1 /******************************************************************************
3 * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
4 * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
5 * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
7 * Portions of this file are derived from the ipw3945 project, as well
8 * as portions of the ieee80211 subsystem header files.
10 * This program is free software; you can redistribute it and/or modify it
11 * under the terms of version 2 of the GNU General Public License as
12 * published by the Free Software Foundation.
14 * This program is distributed in the hope that it will be useful, but WITHOUT
15 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
19 * You should have received a copy of the GNU General Public License along with
20 * this program; if not, write to the Free Software Foundation, Inc.,
21 * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA
23 * The full GNU General Public License is included in this distribution in the
24 * file called LICENSE.
26 * Contact Information:
27 * Intel Linux Wireless <linuxwifi@intel.com>
28 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
30 *****************************************************************************/
31 #include <linux/etherdevice.h>
32 #include <linux/ieee80211.h>
33 #include <linux/slab.h>
34 #include <linux/sched.h>
35 #include <linux/pm_runtime.h>
36 #include <net/ip6_checksum.h>
39 #include "iwl-debug.h"
44 #include "iwl-op-mode.h"
46 /* FIXME: need to abstract out TX command (once we know what it looks like) */
47 #include "dvm/commands.h"
49 #define IWL_TX_CRC_SIZE 4
50 #define IWL_TX_DELIMITER_SIZE 4
52 /*************** DMA-QUEUE-GENERAL-FUNCTIONS *****
57 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
58 * of buffer descriptors, each of which points to one or more data buffers for
59 * the device to read from or fill. Driver and device exchange status of each
60 * queue via "read" and "write" pointers. Driver keeps minimum of 2 empty
61 * entries in each circular buffer, to protect against confusing empty and full
64 * The device reads or writes the data in the queues via the device's several
65 * DMA/FIFO channels. Each queue is mapped to a single DMA channel.
67 * For Tx queue, there are low mark and high mark limits. If, after queuing
68 * the packet for Tx, free space become < low mark, Tx queue stopped. When
69 * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
72 ***************************************************/
74 int iwl_queue_space(const struct iwl_txq *q)
80 * To avoid ambiguity between empty and completely full queues, there
81 * should always be less than TFD_QUEUE_SIZE_MAX elements in the queue.
82 * If q->n_window is smaller than TFD_QUEUE_SIZE_MAX, there is no need
83 * to reserve any queue entries for this purpose.
85 if (q->n_window < TFD_QUEUE_SIZE_MAX)
88 max = TFD_QUEUE_SIZE_MAX - 1;
91 * TFD_QUEUE_SIZE_MAX is a power of 2, so the following is equivalent to
92 * modulo by TFD_QUEUE_SIZE_MAX and is well defined.
94 used = (q->write_ptr - q->read_ptr) & (TFD_QUEUE_SIZE_MAX - 1);
96 if (WARN_ON(used > max))
103 * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
105 static int iwl_queue_init(struct iwl_txq *q, int slots_num)
107 q->n_window = slots_num;
109 /* slots_num must be power-of-two size, otherwise
110 * get_cmd_index is broken. */
111 if (WARN_ON(!is_power_of_2(slots_num)))
114 q->low_mark = q->n_window / 4;
118 q->high_mark = q->n_window / 8;
119 if (q->high_mark < 2)
128 int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
129 struct iwl_dma_ptr *ptr, size_t size)
131 if (WARN_ON(ptr->addr))
134 ptr->addr = dma_alloc_coherent(trans->dev, size,
135 &ptr->dma, GFP_KERNEL);
142 void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr)
144 if (unlikely(!ptr->addr))
147 dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
148 memset(ptr, 0, sizeof(*ptr));
151 static void iwl_pcie_txq_stuck_timer(unsigned long data)
153 struct iwl_txq *txq = (void *)data;
154 struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
155 struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
157 spin_lock(&txq->lock);
158 /* check if triggered erroneously */
159 if (txq->read_ptr == txq->write_ptr) {
160 spin_unlock(&txq->lock);
163 spin_unlock(&txq->lock);
165 iwl_trans_pcie_log_scd_error(trans, txq);
167 iwl_force_nmi(trans);
171 * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
173 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
174 struct iwl_txq *txq, u16 byte_cnt,
177 struct iwlagn_scd_bc_tbl *scd_bc_tbl;
178 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
179 int write_ptr = txq->write_ptr;
180 int txq_id = txq->id;
182 u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
184 struct iwl_tx_cmd *tx_cmd =
185 (void *)txq->entries[txq->write_ptr].cmd->payload;
186 u8 sta_id = tx_cmd->sta_id;
188 scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
190 sec_ctl = tx_cmd->sec_ctl;
192 switch (sec_ctl & TX_CMD_SEC_MSK) {
194 len += IEEE80211_CCMP_MIC_LEN;
196 case TX_CMD_SEC_TKIP:
197 len += IEEE80211_TKIP_ICV_LEN;
200 len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
203 if (trans_pcie->bc_table_dword)
204 len = DIV_ROUND_UP(len, 4);
206 if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
209 bc_ent = cpu_to_le16(len | (sta_id << 12));
211 scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
213 if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
215 tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
218 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
221 struct iwl_trans_pcie *trans_pcie =
222 IWL_TRANS_GET_PCIE_TRANS(trans);
223 struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
224 int txq_id = txq->id;
225 int read_ptr = txq->read_ptr;
228 struct iwl_tx_cmd *tx_cmd =
229 (void *)txq->entries[read_ptr].cmd->payload;
231 WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
233 if (txq_id != trans_pcie->cmd_queue)
234 sta_id = tx_cmd->sta_id;
236 bc_ent = cpu_to_le16(1 | (sta_id << 12));
238 scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
240 if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
242 tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
246 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
248 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
251 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
253 int txq_id = txq->id;
255 lockdep_assert_held(&txq->lock);
258 * explicitly wake up the NIC if:
259 * 1. shadow registers aren't enabled
260 * 2. NIC is woken up for CMD regardless of shadow outside this function
261 * 3. there is a chance that the NIC is asleep
263 if (!trans->cfg->base_params->shadow_reg_enable &&
264 txq_id != trans_pcie->cmd_queue &&
265 test_bit(STATUS_TPOWER_PMI, &trans->status)) {
267 * wake up nic if it's powered down ...
268 * uCode will wake up, and interrupt us again, so next
269 * time we'll skip this part.
271 reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
273 if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
274 IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
276 iwl_set_bit(trans, CSR_GP_CNTRL,
277 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
278 txq->need_update = true;
284 * if not in power-save mode, uCode will never sleep when we're
285 * trying to tx (during RFKILL, we're not trying to tx).
287 IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr);
289 iwl_write32(trans, HBUS_TARG_WRPTR,
290 txq->write_ptr | (txq_id << 8));
293 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
295 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
298 for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) {
299 struct iwl_txq *txq = trans_pcie->txq[i];
301 spin_lock_bh(&txq->lock);
302 if (txq->need_update) {
303 iwl_pcie_txq_inc_wr_ptr(trans, txq);
304 txq->need_update = false;
306 spin_unlock_bh(&txq->lock);
310 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_trans *trans,
314 if (trans->cfg->use_tfh) {
315 struct iwl_tfh_tfd *tfd = _tfd;
316 struct iwl_tfh_tb *tb = &tfd->tbs[idx];
318 return (dma_addr_t)(le64_to_cpu(tb->addr));
320 struct iwl_tfd *tfd = _tfd;
321 struct iwl_tfd_tb *tb = &tfd->tbs[idx];
322 dma_addr_t addr = get_unaligned_le32(&tb->lo);
325 if (sizeof(dma_addr_t) <= sizeof(u32))
328 hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
331 * shift by 16 twice to avoid warnings on 32-bit
332 * (where this code never runs anyway due to the
333 * if statement above)
335 return addr | ((hi_len << 16) << 16);
339 static inline void iwl_pcie_tfd_set_tb(struct iwl_trans *trans, void *tfd,
340 u8 idx, dma_addr_t addr, u16 len)
342 struct iwl_tfd *tfd_fh = (void *)tfd;
343 struct iwl_tfd_tb *tb = &tfd_fh->tbs[idx];
345 u16 hi_n_len = len << 4;
347 put_unaligned_le32(addr, &tb->lo);
348 hi_n_len |= iwl_get_dma_hi_addr(addr);
350 tb->hi_n_len = cpu_to_le16(hi_n_len);
352 tfd_fh->num_tbs = idx + 1;
355 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_trans *trans, void *_tfd)
357 if (trans->cfg->use_tfh) {
358 struct iwl_tfh_tfd *tfd = _tfd;
360 return le16_to_cpu(tfd->num_tbs) & 0x1f;
362 struct iwl_tfd *tfd = _tfd;
364 return tfd->num_tbs & 0x1f;
368 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
369 struct iwl_cmd_meta *meta,
370 struct iwl_txq *txq, int index)
372 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
374 void *tfd = iwl_pcie_get_tfd(trans_pcie, txq, index);
376 /* Sanity check on number of chunks */
377 num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
379 if (num_tbs >= trans_pcie->max_tbs) {
380 IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
381 /* @todo issue fatal error, it is quite serious situation */
385 /* first TB is never freed - it's the bidirectional DMA data */
387 for (i = 1; i < num_tbs; i++) {
388 if (meta->tbs & BIT(i))
389 dma_unmap_page(trans->dev,
390 iwl_pcie_tfd_tb_get_addr(trans, tfd, i),
391 iwl_pcie_tfd_tb_get_len(trans, tfd, i),
394 dma_unmap_single(trans->dev,
395 iwl_pcie_tfd_tb_get_addr(trans, tfd,
397 iwl_pcie_tfd_tb_get_len(trans, tfd,
402 if (trans->cfg->use_tfh) {
403 struct iwl_tfh_tfd *tfd_fh = (void *)tfd;
407 struct iwl_tfd *tfd_fh = (void *)tfd;
415 * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
416 * @trans - transport private data
418 * @dma_dir - the direction of the DMA mapping
420 * Does NOT advance any TFD circular buffer read/write indexes
421 * Does NOT free the TFD itself (which is within circular buffer)
423 void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
425 /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
426 * idx is bounded by n_window
428 int rd_ptr = txq->read_ptr;
429 int idx = get_cmd_index(txq, rd_ptr);
431 lockdep_assert_held(&txq->lock);
433 /* We have only q->n_window txq->entries, but we use
434 * TFD_QUEUE_SIZE_MAX tfds
436 iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, txq, rd_ptr);
442 skb = txq->entries[idx].skb;
444 /* Can be called from irqs-disabled context
445 * If skb is not NULL, it means that the whole queue is being
446 * freed and that the queue is not empty - free the skb
449 iwl_op_mode_free_skb(trans->op_mode, skb);
450 txq->entries[idx].skb = NULL;
455 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
456 dma_addr_t addr, u16 len, bool reset)
458 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
462 tfd = txq->tfds + trans_pcie->tfd_size * txq->write_ptr;
465 memset(tfd, 0, trans_pcie->tfd_size);
467 num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
469 /* Each TFD can point to a maximum max_tbs Tx buffers */
470 if (num_tbs >= trans_pcie->max_tbs) {
471 IWL_ERR(trans, "Error can not send more than %d chunks\n",
472 trans_pcie->max_tbs);
476 if (WARN(addr & ~IWL_TX_DMA_MASK,
477 "Unaligned address = %llx\n", (unsigned long long)addr))
480 iwl_pcie_tfd_set_tb(trans, tfd, num_tbs, addr, len);
485 int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq,
486 int slots_num, bool cmd_queue)
488 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
489 size_t tfd_sz = trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX;
493 if (WARN_ON(txq->entries || txq->tfds))
496 setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer,
498 txq->trans_pcie = trans_pcie;
500 txq->n_window = slots_num;
502 txq->entries = kcalloc(slots_num,
503 sizeof(struct iwl_pcie_txq_entry),
510 for (i = 0; i < slots_num; i++) {
511 txq->entries[i].cmd =
512 kmalloc(sizeof(struct iwl_device_cmd),
514 if (!txq->entries[i].cmd)
518 /* Circular buffer of transmit frame descriptors (TFDs),
519 * shared with device */
520 txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
521 &txq->dma_addr, GFP_KERNEL);
525 BUILD_BUG_ON(IWL_FIRST_TB_SIZE_ALIGN != sizeof(*txq->first_tb_bufs));
527 tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
529 txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
532 if (!txq->first_tb_bufs)
537 dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
539 if (txq->entries && cmd_queue)
540 for (i = 0; i < slots_num; i++)
541 kfree(txq->entries[i].cmd);
549 int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
550 int slots_num, bool cmd_queue)
554 txq->need_update = false;
556 /* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise
557 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
558 BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1));
560 /* Initialize queue's high/low-water marks, and head/tail indexes */
561 ret = iwl_queue_init(txq, slots_num);
565 spin_lock_init(&txq->lock);
568 static struct lock_class_key iwl_pcie_cmd_queue_lock_class;
570 lockdep_set_class(&txq->lock, &iwl_pcie_cmd_queue_lock_class);
573 __skb_queue_head_init(&txq->overflow_q);
578 static void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
581 struct page **page_ptr;
583 page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
586 __free_page(*page_ptr);
591 static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
593 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
595 lockdep_assert_held(&trans_pcie->reg_lock);
597 if (trans_pcie->ref_cmd_in_flight) {
598 trans_pcie->ref_cmd_in_flight = false;
599 IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n");
600 iwl_trans_unref(trans);
603 if (!trans->cfg->base_params->apmg_wake_up_wa)
605 if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
608 trans_pcie->cmd_hold_nic_awake = false;
609 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
610 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
614 * iwl_pcie_txq_unmap - Unmap any remaining DMA mappings and free skb's
616 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
618 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
619 struct iwl_txq *txq = trans_pcie->txq[txq_id];
621 spin_lock_bh(&txq->lock);
622 while (txq->write_ptr != txq->read_ptr) {
623 IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
624 txq_id, txq->read_ptr);
626 if (txq_id != trans_pcie->cmd_queue) {
627 struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
629 if (WARN_ON_ONCE(!skb))
632 iwl_pcie_free_tso_page(trans_pcie, skb);
634 iwl_pcie_txq_free_tfd(trans, txq);
635 txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr);
637 if (txq->read_ptr == txq->write_ptr) {
640 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
641 if (txq_id != trans_pcie->cmd_queue) {
642 IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
644 iwl_trans_unref(trans);
646 iwl_pcie_clear_cmd_in_flight(trans);
648 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
652 while (!skb_queue_empty(&txq->overflow_q)) {
653 struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
655 iwl_op_mode_free_skb(trans->op_mode, skb);
658 spin_unlock_bh(&txq->lock);
660 /* just in case - this queue may have been stopped */
661 iwl_wake_queue(trans, txq);
665 * iwl_pcie_txq_free - Deallocate DMA queue.
666 * @txq: Transmit queue to deallocate.
668 * Empty queue by removing and destroying all BD's.
670 * 0-fill, but do not free "txq" descriptor structure.
672 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
674 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
675 struct iwl_txq *txq = trans_pcie->txq[txq_id];
676 struct device *dev = trans->dev;
682 iwl_pcie_txq_unmap(trans, txq_id);
684 /* De-alloc array of command/tx buffers */
685 if (txq_id == trans_pcie->cmd_queue)
686 for (i = 0; i < txq->n_window; i++) {
687 kzfree(txq->entries[i].cmd);
688 kzfree(txq->entries[i].free_buf);
691 /* De-alloc circular buffer of TFDs */
693 dma_free_coherent(dev,
694 trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX,
695 txq->tfds, txq->dma_addr);
699 dma_free_coherent(dev,
700 sizeof(*txq->first_tb_bufs) * txq->n_window,
701 txq->first_tb_bufs, txq->first_tb_dma);
707 del_timer_sync(&txq->stuck_timer);
709 /* 0-fill queue descriptor structure */
710 memset(txq, 0, sizeof(*txq));
713 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
715 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
716 int nq = trans->cfg->base_params->num_of_queues;
719 int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
720 SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
722 /* make sure all queue are not stopped/used */
723 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
724 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
726 trans_pcie->scd_base_addr =
727 iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
729 WARN_ON(scd_base_addr != 0 &&
730 scd_base_addr != trans_pcie->scd_base_addr);
732 /* reset context data, TX status and translation data */
733 iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
734 SCD_CONTEXT_MEM_LOWER_BOUND,
737 iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
738 trans_pcie->scd_bc_tbls.dma >> 10);
740 /* The chain extension of the SCD doesn't work well. This feature is
741 * enabled by default by the HW, so we need to disable it manually.
743 if (trans->cfg->base_params->scd_chain_ext_wa)
744 iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
746 iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
747 trans_pcie->cmd_fifo,
748 trans_pcie->cmd_q_wdg_timeout);
750 /* Activate all Tx DMA/FIFO channels */
751 iwl_scd_activate_fifos(trans);
753 /* Enable DMA channel */
754 for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
755 iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
756 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
757 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
759 /* Update FH chicken bits */
760 reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
761 iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
762 reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
764 /* Enable L1-Active */
765 if (trans->cfg->device_family < IWL_DEVICE_FAMILY_8000)
766 iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
767 APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
770 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
772 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
776 * we should never get here in gen2 trans mode return early to avoid
777 * having invalid accesses
779 if (WARN_ON_ONCE(trans->cfg->gen2))
782 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
784 struct iwl_txq *txq = trans_pcie->txq[txq_id];
785 if (trans->cfg->use_tfh)
786 iwl_write_direct64(trans,
787 FH_MEM_CBBC_QUEUE(trans, txq_id),
790 iwl_write_direct32(trans,
791 FH_MEM_CBBC_QUEUE(trans, txq_id),
793 iwl_pcie_txq_unmap(trans, txq_id);
798 /* Tell NIC where to find the "keep warm" buffer */
799 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
800 trans_pcie->kw.dma >> 4);
803 * Send 0 as the scd_base_addr since the device may have be reset
804 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
807 iwl_pcie_tx_start(trans, 0);
810 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
812 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
817 spin_lock(&trans_pcie->irq_lock);
819 if (!iwl_trans_grab_nic_access(trans, &flags))
822 /* Stop each Tx DMA channel */
823 for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
824 iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
825 mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
828 /* Wait for DMA channels to be idle */
829 ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
832 "Failing on timeout while stopping DMA channel %d [0x%08x]\n",
833 ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
835 iwl_trans_release_nic_access(trans, &flags);
838 spin_unlock(&trans_pcie->irq_lock);
842 * iwl_pcie_tx_stop - Stop all Tx DMA channels
844 int iwl_pcie_tx_stop(struct iwl_trans *trans)
846 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
849 /* Turn off all Tx DMA fifos */
850 iwl_scd_deactivate_fifos(trans);
852 /* Turn off all Tx DMA channels */
853 iwl_pcie_tx_stop_fh(trans);
856 * This function can be called before the op_mode disabled the
857 * queues. This happens when we have an rfkill interrupt.
858 * Since we stop Tx altogether - mark the queues as stopped.
860 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
861 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
863 /* This can happen: start_hw, stop_device */
864 if (!trans_pcie->txq_memory)
867 /* Unmap DMA from host system and free skb's */
868 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
870 iwl_pcie_txq_unmap(trans, txq_id);
876 * iwl_trans_tx_free - Free TXQ Context
878 * Destroy all TX DMA queues and structures
880 void iwl_pcie_tx_free(struct iwl_trans *trans)
883 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
885 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
888 if (trans_pcie->txq_memory) {
890 txq_id < trans->cfg->base_params->num_of_queues;
892 iwl_pcie_txq_free(trans, txq_id);
893 trans_pcie->txq[txq_id] = NULL;
897 kfree(trans_pcie->txq_memory);
898 trans_pcie->txq_memory = NULL;
900 iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
902 iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
906 * iwl_pcie_tx_alloc - allocate TX context
907 * Allocate all Tx DMA structures and initialize them
909 static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
912 int txq_id, slots_num;
913 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
915 u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues *
916 sizeof(struct iwlagn_scd_bc_tbl);
918 /*It is not allowed to alloc twice, so warn when this happens.
919 * We cannot rely on the previous allocation, so free and fail */
920 if (WARN_ON(trans_pcie->txq_memory)) {
925 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
928 IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
932 /* Alloc keep-warm buffer */
933 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
935 IWL_ERR(trans, "Keep Warm allocation failed\n");
939 trans_pcie->txq_memory = kcalloc(trans->cfg->base_params->num_of_queues,
940 sizeof(struct iwl_txq), GFP_KERNEL);
941 if (!trans_pcie->txq_memory) {
942 IWL_ERR(trans, "Not enough memory for txq\n");
947 /* Alloc and init all Tx queues, including the command queue (#4/#9) */
948 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
950 bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
952 slots_num = cmd_queue ? TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
953 trans_pcie->txq[txq_id] = &trans_pcie->txq_memory[txq_id];
954 ret = iwl_pcie_txq_alloc(trans, trans_pcie->txq[txq_id],
955 slots_num, cmd_queue);
957 IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
960 trans_pcie->txq[txq_id]->id = txq_id;
966 iwl_pcie_tx_free(trans);
971 int iwl_pcie_tx_init(struct iwl_trans *trans)
973 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
975 int txq_id, slots_num;
978 if (!trans_pcie->txq_memory) {
979 ret = iwl_pcie_tx_alloc(trans);
985 spin_lock(&trans_pcie->irq_lock);
987 /* Turn off all Tx DMA fifos */
988 iwl_scd_deactivate_fifos(trans);
990 /* Tell NIC where to find the "keep warm" buffer */
991 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
992 trans_pcie->kw.dma >> 4);
994 spin_unlock(&trans_pcie->irq_lock);
996 /* Alloc and init all Tx queues, including the command queue (#4/#9) */
997 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues;
999 bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
1001 slots_num = cmd_queue ? TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS;
1002 ret = iwl_pcie_txq_init(trans, trans_pcie->txq[txq_id],
1003 slots_num, cmd_queue);
1005 IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1010 * Tell nic where to find circular buffer of TFDs for a
1011 * given Tx queue, and enable the DMA channel used for that
1013 * Circular buffer (TFD queue in DRAM) physical base address
1015 iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
1016 trans_pcie->txq[txq_id]->dma_addr >> 8);
1019 iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
1020 if (trans->cfg->base_params->num_of_queues > 20)
1021 iwl_set_bits_prph(trans, SCD_GP_CTRL,
1022 SCD_GP_CTRL_ENABLE_31_QUEUES);
1026 /*Upon error, free only if we allocated something */
1028 iwl_pcie_tx_free(trans);
1032 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
1034 lockdep_assert_held(&txq->lock);
1036 if (!txq->wd_timeout)
1040 * station is asleep and we send data - that must
1041 * be uAPSD or PS-Poll. Don't rearm the timer.
1047 * if empty delete timer, otherwise move timer forward
1048 * since we're making progress on this queue
1050 if (txq->read_ptr == txq->write_ptr)
1051 del_timer(&txq->stuck_timer);
1053 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1056 /* Frees buffers until index _not_ inclusive */
1057 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1058 struct sk_buff_head *skbs)
1060 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1061 struct iwl_txq *txq = trans_pcie->txq[txq_id];
1062 int tfd_num = ssn & (TFD_QUEUE_SIZE_MAX - 1);
1065 /* This function is not meant to release cmd queue*/
1066 if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1069 spin_lock_bh(&txq->lock);
1071 if (!test_bit(txq_id, trans_pcie->queue_used)) {
1072 IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1077 if (txq->read_ptr == tfd_num)
1080 IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1081 txq_id, txq->read_ptr, tfd_num, ssn);
1083 /*Since we free until index _not_ inclusive, the one before index is
1084 * the last we will free. This one must be used */
1085 last_to_free = iwl_queue_dec_wrap(tfd_num);
1087 if (!iwl_queue_used(txq, last_to_free)) {
1089 "%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1090 __func__, txq_id, last_to_free, TFD_QUEUE_SIZE_MAX,
1091 txq->write_ptr, txq->read_ptr);
1095 if (WARN_ON(!skb_queue_empty(skbs)))
1099 txq->read_ptr != tfd_num;
1100 txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr)) {
1101 struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
1103 if (WARN_ON_ONCE(!skb))
1106 iwl_pcie_free_tso_page(trans_pcie, skb);
1108 __skb_queue_tail(skbs, skb);
1110 txq->entries[txq->read_ptr].skb = NULL;
1112 if (!trans->cfg->use_tfh)
1113 iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1115 iwl_pcie_txq_free_tfd(trans, txq);
1118 iwl_pcie_txq_progress(txq);
1120 if (iwl_queue_space(txq) > txq->low_mark &&
1121 test_bit(txq_id, trans_pcie->queue_stopped)) {
1122 struct sk_buff_head overflow_skbs;
1124 __skb_queue_head_init(&overflow_skbs);
1125 skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1128 * This is tricky: we are in reclaim path which is non
1129 * re-entrant, so noone will try to take the access the
1130 * txq data from that path. We stopped tx, so we can't
1131 * have tx as well. Bottom line, we can unlock and re-lock
1134 spin_unlock_bh(&txq->lock);
1136 while (!skb_queue_empty(&overflow_skbs)) {
1137 struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1138 struct iwl_device_cmd *dev_cmd_ptr;
1140 dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1141 trans_pcie->dev_cmd_offs);
1144 * Note that we can very well be overflowing again.
1145 * In that case, iwl_queue_space will be small again
1146 * and we won't wake mac80211's queue.
1148 iwl_trans_pcie_tx(trans, skb, dev_cmd_ptr, txq_id);
1150 spin_lock_bh(&txq->lock);
1152 if (iwl_queue_space(txq) > txq->low_mark)
1153 iwl_wake_queue(trans, txq);
1156 if (txq->read_ptr == txq->write_ptr) {
1157 IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", txq->id);
1158 iwl_trans_unref(trans);
1162 spin_unlock_bh(&txq->lock);
1165 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1166 const struct iwl_host_cmd *cmd)
1168 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1171 lockdep_assert_held(&trans_pcie->reg_lock);
1173 if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
1174 !trans_pcie->ref_cmd_in_flight) {
1175 trans_pcie->ref_cmd_in_flight = true;
1176 IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
1177 iwl_trans_ref(trans);
1181 * wake up the NIC to make sure that the firmware will see the host
1182 * command - we will let the NIC sleep once all the host commands
1183 * returned. This needs to be done only on NICs that have
1184 * apmg_wake_up_wa set.
1186 if (trans->cfg->base_params->apmg_wake_up_wa &&
1187 !trans_pcie->cmd_hold_nic_awake) {
1188 __iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1189 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1191 ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1192 CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
1193 (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
1194 CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1197 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1198 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
1199 IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1202 trans_pcie->cmd_hold_nic_awake = true;
1209 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1211 * When FW advances 'R' index, all entries between old and new 'R' index
1212 * need to be reclaimed. As result, some free space forms. If there is
1213 * enough free space (> low mark), wake the stack that feeds us.
1215 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1217 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1218 struct iwl_txq *txq = trans_pcie->txq[txq_id];
1219 unsigned long flags;
1222 lockdep_assert_held(&txq->lock);
1224 if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(txq, idx))) {
1226 "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1227 __func__, txq_id, idx, TFD_QUEUE_SIZE_MAX,
1228 txq->write_ptr, txq->read_ptr);
1232 for (idx = iwl_queue_inc_wrap(idx); txq->read_ptr != idx;
1233 txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr)) {
1236 IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1237 idx, txq->write_ptr, txq->read_ptr);
1238 iwl_force_nmi(trans);
1242 if (txq->read_ptr == txq->write_ptr) {
1243 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1244 iwl_pcie_clear_cmd_in_flight(trans);
1245 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1248 iwl_pcie_txq_progress(txq);
1251 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1254 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1259 scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1261 tbl_dw_addr = trans_pcie->scd_base_addr +
1262 SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1264 tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1267 tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1269 tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1271 iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1276 /* Receiver address (actually, Rx station's index into station table),
1277 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1278 #define BUILD_RAxTID(sta_id, tid) (((sta_id) << 4) + (tid))
1280 bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1281 const struct iwl_trans_txq_scd_cfg *cfg,
1282 unsigned int wdg_timeout)
1284 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1285 struct iwl_txq *txq = trans_pcie->txq[txq_id];
1287 bool scd_bug = false;
1289 if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1290 WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1292 txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1297 /* Disable the scheduler prior configuring the cmd queue */
1298 if (txq_id == trans_pcie->cmd_queue &&
1299 trans_pcie->scd_set_active)
1300 iwl_scd_enable_set_active(trans, 0);
1302 /* Stop this Tx queue before configuring it */
1303 iwl_scd_txq_set_inactive(trans, txq_id);
1305 /* Set this queue as a chain-building queue unless it is CMD */
1306 if (txq_id != trans_pcie->cmd_queue)
1307 iwl_scd_txq_set_chain(trans, txq_id);
1309 if (cfg->aggregate) {
1310 u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1312 /* Map receiver-address / traffic-ID to this queue */
1313 iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1315 /* enable aggregations for the queue */
1316 iwl_scd_txq_enable_agg(trans, txq_id);
1320 * disable aggregations for the queue, this will also
1321 * make the ra_tid mapping configuration irrelevant
1322 * since it is now a non-AGG queue.
1324 iwl_scd_txq_disable_agg(trans, txq_id);
1326 ssn = txq->read_ptr;
1330 * If we need to move the SCD write pointer by steps of
1331 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let
1332 * the op_mode know by returning true later.
1333 * Do this only in case cfg is NULL since this trick can
1334 * be done only if we have DQA enabled which is true for mvm
1335 * only. And mvm never sets a cfg pointer.
1336 * This is really ugly, but this is the easiest way out for
1337 * this sad hardware issue.
1338 * This bug has been fixed on devices 9000 and up.
1340 scd_bug = !trans->cfg->mq_rx_supported &&
1341 !((ssn - txq->write_ptr) & 0x3f) &&
1342 (ssn != txq->write_ptr);
1347 /* Place first TFD at index corresponding to start sequence number.
1348 * Assumes that ssn_idx is valid (!= 0xFFF) */
1349 txq->read_ptr = (ssn & 0xff);
1350 txq->write_ptr = (ssn & 0xff);
1351 iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1352 (ssn & 0xff) | (txq_id << 8));
1355 u8 frame_limit = cfg->frame_limit;
1357 iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1359 /* Set up Tx window size and frame limit for this queue */
1360 iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1361 SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1362 iwl_trans_write_mem32(trans,
1363 trans_pcie->scd_base_addr +
1364 SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1365 SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) |
1366 SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit));
1368 /* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1369 iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1370 (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1371 (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1372 (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1373 SCD_QUEUE_STTS_REG_MSK);
1375 /* enable the scheduler for this queue (only) */
1376 if (txq_id == trans_pcie->cmd_queue &&
1377 trans_pcie->scd_set_active)
1378 iwl_scd_enable_set_active(trans, BIT(txq_id));
1380 IWL_DEBUG_TX_QUEUES(trans,
1381 "Activate queue %d on FIFO %d WrPtr: %d\n",
1382 txq_id, fifo, ssn & 0xff);
1384 IWL_DEBUG_TX_QUEUES(trans,
1385 "Activate queue %d WrPtr: %d\n",
1386 txq_id, ssn & 0xff);
1392 void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
1395 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1396 struct iwl_txq *txq = trans_pcie->txq[txq_id];
1398 txq->ampdu = !shared_mode;
1401 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1404 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1405 u32 stts_addr = trans_pcie->scd_base_addr +
1406 SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1407 static const u32 zero_val[4] = {};
1409 trans_pcie->txq[txq_id]->frozen_expiry_remainder = 0;
1410 trans_pcie->txq[txq_id]->frozen = false;
1413 * Upon HW Rfkill - we stop the device, and then stop the queues
1414 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1415 * allow the op_mode to call txq_disable after it already called
1418 if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1419 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1420 "queue %d not used", txq_id);
1424 if (configure_scd) {
1425 iwl_scd_txq_set_inactive(trans, txq_id);
1427 iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1428 ARRAY_SIZE(zero_val));
1431 iwl_pcie_txq_unmap(trans, txq_id);
1432 trans_pcie->txq[txq_id]->ampdu = false;
1434 IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1437 /*************** HOST COMMAND QUEUE FUNCTIONS *****/
1440 * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1441 * @priv: device private data point
1442 * @cmd: a pointer to the ucode command structure
1444 * The function returns < 0 values to indicate the operation
1445 * failed. On success, it returns the index (>= 0) of command in the
1448 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1449 struct iwl_host_cmd *cmd)
1451 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1452 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1453 struct iwl_device_cmd *out_cmd;
1454 struct iwl_cmd_meta *out_meta;
1455 unsigned long flags;
1456 void *dup_buf = NULL;
1457 dma_addr_t phys_addr;
1459 u16 copy_size, cmd_size, tb0_size;
1460 bool had_nocopy = false;
1461 u8 group_id = iwl_cmd_groupid(cmd->id);
1464 const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1465 u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
1467 if (WARN(!trans->wide_cmd_header &&
1468 group_id > IWL_ALWAYS_LONG_GROUP,
1469 "unsupported wide command %#x\n", cmd->id))
1472 if (group_id != 0) {
1473 copy_size = sizeof(struct iwl_cmd_header_wide);
1474 cmd_size = sizeof(struct iwl_cmd_header_wide);
1476 copy_size = sizeof(struct iwl_cmd_header);
1477 cmd_size = sizeof(struct iwl_cmd_header);
1480 /* need one for the header if the first is NOCOPY */
1481 BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1483 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1484 cmddata[i] = cmd->data[i];
1485 cmdlen[i] = cmd->len[i];
1490 /* need at least IWL_FIRST_TB_SIZE copied */
1491 if (copy_size < IWL_FIRST_TB_SIZE) {
1492 int copy = IWL_FIRST_TB_SIZE - copy_size;
1494 if (copy > cmdlen[i])
1501 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1503 if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1507 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1509 * This is also a chunk that isn't copied
1510 * to the static buffer so set had_nocopy.
1514 /* only allowed once */
1515 if (WARN_ON(dup_buf)) {
1520 dup_buf = kmemdup(cmddata[i], cmdlen[i],
1525 /* NOCOPY must not be followed by normal! */
1526 if (WARN_ON(had_nocopy)) {
1530 copy_size += cmdlen[i];
1532 cmd_size += cmd->len[i];
1536 * If any of the command structures end up being larger than
1537 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1538 * allocated into separate TFDs, then we will need to
1539 * increase the size of the buffers.
1541 if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1542 "Command %s (%#x) is too large (%d bytes)\n",
1543 iwl_get_cmd_string(trans, cmd->id),
1544 cmd->id, copy_size)) {
1549 spin_lock_bh(&txq->lock);
1551 if (iwl_queue_space(txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1552 spin_unlock_bh(&txq->lock);
1554 IWL_ERR(trans, "No space in command queue\n");
1555 iwl_op_mode_cmd_queue_full(trans->op_mode);
1560 idx = get_cmd_index(txq, txq->write_ptr);
1561 out_cmd = txq->entries[idx].cmd;
1562 out_meta = &txq->entries[idx].meta;
1564 memset(out_meta, 0, sizeof(*out_meta)); /* re-initialize to NULL */
1565 if (cmd->flags & CMD_WANT_SKB)
1566 out_meta->source = cmd;
1568 /* set up the header */
1569 if (group_id != 0) {
1570 out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1571 out_cmd->hdr_wide.group_id = group_id;
1572 out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1573 out_cmd->hdr_wide.length =
1574 cpu_to_le16(cmd_size -
1575 sizeof(struct iwl_cmd_header_wide));
1576 out_cmd->hdr_wide.reserved = 0;
1577 out_cmd->hdr_wide.sequence =
1578 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1579 INDEX_TO_SEQ(txq->write_ptr));
1581 cmd_pos = sizeof(struct iwl_cmd_header_wide);
1582 copy_size = sizeof(struct iwl_cmd_header_wide);
1584 out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1585 out_cmd->hdr.sequence =
1586 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1587 INDEX_TO_SEQ(txq->write_ptr));
1588 out_cmd->hdr.group_id = 0;
1590 cmd_pos = sizeof(struct iwl_cmd_header);
1591 copy_size = sizeof(struct iwl_cmd_header);
1594 /* and copy the data that needs to be copied */
1595 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1601 /* copy everything if not nocopy/dup */
1602 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1603 IWL_HCMD_DFL_DUP))) {
1606 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1613 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1614 * in total (for bi-directional DMA), but copy up to what
1615 * we can fit into the payload for debug dump purposes.
1617 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1619 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1622 /* However, treat copy_size the proper way, we need it below */
1623 if (copy_size < IWL_FIRST_TB_SIZE) {
1624 copy = IWL_FIRST_TB_SIZE - copy_size;
1626 if (copy > cmd->len[i])
1633 "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1634 iwl_get_cmd_string(trans, cmd->id),
1635 group_id, out_cmd->hdr.cmd,
1636 le16_to_cpu(out_cmd->hdr.sequence),
1637 cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
1639 /* start the TFD with the minimum copy bytes */
1640 tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1641 memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1642 iwl_pcie_txq_build_tfd(trans, txq,
1643 iwl_pcie_get_first_tb_dma(txq, idx),
1646 /* map first command fragment, if any remains */
1647 if (copy_size > tb0_size) {
1648 phys_addr = dma_map_single(trans->dev,
1649 ((u8 *)&out_cmd->hdr) + tb0_size,
1650 copy_size - tb0_size,
1652 if (dma_mapping_error(trans->dev, phys_addr)) {
1653 iwl_pcie_tfd_unmap(trans, out_meta, txq,
1659 iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1660 copy_size - tb0_size, false);
1663 /* map the remaining (adjusted) nocopy/dup fragments */
1664 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1665 const void *data = cmddata[i];
1669 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1672 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1674 phys_addr = dma_map_single(trans->dev, (void *)data,
1675 cmdlen[i], DMA_TO_DEVICE);
1676 if (dma_mapping_error(trans->dev, phys_addr)) {
1677 iwl_pcie_tfd_unmap(trans, out_meta, txq,
1683 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1686 BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
1687 out_meta->flags = cmd->flags;
1688 if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1689 kzfree(txq->entries[idx].free_buf);
1690 txq->entries[idx].free_buf = dup_buf;
1692 trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1694 /* start timer if queue currently empty */
1695 if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
1696 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1698 spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1699 ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1702 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1706 /* Increment and update queue's write index */
1707 txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
1708 iwl_pcie_txq_inc_wr_ptr(trans, txq);
1710 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1713 spin_unlock_bh(&txq->lock);
1721 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1722 * @rxb: Rx buffer to reclaim
1724 void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1725 struct iwl_rx_cmd_buffer *rxb)
1727 struct iwl_rx_packet *pkt = rxb_addr(rxb);
1728 u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1731 int txq_id = SEQ_TO_QUEUE(sequence);
1732 int index = SEQ_TO_INDEX(sequence);
1734 struct iwl_device_cmd *cmd;
1735 struct iwl_cmd_meta *meta;
1736 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1737 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1739 /* If a Tx command is being handled and it isn't in the actual
1740 * command queue then there a command routing bug has been introduced
1741 * in the queue management code. */
1742 if (WARN(txq_id != trans_pcie->cmd_queue,
1743 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1744 txq_id, trans_pcie->cmd_queue, sequence, txq->read_ptr,
1746 iwl_print_hex_error(trans, pkt, 32);
1750 spin_lock_bh(&txq->lock);
1752 cmd_index = get_cmd_index(txq, index);
1753 cmd = txq->entries[cmd_index].cmd;
1754 meta = &txq->entries[cmd_index].meta;
1755 group_id = cmd->hdr.group_id;
1756 cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1758 iwl_pcie_tfd_unmap(trans, meta, txq, index);
1760 /* Input error checking is done when commands are added to queue. */
1761 if (meta->flags & CMD_WANT_SKB) {
1762 struct page *p = rxb_steal_page(rxb);
1764 meta->source->resp_pkt = pkt;
1765 meta->source->_rx_page_addr = (unsigned long)page_address(p);
1766 meta->source->_rx_page_order = trans_pcie->rx_page_order;
1769 if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1770 iwl_op_mode_async_cb(trans->op_mode, cmd);
1772 iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1774 if (!(meta->flags & CMD_ASYNC)) {
1775 if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1777 "HCMD_ACTIVE already clear for command %s\n",
1778 iwl_get_cmd_string(trans, cmd_id));
1780 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1781 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1782 iwl_get_cmd_string(trans, cmd_id));
1783 wake_up(&trans_pcie->wait_command_queue);
1786 if (meta->flags & CMD_MAKE_TRANS_IDLE) {
1787 IWL_DEBUG_INFO(trans, "complete %s - mark trans as idle\n",
1788 iwl_get_cmd_string(trans, cmd->hdr.cmd));
1789 set_bit(STATUS_TRANS_IDLE, &trans->status);
1790 wake_up(&trans_pcie->d0i3_waitq);
1793 if (meta->flags & CMD_WAKE_UP_TRANS) {
1794 IWL_DEBUG_INFO(trans, "complete %s - clear trans idle flag\n",
1795 iwl_get_cmd_string(trans, cmd->hdr.cmd));
1796 clear_bit(STATUS_TRANS_IDLE, &trans->status);
1797 wake_up(&trans_pcie->d0i3_waitq);
1802 spin_unlock_bh(&txq->lock);
1805 #define HOST_COMPLETE_TIMEOUT (2 * HZ)
1807 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1808 struct iwl_host_cmd *cmd)
1812 /* An asynchronous command can not expect an SKB to be set. */
1813 if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1816 ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1819 "Error sending %s: enqueue_hcmd failed: %d\n",
1820 iwl_get_cmd_string(trans, cmd->id), ret);
1826 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1827 struct iwl_host_cmd *cmd)
1829 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1830 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1834 IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1835 iwl_get_cmd_string(trans, cmd->id));
1837 if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1839 "Command %s: a command is already active!\n",
1840 iwl_get_cmd_string(trans, cmd->id)))
1843 IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1844 iwl_get_cmd_string(trans, cmd->id));
1846 if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
1847 ret = wait_event_timeout(trans_pcie->d0i3_waitq,
1848 pm_runtime_active(&trans_pcie->pci_dev->dev),
1849 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
1851 IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
1856 cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1859 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1861 "Error sending %s: enqueue_hcmd failed: %d\n",
1862 iwl_get_cmd_string(trans, cmd->id), ret);
1866 ret = wait_event_timeout(trans_pcie->wait_command_queue,
1867 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1869 HOST_COMPLETE_TIMEOUT);
1871 IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1872 iwl_get_cmd_string(trans, cmd->id),
1873 jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1875 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1876 txq->read_ptr, txq->write_ptr);
1878 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1879 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1880 iwl_get_cmd_string(trans, cmd->id));
1883 iwl_force_nmi(trans);
1884 iwl_trans_fw_error(trans);
1889 if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1890 IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1891 iwl_get_cmd_string(trans, cmd->id));
1897 if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1898 test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1899 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1904 if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1905 IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1906 iwl_get_cmd_string(trans, cmd->id));
1914 if (cmd->flags & CMD_WANT_SKB) {
1916 * Cancel the CMD_WANT_SKB flag for the cmd in the
1917 * TX cmd queue. Otherwise in case the cmd comes
1918 * in later, it will possibly set an invalid
1919 * address (cmd->meta.source).
1921 txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1924 if (cmd->resp_pkt) {
1926 cmd->resp_pkt = NULL;
1932 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1934 if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1935 test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1936 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1941 if (cmd->flags & CMD_ASYNC)
1942 return iwl_pcie_send_hcmd_async(trans, cmd);
1944 /* We still can fail on RFKILL that can be asserted while we wait */
1945 return iwl_pcie_send_hcmd_sync(trans, cmd);
1948 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
1949 struct iwl_txq *txq, u8 hdr_len,
1950 struct iwl_cmd_meta *out_meta,
1951 struct iwl_device_cmd *dev_cmd, u16 tb1_len)
1953 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1958 * Set up TFD's third entry to point directly to remainder
1959 * of skb's head, if any
1961 tb2_len = skb_headlen(skb) - hdr_len;
1964 dma_addr_t tb2_phys = dma_map_single(trans->dev,
1965 skb->data + hdr_len,
1966 tb2_len, DMA_TO_DEVICE);
1967 if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) {
1968 iwl_pcie_tfd_unmap(trans, out_meta, txq,
1972 iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false);
1975 /* set up the remaining entries to point to the data */
1976 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1977 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1981 if (!skb_frag_size(frag))
1984 tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
1985 skb_frag_size(frag), DMA_TO_DEVICE);
1987 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
1988 iwl_pcie_tfd_unmap(trans, out_meta, txq,
1992 tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
1993 skb_frag_size(frag), false);
1995 out_meta->tbs |= BIT(tb_idx);
1998 trace_iwlwifi_dev_tx(trans->dev, skb,
1999 iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr),
2000 trans_pcie->tfd_size,
2001 &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
2003 trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
2008 struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len)
2010 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2011 struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
2016 /* enough room on this page */
2017 if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
2020 /* We don't have enough room on this page, get a new one. */
2021 __free_page(p->page);
2024 p->page = alloc_page(GFP_ATOMIC);
2027 p->pos = page_address(p->page);
2031 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
2032 bool ipv6, unsigned int len)
2035 struct ipv6hdr *iphv6 = iph;
2037 tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
2038 len + tcph->doff * 4,
2041 struct iphdr *iphv4 = iph;
2043 ip_send_check(iphv4);
2044 tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
2045 len + tcph->doff * 4,
2050 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2051 struct iwl_txq *txq, u8 hdr_len,
2052 struct iwl_cmd_meta *out_meta,
2053 struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2055 struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
2056 struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
2057 struct ieee80211_hdr *hdr = (void *)skb->data;
2058 unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
2059 unsigned int mss = skb_shinfo(skb)->gso_size;
2060 u16 length, iv_len, amsdu_pad;
2062 struct iwl_tso_hdr_page *hdr_page;
2063 struct page **page_ptr;
2067 /* if the packet is protected, then it must be CCMP or GCMP */
2068 BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
2069 iv_len = ieee80211_has_protected(hdr->frame_control) ?
2070 IEEE80211_CCMP_HDR_LEN : 0;
2072 trace_iwlwifi_dev_tx(trans->dev, skb,
2073 iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr),
2074 trans_pcie->tfd_size,
2075 &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0);
2077 ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
2078 snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
2079 total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
2082 /* total amount of header we may need for this A-MSDU */
2083 hdr_room = DIV_ROUND_UP(total_len, mss) *
2084 (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
2086 /* Our device supports 9 segments at most, it will fit in 1 page */
2087 hdr_page = get_page_hdr(trans, hdr_room);
2091 get_page(hdr_page->page);
2092 start_hdr = hdr_page->pos;
2093 page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
2094 *page_ptr = hdr_page->page;
2095 memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
2096 hdr_page->pos += iv_len;
2099 * Pull the ieee80211 header + IV to be able to use TSO core,
2100 * we will restore it for the tx_status flow.
2102 skb_pull(skb, hdr_len + iv_len);
2105 * Remove the length of all the headers that we don't actually
2106 * have in the MPDU by themselves, but that we duplicate into
2107 * all the different MSDUs inside the A-MSDU.
2109 le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
2111 tso_start(skb, &tso);
2114 /* this is the data left for this subframe */
2115 unsigned int data_left =
2116 min_t(unsigned int, mss, total_len);
2117 struct sk_buff *csum_skb = NULL;
2118 unsigned int hdr_tb_len;
2119 dma_addr_t hdr_tb_phys;
2120 struct tcphdr *tcph;
2121 u8 *iph, *subf_hdrs_start = hdr_page->pos;
2123 total_len -= data_left;
2125 memset(hdr_page->pos, 0, amsdu_pad);
2126 hdr_page->pos += amsdu_pad;
2127 amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2129 ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2130 hdr_page->pos += ETH_ALEN;
2131 ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2132 hdr_page->pos += ETH_ALEN;
2134 length = snap_ip_tcp_hdrlen + data_left;
2135 *((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2136 hdr_page->pos += sizeof(length);
2139 * This will copy the SNAP as well which will be considered
2142 tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2143 iph = hdr_page->pos + 8;
2144 tcph = (void *)(iph + ip_hdrlen);
2146 /* For testing on current hardware only */
2147 if (trans_pcie->sw_csum_tx) {
2148 csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2155 iwl_compute_pseudo_hdr_csum(iph, tcph,
2160 skb_put_data(csum_skb, tcph, tcp_hdrlen(skb));
2161 skb_reset_transport_header(csum_skb);
2162 csum_skb->csum_start =
2163 (unsigned char *)tcp_hdr(csum_skb) -
2167 hdr_page->pos += snap_ip_tcp_hdrlen;
2169 hdr_tb_len = hdr_page->pos - start_hdr;
2170 hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2171 hdr_tb_len, DMA_TO_DEVICE);
2172 if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2173 dev_kfree_skb(csum_skb);
2177 iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2179 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr,
2181 /* add this subframe's headers' length to the tx_cmd */
2182 le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
2184 /* prepare the start_hdr for the next subframe */
2185 start_hdr = hdr_page->pos;
2187 /* put the payload */
2189 unsigned int size = min_t(unsigned int, tso.size,
2193 if (trans_pcie->sw_csum_tx)
2194 skb_put_data(csum_skb, tso.data, size);
2196 tb_phys = dma_map_single(trans->dev, tso.data,
2197 size, DMA_TO_DEVICE);
2198 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2199 dev_kfree_skb(csum_skb);
2204 iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2206 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
2210 tso_build_data(skb, &tso, size);
2213 /* For testing on early hardware only */
2214 if (trans_pcie->sw_csum_tx) {
2217 csum = skb_checksum(csum_skb,
2218 skb_checksum_start_offset(csum_skb),
2220 skb_checksum_start_offset(csum_skb),
2222 dev_kfree_skb(csum_skb);
2223 dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2224 hdr_tb_len, DMA_TO_DEVICE);
2225 tcph->check = csum_fold(csum);
2226 dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2227 hdr_tb_len, DMA_TO_DEVICE);
2231 /* re -add the WiFi header and IV */
2232 skb_push(skb, hdr_len + iv_len);
2237 iwl_pcie_tfd_unmap(trans, out_meta, txq, txq->write_ptr);
2240 #else /* CONFIG_INET */
2241 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2242 struct iwl_txq *txq, u8 hdr_len,
2243 struct iwl_cmd_meta *out_meta,
2244 struct iwl_device_cmd *dev_cmd, u16 tb1_len)
2246 /* No A-MSDU without CONFIG_INET */
2251 #endif /* CONFIG_INET */
2253 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2254 struct iwl_device_cmd *dev_cmd, int txq_id)
2256 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2257 struct ieee80211_hdr *hdr;
2258 struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2259 struct iwl_cmd_meta *out_meta;
2260 struct iwl_txq *txq;
2261 dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2265 bool wait_write_ptr;
2271 txq = trans_pcie->txq[txq_id];
2273 if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2274 "TX on unused queue %d\n", txq_id))
2277 if (unlikely(trans_pcie->sw_csum_tx &&
2278 skb->ip_summed == CHECKSUM_PARTIAL)) {
2279 int offs = skb_checksum_start_offset(skb);
2280 int csum_offs = offs + skb->csum_offset;
2283 if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2286 csum = skb_checksum(skb, offs, skb->len - offs, 0);
2287 *(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2289 skb->ip_summed = CHECKSUM_UNNECESSARY;
2292 if (skb_is_nonlinear(skb) &&
2293 skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
2294 __skb_linearize(skb))
2297 /* mac80211 always puts the full header into the SKB's head,
2298 * so there's no need to check if it's readable there
2300 hdr = (struct ieee80211_hdr *)skb->data;
2301 fc = hdr->frame_control;
2302 hdr_len = ieee80211_hdrlen(fc);
2304 spin_lock(&txq->lock);
2306 if (iwl_queue_space(txq) < txq->high_mark) {
2307 iwl_stop_queue(trans, txq);
2309 /* don't put the packet on the ring, if there is no room */
2310 if (unlikely(iwl_queue_space(txq) < 3)) {
2311 struct iwl_device_cmd **dev_cmd_ptr;
2313 dev_cmd_ptr = (void *)((u8 *)skb->cb +
2314 trans_pcie->dev_cmd_offs);
2316 *dev_cmd_ptr = dev_cmd;
2317 __skb_queue_tail(&txq->overflow_q, skb);
2319 spin_unlock(&txq->lock);
2324 /* In AGG mode, the index in the ring must correspond to the WiFi
2325 * sequence number. This is a HW requirements to help the SCD to parse
2327 * Check here that the packets are in the right place on the ring.
2329 wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2330 WARN_ONCE(txq->ampdu &&
2331 (wifi_seq & 0xff) != txq->write_ptr,
2332 "Q: %d WiFi Seq %d tfdNum %d",
2333 txq_id, wifi_seq, txq->write_ptr);
2335 /* Set up driver data for this TFD */
2336 txq->entries[txq->write_ptr].skb = skb;
2337 txq->entries[txq->write_ptr].cmd = dev_cmd;
2339 dev_cmd->hdr.sequence =
2340 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2341 INDEX_TO_SEQ(txq->write_ptr)));
2343 tb0_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
2344 scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2345 offsetof(struct iwl_tx_cmd, scratch);
2347 tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2348 tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2350 /* Set up first empty entry in queue's array of Tx/cmd buffers */
2351 out_meta = &txq->entries[txq->write_ptr].meta;
2352 out_meta->flags = 0;
2355 * The second TB (tb1) points to the remainder of the TX command
2356 * and the 802.11 header - dword aligned size
2357 * (This calculation modifies the TX command, so do it before the
2358 * setup of the first TB)
2360 len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2361 hdr_len - IWL_FIRST_TB_SIZE;
2362 /* do not align A-MSDU to dword as the subframe header aligns it */
2363 amsdu = ieee80211_is_data_qos(fc) &&
2364 (*ieee80211_get_qos_ctl(hdr) &
2365 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
2366 if (trans_pcie->sw_csum_tx || !amsdu) {
2367 tb1_len = ALIGN(len, 4);
2368 /* Tell NIC about any 2-byte padding after MAC header */
2370 tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK;
2376 * The first TB points to bi-directional DMA data, we'll
2377 * memcpy the data into it later.
2379 iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2380 IWL_FIRST_TB_SIZE, true);
2382 /* there must be data left over for TB1 or this code must be changed */
2383 BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
2385 /* map the data for TB1 */
2386 tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
2387 tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2388 if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2390 iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2393 if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2397 } else if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2398 out_meta, dev_cmd, tb1_len))) {
2402 /* building the A-MSDU might have changed this data, so memcpy it now */
2403 memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
2406 tfd = iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr);
2407 /* Set up entry for this TFD in Tx byte-count array */
2408 iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len),
2409 iwl_pcie_tfd_get_num_tbs(trans, tfd));
2411 wait_write_ptr = ieee80211_has_morefrags(fc);
2413 /* start timer if queue currently empty */
2414 if (txq->read_ptr == txq->write_ptr) {
2415 if (txq->wd_timeout) {
2417 * If the TXQ is active, then set the timer, if not,
2418 * set the timer in remainder so that the timer will
2419 * be armed with the right value when the station will
2423 mod_timer(&txq->stuck_timer,
2424 jiffies + txq->wd_timeout);
2426 txq->frozen_expiry_remainder = txq->wd_timeout;
2428 IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
2429 iwl_trans_ref(trans);
2432 /* Tell device the write index *just past* this latest filled TFD */
2433 txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
2434 if (!wait_write_ptr)
2435 iwl_pcie_txq_inc_wr_ptr(trans, txq);
2438 * At this point the frame is "transmitted" successfully
2439 * and we will get a TX status notification eventually.
2441 spin_unlock(&txq->lock);
2444 spin_unlock(&txq->lock);