Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
   2/*
   3 * Copyright (C) 2003-2014, 2018-2021, 2023 Intel Corporation
   4 * Copyright (C) 2013-2015 Intel Mobile Communications GmbH
   5 * Copyright (C) 2016-2017 Intel Deutschland GmbH
   6 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   7#include <linux/etherdevice.h>
   8#include <linux/ieee80211.h>
   9#include <linux/slab.h>
  10#include <linux/sched.h>
  11#include <net/ip6_checksum.h>
  12#include <net/tso.h>
  13
  14#include "iwl-debug.h"
  15#include "iwl-csr.h"
  16#include "iwl-prph.h"
  17#include "iwl-io.h"
  18#include "iwl-scd.h"
  19#include "iwl-op-mode.h"
  20#include "internal.h"
  21#include "fw/api/tx.h"
  22
 
 
 
  23/*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
  24 * DMA services
  25 *
  26 * Theory of operation
  27 *
  28 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
  29 * of buffer descriptors, each of which points to one or more data buffers for
  30 * the device to read from or fill.  Driver and device exchange status of each
  31 * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
  32 * entries in each circular buffer, to protect against confusing empty and full
  33 * queue states.
  34 *
  35 * The device reads or writes the data in the queues via the device's several
  36 * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
  37 *
  38 * For Tx queue, there are low mark and high mark limits. If, after queuing
  39 * the packet for Tx, free space become < low mark, Tx queue stopped. When
  40 * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
  41 * Tx queue resumed.
  42 *
  43 ***************************************************/
  44
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  45
  46int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
  47			   struct iwl_dma_ptr *ptr, size_t size)
  48{
  49	if (WARN_ON(ptr->addr))
  50		return -EINVAL;
  51
  52	ptr->addr = dma_alloc_coherent(trans->dev, size,
  53				       &ptr->dma, GFP_KERNEL);
  54	if (!ptr->addr)
  55		return -ENOMEM;
  56	ptr->size = size;
  57	return 0;
  58}
  59
  60void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr)
  61{
  62	if (unlikely(!ptr->addr))
  63		return;
  64
  65	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
  66	memset(ptr, 0, sizeof(*ptr));
  67}
  68
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  69/*
  70 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
  71 */
  72static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
  73				    struct iwl_txq *txq)
  74{
 
  75	u32 reg = 0;
  76	int txq_id = txq->id;
  77
  78	lockdep_assert_held(&txq->lock);
  79
  80	/*
  81	 * explicitly wake up the NIC if:
  82	 * 1. shadow registers aren't enabled
  83	 * 2. NIC is woken up for CMD regardless of shadow outside this function
  84	 * 3. there is a chance that the NIC is asleep
  85	 */
  86	if (!trans->trans_cfg->base_params->shadow_reg_enable &&
  87	    txq_id != trans->txqs.cmd.q_id &&
  88	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
  89		/*
  90		 * wake up nic if it's powered down ...
  91		 * uCode will wake up, and interrupt us again, so next
  92		 * time we'll skip this part.
  93		 */
  94		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
  95
  96		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
  97			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
  98				       txq_id, reg);
  99			iwl_set_bit(trans, CSR_GP_CNTRL,
 100				    CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
 101			txq->need_update = true;
 102			return;
 103		}
 104	}
 105
 106	/*
 107	 * if not in power-save mode, uCode will never sleep when we're
 108	 * trying to tx (during RFKILL, we're not trying to tx).
 109	 */
 110	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr);
 111	if (!txq->block)
 112		iwl_write32(trans, HBUS_TARG_WRPTR,
 113			    txq->write_ptr | (txq_id << 8));
 114}
 115
 116void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
 117{
 
 118	int i;
 119
 120	for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) {
 121		struct iwl_txq *txq = trans->txqs.txq[i];
 122
 123		if (!test_bit(i, trans->txqs.queue_used))
 124			continue;
 125
 126		spin_lock_bh(&txq->lock);
 127		if (txq->need_update) {
 128			iwl_pcie_txq_inc_wr_ptr(trans, txq);
 129			txq->need_update = false;
 130		}
 131		spin_unlock_bh(&txq->lock);
 132	}
 133}
 134
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 135static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
 136				  dma_addr_t addr, u16 len, bool reset)
 137{
 
 138	void *tfd;
 139	u32 num_tbs;
 140
 141	tfd = (u8 *)txq->tfds + trans->txqs.tfd.size * txq->write_ptr;
 142
 143	if (reset)
 144		memset(tfd, 0, trans->txqs.tfd.size);
 145
 146	num_tbs = iwl_txq_gen1_tfd_get_num_tbs(trans, tfd);
 147
 148	/* Each TFD can point to a maximum max_tbs Tx buffers */
 149	if (num_tbs >= trans->txqs.tfd.max_tbs) {
 150		IWL_ERR(trans, "Error can not send more than %d chunks\n",
 151			trans->txqs.tfd.max_tbs);
 152		return -EINVAL;
 153	}
 154
 155	if (WARN(addr & ~IWL_TX_DMA_MASK,
 156		 "Unaligned address = %llx\n", (unsigned long long)addr))
 157		return -EINVAL;
 158
 159	iwl_pcie_gen1_tfd_set_tb(trans, tfd, num_tbs, addr, len);
 160
 161	return num_tbs;
 162}
 163
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 164static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
 165{
 166	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 167
 
 
 168	if (!trans->trans_cfg->base_params->apmg_wake_up_wa)
 169		return;
 170
 171	spin_lock(&trans_pcie->reg_lock);
 172
 173	if (WARN_ON(!trans_pcie->cmd_hold_nic_awake)) {
 174		spin_unlock(&trans_pcie->reg_lock);
 175		return;
 176	}
 177
 178	trans_pcie->cmd_hold_nic_awake = false;
 179	__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
 180				   CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ);
 181	spin_unlock(&trans_pcie->reg_lock);
 182}
 183
 184/*
 185 * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
 186 */
 187static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
 188{
 189	struct iwl_txq *txq = trans->txqs.txq[txq_id];
 190
 191	if (!txq) {
 192		IWL_ERR(trans, "Trying to free a queue that wasn't allocated?\n");
 193		return;
 194	}
 195
 196	spin_lock_bh(&txq->lock);
 197	while (txq->write_ptr != txq->read_ptr) {
 198		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
 199				   txq_id, txq->read_ptr);
 200
 201		if (txq_id != trans->txqs.cmd.q_id) {
 202			struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
 203
 204			if (WARN_ON_ONCE(!skb))
 205				continue;
 206
 207			iwl_txq_free_tso_page(trans, skb);
 208		}
 209		iwl_txq_free_tfd(trans, txq);
 210		txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr);
 
 
 
 211
 212		if (txq->read_ptr == txq->write_ptr &&
 213		    txq_id == trans->txqs.cmd.q_id)
 214			iwl_pcie_clear_cmd_in_flight(trans);
 
 
 215	}
 216
 217	while (!skb_queue_empty(&txq->overflow_q)) {
 218		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
 219
 220		iwl_op_mode_free_skb(trans->op_mode, skb);
 221	}
 222
 223	spin_unlock_bh(&txq->lock);
 224
 225	/* just in case - this queue may have been stopped */
 226	iwl_wake_queue(trans, txq);
 227}
 228
 229/*
 230 * iwl_pcie_txq_free - Deallocate DMA queue.
 231 * @txq: Transmit queue to deallocate.
 232 *
 233 * Empty queue by removing and destroying all BD's.
 234 * Free all buffers.
 235 * 0-fill, but do not free "txq" descriptor structure.
 236 */
 237static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
 238{
 239	struct iwl_txq *txq = trans->txqs.txq[txq_id];
 
 240	struct device *dev = trans->dev;
 241	int i;
 242
 243	if (WARN_ON(!txq))
 244		return;
 245
 246	iwl_pcie_txq_unmap(trans, txq_id);
 247
 248	/* De-alloc array of command/tx buffers */
 249	if (txq_id == trans->txqs.cmd.q_id)
 250		for (i = 0; i < txq->n_window; i++) {
 251			kfree_sensitive(txq->entries[i].cmd);
 252			kfree_sensitive(txq->entries[i].free_buf);
 253		}
 254
 255	/* De-alloc circular buffer of TFDs */
 256	if (txq->tfds) {
 257		dma_free_coherent(dev,
 258				  trans->txqs.tfd.size *
 259				  trans->trans_cfg->base_params->max_tfd_queue_size,
 260				  txq->tfds, txq->dma_addr);
 261		txq->dma_addr = 0;
 262		txq->tfds = NULL;
 263
 264		dma_free_coherent(dev,
 265				  sizeof(*txq->first_tb_bufs) * txq->n_window,
 266				  txq->first_tb_bufs, txq->first_tb_dma);
 267	}
 268
 269	kfree(txq->entries);
 270	txq->entries = NULL;
 271
 272	del_timer_sync(&txq->stuck_timer);
 273
 274	/* 0-fill queue descriptor structure */
 275	memset(txq, 0, sizeof(*txq));
 276}
 277
 278void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
 279{
 280	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 281	int nq = trans->trans_cfg->base_params->num_of_queues;
 282	int chan;
 283	u32 reg_val;
 284	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
 285				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
 286
 287	/* make sure all queue are not stopped/used */
 288	memset(trans->txqs.queue_stopped, 0,
 289	       sizeof(trans->txqs.queue_stopped));
 290	memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used));
 291
 292	trans_pcie->scd_base_addr =
 293		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
 294
 295	WARN_ON(scd_base_addr != 0 &&
 296		scd_base_addr != trans_pcie->scd_base_addr);
 297
 298	/* reset context data, TX status and translation data */
 299	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
 300				   SCD_CONTEXT_MEM_LOWER_BOUND,
 301			    NULL, clear_dwords);
 302
 303	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
 304		       trans->txqs.scd_bc_tbls.dma >> 10);
 305
 306	/* The chain extension of the SCD doesn't work well. This feature is
 307	 * enabled by default by the HW, so we need to disable it manually.
 308	 */
 309	if (trans->trans_cfg->base_params->scd_chain_ext_wa)
 310		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
 311
 312	iwl_trans_ac_txq_enable(trans, trans->txqs.cmd.q_id,
 313				trans->txqs.cmd.fifo,
 314				trans->txqs.cmd.wdg_timeout);
 315
 316	/* Activate all Tx DMA/FIFO channels */
 317	iwl_scd_activate_fifos(trans);
 318
 319	/* Enable DMA channel */
 320	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
 321		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
 322				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
 323				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
 324
 325	/* Update FH chicken bits */
 326	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
 327	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
 328			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
 329
 330	/* Enable L1-Active */
 331	if (trans->trans_cfg->device_family < IWL_DEVICE_FAMILY_8000)
 332		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
 333				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
 334}
 335
 336void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
 337{
 338	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 339	int txq_id;
 340
 341	/*
 342	 * we should never get here in gen2 trans mode return early to avoid
 343	 * having invalid accesses
 344	 */
 345	if (WARN_ON_ONCE(trans->trans_cfg->gen2))
 346		return;
 347
 348	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 349	     txq_id++) {
 350		struct iwl_txq *txq = trans->txqs.txq[txq_id];
 351		if (trans->trans_cfg->gen2)
 352			iwl_write_direct64(trans,
 353					   FH_MEM_CBBC_QUEUE(trans, txq_id),
 354					   txq->dma_addr);
 355		else
 356			iwl_write_direct32(trans,
 357					   FH_MEM_CBBC_QUEUE(trans, txq_id),
 358					   txq->dma_addr >> 8);
 359		iwl_pcie_txq_unmap(trans, txq_id);
 360		txq->read_ptr = 0;
 361		txq->write_ptr = 0;
 362	}
 363
 364	/* Tell NIC where to find the "keep warm" buffer */
 365	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
 366			   trans_pcie->kw.dma >> 4);
 367
 368	/*
 369	 * Send 0 as the scd_base_addr since the device may have be reset
 370	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
 371	 * contain garbage.
 372	 */
 373	iwl_pcie_tx_start(trans, 0);
 374}
 375
 376static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
 377{
 378	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
 379	int ch, ret;
 380	u32 mask = 0;
 381
 382	spin_lock_bh(&trans_pcie->irq_lock);
 383
 384	if (!iwl_trans_grab_nic_access(trans))
 385		goto out;
 386
 387	/* Stop each Tx DMA channel */
 388	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
 389		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
 390		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
 391	}
 392
 393	/* Wait for DMA channels to be idle */
 394	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
 395	if (ret < 0)
 396		IWL_ERR(trans,
 397			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
 398			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
 399
 400	iwl_trans_release_nic_access(trans);
 401
 402out:
 403	spin_unlock_bh(&trans_pcie->irq_lock);
 404}
 405
 406/*
 407 * iwl_pcie_tx_stop - Stop all Tx DMA channels
 408 */
 409int iwl_pcie_tx_stop(struct iwl_trans *trans)
 410{
 411	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 412	int txq_id;
 413
 414	/* Turn off all Tx DMA fifos */
 415	iwl_scd_deactivate_fifos(trans);
 416
 417	/* Turn off all Tx DMA channels */
 418	iwl_pcie_tx_stop_fh(trans);
 419
 420	/*
 421	 * This function can be called before the op_mode disabled the
 422	 * queues. This happens when we have an rfkill interrupt.
 423	 * Since we stop Tx altogether - mark the queues as stopped.
 424	 */
 425	memset(trans->txqs.queue_stopped, 0,
 426	       sizeof(trans->txqs.queue_stopped));
 427	memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used));
 428
 429	/* This can happen: start_hw, stop_device */
 430	if (!trans_pcie->txq_memory)
 431		return 0;
 432
 433	/* Unmap DMA from host system and free skb's */
 434	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 435	     txq_id++)
 436		iwl_pcie_txq_unmap(trans, txq_id);
 437
 438	return 0;
 439}
 440
 441/*
 442 * iwl_trans_tx_free - Free TXQ Context
 443 *
 444 * Destroy all TX DMA queues and structures
 445 */
 446void iwl_pcie_tx_free(struct iwl_trans *trans)
 447{
 448	int txq_id;
 449	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 450
 451	memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used));
 452
 453	/* Tx queues */
 454	if (trans_pcie->txq_memory) {
 455		for (txq_id = 0;
 456		     txq_id < trans->trans_cfg->base_params->num_of_queues;
 457		     txq_id++) {
 458			iwl_pcie_txq_free(trans, txq_id);
 459			trans->txqs.txq[txq_id] = NULL;
 460		}
 461	}
 462
 463	kfree(trans_pcie->txq_memory);
 464	trans_pcie->txq_memory = NULL;
 465
 466	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
 467
 468	iwl_pcie_free_dma_ptr(trans, &trans->txqs.scd_bc_tbls);
 469}
 470
 471/*
 472 * iwl_pcie_tx_alloc - allocate TX context
 473 * Allocate all Tx DMA structures and initialize them
 474 */
 475static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
 476{
 477	int ret;
 478	int txq_id, slots_num;
 479	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 480	u16 bc_tbls_size = trans->trans_cfg->base_params->num_of_queues;
 481
 482	if (WARN_ON(trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210))
 483		return -EINVAL;
 484
 485	bc_tbls_size *= sizeof(struct iwlagn_scd_bc_tbl);
 486
 487	/*It is not allowed to alloc twice, so warn when this happens.
 488	 * We cannot rely on the previous allocation, so free and fail */
 489	if (WARN_ON(trans_pcie->txq_memory)) {
 490		ret = -EINVAL;
 491		goto error;
 492	}
 493
 494	ret = iwl_pcie_alloc_dma_ptr(trans, &trans->txqs.scd_bc_tbls,
 495				     bc_tbls_size);
 496	if (ret) {
 497		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
 498		goto error;
 499	}
 500
 501	/* Alloc keep-warm buffer */
 502	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
 503	if (ret) {
 504		IWL_ERR(trans, "Keep Warm allocation failed\n");
 505		goto error;
 506	}
 507
 508	trans_pcie->txq_memory =
 509		kcalloc(trans->trans_cfg->base_params->num_of_queues,
 510			sizeof(struct iwl_txq), GFP_KERNEL);
 511	if (!trans_pcie->txq_memory) {
 512		IWL_ERR(trans, "Not enough memory for txq\n");
 513		ret = -ENOMEM;
 514		goto error;
 515	}
 516
 517	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
 518	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 519	     txq_id++) {
 520		bool cmd_queue = (txq_id == trans->txqs.cmd.q_id);
 521
 522		if (cmd_queue)
 523			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
 524					  trans->cfg->min_txq_size);
 525		else
 526			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
 527					  trans->cfg->min_ba_txq_size);
 528		trans->txqs.txq[txq_id] = &trans_pcie->txq_memory[txq_id];
 529		ret = iwl_txq_alloc(trans, trans->txqs.txq[txq_id], slots_num,
 530				    cmd_queue);
 531		if (ret) {
 532			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
 533			goto error;
 534		}
 535		trans->txqs.txq[txq_id]->id = txq_id;
 536	}
 537
 538	return 0;
 539
 540error:
 541	iwl_pcie_tx_free(trans);
 542
 543	return ret;
 544}
 545
 546int iwl_pcie_tx_init(struct iwl_trans *trans)
 547{
 548	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 549	int ret;
 550	int txq_id, slots_num;
 551	bool alloc = false;
 552
 553	if (!trans_pcie->txq_memory) {
 554		ret = iwl_pcie_tx_alloc(trans);
 555		if (ret)
 556			goto error;
 557		alloc = true;
 558	}
 559
 560	spin_lock_bh(&trans_pcie->irq_lock);
 561
 562	/* Turn off all Tx DMA fifos */
 563	iwl_scd_deactivate_fifos(trans);
 564
 565	/* Tell NIC where to find the "keep warm" buffer */
 566	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
 567			   trans_pcie->kw.dma >> 4);
 568
 569	spin_unlock_bh(&trans_pcie->irq_lock);
 570
 571	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
 572	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 573	     txq_id++) {
 574		bool cmd_queue = (txq_id == trans->txqs.cmd.q_id);
 575
 576		if (cmd_queue)
 577			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
 578					  trans->cfg->min_txq_size);
 579		else
 580			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
 581					  trans->cfg->min_ba_txq_size);
 582		ret = iwl_txq_init(trans, trans->txqs.txq[txq_id], slots_num,
 583				   cmd_queue);
 584		if (ret) {
 585			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
 586			goto error;
 587		}
 588
 589		/*
 590		 * Tell nic where to find circular buffer of TFDs for a
 591		 * given Tx queue, and enable the DMA channel used for that
 592		 * queue.
 593		 * Circular buffer (TFD queue in DRAM) physical base address
 594		 */
 595		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
 596				   trans->txqs.txq[txq_id]->dma_addr >> 8);
 597	}
 598
 599	iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
 600	if (trans->trans_cfg->base_params->num_of_queues > 20)
 601		iwl_set_bits_prph(trans, SCD_GP_CTRL,
 602				  SCD_GP_CTRL_ENABLE_31_QUEUES);
 603
 604	return 0;
 605error:
 606	/*Upon error, free only if we allocated something */
 607	if (alloc)
 608		iwl_pcie_tx_free(trans);
 609	return ret;
 610}
 611
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 612static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
 613				      const struct iwl_host_cmd *cmd)
 614{
 615	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 
 
 
 616
 617	/* Make sure the NIC is still alive in the bus */
 618	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
 619		return -ENODEV;
 620
 621	if (!trans->trans_cfg->base_params->apmg_wake_up_wa)
 622		return 0;
 623
 624	/*
 625	 * wake up the NIC to make sure that the firmware will see the host
 626	 * command - we will let the NIC sleep once all the host commands
 627	 * returned. This needs to be done only on NICs that have
 628	 * apmg_wake_up_wa set (see above.)
 629	 */
 630	if (!_iwl_trans_pcie_grab_nic_access(trans))
 631		return -EIO;
 632
 633	/*
 634	 * In iwl_trans_grab_nic_access(), we've acquired the reg_lock.
 635	 * There, we also returned immediately if cmd_hold_nic_awake is
 636	 * already true, so it's OK to unconditionally set it to true.
 637	 */
 638	trans_pcie->cmd_hold_nic_awake = true;
 639	spin_unlock(&trans_pcie->reg_lock);
 
 
 
 
 
 
 
 
 640
 641	return 0;
 642}
 643
 644/*
 645 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
 646 *
 647 * When FW advances 'R' index, all entries between old and new 'R' index
 648 * need to be reclaimed. As result, some free space forms.  If there is
 649 * enough free space (> low mark), wake the stack that feeds us.
 650 */
 651static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
 652{
 653	struct iwl_txq *txq = trans->txqs.txq[txq_id];
 
 
 654	int nfreed = 0;
 655	u16 r;
 656
 657	lockdep_assert_held(&txq->lock);
 658
 659	idx = iwl_txq_get_cmd_index(txq, idx);
 660	r = iwl_txq_get_cmd_index(txq, txq->read_ptr);
 661
 662	if (idx >= trans->trans_cfg->base_params->max_tfd_queue_size ||
 663	    (!iwl_txq_used(txq, idx))) {
 664		WARN_ONCE(test_bit(txq_id, trans->txqs.queue_used),
 665			  "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
 666			  __func__, txq_id, idx,
 667			  trans->trans_cfg->base_params->max_tfd_queue_size,
 668			  txq->write_ptr, txq->read_ptr);
 669		return;
 670	}
 671
 672	for (idx = iwl_txq_inc_wrap(trans, idx); r != idx;
 673	     r = iwl_txq_inc_wrap(trans, r)) {
 674		txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr);
 675
 676		if (nfreed++ > 0) {
 677			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
 678				idx, txq->write_ptr, r);
 679			iwl_force_nmi(trans);
 680		}
 681	}
 682
 683	if (txq->read_ptr == txq->write_ptr)
 
 684		iwl_pcie_clear_cmd_in_flight(trans);
 
 
 685
 686	iwl_txq_progress(txq);
 687}
 688
 689static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
 690				 u16 txq_id)
 691{
 692	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 693	u32 tbl_dw_addr;
 694	u32 tbl_dw;
 695	u16 scd_q2ratid;
 696
 697	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
 698
 699	tbl_dw_addr = trans_pcie->scd_base_addr +
 700			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
 701
 702	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
 703
 704	if (txq_id & 0x1)
 705		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
 706	else
 707		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
 708
 709	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
 710
 711	return 0;
 712}
 713
 714/* Receiver address (actually, Rx station's index into station table),
 715 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
 716#define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
 717
 718bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
 719			       const struct iwl_trans_txq_scd_cfg *cfg,
 720			       unsigned int wdg_timeout)
 721{
 722	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 723	struct iwl_txq *txq = trans->txqs.txq[txq_id];
 724	int fifo = -1;
 725	bool scd_bug = false;
 726
 727	if (test_and_set_bit(txq_id, trans->txqs.queue_used))
 728		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
 729
 730	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
 731
 732	if (cfg) {
 733		fifo = cfg->fifo;
 734
 735		/* Disable the scheduler prior configuring the cmd queue */
 736		if (txq_id == trans->txqs.cmd.q_id &&
 737		    trans_pcie->scd_set_active)
 738			iwl_scd_enable_set_active(trans, 0);
 739
 740		/* Stop this Tx queue before configuring it */
 741		iwl_scd_txq_set_inactive(trans, txq_id);
 742
 743		/* Set this queue as a chain-building queue unless it is CMD */
 744		if (txq_id != trans->txqs.cmd.q_id)
 745			iwl_scd_txq_set_chain(trans, txq_id);
 746
 747		if (cfg->aggregate) {
 748			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
 749
 750			/* Map receiver-address / traffic-ID to this queue */
 751			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
 752
 753			/* enable aggregations for the queue */
 754			iwl_scd_txq_enable_agg(trans, txq_id);
 755			txq->ampdu = true;
 756		} else {
 757			/*
 758			 * disable aggregations for the queue, this will also
 759			 * make the ra_tid mapping configuration irrelevant
 760			 * since it is now a non-AGG queue.
 761			 */
 762			iwl_scd_txq_disable_agg(trans, txq_id);
 763
 764			ssn = txq->read_ptr;
 765		}
 766	} else {
 767		/*
 768		 * If we need to move the SCD write pointer by steps of
 769		 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let
 770		 * the op_mode know by returning true later.
 771		 * Do this only in case cfg is NULL since this trick can
 772		 * be done only if we have DQA enabled which is true for mvm
 773		 * only. And mvm never sets a cfg pointer.
 774		 * This is really ugly, but this is the easiest way out for
 775		 * this sad hardware issue.
 776		 * This bug has been fixed on devices 9000 and up.
 777		 */
 778		scd_bug = !trans->trans_cfg->mq_rx_supported &&
 779			!((ssn - txq->write_ptr) & 0x3f) &&
 780			(ssn != txq->write_ptr);
 781		if (scd_bug)
 782			ssn++;
 783	}
 784
 785	/* Place first TFD at index corresponding to start sequence number.
 786	 * Assumes that ssn_idx is valid (!= 0xFFF) */
 787	txq->read_ptr = (ssn & 0xff);
 788	txq->write_ptr = (ssn & 0xff);
 789	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
 790			   (ssn & 0xff) | (txq_id << 8));
 791
 792	if (cfg) {
 793		u8 frame_limit = cfg->frame_limit;
 794
 795		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
 796
 797		/* Set up Tx window size and frame limit for this queue */
 798		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
 799				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
 800		iwl_trans_write_mem32(trans,
 801			trans_pcie->scd_base_addr +
 802			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
 803			SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) |
 804			SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit));
 805
 806		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
 807		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
 808			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
 809			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
 810			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
 811			       SCD_QUEUE_STTS_REG_MSK);
 812
 813		/* enable the scheduler for this queue (only) */
 814		if (txq_id == trans->txqs.cmd.q_id &&
 815		    trans_pcie->scd_set_active)
 816			iwl_scd_enable_set_active(trans, BIT(txq_id));
 817
 818		IWL_DEBUG_TX_QUEUES(trans,
 819				    "Activate queue %d on FIFO %d WrPtr: %d\n",
 820				    txq_id, fifo, ssn & 0xff);
 821	} else {
 822		IWL_DEBUG_TX_QUEUES(trans,
 823				    "Activate queue %d WrPtr: %d\n",
 824				    txq_id, ssn & 0xff);
 825	}
 826
 827	return scd_bug;
 828}
 829
 830void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
 831					bool shared_mode)
 832{
 833	struct iwl_txq *txq = trans->txqs.txq[txq_id];
 
 834
 835	txq->ampdu = !shared_mode;
 836}
 837
 838void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
 839				bool configure_scd)
 840{
 841	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 842	u32 stts_addr = trans_pcie->scd_base_addr +
 843			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
 844	static const u32 zero_val[4] = {};
 845
 846	trans->txqs.txq[txq_id]->frozen_expiry_remainder = 0;
 847	trans->txqs.txq[txq_id]->frozen = false;
 848
 849	/*
 850	 * Upon HW Rfkill - we stop the device, and then stop the queues
 851	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
 852	 * allow the op_mode to call txq_disable after it already called
 853	 * stop_device.
 854	 */
 855	if (!test_and_clear_bit(txq_id, trans->txqs.queue_used)) {
 856		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
 857			  "queue %d not used", txq_id);
 858		return;
 859	}
 860
 861	if (configure_scd) {
 862		iwl_scd_txq_set_inactive(trans, txq_id);
 863
 864		iwl_trans_write_mem(trans, stts_addr, (const void *)zero_val,
 865				    ARRAY_SIZE(zero_val));
 866	}
 867
 868	iwl_pcie_txq_unmap(trans, txq_id);
 869	trans->txqs.txq[txq_id]->ampdu = false;
 870
 871	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
 872}
 873
 874/*************** HOST COMMAND QUEUE FUNCTIONS   *****/
 875
 876static void iwl_trans_pcie_block_txq_ptrs(struct iwl_trans *trans, bool block)
 877{
 878	int i;
 879
 880	for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) {
 881		struct iwl_txq *txq = trans->txqs.txq[i];
 882
 883		if (i == trans->txqs.cmd.q_id)
 884			continue;
 885
 886		/* we skip the command queue (obviously) so it's OK to nest */
 887		spin_lock_nested(&txq->lock, 1);
 888
 889		if (!block && !(WARN_ON_ONCE(!txq->block))) {
 890			txq->block--;
 891			if (!txq->block) {
 892				iwl_write32(trans, HBUS_TARG_WRPTR,
 893					    txq->write_ptr | (i << 8));
 894			}
 895		} else if (block) {
 896			txq->block++;
 897		}
 898
 899		spin_unlock(&txq->lock);
 900	}
 901}
 902
 903/*
 904 * iwl_pcie_enqueue_hcmd - enqueue a uCode command
 905 * @priv: device private data point
 906 * @cmd: a pointer to the ucode command structure
 907 *
 908 * The function returns < 0 values to indicate the operation
 909 * failed. On success, it returns the index (>= 0) of command in the
 910 * command queue.
 911 */
 912int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
 913			  struct iwl_host_cmd *cmd)
 914{
 915	struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id];
 
 916	struct iwl_device_cmd *out_cmd;
 917	struct iwl_cmd_meta *out_meta;
 
 918	void *dup_buf = NULL;
 919	dma_addr_t phys_addr;
 920	int idx;
 921	u16 copy_size, cmd_size, tb0_size;
 922	bool had_nocopy = false;
 923	u8 group_id = iwl_cmd_groupid(cmd->id);
 924	int i, ret;
 925	u32 cmd_pos;
 926	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
 927	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
 928	unsigned long flags;
 929
 930	if (WARN(!trans->wide_cmd_header &&
 931		 group_id > IWL_ALWAYS_LONG_GROUP,
 932		 "unsupported wide command %#x\n", cmd->id))
 933		return -EINVAL;
 934
 935	if (group_id != 0) {
 936		copy_size = sizeof(struct iwl_cmd_header_wide);
 937		cmd_size = sizeof(struct iwl_cmd_header_wide);
 938	} else {
 939		copy_size = sizeof(struct iwl_cmd_header);
 940		cmd_size = sizeof(struct iwl_cmd_header);
 941	}
 942
 943	/* need one for the header if the first is NOCOPY */
 944	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
 945
 946	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
 947		cmddata[i] = cmd->data[i];
 948		cmdlen[i] = cmd->len[i];
 949
 950		if (!cmd->len[i])
 951			continue;
 952
 953		/* need at least IWL_FIRST_TB_SIZE copied */
 954		if (copy_size < IWL_FIRST_TB_SIZE) {
 955			int copy = IWL_FIRST_TB_SIZE - copy_size;
 956
 957			if (copy > cmdlen[i])
 958				copy = cmdlen[i];
 959			cmdlen[i] -= copy;
 960			cmddata[i] += copy;
 961			copy_size += copy;
 962		}
 963
 964		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
 965			had_nocopy = true;
 966			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
 967				idx = -EINVAL;
 968				goto free_dup_buf;
 969			}
 970		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
 971			/*
 972			 * This is also a chunk that isn't copied
 973			 * to the static buffer so set had_nocopy.
 974			 */
 975			had_nocopy = true;
 976
 977			/* only allowed once */
 978			if (WARN_ON(dup_buf)) {
 979				idx = -EINVAL;
 980				goto free_dup_buf;
 981			}
 982
 983			dup_buf = kmemdup(cmddata[i], cmdlen[i],
 984					  GFP_ATOMIC);
 985			if (!dup_buf)
 986				return -ENOMEM;
 987		} else {
 988			/* NOCOPY must not be followed by normal! */
 989			if (WARN_ON(had_nocopy)) {
 990				idx = -EINVAL;
 991				goto free_dup_buf;
 992			}
 993			copy_size += cmdlen[i];
 994		}
 995		cmd_size += cmd->len[i];
 996	}
 997
 998	/*
 999	 * If any of the command structures end up being larger than
1000	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1001	 * allocated into separate TFDs, then we will need to
1002	 * increase the size of the buffers.
1003	 */
1004	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1005		 "Command %s (%#x) is too large (%d bytes)\n",
1006		 iwl_get_cmd_string(trans, cmd->id),
1007		 cmd->id, copy_size)) {
1008		idx = -EINVAL;
1009		goto free_dup_buf;
1010	}
1011
1012	spin_lock_irqsave(&txq->lock, flags);
1013
1014	if (iwl_txq_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1015		spin_unlock_irqrestore(&txq->lock, flags);
1016
1017		IWL_ERR(trans, "No space in command queue\n");
1018		iwl_op_mode_cmd_queue_full(trans->op_mode);
1019		idx = -ENOSPC;
1020		goto free_dup_buf;
1021	}
1022
1023	idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
1024	out_cmd = txq->entries[idx].cmd;
1025	out_meta = &txq->entries[idx].meta;
1026
1027	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1028	if (cmd->flags & CMD_WANT_SKB)
1029		out_meta->source = cmd;
1030
1031	/* set up the header */
1032	if (group_id != 0) {
1033		out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1034		out_cmd->hdr_wide.group_id = group_id;
1035		out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1036		out_cmd->hdr_wide.length =
1037			cpu_to_le16(cmd_size -
1038				    sizeof(struct iwl_cmd_header_wide));
1039		out_cmd->hdr_wide.reserved = 0;
1040		out_cmd->hdr_wide.sequence =
1041			cpu_to_le16(QUEUE_TO_SEQ(trans->txqs.cmd.q_id) |
1042						 INDEX_TO_SEQ(txq->write_ptr));
1043
1044		cmd_pos = sizeof(struct iwl_cmd_header_wide);
1045		copy_size = sizeof(struct iwl_cmd_header_wide);
1046	} else {
1047		out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1048		out_cmd->hdr.sequence =
1049			cpu_to_le16(QUEUE_TO_SEQ(trans->txqs.cmd.q_id) |
1050						 INDEX_TO_SEQ(txq->write_ptr));
1051		out_cmd->hdr.group_id = 0;
1052
1053		cmd_pos = sizeof(struct iwl_cmd_header);
1054		copy_size = sizeof(struct iwl_cmd_header);
1055	}
1056
1057	/* and copy the data that needs to be copied */
1058	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1059		int copy;
1060
1061		if (!cmd->len[i])
1062			continue;
1063
1064		/* copy everything if not nocopy/dup */
1065		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1066					   IWL_HCMD_DFL_DUP))) {
1067			copy = cmd->len[i];
1068
1069			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1070			cmd_pos += copy;
1071			copy_size += copy;
1072			continue;
1073		}
1074
1075		/*
1076		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1077		 * in total (for bi-directional DMA), but copy up to what
1078		 * we can fit into the payload for debug dump purposes.
1079		 */
1080		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1081
1082		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1083		cmd_pos += copy;
1084
1085		/* However, treat copy_size the proper way, we need it below */
1086		if (copy_size < IWL_FIRST_TB_SIZE) {
1087			copy = IWL_FIRST_TB_SIZE - copy_size;
1088
1089			if (copy > cmd->len[i])
1090				copy = cmd->len[i];
1091			copy_size += copy;
1092		}
1093	}
1094
1095	IWL_DEBUG_HC(trans,
1096		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1097		     iwl_get_cmd_string(trans, cmd->id),
1098		     group_id, out_cmd->hdr.cmd,
1099		     le16_to_cpu(out_cmd->hdr.sequence),
1100		     cmd_size, txq->write_ptr, idx, trans->txqs.cmd.q_id);
1101
1102	/* start the TFD with the minimum copy bytes */
1103	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1104	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1105	iwl_pcie_txq_build_tfd(trans, txq,
1106			       iwl_txq_get_first_tb_dma(txq, idx),
1107			       tb0_size, true);
1108
1109	/* map first command fragment, if any remains */
1110	if (copy_size > tb0_size) {
1111		phys_addr = dma_map_single(trans->dev,
1112					   ((u8 *)&out_cmd->hdr) + tb0_size,
1113					   copy_size - tb0_size,
1114					   DMA_TO_DEVICE);
1115		if (dma_mapping_error(trans->dev, phys_addr)) {
1116			iwl_txq_gen1_tfd_unmap(trans, out_meta, txq,
1117					       txq->write_ptr);
1118			idx = -ENOMEM;
1119			goto out;
1120		}
1121
1122		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1123				       copy_size - tb0_size, false);
1124	}
1125
1126	/* map the remaining (adjusted) nocopy/dup fragments */
1127	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1128		void *data = (void *)(uintptr_t)cmddata[i];
1129
1130		if (!cmdlen[i])
1131			continue;
1132		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1133					   IWL_HCMD_DFL_DUP)))
1134			continue;
1135		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1136			data = dup_buf;
1137		phys_addr = dma_map_single(trans->dev, data,
1138					   cmdlen[i], DMA_TO_DEVICE);
1139		if (dma_mapping_error(trans->dev, phys_addr)) {
1140			iwl_txq_gen1_tfd_unmap(trans, out_meta, txq,
1141					       txq->write_ptr);
1142			idx = -ENOMEM;
1143			goto out;
1144		}
1145
1146		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1147	}
1148
1149	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
1150	out_meta->flags = cmd->flags;
1151	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1152		kfree_sensitive(txq->entries[idx].free_buf);
1153	txq->entries[idx].free_buf = dup_buf;
1154
1155	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1156
1157	/* start timer if queue currently empty */
1158	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
1159		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1160
 
1161	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1162	if (ret < 0) {
1163		idx = ret;
 
1164		goto out;
1165	}
1166
1167	if (cmd->flags & CMD_BLOCK_TXQS)
1168		iwl_trans_pcie_block_txq_ptrs(trans, true);
1169
1170	/* Increment and update queue's write index */
1171	txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr);
1172	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1173
 
 
1174 out:
1175	spin_unlock_irqrestore(&txq->lock, flags);
1176 free_dup_buf:
1177	if (idx < 0)
1178		kfree(dup_buf);
1179	return idx;
1180}
1181
1182/*
1183 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1184 * @rxb: Rx buffer to reclaim
1185 */
1186void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1187			    struct iwl_rx_cmd_buffer *rxb)
1188{
1189	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1190	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1191	u8 group_id;
1192	u32 cmd_id;
1193	int txq_id = SEQ_TO_QUEUE(sequence);
1194	int index = SEQ_TO_INDEX(sequence);
1195	int cmd_index;
1196	struct iwl_device_cmd *cmd;
1197	struct iwl_cmd_meta *meta;
1198	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1199	struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id];
1200
1201	/* If a Tx command is being handled and it isn't in the actual
1202	 * command queue then there a command routing bug has been introduced
1203	 * in the queue management code. */
1204	if (WARN(txq_id != trans->txqs.cmd.q_id,
1205		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1206		 txq_id, trans->txqs.cmd.q_id, sequence, txq->read_ptr,
1207		 txq->write_ptr)) {
1208		iwl_print_hex_error(trans, pkt, 32);
1209		return;
1210	}
1211
1212	spin_lock_bh(&txq->lock);
1213
1214	cmd_index = iwl_txq_get_cmd_index(txq, index);
1215	cmd = txq->entries[cmd_index].cmd;
1216	meta = &txq->entries[cmd_index].meta;
1217	group_id = cmd->hdr.group_id;
1218	cmd_id = WIDE_ID(group_id, cmd->hdr.cmd);
1219
1220	if (trans->trans_cfg->gen2)
1221		iwl_txq_gen2_tfd_unmap(trans, meta,
1222				       iwl_txq_get_tfd(trans, txq, index));
1223	else
1224		iwl_txq_gen1_tfd_unmap(trans, meta, txq, index);
1225
1226	/* Input error checking is done when commands are added to queue. */
1227	if (meta->flags & CMD_WANT_SKB) {
1228		struct page *p = rxb_steal_page(rxb);
1229
1230		meta->source->resp_pkt = pkt;
1231		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1232		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1233	}
1234
1235	if (meta->flags & CMD_BLOCK_TXQS)
1236		iwl_trans_pcie_block_txq_ptrs(trans, false);
1237
1238	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1239
1240	if (!(meta->flags & CMD_ASYNC)) {
1241		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1242			IWL_WARN(trans,
1243				 "HCMD_ACTIVE already clear for command %s\n",
1244				 iwl_get_cmd_string(trans, cmd_id));
1245		}
1246		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1247		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1248			       iwl_get_cmd_string(trans, cmd_id));
1249		wake_up(&trans->wait_command_queue);
1250	}
1251
1252	meta->flags = 0;
1253
1254	spin_unlock_bh(&txq->lock);
1255}
1256
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1257static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
1258			     struct iwl_txq *txq, u8 hdr_len,
1259			     struct iwl_cmd_meta *out_meta)
1260{
1261	u16 head_tb_len;
1262	int i;
1263
1264	/*
1265	 * Set up TFD's third entry to point directly to remainder
1266	 * of skb's head, if any
1267	 */
1268	head_tb_len = skb_headlen(skb) - hdr_len;
1269
1270	if (head_tb_len > 0) {
1271		dma_addr_t tb_phys = dma_map_single(trans->dev,
1272						    skb->data + hdr_len,
1273						    head_tb_len, DMA_TO_DEVICE);
1274		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
1275			return -EINVAL;
1276		trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb->data + hdr_len,
1277					tb_phys, head_tb_len);
 
1278		iwl_pcie_txq_build_tfd(trans, txq, tb_phys, head_tb_len, false);
1279	}
1280
1281	/* set up the remaining entries to point to the data */
1282	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1283		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1284		dma_addr_t tb_phys;
1285		int tb_idx;
1286
1287		if (!skb_frag_size(frag))
1288			continue;
1289
1290		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
1291					   skb_frag_size(frag), DMA_TO_DEVICE);
1292
1293		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
1294			return -EINVAL;
1295		trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb_frag_address(frag),
1296					tb_phys, skb_frag_size(frag));
 
1297		tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
1298						skb_frag_size(frag), false);
1299		if (tb_idx < 0)
1300			return tb_idx;
1301
1302		out_meta->tbs |= BIT(tb_idx);
1303	}
1304
1305	return 0;
1306}
1307
1308#ifdef CONFIG_INET
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1309static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
1310				   struct iwl_txq *txq, u8 hdr_len,
1311				   struct iwl_cmd_meta *out_meta,
1312				   struct iwl_device_tx_cmd *dev_cmd,
1313				   u16 tb1_len)
1314{
1315	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
 
1316	struct ieee80211_hdr *hdr = (void *)skb->data;
1317	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
1318	unsigned int mss = skb_shinfo(skb)->gso_size;
1319	u16 length, iv_len, amsdu_pad;
1320	u8 *start_hdr;
1321	struct iwl_tso_hdr_page *hdr_page;
 
1322	struct tso_t tso;
1323
1324	/* if the packet is protected, then it must be CCMP or GCMP */
1325	BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
1326	iv_len = ieee80211_has_protected(hdr->frame_control) ?
1327		IEEE80211_CCMP_HDR_LEN : 0;
1328
1329	trace_iwlwifi_dev_tx(trans->dev, skb,
1330			     iwl_txq_get_tfd(trans, txq, txq->write_ptr),
1331			     trans->txqs.tfd.size,
1332			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0);
1333
1334	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
1335	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
1336	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
1337	amsdu_pad = 0;
1338
1339	/* total amount of header we may need for this A-MSDU */
1340	hdr_room = DIV_ROUND_UP(total_len, mss) *
1341		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
1342
1343	/* Our device supports 9 segments at most, it will fit in 1 page */
1344	hdr_page = get_page_hdr(trans, hdr_room, skb);
1345	if (!hdr_page)
1346		return -ENOMEM;
1347
 
1348	start_hdr = hdr_page->pos;
 
 
1349	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
1350	hdr_page->pos += iv_len;
1351
1352	/*
1353	 * Pull the ieee80211 header + IV to be able to use TSO core,
1354	 * we will restore it for the tx_status flow.
1355	 */
1356	skb_pull(skb, hdr_len + iv_len);
1357
1358	/*
1359	 * Remove the length of all the headers that we don't actually
1360	 * have in the MPDU by themselves, but that we duplicate into
1361	 * all the different MSDUs inside the A-MSDU.
1362	 */
1363	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
1364
1365	tso_start(skb, &tso);
1366
1367	while (total_len) {
1368		/* this is the data left for this subframe */
1369		unsigned int data_left =
1370			min_t(unsigned int, mss, total_len);
 
1371		unsigned int hdr_tb_len;
1372		dma_addr_t hdr_tb_phys;
1373		u8 *subf_hdrs_start = hdr_page->pos;
 
1374
1375		total_len -= data_left;
1376
1377		memset(hdr_page->pos, 0, amsdu_pad);
1378		hdr_page->pos += amsdu_pad;
1379		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
1380				  data_left)) & 0x3;
1381		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
1382		hdr_page->pos += ETH_ALEN;
1383		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
1384		hdr_page->pos += ETH_ALEN;
1385
1386		length = snap_ip_tcp_hdrlen + data_left;
1387		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
1388		hdr_page->pos += sizeof(length);
1389
1390		/*
1391		 * This will copy the SNAP as well which will be considered
1392		 * as MAC header.
1393		 */
1394		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1395
1396		hdr_page->pos += snap_ip_tcp_hdrlen;
1397
1398		hdr_tb_len = hdr_page->pos - start_hdr;
1399		hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
1400					     hdr_tb_len, DMA_TO_DEVICE);
1401		if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys)))
 
1402			return -EINVAL;
 
1403		iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
1404				       hdr_tb_len, false);
1405		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
1406					hdr_tb_phys, hdr_tb_len);
1407		/* add this subframe's headers' length to the tx_cmd */
1408		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
1409
1410		/* prepare the start_hdr for the next subframe */
1411		start_hdr = hdr_page->pos;
1412
1413		/* put the payload */
1414		while (data_left) {
1415			unsigned int size = min_t(unsigned int, tso.size,
1416						  data_left);
1417			dma_addr_t tb_phys;
1418
 
 
 
1419			tb_phys = dma_map_single(trans->dev, tso.data,
1420						 size, DMA_TO_DEVICE);
1421			if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
 
1422				return -EINVAL;
 
1423
1424			iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
1425					       size, false);
1426			trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data,
1427						tb_phys, size);
1428
1429			data_left -= size;
1430			tso_build_data(skb, &tso, size);
1431		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1432	}
1433
1434	/* re -add the WiFi header and IV */
1435	skb_push(skb, hdr_len + iv_len);
1436
1437	return 0;
1438}
1439#else /* CONFIG_INET */
1440static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
1441				   struct iwl_txq *txq, u8 hdr_len,
1442				   struct iwl_cmd_meta *out_meta,
1443				   struct iwl_device_tx_cmd *dev_cmd,
1444				   u16 tb1_len)
1445{
1446	/* No A-MSDU without CONFIG_INET */
1447	WARN_ON(1);
1448
1449	return -1;
1450}
1451#endif /* CONFIG_INET */
1452
1453int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
1454		      struct iwl_device_tx_cmd *dev_cmd, int txq_id)
1455{
 
1456	struct ieee80211_hdr *hdr;
1457	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
1458	struct iwl_cmd_meta *out_meta;
1459	struct iwl_txq *txq;
1460	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
1461	void *tb1_addr;
1462	void *tfd;
1463	u16 len, tb1_len;
1464	bool wait_write_ptr;
1465	__le16 fc;
1466	u8 hdr_len;
1467	u16 wifi_seq;
1468	bool amsdu;
1469
1470	txq = trans->txqs.txq[txq_id];
1471
1472	if (WARN_ONCE(!test_bit(txq_id, trans->txqs.queue_used),
1473		      "TX on unused queue %d\n", txq_id))
1474		return -EINVAL;
1475
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1476	if (skb_is_nonlinear(skb) &&
1477	    skb_shinfo(skb)->nr_frags > IWL_TRANS_MAX_FRAGS(trans) &&
1478	    __skb_linearize(skb))
1479		return -ENOMEM;
1480
1481	/* mac80211 always puts the full header into the SKB's head,
1482	 * so there's no need to check if it's readable there
1483	 */
1484	hdr = (struct ieee80211_hdr *)skb->data;
1485	fc = hdr->frame_control;
1486	hdr_len = ieee80211_hdrlen(fc);
1487
1488	spin_lock(&txq->lock);
1489
1490	if (iwl_txq_space(trans, txq) < txq->high_mark) {
1491		iwl_txq_stop(trans, txq);
1492
1493		/* don't put the packet on the ring, if there is no room */
1494		if (unlikely(iwl_txq_space(trans, txq) < 3)) {
1495			struct iwl_device_tx_cmd **dev_cmd_ptr;
1496
1497			dev_cmd_ptr = (void *)((u8 *)skb->cb +
1498					       trans->txqs.dev_cmd_offs);
1499
1500			*dev_cmd_ptr = dev_cmd;
1501			__skb_queue_tail(&txq->overflow_q, skb);
1502
1503			spin_unlock(&txq->lock);
1504			return 0;
1505		}
1506	}
1507
1508	/* In AGG mode, the index in the ring must correspond to the WiFi
1509	 * sequence number. This is a HW requirements to help the SCD to parse
1510	 * the BA.
1511	 * Check here that the packets are in the right place on the ring.
1512	 */
1513	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
1514	WARN_ONCE(txq->ampdu &&
1515		  (wifi_seq & 0xff) != txq->write_ptr,
1516		  "Q: %d WiFi Seq %d tfdNum %d",
1517		  txq_id, wifi_seq, txq->write_ptr);
1518
1519	/* Set up driver data for this TFD */
1520	txq->entries[txq->write_ptr].skb = skb;
1521	txq->entries[txq->write_ptr].cmd = dev_cmd;
1522
1523	dev_cmd->hdr.sequence =
1524		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
1525			    INDEX_TO_SEQ(txq->write_ptr)));
1526
1527	tb0_phys = iwl_txq_get_first_tb_dma(txq, txq->write_ptr);
1528	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
1529		       offsetof(struct iwl_tx_cmd, scratch);
1530
1531	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
1532	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
1533
1534	/* Set up first empty entry in queue's array of Tx/cmd buffers */
1535	out_meta = &txq->entries[txq->write_ptr].meta;
1536	out_meta->flags = 0;
1537
1538	/*
1539	 * The second TB (tb1) points to the remainder of the TX command
1540	 * and the 802.11 header - dword aligned size
1541	 * (This calculation modifies the TX command, so do it before the
1542	 * setup of the first TB)
1543	 */
1544	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
1545	      hdr_len - IWL_FIRST_TB_SIZE;
1546	/* do not align A-MSDU to dword as the subframe header aligns it */
1547	amsdu = ieee80211_is_data_qos(fc) &&
1548		(*ieee80211_get_qos_ctl(hdr) &
1549		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
1550	if (!amsdu) {
1551		tb1_len = ALIGN(len, 4);
1552		/* Tell NIC about any 2-byte padding after MAC header */
1553		if (tb1_len != len)
1554			tx_cmd->tx_flags |= cpu_to_le32(TX_CMD_FLG_MH_PAD);
1555	} else {
1556		tb1_len = len;
1557	}
1558
1559	/*
1560	 * The first TB points to bi-directional DMA data, we'll
1561	 * memcpy the data into it later.
1562	 */
1563	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
1564			       IWL_FIRST_TB_SIZE, true);
1565
1566	/* there must be data left over for TB1 or this code must be changed */
1567	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
1568	BUILD_BUG_ON(sizeof(struct iwl_cmd_header) +
1569		     offsetofend(struct iwl_tx_cmd, scratch) >
1570		     IWL_FIRST_TB_SIZE);
1571
1572	/* map the data for TB1 */
1573	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
1574	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
1575	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
1576		goto out_err;
1577	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
1578
1579	trace_iwlwifi_dev_tx(trans->dev, skb,
1580			     iwl_txq_get_tfd(trans, txq, txq->write_ptr),
1581			     trans->txqs.tfd.size,
 
1582			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
1583			     hdr_len);
1584
1585	/*
1586	 * If gso_size wasn't set, don't give the frame "amsdu treatment"
1587	 * (adding subframes, etc.).
1588	 * This can happen in some testing flows when the amsdu was already
1589	 * pre-built, and we just need to send the resulting skb.
1590	 */
1591	if (amsdu && skb_shinfo(skb)->gso_size) {
1592		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
1593						     out_meta, dev_cmd,
1594						     tb1_len)))
1595			goto out_err;
1596	} else {
1597		struct sk_buff *frag;
1598
1599		if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
1600					       out_meta)))
1601			goto out_err;
1602
1603		skb_walk_frags(skb, frag) {
1604			if (unlikely(iwl_fill_data_tbs(trans, frag, txq, 0,
1605						       out_meta)))
1606				goto out_err;
1607		}
1608	}
1609
1610	/* building the A-MSDU might have changed this data, so memcpy it now */
1611	memcpy(&txq->first_tb_bufs[txq->write_ptr], dev_cmd, IWL_FIRST_TB_SIZE);
1612
1613	tfd = iwl_txq_get_tfd(trans, txq, txq->write_ptr);
1614	/* Set up entry for this TFD in Tx byte-count array */
1615	iwl_txq_gen1_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len),
1616					 iwl_txq_gen1_tfd_get_num_tbs(trans,
1617								      tfd));
1618
1619	wait_write_ptr = ieee80211_has_morefrags(fc);
1620
1621	/* start timer if queue currently empty */
1622	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) {
1623		/*
1624		 * If the TXQ is active, then set the timer, if not,
1625		 * set the timer in remainder so that the timer will
1626		 * be armed with the right value when the station will
1627		 * wake up.
1628		 */
1629		if (!txq->frozen)
1630			mod_timer(&txq->stuck_timer,
1631				  jiffies + txq->wd_timeout);
1632		else
1633			txq->frozen_expiry_remainder = txq->wd_timeout;
1634	}
1635
1636	/* Tell device the write index *just past* this latest filled TFD */
1637	txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr);
1638	if (!wait_write_ptr)
1639		iwl_pcie_txq_inc_wr_ptr(trans, txq);
1640
1641	/*
1642	 * At this point the frame is "transmitted" successfully
1643	 * and we will get a TX status notification eventually.
1644	 */
1645	spin_unlock(&txq->lock);
1646	return 0;
1647out_err:
1648	iwl_txq_gen1_tfd_unmap(trans, out_meta, txq, txq->write_ptr);
1649	spin_unlock(&txq->lock);
1650	return -1;
1651}
v5.4
   1/******************************************************************************
   2 *
   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.
   5 *
   6 * GPL LICENSE SUMMARY
   7 *
   8 * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
   9 * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  10 * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
  11 * Copyright(c) 2018 - 2019 Intel Corporation
  12 *
  13 * This program is free software; you can redistribute it and/or modify it
  14 * under the terms of version 2 of the GNU General Public License as
  15 * published by the Free Software Foundation.
  16 *
  17 * This program is distributed in the hope that it will be useful, but WITHOUT
  18 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  19 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  20 * more details.
  21 *
  22 * The full GNU General Public License is included in this distribution in the
  23 * file called COPYING.
  24 *
  25 * Contact Information:
  26 *  Intel Linux Wireless <linuxwifi@intel.com>
  27 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  28 *
  29 * BSD LICENSE
  30 *
  31 * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved.
  32 * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH
  33 * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
  34 * Copyright(c) 2018 - 2019 Intel Corporation
  35 * All rights reserved.
  36 *
  37 * Redistribution and use in source and binary forms, with or without
  38 * modification, are permitted provided that the following conditions
  39 * are met:
  40 *
  41 *  * Redistributions of source code must retain the above copyright
  42 *    notice, this list of conditions and the following disclaimer.
  43 *  * Redistributions in binary form must reproduce the above copyright
  44 *    notice, this list of conditions and the following disclaimer in
  45 *    the documentation and/or other materials provided with the
  46 *    distribution.
  47 *  * Neither the name Intel Corporation nor the names of its
  48 *    contributors may be used to endorse or promote products derived
  49 *    from this software without specific prior written permission.
  50 *
  51 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  52 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  53 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  54 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  55 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  56 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  57 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  58 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  59 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  60 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  61 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  62 *
  63 *****************************************************************************/
  64#include <linux/etherdevice.h>
  65#include <linux/ieee80211.h>
  66#include <linux/slab.h>
  67#include <linux/sched.h>
  68#include <net/ip6_checksum.h>
  69#include <net/tso.h>
  70
  71#include "iwl-debug.h"
  72#include "iwl-csr.h"
  73#include "iwl-prph.h"
  74#include "iwl-io.h"
  75#include "iwl-scd.h"
  76#include "iwl-op-mode.h"
  77#include "internal.h"
  78#include "fw/api/tx.h"
  79
  80#define IWL_TX_CRC_SIZE 4
  81#define IWL_TX_DELIMITER_SIZE 4
  82
  83/*************** DMA-QUEUE-GENERAL-FUNCTIONS  *****
  84 * DMA services
  85 *
  86 * Theory of operation
  87 *
  88 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer
  89 * of buffer descriptors, each of which points to one or more data buffers for
  90 * the device to read from or fill.  Driver and device exchange status of each
  91 * queue via "read" and "write" pointers.  Driver keeps minimum of 2 empty
  92 * entries in each circular buffer, to protect against confusing empty and full
  93 * queue states.
  94 *
  95 * The device reads or writes the data in the queues via the device's several
  96 * DMA/FIFO channels.  Each queue is mapped to a single DMA channel.
  97 *
  98 * For Tx queue, there are low mark and high mark limits. If, after queuing
  99 * the packet for Tx, free space become < low mark, Tx queue stopped. When
 100 * reclaiming packets (on 'tx done IRQ), if free space become > high mark,
 101 * Tx queue resumed.
 102 *
 103 ***************************************************/
 104
 105int iwl_queue_space(struct iwl_trans *trans, const struct iwl_txq *q)
 106{
 107	unsigned int max;
 108	unsigned int used;
 109
 110	/*
 111	 * To avoid ambiguity between empty and completely full queues, there
 112	 * should always be less than max_tfd_queue_size elements in the queue.
 113	 * If q->n_window is smaller than max_tfd_queue_size, there is no need
 114	 * to reserve any queue entries for this purpose.
 115	 */
 116	if (q->n_window < trans->trans_cfg->base_params->max_tfd_queue_size)
 117		max = q->n_window;
 118	else
 119		max = trans->trans_cfg->base_params->max_tfd_queue_size - 1;
 120
 121	/*
 122	 * max_tfd_queue_size is a power of 2, so the following is equivalent to
 123	 * modulo by max_tfd_queue_size and is well defined.
 124	 */
 125	used = (q->write_ptr - q->read_ptr) &
 126		(trans->trans_cfg->base_params->max_tfd_queue_size - 1);
 127
 128	if (WARN_ON(used > max))
 129		return 0;
 130
 131	return max - used;
 132}
 133
 134/*
 135 * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
 136 */
 137static int iwl_queue_init(struct iwl_txq *q, int slots_num)
 138{
 139	q->n_window = slots_num;
 140
 141	/* slots_num must be power-of-two size, otherwise
 142	 * iwl_pcie_get_cmd_index is broken. */
 143	if (WARN_ON(!is_power_of_2(slots_num)))
 144		return -EINVAL;
 145
 146	q->low_mark = q->n_window / 4;
 147	if (q->low_mark < 4)
 148		q->low_mark = 4;
 149
 150	q->high_mark = q->n_window / 8;
 151	if (q->high_mark < 2)
 152		q->high_mark = 2;
 153
 154	q->write_ptr = 0;
 155	q->read_ptr = 0;
 156
 157	return 0;
 158}
 159
 160int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans,
 161			   struct iwl_dma_ptr *ptr, size_t size)
 162{
 163	if (WARN_ON(ptr->addr))
 164		return -EINVAL;
 165
 166	ptr->addr = dma_alloc_coherent(trans->dev, size,
 167				       &ptr->dma, GFP_KERNEL);
 168	if (!ptr->addr)
 169		return -ENOMEM;
 170	ptr->size = size;
 171	return 0;
 172}
 173
 174void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr)
 175{
 176	if (unlikely(!ptr->addr))
 177		return;
 178
 179	dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma);
 180	memset(ptr, 0, sizeof(*ptr));
 181}
 182
 183static void iwl_pcie_txq_stuck_timer(struct timer_list *t)
 184{
 185	struct iwl_txq *txq = from_timer(txq, t, stuck_timer);
 186	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
 187	struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie);
 188
 189	spin_lock(&txq->lock);
 190	/* check if triggered erroneously */
 191	if (txq->read_ptr == txq->write_ptr) {
 192		spin_unlock(&txq->lock);
 193		return;
 194	}
 195	spin_unlock(&txq->lock);
 196
 197	iwl_trans_pcie_log_scd_error(trans, txq);
 198
 199	iwl_force_nmi(trans);
 200}
 201
 202/*
 203 * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array
 204 */
 205static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
 206					     struct iwl_txq *txq, u16 byte_cnt,
 207					     int num_tbs)
 208{
 209	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
 210	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 211	int write_ptr = txq->write_ptr;
 212	int txq_id = txq->id;
 213	u8 sec_ctl = 0;
 214	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
 215	__le16 bc_ent;
 216	struct iwl_tx_cmd *tx_cmd =
 217		(void *)txq->entries[txq->write_ptr].cmd->payload;
 218	u8 sta_id = tx_cmd->sta_id;
 219
 220	scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
 221
 222	sec_ctl = tx_cmd->sec_ctl;
 223
 224	switch (sec_ctl & TX_CMD_SEC_MSK) {
 225	case TX_CMD_SEC_CCM:
 226		len += IEEE80211_CCMP_MIC_LEN;
 227		break;
 228	case TX_CMD_SEC_TKIP:
 229		len += IEEE80211_TKIP_ICV_LEN;
 230		break;
 231	case TX_CMD_SEC_WEP:
 232		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
 233		break;
 234	}
 235	if (trans_pcie->bc_table_dword)
 236		len = DIV_ROUND_UP(len, 4);
 237
 238	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
 239		return;
 240
 241	bc_ent = cpu_to_le16(len | (sta_id << 12));
 242
 243	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
 244
 245	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
 246		scd_bc_tbl[txq_id].
 247			tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent;
 248}
 249
 250static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
 251					    struct iwl_txq *txq)
 252{
 253	struct iwl_trans_pcie *trans_pcie =
 254		IWL_TRANS_GET_PCIE_TRANS(trans);
 255	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr;
 256	int txq_id = txq->id;
 257	int read_ptr = txq->read_ptr;
 258	u8 sta_id = 0;
 259	__le16 bc_ent;
 260	struct iwl_tx_cmd *tx_cmd =
 261		(void *)txq->entries[read_ptr].cmd->payload;
 262
 263	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
 264
 265	if (txq_id != trans_pcie->cmd_queue)
 266		sta_id = tx_cmd->sta_id;
 267
 268	bc_ent = cpu_to_le16(1 | (sta_id << 12));
 269
 270	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
 271
 272	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
 273		scd_bc_tbl[txq_id].
 274			tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent;
 275}
 276
 277/*
 278 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware
 279 */
 280static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans,
 281				    struct iwl_txq *txq)
 282{
 283	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 284	u32 reg = 0;
 285	int txq_id = txq->id;
 286
 287	lockdep_assert_held(&txq->lock);
 288
 289	/*
 290	 * explicitly wake up the NIC if:
 291	 * 1. shadow registers aren't enabled
 292	 * 2. NIC is woken up for CMD regardless of shadow outside this function
 293	 * 3. there is a chance that the NIC is asleep
 294	 */
 295	if (!trans->trans_cfg->base_params->shadow_reg_enable &&
 296	    txq_id != trans_pcie->cmd_queue &&
 297	    test_bit(STATUS_TPOWER_PMI, &trans->status)) {
 298		/*
 299		 * wake up nic if it's powered down ...
 300		 * uCode will wake up, and interrupt us again, so next
 301		 * time we'll skip this part.
 302		 */
 303		reg = iwl_read32(trans, CSR_UCODE_DRV_GP1);
 304
 305		if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) {
 306			IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n",
 307				       txq_id, reg);
 308			iwl_set_bit(trans, CSR_GP_CNTRL,
 309				    BIT(trans->trans_cfg->csr->flag_mac_access_req));
 310			txq->need_update = true;
 311			return;
 312		}
 313	}
 314
 315	/*
 316	 * if not in power-save mode, uCode will never sleep when we're
 317	 * trying to tx (during RFKILL, we're not trying to tx).
 318	 */
 319	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr);
 320	if (!txq->block)
 321		iwl_write32(trans, HBUS_TARG_WRPTR,
 322			    txq->write_ptr | (txq_id << 8));
 323}
 324
 325void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans)
 326{
 327	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 328	int i;
 329
 330	for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) {
 331		struct iwl_txq *txq = trans_pcie->txq[i];
 332
 333		if (!test_bit(i, trans_pcie->queue_used))
 334			continue;
 335
 336		spin_lock_bh(&txq->lock);
 337		if (txq->need_update) {
 338			iwl_pcie_txq_inc_wr_ptr(trans, txq);
 339			txq->need_update = false;
 340		}
 341		spin_unlock_bh(&txq->lock);
 342	}
 343}
 344
 345static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_trans *trans,
 346						  void *_tfd, u8 idx)
 347{
 348
 349	if (trans->trans_cfg->use_tfh) {
 350		struct iwl_tfh_tfd *tfd = _tfd;
 351		struct iwl_tfh_tb *tb = &tfd->tbs[idx];
 352
 353		return (dma_addr_t)(le64_to_cpu(tb->addr));
 354	} else {
 355		struct iwl_tfd *tfd = _tfd;
 356		struct iwl_tfd_tb *tb = &tfd->tbs[idx];
 357		dma_addr_t addr = get_unaligned_le32(&tb->lo);
 358		dma_addr_t hi_len;
 359
 360		if (sizeof(dma_addr_t) <= sizeof(u32))
 361			return addr;
 362
 363		hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
 364
 365		/*
 366		 * shift by 16 twice to avoid warnings on 32-bit
 367		 * (where this code never runs anyway due to the
 368		 * if statement above)
 369		 */
 370		return addr | ((hi_len << 16) << 16);
 371	}
 372}
 373
 374static inline void iwl_pcie_tfd_set_tb(struct iwl_trans *trans, void *tfd,
 375				       u8 idx, dma_addr_t addr, u16 len)
 376{
 377	struct iwl_tfd *tfd_fh = (void *)tfd;
 378	struct iwl_tfd_tb *tb = &tfd_fh->tbs[idx];
 379
 380	u16 hi_n_len = len << 4;
 381
 382	put_unaligned_le32(addr, &tb->lo);
 383	hi_n_len |= iwl_get_dma_hi_addr(addr);
 384
 385	tb->hi_n_len = cpu_to_le16(hi_n_len);
 386
 387	tfd_fh->num_tbs = idx + 1;
 388}
 389
 390static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_trans *trans, void *_tfd)
 391{
 392	if (trans->trans_cfg->use_tfh) {
 393		struct iwl_tfh_tfd *tfd = _tfd;
 394
 395		return le16_to_cpu(tfd->num_tbs) & 0x1f;
 396	} else {
 397		struct iwl_tfd *tfd = _tfd;
 398
 399		return tfd->num_tbs & 0x1f;
 400	}
 401}
 402
 403static void iwl_pcie_tfd_unmap(struct iwl_trans *trans,
 404			       struct iwl_cmd_meta *meta,
 405			       struct iwl_txq *txq, int index)
 406{
 407	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 408	int i, num_tbs;
 409	void *tfd = iwl_pcie_get_tfd(trans, txq, index);
 410
 411	/* Sanity check on number of chunks */
 412	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
 413
 414	if (num_tbs > trans_pcie->max_tbs) {
 415		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
 416		/* @todo issue fatal error, it is quite serious situation */
 417		return;
 418	}
 419
 420	/* first TB is never freed - it's the bidirectional DMA data */
 421
 422	for (i = 1; i < num_tbs; i++) {
 423		if (meta->tbs & BIT(i))
 424			dma_unmap_page(trans->dev,
 425				       iwl_pcie_tfd_tb_get_addr(trans, tfd, i),
 426				       iwl_pcie_tfd_tb_get_len(trans, tfd, i),
 427				       DMA_TO_DEVICE);
 428		else
 429			dma_unmap_single(trans->dev,
 430					 iwl_pcie_tfd_tb_get_addr(trans, tfd,
 431								  i),
 432					 iwl_pcie_tfd_tb_get_len(trans, tfd,
 433								 i),
 434					 DMA_TO_DEVICE);
 435	}
 436
 437	meta->tbs = 0;
 438
 439	if (trans->trans_cfg->use_tfh) {
 440		struct iwl_tfh_tfd *tfd_fh = (void *)tfd;
 441
 442		tfd_fh->num_tbs = 0;
 443	} else {
 444		struct iwl_tfd *tfd_fh = (void *)tfd;
 445
 446		tfd_fh->num_tbs = 0;
 447	}
 448
 449}
 450
 451/*
 452 * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
 453 * @trans - transport private data
 454 * @txq - tx queue
 455 * @dma_dir - the direction of the DMA mapping
 456 *
 457 * Does NOT advance any TFD circular buffer read/write indexes
 458 * Does NOT free the TFD itself (which is within circular buffer)
 459 */
 460void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
 461{
 462	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
 463	 * idx is bounded by n_window
 464	 */
 465	int rd_ptr = txq->read_ptr;
 466	int idx = iwl_pcie_get_cmd_index(txq, rd_ptr);
 467
 468	lockdep_assert_held(&txq->lock);
 469
 470	/* We have only q->n_window txq->entries, but we use
 471	 * TFD_QUEUE_SIZE_MAX tfds
 472	 */
 473	iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, txq, rd_ptr);
 474
 475	/* free SKB */
 476	if (txq->entries) {
 477		struct sk_buff *skb;
 478
 479		skb = txq->entries[idx].skb;
 480
 481		/* Can be called from irqs-disabled context
 482		 * If skb is not NULL, it means that the whole queue is being
 483		 * freed and that the queue is not empty - free the skb
 484		 */
 485		if (skb) {
 486			iwl_op_mode_free_skb(trans->op_mode, skb);
 487			txq->entries[idx].skb = NULL;
 488		}
 489	}
 490}
 491
 492static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq,
 493				  dma_addr_t addr, u16 len, bool reset)
 494{
 495	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 496	void *tfd;
 497	u32 num_tbs;
 498
 499	tfd = txq->tfds + trans_pcie->tfd_size * txq->write_ptr;
 500
 501	if (reset)
 502		memset(tfd, 0, trans_pcie->tfd_size);
 503
 504	num_tbs = iwl_pcie_tfd_get_num_tbs(trans, tfd);
 505
 506	/* Each TFD can point to a maximum max_tbs Tx buffers */
 507	if (num_tbs >= trans_pcie->max_tbs) {
 508		IWL_ERR(trans, "Error can not send more than %d chunks\n",
 509			trans_pcie->max_tbs);
 510		return -EINVAL;
 511	}
 512
 513	if (WARN(addr & ~IWL_TX_DMA_MASK,
 514		 "Unaligned address = %llx\n", (unsigned long long)addr))
 515		return -EINVAL;
 516
 517	iwl_pcie_tfd_set_tb(trans, tfd, num_tbs, addr, len);
 518
 519	return num_tbs;
 520}
 521
 522int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq,
 523		       int slots_num, bool cmd_queue)
 524{
 525	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 526	size_t tfd_sz = trans_pcie->tfd_size *
 527		trans->trans_cfg->base_params->max_tfd_queue_size;
 528	size_t tb0_buf_sz;
 529	int i;
 530
 531	if (WARN_ON(txq->entries || txq->tfds))
 532		return -EINVAL;
 533
 534	if (trans->trans_cfg->use_tfh)
 535		tfd_sz = trans_pcie->tfd_size * slots_num;
 536
 537	timer_setup(&txq->stuck_timer, iwl_pcie_txq_stuck_timer, 0);
 538	txq->trans_pcie = trans_pcie;
 539
 540	txq->n_window = slots_num;
 541
 542	txq->entries = kcalloc(slots_num,
 543			       sizeof(struct iwl_pcie_txq_entry),
 544			       GFP_KERNEL);
 545
 546	if (!txq->entries)
 547		goto error;
 548
 549	if (cmd_queue)
 550		for (i = 0; i < slots_num; i++) {
 551			txq->entries[i].cmd =
 552				kmalloc(sizeof(struct iwl_device_cmd),
 553					GFP_KERNEL);
 554			if (!txq->entries[i].cmd)
 555				goto error;
 556		}
 557
 558	/* Circular buffer of transmit frame descriptors (TFDs),
 559	 * shared with device */
 560	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
 561				       &txq->dma_addr, GFP_KERNEL);
 562	if (!txq->tfds)
 563		goto error;
 564
 565	BUILD_BUG_ON(IWL_FIRST_TB_SIZE_ALIGN != sizeof(*txq->first_tb_bufs));
 566
 567	tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
 568
 569	txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
 570					      &txq->first_tb_dma,
 571					      GFP_KERNEL);
 572	if (!txq->first_tb_bufs)
 573		goto err_free_tfds;
 574
 575	return 0;
 576err_free_tfds:
 577	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
 578error:
 579	if (txq->entries && cmd_queue)
 580		for (i = 0; i < slots_num; i++)
 581			kfree(txq->entries[i].cmd);
 582	kfree(txq->entries);
 583	txq->entries = NULL;
 584
 585	return -ENOMEM;
 586
 587}
 588
 589int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq,
 590		      int slots_num, bool cmd_queue)
 591{
 592	int ret;
 593	u32 tfd_queue_max_size =
 594		trans->trans_cfg->base_params->max_tfd_queue_size;
 595
 596	txq->need_update = false;
 597
 598	/* max_tfd_queue_size must be power-of-two size, otherwise
 599	 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */
 600	if (WARN_ONCE(tfd_queue_max_size & (tfd_queue_max_size - 1),
 601		      "Max tfd queue size must be a power of two, but is %d",
 602		      tfd_queue_max_size))
 603		return -EINVAL;
 604
 605	/* Initialize queue's high/low-water marks, and head/tail indexes */
 606	ret = iwl_queue_init(txq, slots_num);
 607	if (ret)
 608		return ret;
 609
 610	spin_lock_init(&txq->lock);
 611
 612	if (cmd_queue) {
 613		static struct lock_class_key iwl_pcie_cmd_queue_lock_class;
 614
 615		lockdep_set_class(&txq->lock, &iwl_pcie_cmd_queue_lock_class);
 616	}
 617
 618	__skb_queue_head_init(&txq->overflow_q);
 619
 620	return 0;
 621}
 622
 623void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
 624			    struct sk_buff *skb)
 625{
 626	struct page **page_ptr;
 627
 628	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
 629
 630	if (*page_ptr) {
 631		__free_page(*page_ptr);
 632		*page_ptr = NULL;
 633	}
 634}
 635
 636static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans)
 637{
 638	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 639
 640	lockdep_assert_held(&trans_pcie->reg_lock);
 641
 642	if (!trans->trans_cfg->base_params->apmg_wake_up_wa)
 643		return;
 644	if (WARN_ON(!trans_pcie->cmd_hold_nic_awake))
 
 
 
 
 645		return;
 
 646
 647	trans_pcie->cmd_hold_nic_awake = false;
 648	__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
 649				   BIT(trans->trans_cfg->csr->flag_mac_access_req));
 
 650}
 651
 652/*
 653 * iwl_pcie_txq_unmap -  Unmap any remaining DMA mappings and free skb's
 654 */
 655static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id)
 656{
 657	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 658	struct iwl_txq *txq = trans_pcie->txq[txq_id];
 
 
 
 
 659
 660	spin_lock_bh(&txq->lock);
 661	while (txq->write_ptr != txq->read_ptr) {
 662		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
 663				   txq_id, txq->read_ptr);
 664
 665		if (txq_id != trans_pcie->cmd_queue) {
 666			struct sk_buff *skb = txq->entries[txq->read_ptr].skb;
 667
 668			if (WARN_ON_ONCE(!skb))
 669				continue;
 670
 671			iwl_pcie_free_tso_page(trans_pcie, skb);
 672		}
 673		iwl_pcie_txq_free_tfd(trans, txq);
 674		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
 675
 676		if (txq->read_ptr == txq->write_ptr) {
 677			unsigned long flags;
 678
 679			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
 680			if (txq_id == trans_pcie->cmd_queue)
 681				iwl_pcie_clear_cmd_in_flight(trans);
 682			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
 683		}
 684	}
 685
 686	while (!skb_queue_empty(&txq->overflow_q)) {
 687		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
 688
 689		iwl_op_mode_free_skb(trans->op_mode, skb);
 690	}
 691
 692	spin_unlock_bh(&txq->lock);
 693
 694	/* just in case - this queue may have been stopped */
 695	iwl_wake_queue(trans, txq);
 696}
 697
 698/*
 699 * iwl_pcie_txq_free - Deallocate DMA queue.
 700 * @txq: Transmit queue to deallocate.
 701 *
 702 * Empty queue by removing and destroying all BD's.
 703 * Free all buffers.
 704 * 0-fill, but do not free "txq" descriptor structure.
 705 */
 706static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id)
 707{
 708	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 709	struct iwl_txq *txq = trans_pcie->txq[txq_id];
 710	struct device *dev = trans->dev;
 711	int i;
 712
 713	if (WARN_ON(!txq))
 714		return;
 715
 716	iwl_pcie_txq_unmap(trans, txq_id);
 717
 718	/* De-alloc array of command/tx buffers */
 719	if (txq_id == trans_pcie->cmd_queue)
 720		for (i = 0; i < txq->n_window; i++) {
 721			kzfree(txq->entries[i].cmd);
 722			kzfree(txq->entries[i].free_buf);
 723		}
 724
 725	/* De-alloc circular buffer of TFDs */
 726	if (txq->tfds) {
 727		dma_free_coherent(dev,
 728				  trans_pcie->tfd_size *
 729				  trans->trans_cfg->base_params->max_tfd_queue_size,
 730				  txq->tfds, txq->dma_addr);
 731		txq->dma_addr = 0;
 732		txq->tfds = NULL;
 733
 734		dma_free_coherent(dev,
 735				  sizeof(*txq->first_tb_bufs) * txq->n_window,
 736				  txq->first_tb_bufs, txq->first_tb_dma);
 737	}
 738
 739	kfree(txq->entries);
 740	txq->entries = NULL;
 741
 742	del_timer_sync(&txq->stuck_timer);
 743
 744	/* 0-fill queue descriptor structure */
 745	memset(txq, 0, sizeof(*txq));
 746}
 747
 748void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr)
 749{
 750	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 751	int nq = trans->trans_cfg->base_params->num_of_queues;
 752	int chan;
 753	u32 reg_val;
 754	int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) -
 755				SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32);
 756
 757	/* make sure all queue are not stopped/used */
 758	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
 759	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
 
 760
 761	trans_pcie->scd_base_addr =
 762		iwl_read_prph(trans, SCD_SRAM_BASE_ADDR);
 763
 764	WARN_ON(scd_base_addr != 0 &&
 765		scd_base_addr != trans_pcie->scd_base_addr);
 766
 767	/* reset context data, TX status and translation data */
 768	iwl_trans_write_mem(trans, trans_pcie->scd_base_addr +
 769				   SCD_CONTEXT_MEM_LOWER_BOUND,
 770			    NULL, clear_dwords);
 771
 772	iwl_write_prph(trans, SCD_DRAM_BASE_ADDR,
 773		       trans_pcie->scd_bc_tbls.dma >> 10);
 774
 775	/* The chain extension of the SCD doesn't work well. This feature is
 776	 * enabled by default by the HW, so we need to disable it manually.
 777	 */
 778	if (trans->trans_cfg->base_params->scd_chain_ext_wa)
 779		iwl_write_prph(trans, SCD_CHAINEXT_EN, 0);
 780
 781	iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue,
 782				trans_pcie->cmd_fifo,
 783				trans_pcie->cmd_q_wdg_timeout);
 784
 785	/* Activate all Tx DMA/FIFO channels */
 786	iwl_scd_activate_fifos(trans);
 787
 788	/* Enable DMA channel */
 789	for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++)
 790		iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan),
 791				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE |
 792				   FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE);
 793
 794	/* Update FH chicken bits */
 795	reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG);
 796	iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG,
 797			   reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN);
 798
 799	/* Enable L1-Active */
 800	if (trans->trans_cfg->device_family < IWL_DEVICE_FAMILY_8000)
 801		iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG,
 802				    APMG_PCIDEV_STT_VAL_L1_ACT_DIS);
 803}
 804
 805void iwl_trans_pcie_tx_reset(struct iwl_trans *trans)
 806{
 807	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 808	int txq_id;
 809
 810	/*
 811	 * we should never get here in gen2 trans mode return early to avoid
 812	 * having invalid accesses
 813	 */
 814	if (WARN_ON_ONCE(trans->trans_cfg->gen2))
 815		return;
 816
 817	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 818	     txq_id++) {
 819		struct iwl_txq *txq = trans_pcie->txq[txq_id];
 820		if (trans->trans_cfg->use_tfh)
 821			iwl_write_direct64(trans,
 822					   FH_MEM_CBBC_QUEUE(trans, txq_id),
 823					   txq->dma_addr);
 824		else
 825			iwl_write_direct32(trans,
 826					   FH_MEM_CBBC_QUEUE(trans, txq_id),
 827					   txq->dma_addr >> 8);
 828		iwl_pcie_txq_unmap(trans, txq_id);
 829		txq->read_ptr = 0;
 830		txq->write_ptr = 0;
 831	}
 832
 833	/* Tell NIC where to find the "keep warm" buffer */
 834	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
 835			   trans_pcie->kw.dma >> 4);
 836
 837	/*
 838	 * Send 0 as the scd_base_addr since the device may have be reset
 839	 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will
 840	 * contain garbage.
 841	 */
 842	iwl_pcie_tx_start(trans, 0);
 843}
 844
 845static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans)
 846{
 847	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 848	unsigned long flags;
 849	int ch, ret;
 850	u32 mask = 0;
 851
 852	spin_lock(&trans_pcie->irq_lock);
 853
 854	if (!iwl_trans_grab_nic_access(trans, &flags))
 855		goto out;
 856
 857	/* Stop each Tx DMA channel */
 858	for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) {
 859		iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0);
 860		mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch);
 861	}
 862
 863	/* Wait for DMA channels to be idle */
 864	ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000);
 865	if (ret < 0)
 866		IWL_ERR(trans,
 867			"Failing on timeout while stopping DMA channel %d [0x%08x]\n",
 868			ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG));
 869
 870	iwl_trans_release_nic_access(trans, &flags);
 871
 872out:
 873	spin_unlock(&trans_pcie->irq_lock);
 874}
 875
 876/*
 877 * iwl_pcie_tx_stop - Stop all Tx DMA channels
 878 */
 879int iwl_pcie_tx_stop(struct iwl_trans *trans)
 880{
 881	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 882	int txq_id;
 883
 884	/* Turn off all Tx DMA fifos */
 885	iwl_scd_deactivate_fifos(trans);
 886
 887	/* Turn off all Tx DMA channels */
 888	iwl_pcie_tx_stop_fh(trans);
 889
 890	/*
 891	 * This function can be called before the op_mode disabled the
 892	 * queues. This happens when we have an rfkill interrupt.
 893	 * Since we stop Tx altogether - mark the queues as stopped.
 894	 */
 895	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
 896	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
 
 897
 898	/* This can happen: start_hw, stop_device */
 899	if (!trans_pcie->txq_memory)
 900		return 0;
 901
 902	/* Unmap DMA from host system and free skb's */
 903	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 904	     txq_id++)
 905		iwl_pcie_txq_unmap(trans, txq_id);
 906
 907	return 0;
 908}
 909
 910/*
 911 * iwl_trans_tx_free - Free TXQ Context
 912 *
 913 * Destroy all TX DMA queues and structures
 914 */
 915void iwl_pcie_tx_free(struct iwl_trans *trans)
 916{
 917	int txq_id;
 918	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 919
 920	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
 921
 922	/* Tx queues */
 923	if (trans_pcie->txq_memory) {
 924		for (txq_id = 0;
 925		     txq_id < trans->trans_cfg->base_params->num_of_queues;
 926		     txq_id++) {
 927			iwl_pcie_txq_free(trans, txq_id);
 928			trans_pcie->txq[txq_id] = NULL;
 929		}
 930	}
 931
 932	kfree(trans_pcie->txq_memory);
 933	trans_pcie->txq_memory = NULL;
 934
 935	iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw);
 936
 937	iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls);
 938}
 939
 940/*
 941 * iwl_pcie_tx_alloc - allocate TX context
 942 * Allocate all Tx DMA structures and initialize them
 943 */
 944static int iwl_pcie_tx_alloc(struct iwl_trans *trans)
 945{
 946	int ret;
 947	int txq_id, slots_num;
 948	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 949	u16 bc_tbls_size = trans->trans_cfg->base_params->num_of_queues;
 950
 951	bc_tbls_size *= (trans->trans_cfg->device_family >=
 952			 IWL_DEVICE_FAMILY_22560) ?
 953		sizeof(struct iwl_gen3_bc_tbl) :
 954		sizeof(struct iwlagn_scd_bc_tbl);
 955
 956	/*It is not allowed to alloc twice, so warn when this happens.
 957	 * We cannot rely on the previous allocation, so free and fail */
 958	if (WARN_ON(trans_pcie->txq_memory)) {
 959		ret = -EINVAL;
 960		goto error;
 961	}
 962
 963	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls,
 964				     bc_tbls_size);
 965	if (ret) {
 966		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
 967		goto error;
 968	}
 969
 970	/* Alloc keep-warm buffer */
 971	ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE);
 972	if (ret) {
 973		IWL_ERR(trans, "Keep Warm allocation failed\n");
 974		goto error;
 975	}
 976
 977	trans_pcie->txq_memory =
 978		kcalloc(trans->trans_cfg->base_params->num_of_queues,
 979			sizeof(struct iwl_txq), GFP_KERNEL);
 980	if (!trans_pcie->txq_memory) {
 981		IWL_ERR(trans, "Not enough memory for txq\n");
 982		ret = -ENOMEM;
 983		goto error;
 984	}
 985
 986	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
 987	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
 988	     txq_id++) {
 989		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
 990
 991		if (cmd_queue)
 992			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
 993					  trans->cfg->min_txq_size);
 994		else
 995			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
 996					  trans->cfg->min_256_ba_txq_size);
 997		trans_pcie->txq[txq_id] = &trans_pcie->txq_memory[txq_id];
 998		ret = iwl_pcie_txq_alloc(trans, trans_pcie->txq[txq_id],
 999					 slots_num, cmd_queue);
1000		if (ret) {
1001			IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1002			goto error;
1003		}
1004		trans_pcie->txq[txq_id]->id = txq_id;
1005	}
1006
1007	return 0;
1008
1009error:
1010	iwl_pcie_tx_free(trans);
1011
1012	return ret;
1013}
1014
1015int iwl_pcie_tx_init(struct iwl_trans *trans)
1016{
1017	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1018	int ret;
1019	int txq_id, slots_num;
1020	bool alloc = false;
1021
1022	if (!trans_pcie->txq_memory) {
1023		ret = iwl_pcie_tx_alloc(trans);
1024		if (ret)
1025			goto error;
1026		alloc = true;
1027	}
1028
1029	spin_lock(&trans_pcie->irq_lock);
1030
1031	/* Turn off all Tx DMA fifos */
1032	iwl_scd_deactivate_fifos(trans);
1033
1034	/* Tell NIC where to find the "keep warm" buffer */
1035	iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG,
1036			   trans_pcie->kw.dma >> 4);
1037
1038	spin_unlock(&trans_pcie->irq_lock);
1039
1040	/* Alloc and init all Tx queues, including the command queue (#4/#9) */
1041	for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues;
1042	     txq_id++) {
1043		bool cmd_queue = (txq_id == trans_pcie->cmd_queue);
1044
1045		if (cmd_queue)
1046			slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE,
1047					  trans->cfg->min_txq_size);
1048		else
1049			slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE,
1050					  trans->cfg->min_256_ba_txq_size);
1051		ret = iwl_pcie_txq_init(trans, trans_pcie->txq[txq_id],
1052					slots_num, cmd_queue);
1053		if (ret) {
1054			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1055			goto error;
1056		}
1057
1058		/*
1059		 * Tell nic where to find circular buffer of TFDs for a
1060		 * given Tx queue, and enable the DMA channel used for that
1061		 * queue.
1062		 * Circular buffer (TFD queue in DRAM) physical base address
1063		 */
1064		iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id),
1065				   trans_pcie->txq[txq_id]->dma_addr >> 8);
1066	}
1067
1068	iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE);
1069	if (trans->trans_cfg->base_params->num_of_queues > 20)
1070		iwl_set_bits_prph(trans, SCD_GP_CTRL,
1071				  SCD_GP_CTRL_ENABLE_31_QUEUES);
1072
1073	return 0;
1074error:
1075	/*Upon error, free only if we allocated something */
1076	if (alloc)
1077		iwl_pcie_tx_free(trans);
1078	return ret;
1079}
1080
1081static inline void iwl_pcie_txq_progress(struct iwl_txq *txq)
1082{
1083	lockdep_assert_held(&txq->lock);
1084
1085	if (!txq->wd_timeout)
1086		return;
1087
1088	/*
1089	 * station is asleep and we send data - that must
1090	 * be uAPSD or PS-Poll. Don't rearm the timer.
1091	 */
1092	if (txq->frozen)
1093		return;
1094
1095	/*
1096	 * if empty delete timer, otherwise move timer forward
1097	 * since we're making progress on this queue
1098	 */
1099	if (txq->read_ptr == txq->write_ptr)
1100		del_timer(&txq->stuck_timer);
1101	else
1102		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1103}
1104
1105/* Frees buffers until index _not_ inclusive */
1106void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1107			    struct sk_buff_head *skbs)
1108{
1109	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1110	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1111	int tfd_num = iwl_pcie_get_cmd_index(txq, ssn);
1112	int read_ptr = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1113	int last_to_free;
1114
1115	/* This function is not meant to release cmd queue*/
1116	if (WARN_ON(txq_id == trans_pcie->cmd_queue))
1117		return;
1118
1119	spin_lock_bh(&txq->lock);
1120
1121	if (!test_bit(txq_id, trans_pcie->queue_used)) {
1122		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1123				    txq_id, ssn);
1124		goto out;
1125	}
1126
1127	if (read_ptr == tfd_num)
1128		goto out;
1129
1130	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1131			   txq_id, txq->read_ptr, tfd_num, ssn);
1132
1133	/*Since we free until index _not_ inclusive, the one before index is
1134	 * the last we will free. This one must be used */
1135	last_to_free = iwl_queue_dec_wrap(trans, tfd_num);
1136
1137	if (!iwl_queue_used(txq, last_to_free)) {
1138		IWL_ERR(trans,
1139			"%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1140			__func__, txq_id, last_to_free,
1141			trans->trans_cfg->base_params->max_tfd_queue_size,
1142			txq->write_ptr, txq->read_ptr);
1143		goto out;
1144	}
1145
1146	if (WARN_ON(!skb_queue_empty(skbs)))
1147		goto out;
1148
1149	for (;
1150	     read_ptr != tfd_num;
1151	     txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr),
1152	     read_ptr = iwl_pcie_get_cmd_index(txq, txq->read_ptr)) {
1153		struct sk_buff *skb = txq->entries[read_ptr].skb;
1154
1155		if (WARN_ON_ONCE(!skb))
1156			continue;
1157
1158		iwl_pcie_free_tso_page(trans_pcie, skb);
1159
1160		__skb_queue_tail(skbs, skb);
1161
1162		txq->entries[read_ptr].skb = NULL;
1163
1164		if (!trans->trans_cfg->use_tfh)
1165			iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq);
1166
1167		iwl_pcie_txq_free_tfd(trans, txq);
1168	}
1169
1170	iwl_pcie_txq_progress(txq);
1171
1172	if (iwl_queue_space(trans, txq) > txq->low_mark &&
1173	    test_bit(txq_id, trans_pcie->queue_stopped)) {
1174		struct sk_buff_head overflow_skbs;
1175
1176		__skb_queue_head_init(&overflow_skbs);
1177		skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
1178
1179		/*
1180		 * We are going to transmit from the overflow queue.
1181		 * Remember this state so that wait_for_txq_empty will know we
1182		 * are adding more packets to the TFD queue. It cannot rely on
1183		 * the state of &txq->overflow_q, as we just emptied it, but
1184		 * haven't TXed the content yet.
1185		 */
1186		txq->overflow_tx = true;
1187
1188		/*
1189		 * This is tricky: we are in reclaim path which is non
1190		 * re-entrant, so noone will try to take the access the
1191		 * txq data from that path. We stopped tx, so we can't
1192		 * have tx as well. Bottom line, we can unlock and re-lock
1193		 * later.
1194		 */
1195		spin_unlock_bh(&txq->lock);
1196
1197		while (!skb_queue_empty(&overflow_skbs)) {
1198			struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
1199			struct iwl_device_cmd *dev_cmd_ptr;
1200
1201			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1202						 trans_pcie->dev_cmd_offs);
1203
1204			/*
1205			 * Note that we can very well be overflowing again.
1206			 * In that case, iwl_queue_space will be small again
1207			 * and we won't wake mac80211's queue.
1208			 */
1209			iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id);
1210		}
1211
1212		if (iwl_queue_space(trans, txq) > txq->low_mark)
1213			iwl_wake_queue(trans, txq);
1214
1215		spin_lock_bh(&txq->lock);
1216		txq->overflow_tx = false;
1217	}
1218
1219out:
1220	spin_unlock_bh(&txq->lock);
1221}
1222
1223/* Set wr_ptr of specific device and txq  */
1224void iwl_trans_pcie_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr)
1225{
1226	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1227	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1228
1229	spin_lock_bh(&txq->lock);
1230
1231	txq->write_ptr = ptr;
1232	txq->read_ptr = txq->write_ptr;
1233
1234	spin_unlock_bh(&txq->lock);
1235}
1236
1237static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans,
1238				      const struct iwl_host_cmd *cmd)
1239{
1240	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1241	int ret;
1242
1243	lockdep_assert_held(&trans_pcie->reg_lock);
1244
1245	/* Make sure the NIC is still alive in the bus */
1246	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1247		return -ENODEV;
1248
 
 
 
1249	/*
1250	 * wake up the NIC to make sure that the firmware will see the host
1251	 * command - we will let the NIC sleep once all the host commands
1252	 * returned. This needs to be done only on NICs that have
1253	 * apmg_wake_up_wa set.
1254	 */
1255	if (trans->trans_cfg->base_params->apmg_wake_up_wa &&
1256	    !trans_pcie->cmd_hold_nic_awake) {
1257		__iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL,
1258					 BIT(trans->trans_cfg->csr->flag_mac_access_req));
1259
1260		ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
1261				   BIT(trans->trans_cfg->csr->flag_val_mac_access_en),
1262				   (BIT(trans->trans_cfg->csr->flag_mac_clock_ready) |
1263				    CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP),
1264				   15000);
1265		if (ret < 0) {
1266			__iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL,
1267					BIT(trans->trans_cfg->csr->flag_mac_access_req));
1268			IWL_ERR(trans, "Failed to wake NIC for hcmd\n");
1269			return -EIO;
1270		}
1271		trans_pcie->cmd_hold_nic_awake = true;
1272	}
1273
1274	return 0;
1275}
1276
1277/*
1278 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd
1279 *
1280 * When FW advances 'R' index, all entries between old and new 'R' index
1281 * need to be reclaimed. As result, some free space forms.  If there is
1282 * enough free space (> low mark), wake the stack that feeds us.
1283 */
1284void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx)
1285{
1286	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1287	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1288	unsigned long flags;
1289	int nfreed = 0;
1290	u16 r;
1291
1292	lockdep_assert_held(&txq->lock);
1293
1294	idx = iwl_pcie_get_cmd_index(txq, idx);
1295	r = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
1296
1297	if (idx >= trans->trans_cfg->base_params->max_tfd_queue_size ||
1298	    (!iwl_queue_used(txq, idx))) {
1299		WARN_ONCE(test_bit(txq_id, trans_pcie->queue_used),
1300			  "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n",
1301			  __func__, txq_id, idx,
1302			  trans->trans_cfg->base_params->max_tfd_queue_size,
1303			  txq->write_ptr, txq->read_ptr);
1304		return;
1305	}
1306
1307	for (idx = iwl_queue_inc_wrap(trans, idx); r != idx;
1308	     r = iwl_queue_inc_wrap(trans, r)) {
1309		txq->read_ptr = iwl_queue_inc_wrap(trans, txq->read_ptr);
1310
1311		if (nfreed++ > 0) {
1312			IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n",
1313				idx, txq->write_ptr, r);
1314			iwl_force_nmi(trans);
1315		}
1316	}
1317
1318	if (txq->read_ptr == txq->write_ptr) {
1319		spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1320		iwl_pcie_clear_cmd_in_flight(trans);
1321		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1322	}
1323
1324	iwl_pcie_txq_progress(txq);
1325}
1326
1327static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid,
1328				 u16 txq_id)
1329{
1330	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1331	u32 tbl_dw_addr;
1332	u32 tbl_dw;
1333	u16 scd_q2ratid;
1334
1335	scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK;
1336
1337	tbl_dw_addr = trans_pcie->scd_base_addr +
1338			SCD_TRANS_TBL_OFFSET_QUEUE(txq_id);
1339
1340	tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr);
1341
1342	if (txq_id & 0x1)
1343		tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF);
1344	else
1345		tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000);
1346
1347	iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw);
1348
1349	return 0;
1350}
1351
1352/* Receiver address (actually, Rx station's index into station table),
1353 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */
1354#define BUILD_RAxTID(sta_id, tid)	(((sta_id) << 4) + (tid))
1355
1356bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn,
1357			       const struct iwl_trans_txq_scd_cfg *cfg,
1358			       unsigned int wdg_timeout)
1359{
1360	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1361	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1362	int fifo = -1;
1363	bool scd_bug = false;
1364
1365	if (test_and_set_bit(txq_id, trans_pcie->queue_used))
1366		WARN_ONCE(1, "queue %d already used - expect issues", txq_id);
1367
1368	txq->wd_timeout = msecs_to_jiffies(wdg_timeout);
1369
1370	if (cfg) {
1371		fifo = cfg->fifo;
1372
1373		/* Disable the scheduler prior configuring the cmd queue */
1374		if (txq_id == trans_pcie->cmd_queue &&
1375		    trans_pcie->scd_set_active)
1376			iwl_scd_enable_set_active(trans, 0);
1377
1378		/* Stop this Tx queue before configuring it */
1379		iwl_scd_txq_set_inactive(trans, txq_id);
1380
1381		/* Set this queue as a chain-building queue unless it is CMD */
1382		if (txq_id != trans_pcie->cmd_queue)
1383			iwl_scd_txq_set_chain(trans, txq_id);
1384
1385		if (cfg->aggregate) {
1386			u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid);
1387
1388			/* Map receiver-address / traffic-ID to this queue */
1389			iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id);
1390
1391			/* enable aggregations for the queue */
1392			iwl_scd_txq_enable_agg(trans, txq_id);
1393			txq->ampdu = true;
1394		} else {
1395			/*
1396			 * disable aggregations for the queue, this will also
1397			 * make the ra_tid mapping configuration irrelevant
1398			 * since it is now a non-AGG queue.
1399			 */
1400			iwl_scd_txq_disable_agg(trans, txq_id);
1401
1402			ssn = txq->read_ptr;
1403		}
1404	} else {
1405		/*
1406		 * If we need to move the SCD write pointer by steps of
1407		 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let
1408		 * the op_mode know by returning true later.
1409		 * Do this only in case cfg is NULL since this trick can
1410		 * be done only if we have DQA enabled which is true for mvm
1411		 * only. And mvm never sets a cfg pointer.
1412		 * This is really ugly, but this is the easiest way out for
1413		 * this sad hardware issue.
1414		 * This bug has been fixed on devices 9000 and up.
1415		 */
1416		scd_bug = !trans->trans_cfg->mq_rx_supported &&
1417			!((ssn - txq->write_ptr) & 0x3f) &&
1418			(ssn != txq->write_ptr);
1419		if (scd_bug)
1420			ssn++;
1421	}
1422
1423	/* Place first TFD at index corresponding to start sequence number.
1424	 * Assumes that ssn_idx is valid (!= 0xFFF) */
1425	txq->read_ptr = (ssn & 0xff);
1426	txq->write_ptr = (ssn & 0xff);
1427	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1428			   (ssn & 0xff) | (txq_id << 8));
1429
1430	if (cfg) {
1431		u8 frame_limit = cfg->frame_limit;
1432
1433		iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn);
1434
1435		/* Set up Tx window size and frame limit for this queue */
1436		iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr +
1437				SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0);
1438		iwl_trans_write_mem32(trans,
1439			trans_pcie->scd_base_addr +
1440			SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32),
1441			SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) |
1442			SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit));
1443
1444		/* Set up status area in SRAM, map to Tx DMA/FIFO, activate */
1445		iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id),
1446			       (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) |
1447			       (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) |
1448			       (1 << SCD_QUEUE_STTS_REG_POS_WSL) |
1449			       SCD_QUEUE_STTS_REG_MSK);
1450
1451		/* enable the scheduler for this queue (only) */
1452		if (txq_id == trans_pcie->cmd_queue &&
1453		    trans_pcie->scd_set_active)
1454			iwl_scd_enable_set_active(trans, BIT(txq_id));
1455
1456		IWL_DEBUG_TX_QUEUES(trans,
1457				    "Activate queue %d on FIFO %d WrPtr: %d\n",
1458				    txq_id, fifo, ssn & 0xff);
1459	} else {
1460		IWL_DEBUG_TX_QUEUES(trans,
1461				    "Activate queue %d WrPtr: %d\n",
1462				    txq_id, ssn & 0xff);
1463	}
1464
1465	return scd_bug;
1466}
1467
1468void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
1469					bool shared_mode)
1470{
1471	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1472	struct iwl_txq *txq = trans_pcie->txq[txq_id];
1473
1474	txq->ampdu = !shared_mode;
1475}
1476
1477void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id,
1478				bool configure_scd)
1479{
1480	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1481	u32 stts_addr = trans_pcie->scd_base_addr +
1482			SCD_TX_STTS_QUEUE_OFFSET(txq_id);
1483	static const u32 zero_val[4] = {};
1484
1485	trans_pcie->txq[txq_id]->frozen_expiry_remainder = 0;
1486	trans_pcie->txq[txq_id]->frozen = false;
1487
1488	/*
1489	 * Upon HW Rfkill - we stop the device, and then stop the queues
1490	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1491	 * allow the op_mode to call txq_disable after it already called
1492	 * stop_device.
1493	 */
1494	if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) {
1495		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1496			  "queue %d not used", txq_id);
1497		return;
1498	}
1499
1500	if (configure_scd) {
1501		iwl_scd_txq_set_inactive(trans, txq_id);
1502
1503		iwl_trans_write_mem(trans, stts_addr, (void *)zero_val,
1504				    ARRAY_SIZE(zero_val));
1505	}
1506
1507	iwl_pcie_txq_unmap(trans, txq_id);
1508	trans_pcie->txq[txq_id]->ampdu = false;
1509
1510	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id);
1511}
1512
1513/*************** HOST COMMAND QUEUE FUNCTIONS   *****/
1514
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1515/*
1516 * iwl_pcie_enqueue_hcmd - enqueue a uCode command
1517 * @priv: device private data point
1518 * @cmd: a pointer to the ucode command structure
1519 *
1520 * The function returns < 0 values to indicate the operation
1521 * failed. On success, it returns the index (>= 0) of command in the
1522 * command queue.
1523 */
1524static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans,
1525				 struct iwl_host_cmd *cmd)
1526{
1527	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1528	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1529	struct iwl_device_cmd *out_cmd;
1530	struct iwl_cmd_meta *out_meta;
1531	unsigned long flags;
1532	void *dup_buf = NULL;
1533	dma_addr_t phys_addr;
1534	int idx;
1535	u16 copy_size, cmd_size, tb0_size;
1536	bool had_nocopy = false;
1537	u8 group_id = iwl_cmd_groupid(cmd->id);
1538	int i, ret;
1539	u32 cmd_pos;
1540	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
1541	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
 
1542
1543	if (WARN(!trans->wide_cmd_header &&
1544		 group_id > IWL_ALWAYS_LONG_GROUP,
1545		 "unsupported wide command %#x\n", cmd->id))
1546		return -EINVAL;
1547
1548	if (group_id != 0) {
1549		copy_size = sizeof(struct iwl_cmd_header_wide);
1550		cmd_size = sizeof(struct iwl_cmd_header_wide);
1551	} else {
1552		copy_size = sizeof(struct iwl_cmd_header);
1553		cmd_size = sizeof(struct iwl_cmd_header);
1554	}
1555
1556	/* need one for the header if the first is NOCOPY */
1557	BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1);
1558
1559	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1560		cmddata[i] = cmd->data[i];
1561		cmdlen[i] = cmd->len[i];
1562
1563		if (!cmd->len[i])
1564			continue;
1565
1566		/* need at least IWL_FIRST_TB_SIZE copied */
1567		if (copy_size < IWL_FIRST_TB_SIZE) {
1568			int copy = IWL_FIRST_TB_SIZE - copy_size;
1569
1570			if (copy > cmdlen[i])
1571				copy = cmdlen[i];
1572			cmdlen[i] -= copy;
1573			cmddata[i] += copy;
1574			copy_size += copy;
1575		}
1576
1577		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
1578			had_nocopy = true;
1579			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
1580				idx = -EINVAL;
1581				goto free_dup_buf;
1582			}
1583		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
1584			/*
1585			 * This is also a chunk that isn't copied
1586			 * to the static buffer so set had_nocopy.
1587			 */
1588			had_nocopy = true;
1589
1590			/* only allowed once */
1591			if (WARN_ON(dup_buf)) {
1592				idx = -EINVAL;
1593				goto free_dup_buf;
1594			}
1595
1596			dup_buf = kmemdup(cmddata[i], cmdlen[i],
1597					  GFP_ATOMIC);
1598			if (!dup_buf)
1599				return -ENOMEM;
1600		} else {
1601			/* NOCOPY must not be followed by normal! */
1602			if (WARN_ON(had_nocopy)) {
1603				idx = -EINVAL;
1604				goto free_dup_buf;
1605			}
1606			copy_size += cmdlen[i];
1607		}
1608		cmd_size += cmd->len[i];
1609	}
1610
1611	/*
1612	 * If any of the command structures end up being larger than
1613	 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically
1614	 * allocated into separate TFDs, then we will need to
1615	 * increase the size of the buffers.
1616	 */
1617	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
1618		 "Command %s (%#x) is too large (%d bytes)\n",
1619		 iwl_get_cmd_string(trans, cmd->id),
1620		 cmd->id, copy_size)) {
1621		idx = -EINVAL;
1622		goto free_dup_buf;
1623	}
1624
1625	spin_lock_bh(&txq->lock);
1626
1627	if (iwl_queue_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
1628		spin_unlock_bh(&txq->lock);
1629
1630		IWL_ERR(trans, "No space in command queue\n");
1631		iwl_op_mode_cmd_queue_full(trans->op_mode);
1632		idx = -ENOSPC;
1633		goto free_dup_buf;
1634	}
1635
1636	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
1637	out_cmd = txq->entries[idx].cmd;
1638	out_meta = &txq->entries[idx].meta;
1639
1640	memset(out_meta, 0, sizeof(*out_meta));	/* re-initialize to NULL */
1641	if (cmd->flags & CMD_WANT_SKB)
1642		out_meta->source = cmd;
1643
1644	/* set up the header */
1645	if (group_id != 0) {
1646		out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
1647		out_cmd->hdr_wide.group_id = group_id;
1648		out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
1649		out_cmd->hdr_wide.length =
1650			cpu_to_le16(cmd_size -
1651				    sizeof(struct iwl_cmd_header_wide));
1652		out_cmd->hdr_wide.reserved = 0;
1653		out_cmd->hdr_wide.sequence =
1654			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1655						 INDEX_TO_SEQ(txq->write_ptr));
1656
1657		cmd_pos = sizeof(struct iwl_cmd_header_wide);
1658		copy_size = sizeof(struct iwl_cmd_header_wide);
1659	} else {
1660		out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id);
1661		out_cmd->hdr.sequence =
1662			cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
1663						 INDEX_TO_SEQ(txq->write_ptr));
1664		out_cmd->hdr.group_id = 0;
1665
1666		cmd_pos = sizeof(struct iwl_cmd_header);
1667		copy_size = sizeof(struct iwl_cmd_header);
1668	}
1669
1670	/* and copy the data that needs to be copied */
1671	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1672		int copy;
1673
1674		if (!cmd->len[i])
1675			continue;
1676
1677		/* copy everything if not nocopy/dup */
1678		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1679					   IWL_HCMD_DFL_DUP))) {
1680			copy = cmd->len[i];
1681
1682			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1683			cmd_pos += copy;
1684			copy_size += copy;
1685			continue;
1686		}
1687
1688		/*
1689		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
1690		 * in total (for bi-directional DMA), but copy up to what
1691		 * we can fit into the payload for debug dump purposes.
1692		 */
1693		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
1694
1695		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
1696		cmd_pos += copy;
1697
1698		/* However, treat copy_size the proper way, we need it below */
1699		if (copy_size < IWL_FIRST_TB_SIZE) {
1700			copy = IWL_FIRST_TB_SIZE - copy_size;
1701
1702			if (copy > cmd->len[i])
1703				copy = cmd->len[i];
1704			copy_size += copy;
1705		}
1706	}
1707
1708	IWL_DEBUG_HC(trans,
1709		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
1710		     iwl_get_cmd_string(trans, cmd->id),
1711		     group_id, out_cmd->hdr.cmd,
1712		     le16_to_cpu(out_cmd->hdr.sequence),
1713		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
1714
1715	/* start the TFD with the minimum copy bytes */
1716	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
1717	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
1718	iwl_pcie_txq_build_tfd(trans, txq,
1719			       iwl_pcie_get_first_tb_dma(txq, idx),
1720			       tb0_size, true);
1721
1722	/* map first command fragment, if any remains */
1723	if (copy_size > tb0_size) {
1724		phys_addr = dma_map_single(trans->dev,
1725					   ((u8 *)&out_cmd->hdr) + tb0_size,
1726					   copy_size - tb0_size,
1727					   DMA_TO_DEVICE);
1728		if (dma_mapping_error(trans->dev, phys_addr)) {
1729			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1730					   txq->write_ptr);
1731			idx = -ENOMEM;
1732			goto out;
1733		}
1734
1735		iwl_pcie_txq_build_tfd(trans, txq, phys_addr,
1736				       copy_size - tb0_size, false);
1737	}
1738
1739	/* map the remaining (adjusted) nocopy/dup fragments */
1740	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
1741		const void *data = cmddata[i];
1742
1743		if (!cmdlen[i])
1744			continue;
1745		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
1746					   IWL_HCMD_DFL_DUP)))
1747			continue;
1748		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
1749			data = dup_buf;
1750		phys_addr = dma_map_single(trans->dev, (void *)data,
1751					   cmdlen[i], DMA_TO_DEVICE);
1752		if (dma_mapping_error(trans->dev, phys_addr)) {
1753			iwl_pcie_tfd_unmap(trans, out_meta, txq,
1754					   txq->write_ptr);
1755			idx = -ENOMEM;
1756			goto out;
1757		}
1758
1759		iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false);
1760	}
1761
1762	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
1763	out_meta->flags = cmd->flags;
1764	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
1765		kzfree(txq->entries[idx].free_buf);
1766	txq->entries[idx].free_buf = dup_buf;
1767
1768	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
1769
1770	/* start timer if queue currently empty */
1771	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
1772		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1773
1774	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
1775	ret = iwl_pcie_set_cmd_in_flight(trans, cmd);
1776	if (ret < 0) {
1777		idx = ret;
1778		spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1779		goto out;
1780	}
1781
 
 
 
1782	/* Increment and update queue's write index */
1783	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
1784	iwl_pcie_txq_inc_wr_ptr(trans, txq);
1785
1786	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
1787
1788 out:
1789	spin_unlock_bh(&txq->lock);
1790 free_dup_buf:
1791	if (idx < 0)
1792		kfree(dup_buf);
1793	return idx;
1794}
1795
1796/*
1797 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them
1798 * @rxb: Rx buffer to reclaim
1799 */
1800void iwl_pcie_hcmd_complete(struct iwl_trans *trans,
1801			    struct iwl_rx_cmd_buffer *rxb)
1802{
1803	struct iwl_rx_packet *pkt = rxb_addr(rxb);
1804	u16 sequence = le16_to_cpu(pkt->hdr.sequence);
1805	u8 group_id;
1806	u32 cmd_id;
1807	int txq_id = SEQ_TO_QUEUE(sequence);
1808	int index = SEQ_TO_INDEX(sequence);
1809	int cmd_index;
1810	struct iwl_device_cmd *cmd;
1811	struct iwl_cmd_meta *meta;
1812	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1813	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1814
1815	/* If a Tx command is being handled and it isn't in the actual
1816	 * command queue then there a command routing bug has been introduced
1817	 * in the queue management code. */
1818	if (WARN(txq_id != trans_pcie->cmd_queue,
1819		 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n",
1820		 txq_id, trans_pcie->cmd_queue, sequence, txq->read_ptr,
1821		 txq->write_ptr)) {
1822		iwl_print_hex_error(trans, pkt, 32);
1823		return;
1824	}
1825
1826	spin_lock_bh(&txq->lock);
1827
1828	cmd_index = iwl_pcie_get_cmd_index(txq, index);
1829	cmd = txq->entries[cmd_index].cmd;
1830	meta = &txq->entries[cmd_index].meta;
1831	group_id = cmd->hdr.group_id;
1832	cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0);
1833
1834	iwl_pcie_tfd_unmap(trans, meta, txq, index);
 
 
 
 
1835
1836	/* Input error checking is done when commands are added to queue. */
1837	if (meta->flags & CMD_WANT_SKB) {
1838		struct page *p = rxb_steal_page(rxb);
1839
1840		meta->source->resp_pkt = pkt;
1841		meta->source->_rx_page_addr = (unsigned long)page_address(p);
1842		meta->source->_rx_page_order = trans_pcie->rx_page_order;
1843	}
1844
1845	if (meta->flags & CMD_WANT_ASYNC_CALLBACK)
1846		iwl_op_mode_async_cb(trans->op_mode, cmd);
1847
1848	iwl_pcie_cmdq_reclaim(trans, txq_id, index);
1849
1850	if (!(meta->flags & CMD_ASYNC)) {
1851		if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) {
1852			IWL_WARN(trans,
1853				 "HCMD_ACTIVE already clear for command %s\n",
1854				 iwl_get_cmd_string(trans, cmd_id));
1855		}
1856		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1857		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1858			       iwl_get_cmd_string(trans, cmd_id));
1859		wake_up(&trans_pcie->wait_command_queue);
1860	}
1861
1862	meta->flags = 0;
1863
1864	spin_unlock_bh(&txq->lock);
1865}
1866
1867#define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1868
1869static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans,
1870				    struct iwl_host_cmd *cmd)
1871{
1872	int ret;
1873
1874	/* An asynchronous command can not expect an SKB to be set. */
1875	if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1876		return -EINVAL;
1877
1878	ret = iwl_pcie_enqueue_hcmd(trans, cmd);
1879	if (ret < 0) {
1880		IWL_ERR(trans,
1881			"Error sending %s: enqueue_hcmd failed: %d\n",
1882			iwl_get_cmd_string(trans, cmd->id), ret);
1883		return ret;
1884	}
1885	return 0;
1886}
1887
1888static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans,
1889				   struct iwl_host_cmd *cmd)
1890{
1891	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1892	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
1893	int cmd_idx;
1894	int ret;
1895
1896	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n",
1897		       iwl_get_cmd_string(trans, cmd->id));
1898
1899	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1900				  &trans->status),
1901		 "Command %s: a command is already active!\n",
1902		 iwl_get_cmd_string(trans, cmd->id)))
1903		return -EIO;
1904
1905	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n",
1906		       iwl_get_cmd_string(trans, cmd->id));
1907
1908	cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd);
1909	if (cmd_idx < 0) {
1910		ret = cmd_idx;
1911		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1912		IWL_ERR(trans,
1913			"Error sending %s: enqueue_hcmd failed: %d\n",
1914			iwl_get_cmd_string(trans, cmd->id), ret);
1915		return ret;
1916	}
1917
1918	ret = wait_event_timeout(trans_pcie->wait_command_queue,
1919				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1920					   &trans->status),
1921				 HOST_COMPLETE_TIMEOUT);
1922	if (!ret) {
1923		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1924			iwl_get_cmd_string(trans, cmd->id),
1925			jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1926
1927		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1928			txq->read_ptr, txq->write_ptr);
1929
1930		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1931		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1932			       iwl_get_cmd_string(trans, cmd->id));
1933		ret = -ETIMEDOUT;
1934
1935		iwl_trans_pcie_sync_nmi(trans);
1936		goto cancel;
1937	}
1938
1939	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1940		iwl_trans_pcie_dump_regs(trans);
1941		IWL_ERR(trans, "FW error in SYNC CMD %s\n",
1942			iwl_get_cmd_string(trans, cmd->id));
1943		dump_stack();
1944		ret = -EIO;
1945		goto cancel;
1946	}
1947
1948	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1949	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1950		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1951		ret = -ERFKILL;
1952		goto cancel;
1953	}
1954
1955	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1956		IWL_ERR(trans, "Error: Response NULL in '%s'\n",
1957			iwl_get_cmd_string(trans, cmd->id));
1958		ret = -EIO;
1959		goto cancel;
1960	}
1961
1962	return 0;
1963
1964cancel:
1965	if (cmd->flags & CMD_WANT_SKB) {
1966		/*
1967		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1968		 * TX cmd queue. Otherwise in case the cmd comes
1969		 * in later, it will possibly set an invalid
1970		 * address (cmd->meta.source).
1971		 */
1972		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1973	}
1974
1975	if (cmd->resp_pkt) {
1976		iwl_free_resp(cmd);
1977		cmd->resp_pkt = NULL;
1978	}
1979
1980	return ret;
1981}
1982
1983int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd)
1984{
1985	/* Make sure the NIC is still alive in the bus */
1986	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1987		return -ENODEV;
1988
1989	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1990	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1991		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1992				  cmd->id);
1993		return -ERFKILL;
1994	}
1995
1996	if (cmd->flags & CMD_ASYNC)
1997		return iwl_pcie_send_hcmd_async(trans, cmd);
1998
1999	/* We still can fail on RFKILL that can be asserted while we wait */
2000	return iwl_pcie_send_hcmd_sync(trans, cmd);
2001}
2002
2003static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
2004			     struct iwl_txq *txq, u8 hdr_len,
2005			     struct iwl_cmd_meta *out_meta)
2006{
2007	u16 head_tb_len;
2008	int i;
2009
2010	/*
2011	 * Set up TFD's third entry to point directly to remainder
2012	 * of skb's head, if any
2013	 */
2014	head_tb_len = skb_headlen(skb) - hdr_len;
2015
2016	if (head_tb_len > 0) {
2017		dma_addr_t tb_phys = dma_map_single(trans->dev,
2018						    skb->data + hdr_len,
2019						    head_tb_len, DMA_TO_DEVICE);
2020		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
2021			return -EINVAL;
2022		trace_iwlwifi_dev_tx_tb(trans->dev, skb,
2023					skb->data + hdr_len,
2024					head_tb_len);
2025		iwl_pcie_txq_build_tfd(trans, txq, tb_phys, head_tb_len, false);
2026	}
2027
2028	/* set up the remaining entries to point to the data */
2029	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2030		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2031		dma_addr_t tb_phys;
2032		int tb_idx;
2033
2034		if (!skb_frag_size(frag))
2035			continue;
2036
2037		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
2038					   skb_frag_size(frag), DMA_TO_DEVICE);
2039
2040		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
2041			return -EINVAL;
2042		trace_iwlwifi_dev_tx_tb(trans->dev, skb,
2043					skb_frag_address(frag),
2044					skb_frag_size(frag));
2045		tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2046						skb_frag_size(frag), false);
2047		if (tb_idx < 0)
2048			return tb_idx;
2049
2050		out_meta->tbs |= BIT(tb_idx);
2051	}
2052
2053	return 0;
2054}
2055
2056#ifdef CONFIG_INET
2057struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len)
2058{
2059	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2060	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page);
2061
2062	if (!p->page)
2063		goto alloc;
2064
2065	/* enough room on this page */
2066	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
2067		return p;
2068
2069	/* We don't have enough room on this page, get a new one. */
2070	__free_page(p->page);
2071
2072alloc:
2073	p->page = alloc_page(GFP_ATOMIC);
2074	if (!p->page)
2075		return NULL;
2076	p->pos = page_address(p->page);
2077	return p;
2078}
2079
2080static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
2081					bool ipv6, unsigned int len)
2082{
2083	if (ipv6) {
2084		struct ipv6hdr *iphv6 = iph;
2085
2086		tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr,
2087					       len + tcph->doff * 4,
2088					       IPPROTO_TCP, 0);
2089	} else {
2090		struct iphdr *iphv4 = iph;
2091
2092		ip_send_check(iphv4);
2093		tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr,
2094						 len + tcph->doff * 4,
2095						 IPPROTO_TCP, 0);
2096	}
2097}
2098
2099static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2100				   struct iwl_txq *txq, u8 hdr_len,
2101				   struct iwl_cmd_meta *out_meta,
2102				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
 
2103{
2104	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
2105	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
2106	struct ieee80211_hdr *hdr = (void *)skb->data;
2107	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
2108	unsigned int mss = skb_shinfo(skb)->gso_size;
2109	u16 length, iv_len, amsdu_pad;
2110	u8 *start_hdr;
2111	struct iwl_tso_hdr_page *hdr_page;
2112	struct page **page_ptr;
2113	struct tso_t tso;
2114
2115	/* if the packet is protected, then it must be CCMP or GCMP */
2116	BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN);
2117	iv_len = ieee80211_has_protected(hdr->frame_control) ?
2118		IEEE80211_CCMP_HDR_LEN : 0;
2119
2120	trace_iwlwifi_dev_tx(trans->dev, skb,
2121			     iwl_pcie_get_tfd(trans, txq, txq->write_ptr),
2122			     trans_pcie->tfd_size,
2123			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0);
2124
2125	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
2126	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
2127	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
2128	amsdu_pad = 0;
2129
2130	/* total amount of header we may need for this A-MSDU */
2131	hdr_room = DIV_ROUND_UP(total_len, mss) *
2132		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
2133
2134	/* Our device supports 9 segments at most, it will fit in 1 page */
2135	hdr_page = get_page_hdr(trans, hdr_room);
2136	if (!hdr_page)
2137		return -ENOMEM;
2138
2139	get_page(hdr_page->page);
2140	start_hdr = hdr_page->pos;
2141	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
2142	*page_ptr = hdr_page->page;
2143	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
2144	hdr_page->pos += iv_len;
2145
2146	/*
2147	 * Pull the ieee80211 header + IV to be able to use TSO core,
2148	 * we will restore it for the tx_status flow.
2149	 */
2150	skb_pull(skb, hdr_len + iv_len);
2151
2152	/*
2153	 * Remove the length of all the headers that we don't actually
2154	 * have in the MPDU by themselves, but that we duplicate into
2155	 * all the different MSDUs inside the A-MSDU.
2156	 */
2157	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
2158
2159	tso_start(skb, &tso);
2160
2161	while (total_len) {
2162		/* this is the data left for this subframe */
2163		unsigned int data_left =
2164			min_t(unsigned int, mss, total_len);
2165		struct sk_buff *csum_skb = NULL;
2166		unsigned int hdr_tb_len;
2167		dma_addr_t hdr_tb_phys;
2168		struct tcphdr *tcph;
2169		u8 *iph, *subf_hdrs_start = hdr_page->pos;
2170
2171		total_len -= data_left;
2172
2173		memset(hdr_page->pos, 0, amsdu_pad);
2174		hdr_page->pos += amsdu_pad;
2175		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
2176				  data_left)) & 0x3;
2177		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
2178		hdr_page->pos += ETH_ALEN;
2179		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
2180		hdr_page->pos += ETH_ALEN;
2181
2182		length = snap_ip_tcp_hdrlen + data_left;
2183		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
2184		hdr_page->pos += sizeof(length);
2185
2186		/*
2187		 * This will copy the SNAP as well which will be considered
2188		 * as MAC header.
2189		 */
2190		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
2191		iph = hdr_page->pos + 8;
2192		tcph = (void *)(iph + ip_hdrlen);
2193
2194		/* For testing on current hardware only */
2195		if (trans_pcie->sw_csum_tx) {
2196			csum_skb = alloc_skb(data_left + tcp_hdrlen(skb),
2197					     GFP_ATOMIC);
2198			if (!csum_skb)
2199				return -ENOMEM;
2200
2201			iwl_compute_pseudo_hdr_csum(iph, tcph,
2202						    skb->protocol ==
2203							htons(ETH_P_IPV6),
2204						    data_left);
2205
2206			skb_put_data(csum_skb, tcph, tcp_hdrlen(skb));
2207			skb_reset_transport_header(csum_skb);
2208			csum_skb->csum_start =
2209				(unsigned char *)tcp_hdr(csum_skb) -
2210						 csum_skb->head;
2211		}
2212
2213		hdr_page->pos += snap_ip_tcp_hdrlen;
2214
2215		hdr_tb_len = hdr_page->pos - start_hdr;
2216		hdr_tb_phys = dma_map_single(trans->dev, start_hdr,
2217					     hdr_tb_len, DMA_TO_DEVICE);
2218		if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) {
2219			dev_kfree_skb(csum_skb);
2220			return -EINVAL;
2221		}
2222		iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys,
2223				       hdr_tb_len, false);
2224		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
2225					hdr_tb_len);
2226		/* add this subframe's headers' length to the tx_cmd */
2227		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
2228
2229		/* prepare the start_hdr for the next subframe */
2230		start_hdr = hdr_page->pos;
2231
2232		/* put the payload */
2233		while (data_left) {
2234			unsigned int size = min_t(unsigned int, tso.size,
2235						  data_left);
2236			dma_addr_t tb_phys;
2237
2238			if (trans_pcie->sw_csum_tx)
2239				skb_put_data(csum_skb, tso.data, size);
2240
2241			tb_phys = dma_map_single(trans->dev, tso.data,
2242						 size, DMA_TO_DEVICE);
2243			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
2244				dev_kfree_skb(csum_skb);
2245				return -EINVAL;
2246			}
2247
2248			iwl_pcie_txq_build_tfd(trans, txq, tb_phys,
2249					       size, false);
2250			trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data,
2251						size);
2252
2253			data_left -= size;
2254			tso_build_data(skb, &tso, size);
2255		}
2256
2257		/* For testing on early hardware only */
2258		if (trans_pcie->sw_csum_tx) {
2259			__wsum csum;
2260
2261			csum = skb_checksum(csum_skb,
2262					    skb_checksum_start_offset(csum_skb),
2263					    csum_skb->len -
2264					    skb_checksum_start_offset(csum_skb),
2265					    0);
2266			dev_kfree_skb(csum_skb);
2267			dma_sync_single_for_cpu(trans->dev, hdr_tb_phys,
2268						hdr_tb_len, DMA_TO_DEVICE);
2269			tcph->check = csum_fold(csum);
2270			dma_sync_single_for_device(trans->dev, hdr_tb_phys,
2271						   hdr_tb_len, DMA_TO_DEVICE);
2272		}
2273	}
2274
2275	/* re -add the WiFi header and IV */
2276	skb_push(skb, hdr_len + iv_len);
2277
2278	return 0;
2279}
2280#else /* CONFIG_INET */
2281static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
2282				   struct iwl_txq *txq, u8 hdr_len,
2283				   struct iwl_cmd_meta *out_meta,
2284				   struct iwl_device_cmd *dev_cmd, u16 tb1_len)
 
2285{
2286	/* No A-MSDU without CONFIG_INET */
2287	WARN_ON(1);
2288
2289	return -1;
2290}
2291#endif /* CONFIG_INET */
2292
2293int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
2294		      struct iwl_device_cmd *dev_cmd, int txq_id)
2295{
2296	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
2297	struct ieee80211_hdr *hdr;
2298	struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload;
2299	struct iwl_cmd_meta *out_meta;
2300	struct iwl_txq *txq;
2301	dma_addr_t tb0_phys, tb1_phys, scratch_phys;
2302	void *tb1_addr;
2303	void *tfd;
2304	u16 len, tb1_len;
2305	bool wait_write_ptr;
2306	__le16 fc;
2307	u8 hdr_len;
2308	u16 wifi_seq;
2309	bool amsdu;
2310
2311	txq = trans_pcie->txq[txq_id];
2312
2313	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
2314		      "TX on unused queue %d\n", txq_id))
2315		return -EINVAL;
2316
2317	if (unlikely(trans_pcie->sw_csum_tx &&
2318		     skb->ip_summed == CHECKSUM_PARTIAL)) {
2319		int offs = skb_checksum_start_offset(skb);
2320		int csum_offs = offs + skb->csum_offset;
2321		__wsum csum;
2322
2323		if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16)))
2324			return -1;
2325
2326		csum = skb_checksum(skb, offs, skb->len - offs, 0);
2327		*(__sum16 *)(skb->data + csum_offs) = csum_fold(csum);
2328
2329		skb->ip_summed = CHECKSUM_UNNECESSARY;
2330	}
2331
2332	if (skb_is_nonlinear(skb) &&
2333	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
2334	    __skb_linearize(skb))
2335		return -ENOMEM;
2336
2337	/* mac80211 always puts the full header into the SKB's head,
2338	 * so there's no need to check if it's readable there
2339	 */
2340	hdr = (struct ieee80211_hdr *)skb->data;
2341	fc = hdr->frame_control;
2342	hdr_len = ieee80211_hdrlen(fc);
2343
2344	spin_lock(&txq->lock);
2345
2346	if (iwl_queue_space(trans, txq) < txq->high_mark) {
2347		iwl_stop_queue(trans, txq);
2348
2349		/* don't put the packet on the ring, if there is no room */
2350		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
2351			struct iwl_device_cmd **dev_cmd_ptr;
2352
2353			dev_cmd_ptr = (void *)((u8 *)skb->cb +
2354					       trans_pcie->dev_cmd_offs);
2355
2356			*dev_cmd_ptr = dev_cmd;
2357			__skb_queue_tail(&txq->overflow_q, skb);
2358
2359			spin_unlock(&txq->lock);
2360			return 0;
2361		}
2362	}
2363
2364	/* In AGG mode, the index in the ring must correspond to the WiFi
2365	 * sequence number. This is a HW requirements to help the SCD to parse
2366	 * the BA.
2367	 * Check here that the packets are in the right place on the ring.
2368	 */
2369	wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl));
2370	WARN_ONCE(txq->ampdu &&
2371		  (wifi_seq & 0xff) != txq->write_ptr,
2372		  "Q: %d WiFi Seq %d tfdNum %d",
2373		  txq_id, wifi_seq, txq->write_ptr);
2374
2375	/* Set up driver data for this TFD */
2376	txq->entries[txq->write_ptr].skb = skb;
2377	txq->entries[txq->write_ptr].cmd = dev_cmd;
2378
2379	dev_cmd->hdr.sequence =
2380		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
2381			    INDEX_TO_SEQ(txq->write_ptr)));
2382
2383	tb0_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
2384	scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) +
2385		       offsetof(struct iwl_tx_cmd, scratch);
2386
2387	tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys);
2388	tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys);
2389
2390	/* Set up first empty entry in queue's array of Tx/cmd buffers */
2391	out_meta = &txq->entries[txq->write_ptr].meta;
2392	out_meta->flags = 0;
2393
2394	/*
2395	 * The second TB (tb1) points to the remainder of the TX command
2396	 * and the 802.11 header - dword aligned size
2397	 * (This calculation modifies the TX command, so do it before the
2398	 * setup of the first TB)
2399	 */
2400	len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) +
2401	      hdr_len - IWL_FIRST_TB_SIZE;
2402	/* do not align A-MSDU to dword as the subframe header aligns it */
2403	amsdu = ieee80211_is_data_qos(fc) &&
2404		(*ieee80211_get_qos_ctl(hdr) &
2405		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
2406	if (trans_pcie->sw_csum_tx || !amsdu) {
2407		tb1_len = ALIGN(len, 4);
2408		/* Tell NIC about any 2-byte padding after MAC header */
2409		if (tb1_len != len)
2410			tx_cmd->tx_flags |= cpu_to_le32(TX_CMD_FLG_MH_PAD);
2411	} else {
2412		tb1_len = len;
2413	}
2414
2415	/*
2416	 * The first TB points to bi-directional DMA data, we'll
2417	 * memcpy the data into it later.
2418	 */
2419	iwl_pcie_txq_build_tfd(trans, txq, tb0_phys,
2420			       IWL_FIRST_TB_SIZE, true);
2421
2422	/* there must be data left over for TB1 or this code must be changed */
2423	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_FIRST_TB_SIZE);
 
 
 
2424
2425	/* map the data for TB1 */
2426	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
2427	tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
2428	if (unlikely(dma_mapping_error(trans->dev, tb1_phys)))
2429		goto out_err;
2430	iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false);
2431
2432	trace_iwlwifi_dev_tx(trans->dev, skb,
2433			     iwl_pcie_get_tfd(trans, txq,
2434					      txq->write_ptr),
2435			     trans_pcie->tfd_size,
2436			     &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len,
2437			     hdr_len);
2438
2439	/*
2440	 * If gso_size wasn't set, don't give the frame "amsdu treatment"
2441	 * (adding subframes, etc.).
2442	 * This can happen in some testing flows when the amsdu was already
2443	 * pre-built, and we just need to send the resulting skb.
2444	 */
2445	if (amsdu && skb_shinfo(skb)->gso_size) {
2446		if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len,
2447						     out_meta, dev_cmd,
2448						     tb1_len)))
2449			goto out_err;
2450	} else {
2451		struct sk_buff *frag;
2452
2453		if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len,
2454					       out_meta)))
2455			goto out_err;
2456
2457		skb_walk_frags(skb, frag) {
2458			if (unlikely(iwl_fill_data_tbs(trans, frag, txq, 0,
2459						       out_meta)))
2460				goto out_err;
2461		}
2462	}
2463
2464	/* building the A-MSDU might have changed this data, so memcpy it now */
2465	memcpy(&txq->first_tb_bufs[txq->write_ptr], dev_cmd, IWL_FIRST_TB_SIZE);
2466
2467	tfd = iwl_pcie_get_tfd(trans, txq, txq->write_ptr);
2468	/* Set up entry for this TFD in Tx byte-count array */
2469	iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len),
2470					 iwl_pcie_tfd_get_num_tbs(trans, tfd));
 
2471
2472	wait_write_ptr = ieee80211_has_morefrags(fc);
2473
2474	/* start timer if queue currently empty */
2475	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) {
2476		/*
2477		 * If the TXQ is active, then set the timer, if not,
2478		 * set the timer in remainder so that the timer will
2479		 * be armed with the right value when the station will
2480		 * wake up.
2481		 */
2482		if (!txq->frozen)
2483			mod_timer(&txq->stuck_timer,
2484				  jiffies + txq->wd_timeout);
2485		else
2486			txq->frozen_expiry_remainder = txq->wd_timeout;
2487	}
2488
2489	/* Tell device the write index *just past* this latest filled TFD */
2490	txq->write_ptr = iwl_queue_inc_wrap(trans, txq->write_ptr);
2491	if (!wait_write_ptr)
2492		iwl_pcie_txq_inc_wr_ptr(trans, txq);
2493
2494	/*
2495	 * At this point the frame is "transmitted" successfully
2496	 * and we will get a TX status notification eventually.
2497	 */
2498	spin_unlock(&txq->lock);
2499	return 0;
2500out_err:
2501	iwl_pcie_tfd_unmap(trans, out_meta, txq, txq->write_ptr);
2502	spin_unlock(&txq->lock);
2503	return -1;
2504}