1094                 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1095                                         CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1096
1097         return 0;
1098 }
1099
1100 /*
1101  * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1102  *
1103  * When FW advances 'R' index, all entries between old and new 'R' index
1104  * need to be reclaimed. As result, some free space forms.  If there is
1105  * enough free space (> low mark), wake the stack that feeds us.
1106  */
1107 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1108 {
1109         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1110         struct iwl_txq *txq = &trans_pcie->txq[txq_id];
1111         struct iwl_queue *q = &txq->q;
1112         unsigned long flags;
1113         int nfreed = 0;
1114
1115         lockdep_assert_held(&txq->lock);
1116
1117         if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(q, idx))) {
1118                 IWL_ERR(trans,
1119                         "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1120                         __func__, txq_id, idx, TFD_QUEUE_SIZE_MAX,
1121                         q->write_ptr, q->read_ptr);
1122                 return;
1123         }
1124
1125         for (idx = iwl_queue_inc_wrap(idx); q->read_ptr != idx;
1126              q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) {
1127
1128                 if (nfreed++ > 0) {
1129                         IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1130                                 idx, q->write_ptr, q->read_ptr);
1131                         iwl_force_nmi(trans);
1132                 }
1133         }
1134
1135         if (q->read_ptr == q->write_ptr) {
1136                 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1137                 iwl_pcie_clear_cmd_in_flight(trans);
1138                 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1139         }
1140
1141         iwl_pcie_txq_progress(txq);
1142 }
1143
1144 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1145                                  u16 txq_id)
1146 {
1147         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1148         u32 tbl_dw_addr;
1149         u32 tbl_dw;
1150         u16 scd_q2ratid;
1151
1152         scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1153
1154         tbl_dw_addr = trans_pcie->scd_base_addr +
1155                         SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1156
1157         tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1158
1159         if (txq_id & 0x1)
1160                 tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1161         else
1162                 tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1163
1164         iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1165
1166         return 0;
1167 }
1168
1169 /* Receiver address (actually, Rx station's index into station table),
1170  * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1171 #define BUILD_RAxTID(sta_id, tid)       (((sta_id) << 4) + (tid))
1172
1173 void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1174                                const struct iwl_trans_txq_scd_cfg *cfg,
1175                                unsigned int wdg_timeout)
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         int fifo = -1;
1180
1181         if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1182                 WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1183
1184         txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1185
1186         if (cfg) {
1187                 fifo = cfg->fifo;
1188
1189                 /* Disable the scheduler prior configuring the cmd queue */
1190                 if (txq_id == trans_pcie->cmd_queue &&
1191                     trans_pcie->scd_set_active)
1192                         iwl_scd_enable_set_active(trans, 0);
1193
1194                 /* Stop this Tx queue before configuring it */
1195                 iwl_scd_txq_set_inactive(trans, txq_id);
1196
1197                 /* Set this queue as a chain-building queue unless it is CMD */
1198                 if (txq_id != trans_pcie->cmd_queue)
1199                         iwl_scd_txq_set_chain(trans, txq_id);
1200
1201                 if (cfg->aggregate) {
1202                         u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1203
1204                         /* Map receiver-address / traffic-ID to this queue */
1205                         iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1206
1207                         /* enable aggregations for the queue */
1208                         iwl_scd_txq_enable_agg(trans, txq_id);
1209                         txq->ampdu = true;
1210                 } else {
1211                         /*
1212                          * disable aggregations for the queue, this will also
1213                          * make the ra_tid mapping configuration irrelevant
1214                          * since it is now a non-AGG queue.
1215                          */
1216                         iwl_scd_txq_disable_agg(trans, txq_id);
1217
1218                         ssn = txq->q.read_ptr;
1219                 }
1220         }
1221
1222         /* Place first TFD at index corresponding to start sequence number.
1223          * Assumes that ssn_idx is valid (!= 0xFFF) */
1224         txq->q.read_ptr = (ssn & 0xff);
1225         txq->q.write_ptr = (ssn & 0xff);
1226         iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1227                            (ssn & 0xff) | (txq_id << 8));
1228
1229         if (cfg) {
1230                 u8 frame_limit = cfg->frame_limit;
1231
1232                 iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1233
1234                 /* Set up Tx window size and frame limit for this queue */
1235                 iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1236                                 SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1237                 iwl_trans_write_mem32(trans,
1238                         trans_pcie->scd_base_addr +
1239                         SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1240                         ((frame_limit << SCD_QUEUE_CTX_REG2_WIN_SIZE_POS) &
1241                                         SCD_QUEUE_CTX_REG2_WIN_SIZE_MSK) |
1242                         ((frame_limit << SCD_QUEUE_CTX_REG2_FRAME_LIMIT_POS) &
1243                                         SCD_QUEUE_CTX_REG2_FRAME_LIMIT_MSK));
1244
1245                 /* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1246                 iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1247                                (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1248                                (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1249                                (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1250                                SCD_QUEUE_STTS_REG_MSK);
1251
1252                 /* enable the scheduler for this queue (only) */
1253                 if (txq_id == trans_pcie->cmd_queue &&
1254                     trans_pcie->scd_set_active)
1255                         iwl_scd_enable_set_active(trans, BIT(txq_id));
1256
1257                 IWL_DEBUG_TX_QUEUES(trans,
1258                                     "Activate queue %d on FIFO %d WrPtr: %d\n",
1259                                     txq_id, fifo, ssn & 0xff);
1260         } else {
1261                 IWL_DEBUG_TX_QUEUES(trans,
1262                                     "Activate queue %d WrPtr: %d\n",
1263                                     txq_id, ssn & 0xff);
1264         }
1265
1266         txq->active = true;
1267 }
1268
1269 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1270                                 bool configure_scd)
1271 {
1272         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1273         u32 stts_addr = trans_pcie->scd_base_addr +
1274                         SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1275         static const u32 zero_val[4] = {};
1276
1277         trans_pcie->txq[txq_id].frozen_expiry_remainder = 0;
1278         trans_pcie->txq[txq_id].frozen = false;
1279
1280         /*
1281          * Upon HW Rfkill - we stop the device, and then stop the queues
1282          * in the op_mode. Just for the sake of the simplicity of the op_mode,
1283          * allow the op_mode to call txq_disable after it already called
1284          * stop_device.
1285          */
1286         if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1287                 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1288                           "queue %d not used", txq_id);
1289                 return;
1290         }
1291
1292         if (configure_scd) {
1293                 iwl_scd_txq_set_inactive(trans, txq_id);
1294
1295                 iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1296                                     ARRAY_SIZE(zero_val));
1297         }
1298
1299         iwl_pcie_txq_unmap(trans, txq_id);
1300         trans_pcie->txq[txq_id].ampdu = false;
1301
1302         IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1303 }
1304
1305 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1306
1307 /*
1308  * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1309  * @priv: device private data point
1310  * @cmd: a pointer to the ucode command structure
1311  *
1312  * The function returns < 0 values to indicate the operation
1313  * failed. On success, it returns the index (>= 0) of command in the
1314  * command queue.
1315  */
1316 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1317                                  struct iwl_host_cmd *cmd)
1318 {
1319         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1320         struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1321         struct iwl_queue *q = &txq->q;
1322         struct iwl_device_cmd *out_cmd;
1323         struct iwl_cmd_meta *out_meta;
1324         unsigned long flags;
1325         void *dup_buf = NULL;
1326         dma_addr_t phys_addr;
1327         int idx;
1328         u16 copy_size, cmd_size, scratch_size;
1329         bool had_nocopy = false;
1330         int i, ret;
1331         u32 cmd_pos;
1332         const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1333         u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1334
1335         copy_size = sizeof(out_cmd->hdr);
1336         cmd_size = sizeof(out_cmd->hdr);
1337
1338         /* need one for the header if the first is NOCOPY */
1339         BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1340
1341         for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1342                 cmddata[i] = cmd->data[i];
1343                 cmdlen[i] = cmd->len[i];
1344
1345                 if (!cmd->len[i])
1346                         continue;
1347
1348                 /* need at least IWL_HCMD_SCRATCHBUF_SIZE copied */
1349                 if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) {
1350                         int copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size;
1351
1352                         if (copy > cmdlen[i])
1353                                 copy = cmdlen[i];
1354                         cmdlen[i] -= copy;
1355                         cmddata[i] += copy;
1356                         copy_size += copy;
1357                 }
1358
1359                 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1360                         had_nocopy = true;
1361                         if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1362                                 idx = -EINVAL;
1363                                 goto free_dup_buf;
1364                         }
1365                 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1366                         /*
1367                          * This is also a chunk that isn't copied
1368                          * to the static buffer so set had_nocopy.
1369                          */
1370                         had_nocopy = true;
1371
1372                         /* only allowed once */
1373                         if (WARN_ON(dup_buf)) {
1374                                 idx = -EINVAL;
1375                                 goto free_dup_buf;
1376                         }
1377
1378                         dup_buf = kmemdup(cmddata[i], cmdlen[i],
1379                                           GFP_ATOMIC);
1380                         if (!dup_buf)
1381                                 return -ENOMEM;
1382                 } else {
1383                         /* NOCOPY must not be followed by normal! */
1384                         if (WARN_ON(had_nocopy)) {
1385                                 idx = -EINVAL;
1386                                 goto free_dup_buf;
1387                         }
1388                         copy_size += cmdlen[i];
1389                 }
1390                 cmd_size += cmd->len[i];
1391         }
1392
1393         /*
1394          * If any of the command structures end up being larger than
1395          * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1396          * allocated into separate TFDs, then we will need to
1397          * increase the size of the buffers.
1398          */
1399         if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1400                  "Command %s (%#x) is too large (%d bytes)\n",
1401                  get_cmd_string(trans_pcie, cmd->id), cmd->id, copy_size)) {
1402                 idx = -EINVAL;
1403                 goto free_dup_buf;
1404         }
1405
1406         spin_lock_bh(&txq->lock);
1407
1408         if (iwl_queue_space(q) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1409                 spin_unlock_bh(&txq->lock);
1410
1411                 IWL_ERR(trans, "No space in command queue\n");
1412                 iwl_op_mode_cmd_queue_full(trans->op_mode);
1413                 idx = -ENOSPC;
1414                 goto free_dup_buf;
1415         }
1416
1417         idx = get_cmd_index(q, q->write_ptr);
1418         out_cmd = txq->entries[idx].cmd;
1419         out_meta = &txq->entries[idx].meta;
1420
1421         memset(out_meta, 0, sizeof(*out_meta)); /* re-initialize to NULL */
1422         if (cmd->flags & CMD_WANT_SKB)
1423                 out_meta->source = cmd;
1424
1425         /* set up the header */
1426
1427         out_cmd->hdr.cmd = cmd->id;
1428         out_cmd->hdr.flags = 0;
1429         out_cmd->hdr.sequence =
1430                 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1431                                          INDEX_TO_SEQ(q->write_ptr));
1432
1433         /* and copy the data that needs to be copied */
1434         cmd_pos = offsetof(struct iwl_device_cmd, payload);
1435         copy_size = sizeof(out_cmd->hdr);
1436         for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1437                 int copy;
1438
1439                 if (!cmd->len[i])
1440                         continue;
1441
1442                 /* copy everything if not nocopy/dup */
1443                 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1444                                            IWL_HCMD_DFL_DUP))) {
1445                         copy = cmd->len[i];
1446
1447                         memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1448                         cmd_pos += copy;
1449                         copy_size += copy;
1450                         continue;
1451                 }
1452
1453                 /*
1454                  * Otherwise we need at least IWL_HCMD_SCRATCHBUF_SIZE copied
1455                  * in total (for the scratchbuf handling), but copy up to what
1456                  * we can fit into the payload for debug dump purposes.
1457                  */
1458                 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1459
1460                 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1461                 cmd_pos += copy;
1462
1463                 /* However, treat copy_size the proper way, we need it below */
1464                 if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) {
1465                         copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size;
1466
1467                         if (copy > cmd->len[i])
1468                                 copy = cmd->len[i];
1469                         copy_size += copy;
1470                 }
1471         }
1472
1473         IWL_DEBUG_HC(trans,
1474                      "Sending command %s (#%x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1475                      get_cmd_string(trans_pcie, out_cmd->hdr.cmd),
1476                      out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
1477                      cmd_size, q->write_ptr, idx, trans_pcie->cmd_queue);
1478
1479         /* start the TFD with the scratchbuf */
1480         scratch_size = min_t(int, copy_size, IWL_HCMD_SCRATCHBUF_SIZE);
1481         memcpy(&txq->scratchbufs[q->write_ptr], &out_cmd->hdr, scratch_size);
1482         iwl_pcie_txq_build_tfd(trans, txq,
1483                                iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr),
1484                                scratch_size, true);
1485
1486         /* map first command fragment, if any remains */
1487         if (copy_size > scratch_size) {
1488                 phys_addr = dma_map_single(trans->dev,
1489                                            ((u8 *)&out_cmd->hdr) + scratch_size,
1490                                            copy_size - scratch_size,
1491                                            DMA_TO_DEVICE);
1492                 if (dma_mapping_error(trans->dev, phys_addr)) {
1493                         iwl_pcie_tfd_unmap(trans, out_meta,
1494                                            &txq->tfds[q->write_ptr]);
1495                         idx = -ENOMEM;
1496                         goto out;
1497                 }
1498
1499                 iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1500                                        copy_size - scratch_size, false);
1501         }
1502
1503         /* map the remaining (adjusted) nocopy/dup fragments */
1504         for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1505                 const void *data = cmddata[i];
1506
1507                 if (!cmdlen[i])
1508                         continue;
1509                 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1510                                            IWL_HCMD_DFL_DUP)))
1511                         continue;
1512                 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1513                         data = dup_buf;
1514                 phys_addr = dma_map_single(trans->dev, (void *)data,
1515                                            cmdlen[i], DMA_TO_DEVICE);
1516                 if (dma_mapping_error(trans->dev, phys_addr)) {
1517                         iwl_pcie_tfd_unmap(trans, out_meta,
1518                                            &txq->tfds[q->write_ptr]);
1519                         idx = -ENOMEM;
1520                         goto out;
1521                 }
1522
1523                 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1524         }
1525
1526         out_meta->flags = cmd->flags;
1527         if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1528                 kzfree(txq->entries[idx].free_buf);
1529         txq->entries[idx].free_buf = dup_buf;
1530
1531         trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr);
1532
1533         /* start timer if queue currently empty */
1534         if (q->read_ptr == q->write_ptr && txq->wd_timeout)
1535                 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1536
1537         spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1538         ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1539         if (ret < 0) {
1540                 idx = ret;
1541                 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1542                 goto out;
1543         }
1544
1545         /* Increment and update queue's write index */
1546         q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
1547         iwl_pcie_txq_inc_wr_ptr(trans, txq);
1548
1549         spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1550
1551  out:
1552         spin_unlock_bh(&txq->lock);
1553  free_dup_buf:
1554         if (idx < 0)
1555                 kfree(dup_buf);
1556         return idx;
1557 }
1558
1559 /*
1560  * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1561  * @rxb: Rx buffer to reclaim
1562  * @handler_status: return value of the handler of the command
1563  *      (put in setup_rx_handlers)
1564  *
1565  * If an Rx buffer has an async callback associated with it the callback
1566  * will be executed.  The attached skb (if present) will only be freed
1567  * if the callback returns 1
1568  */
1569 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1570                             struct iwl_rx_cmd_buffer *rxb, int handler_status)
1571 {
1572         struct iwl_rx_packet *pkt = rxb_addr(rxb);
1573         u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1574         int txq_id = SEQ_TO_QUEUE(sequence);
1575         int index = SEQ_TO_INDEX(sequence);
1576         int cmd_index;
1577         struct iwl_device_cmd *cmd;
1578         struct iwl_cmd_meta *meta;
1579         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1580         struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1581
1582         /* If a Tx command is being handled and it isn't in the actual
1583          * command queue then there a command routing bug has been introduced
1584          * in the queue management code. */
1585         if (WARN(txq_id != trans_pcie->cmd_queue,
1586                  "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1587                  txq_id, trans_pcie->cmd_queue, sequence,
1588                  trans_pcie->txq[trans_pcie->cmd_queue].q.read_ptr,
1589                  trans_pcie->txq[trans_pcie->cmd_queue].q.write_ptr)) {
1590                 iwl_print_hex_error(trans, pkt, 32);
1591                 return;
1592         }
1593
1594         spin_lock_bh(&txq->lock);
1595
1596         cmd_index = get_cmd_index(&txq->q, index);
1597         cmd = txq->entries[cmd_index].cmd;
1598         meta = &txq->entries[cmd_index].meta;
1599
1600         iwl_pcie_tfd_unmap(trans, meta, &txq->tfds[index]);
1601
1602         /* Input error checking is done when commands are added to queue. */
1603         if (meta->flags & CMD_WANT_SKB) {
1604                 struct page *p = rxb_steal_page(rxb);
1605
1606                 meta->source->resp_pkt = pkt;
1607                 meta->source->_rx_page_addr = (unsigned long)page_address(p);
1608                 meta->source->_rx_page_order = trans_pcie->rx_page_order;
1609                 meta->source->handler_status = handler_status;
1610         }
1611
1612         iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1613
1614         if (!(meta->flags & CMD_ASYNC)) {
1615                 if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1616                         IWL_WARN(trans,
1617                                  "HCMD_ACTIVE already clear for command %s\n",
1618                                  get_cmd_string(trans_pcie, cmd->hdr.cmd));
1619                 }
1620                 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1621                 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1622                                get_cmd_string(trans_pcie, cmd->hdr.cmd));
1623                 wake_up(&trans_pcie->wait_command_queue);
1624         }
1625
1626         meta->flags = 0;
1627
1628         spin_unlock_bh(&txq->lock);
1629 }
1630
1631 #define HOST_COMPLETE_TIMEOUT   (2 * HZ)
1632
1633 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1634                                     struct iwl_host_cmd *cmd)
1635 {
1636         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1637         int ret;
1638
1639         /* An asynchronous command can not expect an SKB to be set. */
1640         if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1641                 return -EINVAL;
1642
1643         ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1644         if (ret < 0) {
1645                 IWL_ERR(trans,
1646                         "Error sending %s: enqueue_hcmd failed: %d\n",
1647                         get_cmd_string(trans_pcie, cmd->id), ret);
1648                 return ret;
1649         }
1650         return 0;
1651 }
1652
1653 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1654                                    struct iwl_host_cmd *cmd)
1655 {
1656         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1657         int cmd_idx;
1658         int ret;
1659
1660         IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1661                        get_cmd_string(trans_pcie, cmd->id));
1662
1663         if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1664                                   &trans->status),
1665                  "Command %s: a command is already active!\n",
1666                  get_cmd_string(trans_pcie, cmd->id)))
1667                 return -EIO;
1668
1669         IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1670                        get_cmd_string(trans_pcie, cmd->id));
1671
1672         cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1673         if (cmd_idx < 0) {
1674                 ret = cmd_idx;
1675                 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1676                 IWL_ERR(trans,
1677                         "Error sending %s: enqueue_hcmd failed: %d\n",
1678                         get_cmd_string(trans_pcie, cmd->id), ret);
1679                 return ret;
1680         }
1681
1682         ret = wait_event_timeout(trans_pcie->wait_command_queue,
1683                                  !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1684                                            &trans->status),
1685                                  HOST_COMPLETE_TIMEOUT);
1686         if (!ret) {
1687                 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue];
1688                 struct iwl_queue *q = &txq->q;
1689
1690                 IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1691                         get_cmd_string(trans_pcie, cmd->id),
1692                         jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1693
1694                 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1695                         q->read_ptr, q->write_ptr);
1696
1697                 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1698                 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1699                                get_cmd_string(trans_pcie, cmd->id));
1700                 ret = -ETIMEDOUT;
1701
1702                 iwl_force_nmi(trans);
1703                 iwl_trans_fw_error(trans);
1704
1705                 goto cancel;
1706         }
1707
1708         if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1709                 IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1710                         get_cmd_string(trans_pcie, cmd->id));
1711                 dump_stack();
1712                 ret = -EIO;
1713                 goto cancel;
1714         }
1715
1716         if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1717             test_bit(STATUS_RFKILL, &trans->status)) {
1718                 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1719                 ret = -ERFKILL;
1720                 goto cancel;
1721         }
1722
1723         if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1724                 IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1725                         get_cmd_string(trans_pcie, cmd->id));
1726                 ret = -EIO;
1727                 goto cancel;
1728         }
1729
1730         return 0;
1731
1732 cancel:
1733         if (cmd->flags & CMD_WANT_SKB) {
1734                 /*
1735                  * Cancel the CMD_WANT_SKB flag for the cmd in the
1736                  * TX cmd queue. Otherwise in case the cmd comes
1737                  * in later, it will possibly set an invalid
1738                  * address (cmd->meta.source).
1739                  */
1740                 trans_pcie->txq[trans_pcie->cmd_queue].
1741                         entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1742         }
1743
1744         if (cmd->resp_pkt) {
1745                 iwl_free_resp(cmd);
1746                 cmd->resp_pkt = NULL;
1747         }
1748
1749         return ret;
1750 }
1751
1752 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1753 {
1754         if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1755             test_bit(STATUS_RFKILL, &trans->status)) {
1756                 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1757                                   cmd->id);
1758                 return -ERFKILL;
1759         }
1760
1761         if (cmd->flags & CMD_ASYNC)
1762                 return iwl_pcie_send_hcmd_async(trans, cmd);
1763
1764         /* We still can fail on RFKILL that can be asserted while we wait */
1765         return iwl_pcie_send_hcmd_sync(trans, cmd);
1766 }
1767
1768 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
1769                       struct iwl_device_cmd *dev_cmd, int txq_id)
1770 {
1771         struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1772         struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
1773         struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
1774         struct iwl_cmd_meta *out_meta;
1775         struct iwl_txq *txq;
1776         struct iwl_queue *q;
1777         dma_addr_t tb0_phys, tb1_phys, scratch_phys;
1778         void *tb1_addr;
1779         u16 len, tb1_len, tb2_len;
1780         bool wait_write_ptr;
1781         __le16 fc = hdr->frame_control;
1782         u8 hdr_len = ieee80211_hdrlen(fc);
1783         u16 wifi_seq;
1784
1785         txq = &trans_pcie->txq[txq_id];
1786         q = &txq->q;
1787
1788         if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
1789                       "TX on unused queue %d\n", txq_id))
1790                 return -EINVAL;
1791
1792         spin_lock(&txq->lock);
1793
1794         /* In AGG mode, the index in the ring must correspond to the WiFi
1795          * sequence number. This is a HW requirements to help the SCD to parse
1796          * the BA.
1797          * Check here that the packets are in the right place on the ring.
1798          */
1799         wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
1800         WARN_ONCE(txq->ampdu &&
1801                   (wifi_seq & 0xff) != q->write_ptr,
1802                   "Q: %d WiFi Seq %d tfdNum %d",
1803                   txq_id, wifi_seq, q->write_ptr);
1804
1805         /* Set up driver data for this TFD */
1806         txq->entries[q->write_ptr].skb = skb;
1807         txq->entries[q->write_ptr].cmd = dev_cmd;
1808
1809         dev_cmd->hdr.sequence =
1810                 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
1811                             INDEX_TO_SEQ(q->write_ptr)));
1812
1813         tb0_phys = iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr);
1814         scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
1815                        offsetof(struct iwl_tx_cmd, scratch);
1816
1817         tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
1818         tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
1819
1820         /* Set up first empty entry in queue's array of Tx/cmd buffers */
1821         out_meta = &txq->entries[q->write_ptr].meta;
1822
1823         /*
1824          * The second TB (tb1) points to the remainder of the TX command
1825          * and the 802.11 header - dword aligned size
1826          * (This calculation modifies the TX command, so do it before the
1827          * setup of the first TB)
1828          */
1829         len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
1830               hdr_len - IWL_HCMD_SCRATCHBUF_SIZE;
1831         tb1_len = ALIGN(len, 4);
1832
1833         /* Tell NIC about any 2-byte padding after MAC header */
1834         if (tb1_len != len)
1835                 tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
1836
1837         /* The first TB points to the scratchbuf data - min_copy bytes */
1838         memcpy(&txq->scratchbufs[q->write_ptr], &dev_cmd->hdr,
1839                IWL_HCMD_SCRATCHBUF_SIZE);
1840         iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
1841                                IWL_HCMD_SCRATCHBUF_SIZE, true);
1842
1843         /* there must be data left over for TB1 or this code must be changed */
1844         BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_HCMD_SCRATCHBUF_SIZE);
1845
1846         /* map the data for TB1 */
1847         tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_HCMD_SCRATCHBUF_SIZE;
1848         tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
1849         if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
1850                 goto out_err;
1851         iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
1852
1853         /*
1854          * Set up TFD's third entry to point directly to remainder
1855          * of skb, if any (802.11 null frames have no payload).
1856          */
1857         tb2_len = skb->len - hdr_len;
1858         if (tb2_len > 0) {
1859                 dma_addr_t tb2_phys = dma_map_single(trans->dev,
1860                                                      skb->data + hdr_len,
1861                                                      tb2_len, DMA_TO_DEVICE);
1862                 if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) {
1863                         iwl_pcie_tfd_unmap(trans, out_meta,
1864                                            &txq->tfds[q->write_ptr]);
1865                         goto out_err;
1866                 }
1867                 iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false);
1868         }
1869
1870         /* Set up entry for this TFD in Tx byte-count array */
1871         iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len));
1872
1873         trace_iwlwifi_dev_tx(trans->dev, skb,
1874                              &txq->tfds[txq->q.write_ptr],
1875                              sizeof(struct iwl_tfd),
1876                              &dev_cmd->hdr, IWL_HCMD_SCRATCHBUF_SIZE + tb1_len,
1877                              skb->data + hdr_len, tb2_len);
1878         trace_iwlwifi_dev_tx_data(trans->dev, skb,
1879                                   skb->data + hdr_len, tb2_len);
1880
1881         wait_write_ptr = ieee80211_has_morefrags(fc);
1882
1883         /* start timer if queue currently empty */
1884         if (q->read_ptr == q->write_ptr) {
1885                 if (txq->wd_timeout)
1886                         mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1887                 IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", q->id);
1888                 iwl_trans_pcie_ref(trans);
1889         }
1890
1891         /* Tell device the write index *just past* this latest filled TFD */
1892         q->write_ptr = iwl_queue_inc_wrap(q->write_ptr);
1893         if (!wait_write_ptr)
1894                 iwl_pcie_txq_inc_wr_ptr(trans, txq);
1895
1896         /*
1897          * At this point the frame is "transmitted" successfully
1898          * and we will get a TX status notification eventually.
1899          */
1900         if (iwl_queue_space(q) < q->high_mark) {
1901                 if (wait_write_ptr)
1902                         iwl_pcie_txq_inc_wr_ptr(trans, txq);
1903                 else
1904                         iwl_stop_queue(trans, txq);
1905         }
1906         spin_unlock(&txq->lock);
1907         return 0;
1908 out_err:
1909         spin_unlock(&txq->lock);
1910         return -1;
1911 }