1 /******************************************************************************
3 * This file is provided under a dual BSD/GPLv2 license. When using or
4 * redistributing this file, you may do so under either license.
8 * Copyright(c) 2017 Intel Deutschland GmbH
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of version 2 of the GNU General Public License as
12 * published by the Free Software Foundation.
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
21 * Copyright(c) 2017 Intel Deutschland GmbH
22 * All rights reserved.
24 * Redistribution and use in source and binary forms, with or without
25 * modification, are permitted provided that the following conditions
28 * * Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * * Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in
32 * the documentation and/or other materials provided with the
34 * * Neither the name Intel Corporation nor the names of its
35 * contributors may be used to endorse or promote products derived
36 * from this software without specific prior written permission.
38 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
39 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
40 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
41 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
42 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
43 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
44 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
45 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
46 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
47 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
48 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
50 *****************************************************************************/
51 #include <linux/pm_runtime.h>
54 #include "iwl-debug.h"
58 #include "fw/api/tx.h"
61 * iwl_pcie_gen2_tx_stop - Stop all Tx DMA channels
63 void iwl_pcie_gen2_tx_stop(struct iwl_trans *trans)
65 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
69 * This function can be called before the op_mode disabled the
70 * queues. This happens when we have an rfkill interrupt.
71 * Since we stop Tx altogether - mark the queues as stopped.
73 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
74 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
76 /* Unmap DMA from host system and free skb's */
77 for (txq_id = 0; txq_id < ARRAY_SIZE(trans_pcie->txq); txq_id++) {
78 if (!trans_pcie->txq[txq_id])
80 iwl_pcie_gen2_txq_unmap(trans, txq_id);
85 * iwl_pcie_txq_update_byte_tbl - Set up entry in Tx byte-count array
87 static void iwl_pcie_gen2_update_byte_tbl(struct iwl_txq *txq, u16 byte_cnt,
90 struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr;
91 int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
92 u8 filled_tfd_size, num_fetch_chunks;
96 len = DIV_ROUND_UP(len, 4);
98 if (WARN_ON(len > 0xFFF || idx >= txq->n_window))
101 filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) +
102 num_tbs * sizeof(struct iwl_tfh_tb);
104 * filled_tfd_size contains the number of filled bytes in the TFD.
105 * Dividing it by 64 will give the number of chunks to fetch
106 * to SRAM- 0 for one chunk, 1 for 2 and so on.
107 * If, for example, TFD contains only 3 TBs then 32 bytes
108 * of the TFD are used, and only one chunk of 64 bytes should
111 num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1;
113 bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12));
114 scd_bc_tbl->tfd_offset[idx] = bc_ent;
118 * iwl_pcie_gen2_txq_inc_wr_ptr - Send new write index to hardware
120 static void iwl_pcie_gen2_txq_inc_wr_ptr(struct iwl_trans *trans,
123 lockdep_assert_held(&txq->lock);
125 IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr);
128 * if not in power-save mode, uCode will never sleep when we're
129 * trying to tx (during RFKILL, we're not trying to tx).
131 iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16));
134 static u8 iwl_pcie_gen2_get_num_tbs(struct iwl_trans *trans,
135 struct iwl_tfh_tfd *tfd)
137 return le16_to_cpu(tfd->num_tbs) & 0x1f;
140 static void iwl_pcie_gen2_tfd_unmap(struct iwl_trans *trans,
141 struct iwl_cmd_meta *meta,
142 struct iwl_tfh_tfd *tfd)
144 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
147 /* Sanity check on number of chunks */
148 num_tbs = iwl_pcie_gen2_get_num_tbs(trans, tfd);
150 if (num_tbs >= trans_pcie->max_tbs) {
151 IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
155 /* first TB is never freed - it's the bidirectional DMA data */
156 for (i = 1; i < num_tbs; i++) {
157 if (meta->tbs & BIT(i))
158 dma_unmap_page(trans->dev,
159 le64_to_cpu(tfd->tbs[i].addr),
160 le16_to_cpu(tfd->tbs[i].tb_len),
163 dma_unmap_single(trans->dev,
164 le64_to_cpu(tfd->tbs[i].addr),
165 le16_to_cpu(tfd->tbs[i].tb_len),
172 static void iwl_pcie_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
174 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
176 /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
177 * idx is bounded by n_window
179 int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
181 lockdep_assert_held(&txq->lock);
183 iwl_pcie_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
184 iwl_pcie_get_tfd(trans_pcie, txq, idx));
190 skb = txq->entries[idx].skb;
192 /* Can be called from irqs-disabled context
193 * If skb is not NULL, it means that the whole queue is being
194 * freed and that the queue is not empty - free the skb
197 iwl_op_mode_free_skb(trans->op_mode, skb);
198 txq->entries[idx].skb = NULL;
203 static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
204 struct iwl_tfh_tfd *tfd, dma_addr_t addr,
207 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
208 int idx = iwl_pcie_gen2_get_num_tbs(trans, tfd);
209 struct iwl_tfh_tb *tb = &tfd->tbs[idx];
211 /* Each TFD can point to a maximum max_tbs Tx buffers */
212 if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) {
213 IWL_ERR(trans, "Error can not send more than %d chunks\n",
214 trans_pcie->max_tbs);
218 put_unaligned_le64(addr, &tb->addr);
219 tb->tb_len = cpu_to_le16(len);
221 tfd->num_tbs = cpu_to_le16(idx + 1);
226 static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
228 struct iwl_tfh_tfd *tfd, int start_len,
229 u8 hdr_len, struct iwl_device_cmd *dev_cmd)
232 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
233 struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
234 struct ieee80211_hdr *hdr = (void *)skb->data;
235 unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
236 unsigned int mss = skb_shinfo(skb)->gso_size;
237 u16 length, iv_len, amsdu_pad;
239 struct iwl_tso_hdr_page *hdr_page;
240 struct page **page_ptr;
243 /* if the packet is protected, then it must be CCMP or GCMP */
244 iv_len = ieee80211_has_protected(hdr->frame_control) ?
245 IEEE80211_CCMP_HDR_LEN : 0;
247 trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
248 &dev_cmd->hdr, start_len, 0);
250 ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
251 snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
252 total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
255 /* total amount of header we may need for this A-MSDU */
256 hdr_room = DIV_ROUND_UP(total_len, mss) *
257 (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
259 /* Our device supports 9 segments at most, it will fit in 1 page */
260 hdr_page = get_page_hdr(trans, hdr_room);
264 get_page(hdr_page->page);
265 start_hdr = hdr_page->pos;
266 page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
267 *page_ptr = hdr_page->page;
268 memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
269 hdr_page->pos += iv_len;
272 * Pull the ieee80211 header + IV to be able to use TSO core,
273 * we will restore it for the tx_status flow.
275 skb_pull(skb, hdr_len + iv_len);
278 * Remove the length of all the headers that we don't actually
279 * have in the MPDU by themselves, but that we duplicate into
280 * all the different MSDUs inside the A-MSDU.
282 le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
284 tso_start(skb, &tso);
287 /* this is the data left for this subframe */
288 unsigned int data_left = min_t(unsigned int, mss, total_len);
289 struct sk_buff *csum_skb = NULL;
292 u8 *subf_hdrs_start = hdr_page->pos;
294 total_len -= data_left;
296 memset(hdr_page->pos, 0, amsdu_pad);
297 hdr_page->pos += amsdu_pad;
298 amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
300 ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
301 hdr_page->pos += ETH_ALEN;
302 ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
303 hdr_page->pos += ETH_ALEN;
305 length = snap_ip_tcp_hdrlen + data_left;
306 *((__be16 *)hdr_page->pos) = cpu_to_be16(length);
307 hdr_page->pos += sizeof(length);
310 * This will copy the SNAP as well which will be considered
313 tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
315 hdr_page->pos += snap_ip_tcp_hdrlen;
317 tb_len = hdr_page->pos - start_hdr;
318 tb_phys = dma_map_single(trans->dev, start_hdr,
319 tb_len, DMA_TO_DEVICE);
320 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
321 dev_kfree_skb(csum_skb);
324 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
325 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr, tb_len);
326 /* add this subframe's headers' length to the tx_cmd */
327 le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
329 /* prepare the start_hdr for the next subframe */
330 start_hdr = hdr_page->pos;
332 /* put the payload */
334 tb_len = min_t(unsigned int, tso.size, data_left);
335 tb_phys = dma_map_single(trans->dev, tso.data,
336 tb_len, DMA_TO_DEVICE);
337 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
338 dev_kfree_skb(csum_skb);
341 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
342 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
346 tso_build_data(skb, &tso, tb_len);
350 /* re -add the WiFi header and IV */
351 skb_push(skb, hdr_len + iv_len);
361 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
363 struct iwl_device_cmd *dev_cmd,
365 struct iwl_cmd_meta *out_meta)
367 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
368 struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
369 int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
370 struct iwl_tfh_tfd *tfd =
371 iwl_pcie_get_tfd(trans_pcie, txq, idx);
374 int i, len, tb1_len, tb2_len, hdr_len;
377 memset(tfd, 0, sizeof(*tfd));
379 amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
380 (*ieee80211_get_qos_ctl(hdr) &
381 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
383 tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
384 /* The first TB points to bi-directional DMA data */
386 memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr,
389 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
391 /* there must be data left over for TB1 or this code must be changed */
392 BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
395 * The second TB (tb1) points to the remainder of the TX command
396 * and the 802.11 header - dword aligned size
397 * (This calculation modifies the TX command, so do it before the
398 * setup of the first TB)
400 len = sizeof(struct iwl_tx_cmd_gen2) + sizeof(struct iwl_cmd_header) +
401 ieee80211_hdrlen(hdr->frame_control) - IWL_FIRST_TB_SIZE;
403 /* do not align A-MSDU to dword as the subframe header aligns it */
407 tb1_len = ALIGN(len, 4);
409 /* map the data for TB1 */
410 tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
411 tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
412 if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
414 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
416 hdr_len = ieee80211_hdrlen(hdr->frame_control);
419 if (iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
420 tb1_len + IWL_FIRST_TB_SIZE,
425 * building the A-MSDU might have changed this data, so memcpy
428 memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr,
433 /* set up TFD's third entry to point to remainder of skb's head */
434 tb2_len = skb_headlen(skb) - hdr_len;
437 tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
438 tb2_len, DMA_TO_DEVICE);
439 if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
441 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
444 /* set up the remaining entries to point to the data */
445 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
446 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
449 if (!skb_frag_size(frag))
452 tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
453 skb_frag_size(frag), DMA_TO_DEVICE);
455 if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
457 tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
458 skb_frag_size(frag));
460 out_meta->tbs |= BIT(tb_idx);
463 trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
464 IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
465 trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
470 iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
474 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
475 struct iwl_device_cmd *dev_cmd, int txq_id)
477 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
478 struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
479 struct iwl_cmd_meta *out_meta;
480 struct iwl_txq *txq = trans_pcie->txq[txq_id];
484 if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
485 "TX on unused queue %d\n", txq_id))
488 if (skb_is_nonlinear(skb) &&
489 skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
490 __skb_linearize(skb))
493 spin_lock(&txq->lock);
495 idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
497 /* Set up driver data for this TFD */
498 txq->entries[idx].skb = skb;
499 txq->entries[idx].cmd = dev_cmd;
501 dev_cmd->hdr.sequence =
502 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
505 /* Set up first empty entry in queue's array of Tx/cmd buffers */
506 out_meta = &txq->entries[idx].meta;
509 tfd = iwl_pcie_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
511 spin_unlock(&txq->lock);
515 /* Set up entry for this TFD in Tx byte-count array */
516 iwl_pcie_gen2_update_byte_tbl(txq, le16_to_cpu(tx_cmd->len),
517 iwl_pcie_gen2_get_num_tbs(trans, tfd));
519 /* start timer if queue currently empty */
520 if (txq->read_ptr == txq->write_ptr) {
522 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
523 IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
524 iwl_trans_ref(trans);
527 /* Tell device the write index *just past* this latest filled TFD */
528 txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
529 iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
530 if (iwl_queue_space(txq) < txq->high_mark)
531 iwl_stop_queue(trans, txq);
534 * At this point the frame is "transmitted" successfully
535 * and we will get a TX status notification eventually.
537 spin_unlock(&txq->lock);
541 /*************** HOST COMMAND QUEUE FUNCTIONS *****/
544 * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command
545 * @priv: device private data point
546 * @cmd: a pointer to the ucode command structure
548 * The function returns < 0 values to indicate the operation
549 * failed. On success, it returns the index (>= 0) of command in the
552 static int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans,
553 struct iwl_host_cmd *cmd)
555 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
556 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
557 struct iwl_device_cmd *out_cmd;
558 struct iwl_cmd_meta *out_meta;
560 void *dup_buf = NULL;
561 dma_addr_t phys_addr;
562 int i, cmd_pos, idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
563 u16 copy_size, cmd_size, tb0_size;
564 bool had_nocopy = false;
565 u8 group_id = iwl_cmd_groupid(cmd->id);
566 const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
567 u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
568 struct iwl_tfh_tfd *tfd =
569 iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr);
571 memset(tfd, 0, sizeof(*tfd));
573 copy_size = sizeof(struct iwl_cmd_header_wide);
574 cmd_size = sizeof(struct iwl_cmd_header_wide);
576 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
577 cmddata[i] = cmd->data[i];
578 cmdlen[i] = cmd->len[i];
583 /* need at least IWL_FIRST_TB_SIZE copied */
584 if (copy_size < IWL_FIRST_TB_SIZE) {
585 int copy = IWL_FIRST_TB_SIZE - copy_size;
587 if (copy > cmdlen[i])
594 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
596 if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
600 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
602 * This is also a chunk that isn't copied
603 * to the static buffer so set had_nocopy.
607 /* only allowed once */
608 if (WARN_ON(dup_buf)) {
613 dup_buf = kmemdup(cmddata[i], cmdlen[i],
618 /* NOCOPY must not be followed by normal! */
619 if (WARN_ON(had_nocopy)) {
623 copy_size += cmdlen[i];
625 cmd_size += cmd->len[i];
629 * If any of the command structures end up being larger than the
630 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into
631 * separate TFDs, then we will need to increase the size of the buffers
633 if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
634 "Command %s (%#x) is too large (%d bytes)\n",
635 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) {
640 spin_lock_bh(&txq->lock);
642 if (iwl_queue_space(txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
643 spin_unlock_bh(&txq->lock);
645 IWL_ERR(trans, "No space in command queue\n");
646 iwl_op_mode_cmd_queue_full(trans->op_mode);
651 out_cmd = txq->entries[idx].cmd;
652 out_meta = &txq->entries[idx].meta;
654 /* re-initialize to NULL */
655 memset(out_meta, 0, sizeof(*out_meta));
656 if (cmd->flags & CMD_WANT_SKB)
657 out_meta->source = cmd;
659 /* set up the header */
660 out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
661 out_cmd->hdr_wide.group_id = group_id;
662 out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
663 out_cmd->hdr_wide.length =
664 cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
665 out_cmd->hdr_wide.reserved = 0;
666 out_cmd->hdr_wide.sequence =
667 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
668 INDEX_TO_SEQ(txq->write_ptr));
670 cmd_pos = sizeof(struct iwl_cmd_header_wide);
671 copy_size = sizeof(struct iwl_cmd_header_wide);
673 /* and copy the data that needs to be copied */
674 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
680 /* copy everything if not nocopy/dup */
681 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
682 IWL_HCMD_DFL_DUP))) {
685 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
692 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
693 * in total (for bi-directional DMA), but copy up to what
694 * we can fit into the payload for debug dump purposes.
696 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
698 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
701 /* However, treat copy_size the proper way, we need it below */
702 if (copy_size < IWL_FIRST_TB_SIZE) {
703 copy = IWL_FIRST_TB_SIZE - copy_size;
705 if (copy > cmd->len[i])
712 "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
713 iwl_get_cmd_string(trans, cmd->id), group_id,
714 out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
715 cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
717 /* start the TFD with the minimum copy bytes */
718 tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
719 memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
720 iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx),
723 /* map first command fragment, if any remains */
724 if (copy_size > tb0_size) {
725 phys_addr = dma_map_single(trans->dev,
726 ((u8 *)&out_cmd->hdr) + tb0_size,
727 copy_size - tb0_size,
729 if (dma_mapping_error(trans->dev, phys_addr)) {
731 iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
734 iwl_pcie_gen2_set_tb(trans, tfd, phys_addr,
735 copy_size - tb0_size);
738 /* map the remaining (adjusted) nocopy/dup fragments */
739 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
740 const void *data = cmddata[i];
744 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
747 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
749 phys_addr = dma_map_single(trans->dev, (void *)data,
750 cmdlen[i], DMA_TO_DEVICE);
751 if (dma_mapping_error(trans->dev, phys_addr)) {
753 iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
756 iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
759 BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
760 out_meta->flags = cmd->flags;
761 if (WARN_ON_ONCE(txq->entries[idx].free_buf))
762 kzfree(txq->entries[idx].free_buf);
763 txq->entries[idx].free_buf = dup_buf;
765 trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
767 /* start timer if queue currently empty */
768 if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
769 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
771 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
772 if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
773 !trans_pcie->ref_cmd_in_flight) {
774 trans_pcie->ref_cmd_in_flight = true;
775 IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
776 iwl_trans_ref(trans);
778 /* Increment and update queue's write index */
779 txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
780 iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
781 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
784 spin_unlock_bh(&txq->lock);
791 #define HOST_COMPLETE_TIMEOUT (2 * HZ)
793 static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans,
794 struct iwl_host_cmd *cmd)
796 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
797 const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
798 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
802 IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
804 if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
806 "Command %s: a command is already active!\n", cmd_str))
809 IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
811 if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
812 ret = wait_event_timeout(trans_pcie->d0i3_waitq,
813 pm_runtime_active(&trans_pcie->pci_dev->dev),
814 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
816 IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
821 cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
824 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
825 IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
830 ret = wait_event_timeout(trans_pcie->wait_command_queue,
831 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
833 HOST_COMPLETE_TIMEOUT);
835 IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
836 cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
838 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
839 txq->read_ptr, txq->write_ptr);
841 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
842 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
846 iwl_force_nmi(trans);
847 iwl_trans_fw_error(trans);
852 if (test_bit(STATUS_FW_ERROR, &trans->status)) {
853 IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
859 if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
860 test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
861 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
866 if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
867 IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
875 if (cmd->flags & CMD_WANT_SKB) {
877 * Cancel the CMD_WANT_SKB flag for the cmd in the
878 * TX cmd queue. Otherwise in case the cmd comes
879 * in later, it will possibly set an invalid
880 * address (cmd->meta.source).
882 txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
887 cmd->resp_pkt = NULL;
893 int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans,
894 struct iwl_host_cmd *cmd)
896 if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
897 test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
898 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
903 if (cmd->flags & CMD_ASYNC) {
906 /* An asynchronous command can not expect an SKB to be set. */
907 if (WARN_ON(cmd->flags & CMD_WANT_SKB))
910 ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
913 "Error sending %s: enqueue_hcmd failed: %d\n",
914 iwl_get_cmd_string(trans, cmd->id), ret);
920 return iwl_pcie_gen2_send_hcmd_sync(trans, cmd);
924 * iwl_pcie_gen2_txq_unmap - Unmap any remaining DMA mappings and free skb's
926 void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id)
928 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
929 struct iwl_txq *txq = trans_pcie->txq[txq_id];
931 spin_lock_bh(&txq->lock);
932 while (txq->write_ptr != txq->read_ptr) {
933 IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
934 txq_id, txq->read_ptr);
936 if (txq_id != trans_pcie->cmd_queue) {
937 int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
938 struct sk_buff *skb = txq->entries[idx].skb;
940 if (WARN_ON_ONCE(!skb))
943 iwl_pcie_free_tso_page(trans_pcie, skb);
945 iwl_pcie_gen2_free_tfd(trans, txq);
946 txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr);
948 if (txq->read_ptr == txq->write_ptr) {
951 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
952 if (txq_id != trans_pcie->cmd_queue) {
953 IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
955 iwl_trans_unref(trans);
956 } else if (trans_pcie->ref_cmd_in_flight) {
957 trans_pcie->ref_cmd_in_flight = false;
959 "clear ref_cmd_in_flight\n");
960 iwl_trans_unref(trans);
962 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
965 spin_unlock_bh(&txq->lock);
967 /* just in case - this queue may have been stopped */
968 iwl_wake_queue(trans, txq);
971 static void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans,
974 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
975 struct device *dev = trans->dev;
977 /* De-alloc circular buffer of TFDs */
979 dma_free_coherent(dev,
980 trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX,
981 txq->tfds, txq->dma_addr);
982 dma_free_coherent(dev,
983 sizeof(*txq->first_tb_bufs) * txq->n_window,
984 txq->first_tb_bufs, txq->first_tb_dma);
988 iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl);
993 * iwl_pcie_txq_free - Deallocate DMA queue.
994 * @txq: Transmit queue to deallocate.
996 * Empty queue by removing and destroying all BD's.
998 * 0-fill, but do not free "txq" descriptor structure.
1000 static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id)
1002 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1003 struct iwl_txq *txq = trans_pcie->txq[txq_id];
1009 iwl_pcie_gen2_txq_unmap(trans, txq_id);
1011 /* De-alloc array of command/tx buffers */
1012 if (txq_id == trans_pcie->cmd_queue)
1013 for (i = 0; i < txq->n_window; i++) {
1014 kzfree(txq->entries[i].cmd);
1015 kzfree(txq->entries[i].free_buf);
1017 del_timer_sync(&txq->stuck_timer);
1019 iwl_pcie_gen2_txq_free_memory(trans, txq);
1021 trans_pcie->txq[txq_id] = NULL;
1023 clear_bit(txq_id, trans_pcie->queue_used);
1026 int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
1027 struct iwl_tx_queue_cfg_cmd *cmd,
1029 unsigned int timeout)
1031 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1032 struct iwl_tx_queue_cfg_rsp *rsp;
1033 struct iwl_txq *txq;
1034 struct iwl_host_cmd hcmd = {
1036 .len = { sizeof(*cmd) },
1038 .flags = CMD_WANT_SKB,
1043 txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1046 ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl,
1047 sizeof(struct iwlagn_scd_bc_tbl));
1049 IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1054 ret = iwl_pcie_txq_alloc(trans, txq, TFD_TX_CMD_SLOTS, false);
1056 IWL_ERR(trans, "Tx queue alloc failed\n");
1059 ret = iwl_pcie_txq_init(trans, txq, TFD_TX_CMD_SLOTS, false);
1061 IWL_ERR(trans, "Tx queue init failed\n");
1065 txq->wd_timeout = msecs_to_jiffies(timeout);
1067 cmd->tfdq_addr = cpu_to_le64(txq->dma_addr);
1068 cmd->byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1069 cmd->cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(TFD_TX_CMD_SLOTS));
1071 ret = iwl_trans_send_cmd(trans, &hcmd);
1075 if (WARN_ON(iwl_rx_packet_payload_len(hcmd.resp_pkt) != sizeof(*rsp))) {
1077 goto error_free_resp;
1080 rsp = (void *)hcmd.resp_pkt->data;
1081 qid = le16_to_cpu(rsp->queue_number);
1082 wr_ptr = le16_to_cpu(rsp->write_pointer);
1084 if (qid >= ARRAY_SIZE(trans_pcie->txq)) {
1085 WARN_ONCE(1, "queue index %d unsupported", qid);
1087 goto error_free_resp;
1090 if (test_and_set_bit(qid, trans_pcie->queue_used)) {
1091 WARN_ONCE(1, "queue %d already used", qid);
1093 goto error_free_resp;
1097 trans_pcie->txq[qid] = txq;
1098 wr_ptr &= (TFD_QUEUE_SIZE_MAX - 1);
1100 /* Place first TFD at index corresponding to start sequence number */
1101 txq->read_ptr = wr_ptr;
1102 txq->write_ptr = wr_ptr;
1103 iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1104 (txq->write_ptr) | (qid << 16));
1105 IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1107 iwl_free_resp(&hcmd);
1111 iwl_free_resp(&hcmd);
1113 iwl_pcie_gen2_txq_free_memory(trans, txq);
1117 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue)
1119 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1122 * Upon HW Rfkill - we stop the device, and then stop the queues
1123 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1124 * allow the op_mode to call txq_disable after it already called
1127 if (!test_and_clear_bit(queue, trans_pcie->queue_used)) {
1128 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1129 "queue %d not used", queue);
1133 iwl_pcie_gen2_txq_unmap(trans, queue);
1135 IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1138 void iwl_pcie_gen2_tx_free(struct iwl_trans *trans)
1140 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1143 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
1145 /* Free all TX queues */
1146 for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) {
1147 if (!trans_pcie->txq[i])
1150 iwl_pcie_gen2_txq_free(trans, i);
1154 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans)
1156 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1157 struct iwl_txq *cmd_queue;
1158 int txq_id = trans_pcie->cmd_queue, ret;
1160 iwl_pcie_set_tx_cmd_queue_size(trans);
1162 /* alloc and init the command queue */
1163 if (!trans_pcie->txq[txq_id]) {
1164 cmd_queue = kzalloc(sizeof(*cmd_queue), GFP_KERNEL);
1166 IWL_ERR(trans, "Not enough memory for command queue\n");
1169 trans_pcie->txq[txq_id] = cmd_queue;
1170 ret = iwl_pcie_txq_alloc(trans, cmd_queue,
1171 trans_pcie->tx_cmd_queue_size, true);
1173 IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1177 cmd_queue = trans_pcie->txq[txq_id];
1180 ret = iwl_pcie_txq_init(trans, cmd_queue,
1181 trans_pcie->tx_cmd_queue_size, true);
1183 IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1186 trans_pcie->txq[txq_id]->id = txq_id;
1187 set_bit(txq_id, trans_pcie->queue_used);
1192 iwl_pcie_gen2_tx_free(trans);