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) 2017 Intel Deutschland GmbH
  4 * Copyright (C) 2018-2020, 2023 Intel Corporation
  5 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  6#include <net/tso.h>
  7#include <linux/tcp.h>
  8
  9#include "iwl-debug.h"
 10#include "iwl-csr.h"
 11#include "iwl-io.h"
 12#include "internal.h"
 13#include "fw/api/tx.h"
 14#include "queue/tx.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 15
 16/*************** HOST COMMAND QUEUE FUNCTIONS   *****/
 17
 18/*
 19 * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command
 20 * @priv: device private data point
 21 * @cmd: a pointer to the ucode command structure
 22 *
 23 * The function returns < 0 values to indicate the operation
 24 * failed. On success, it returns the index (>= 0) of command in the
 25 * command queue.
 26 */
 27int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans,
 28			       struct iwl_host_cmd *cmd)
 29{
 30	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 31	struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id];
 32	struct iwl_device_cmd *out_cmd;
 33	struct iwl_cmd_meta *out_meta;
 
 34	void *dup_buf = NULL;
 35	dma_addr_t phys_addr;
 36	int i, cmd_pos, idx;
 37	u16 copy_size, cmd_size, tb0_size;
 38	bool had_nocopy = false;
 39	u8 group_id = iwl_cmd_groupid(cmd->id);
 40	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
 41	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
 42	struct iwl_tfh_tfd *tfd;
 43	unsigned long flags;
 44
 45	if (WARN_ON(cmd->flags & CMD_BLOCK_TXQS))
 46		return -EINVAL;
 47
 48	copy_size = sizeof(struct iwl_cmd_header_wide);
 49	cmd_size = sizeof(struct iwl_cmd_header_wide);
 50
 51	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
 52		cmddata[i] = cmd->data[i];
 53		cmdlen[i] = cmd->len[i];
 54
 55		if (!cmd->len[i])
 56			continue;
 57
 58		/* need at least IWL_FIRST_TB_SIZE copied */
 59		if (copy_size < IWL_FIRST_TB_SIZE) {
 60			int copy = IWL_FIRST_TB_SIZE - copy_size;
 61
 62			if (copy > cmdlen[i])
 63				copy = cmdlen[i];
 64			cmdlen[i] -= copy;
 65			cmddata[i] += copy;
 66			copy_size += copy;
 67		}
 68
 69		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
 70			had_nocopy = true;
 71			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
 72				idx = -EINVAL;
 73				goto free_dup_buf;
 74			}
 75		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
 76			/*
 77			 * This is also a chunk that isn't copied
 78			 * to the static buffer so set had_nocopy.
 79			 */
 80			had_nocopy = true;
 81
 82			/* only allowed once */
 83			if (WARN_ON(dup_buf)) {
 84				idx = -EINVAL;
 85				goto free_dup_buf;
 86			}
 87
 88			dup_buf = kmemdup(cmddata[i], cmdlen[i],
 89					  GFP_ATOMIC);
 90			if (!dup_buf)
 91				return -ENOMEM;
 92		} else {
 93			/* NOCOPY must not be followed by normal! */
 94			if (WARN_ON(had_nocopy)) {
 95				idx = -EINVAL;
 96				goto free_dup_buf;
 97			}
 98			copy_size += cmdlen[i];
 99		}
100		cmd_size += cmd->len[i];
101	}
102
103	/*
104	 * If any of the command structures end up being larger than the
105	 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into
106	 * separate TFDs, then we will need to increase the size of the buffers
107	 */
108	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
109		 "Command %s (%#x) is too large (%d bytes)\n",
110		 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) {
111		idx = -EINVAL;
112		goto free_dup_buf;
113	}
114
115	spin_lock_irqsave(&txq->lock, flags);
116
117	idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
118	tfd = iwl_txq_get_tfd(trans, txq, txq->write_ptr);
119	memset(tfd, 0, sizeof(*tfd));
120
121	if (iwl_txq_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
122		spin_unlock_irqrestore(&txq->lock, flags);
123
124		IWL_ERR(trans, "No space in command queue\n");
125		iwl_op_mode_cmd_queue_full(trans->op_mode);
126		idx = -ENOSPC;
127		goto free_dup_buf;
128	}
129
130	out_cmd = txq->entries[idx].cmd;
131	out_meta = &txq->entries[idx].meta;
132
133	/* re-initialize to NULL */
134	memset(out_meta, 0, sizeof(*out_meta));
135	if (cmd->flags & CMD_WANT_SKB)
136		out_meta->source = cmd;
137
138	/* set up the header */
139	out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
140	out_cmd->hdr_wide.group_id = group_id;
141	out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
142	out_cmd->hdr_wide.length =
143		cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
144	out_cmd->hdr_wide.reserved = 0;
145	out_cmd->hdr_wide.sequence =
146		cpu_to_le16(QUEUE_TO_SEQ(trans->txqs.cmd.q_id) |
147					 INDEX_TO_SEQ(txq->write_ptr));
148
149	cmd_pos = sizeof(struct iwl_cmd_header_wide);
150	copy_size = sizeof(struct iwl_cmd_header_wide);
151
152	/* and copy the data that needs to be copied */
153	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
154		int copy;
155
156		if (!cmd->len[i])
157			continue;
158
159		/* copy everything if not nocopy/dup */
160		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
161					   IWL_HCMD_DFL_DUP))) {
162			copy = cmd->len[i];
163
164			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
165			cmd_pos += copy;
166			copy_size += copy;
167			continue;
168		}
169
170		/*
171		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
172		 * in total (for bi-directional DMA), but copy up to what
173		 * we can fit into the payload for debug dump purposes.
174		 */
175		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
176
177		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
178		cmd_pos += copy;
179
180		/* However, treat copy_size the proper way, we need it below */
181		if (copy_size < IWL_FIRST_TB_SIZE) {
182			copy = IWL_FIRST_TB_SIZE - copy_size;
183
184			if (copy > cmd->len[i])
185				copy = cmd->len[i];
186			copy_size += copy;
187		}
188	}
189
190	IWL_DEBUG_HC(trans,
191		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
192		     iwl_get_cmd_string(trans, cmd->id), group_id,
193		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
194		     cmd_size, txq->write_ptr, idx, trans->txqs.cmd.q_id);
195
196	/* start the TFD with the minimum copy bytes */
197	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
198	memcpy(&txq->first_tb_bufs[idx], out_cmd, tb0_size);
199	iwl_txq_gen2_set_tb(trans, tfd, iwl_txq_get_first_tb_dma(txq, idx),
200			    tb0_size);
201
202	/* map first command fragment, if any remains */
203	if (copy_size > tb0_size) {
204		phys_addr = dma_map_single(trans->dev,
205					   (u8 *)out_cmd + tb0_size,
206					   copy_size - tb0_size,
207					   DMA_TO_DEVICE);
208		if (dma_mapping_error(trans->dev, phys_addr)) {
209			idx = -ENOMEM;
210			iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd);
211			goto out;
212		}
213		iwl_txq_gen2_set_tb(trans, tfd, phys_addr,
214				    copy_size - tb0_size);
215	}
216
217	/* map the remaining (adjusted) nocopy/dup fragments */
218	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
219		void *data = (void *)(uintptr_t)cmddata[i];
220
221		if (!cmdlen[i])
222			continue;
223		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
224					   IWL_HCMD_DFL_DUP)))
225			continue;
226		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
227			data = dup_buf;
228		phys_addr = dma_map_single(trans->dev, data,
229					   cmdlen[i], DMA_TO_DEVICE);
230		if (dma_mapping_error(trans->dev, phys_addr)) {
231			idx = -ENOMEM;
232			iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd);
233			goto out;
234		}
235		iwl_txq_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
236	}
237
238	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
239	out_meta->flags = cmd->flags;
240	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
241		kfree_sensitive(txq->entries[idx].free_buf);
242	txq->entries[idx].free_buf = dup_buf;
243
244	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
245
246	/* start timer if queue currently empty */
247	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
248		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
249
250	spin_lock(&trans_pcie->reg_lock);
 
 
 
 
 
 
251	/* Increment and update queue's write index */
252	txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr);
253	iwl_txq_inc_wr_ptr(trans, txq);
254	spin_unlock(&trans_pcie->reg_lock);
255
256out:
257	spin_unlock_irqrestore(&txq->lock, flags);
258free_dup_buf:
259	if (idx < 0)
260		kfree(dup_buf);
261	return idx;
262}
v4.17
   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) 2017 Intel Deutschland GmbH
   9 *
  10 * This program is free software; you can redistribute it and/or modify
  11 * it under the terms of version 2 of the GNU General Public License as
  12 * published by the Free Software Foundation.
  13 *
  14 * This program is distributed in the hope that it will be useful, but
  15 * WITHOUT ANY WARRANTY; without even the implied warranty of
  16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17 * General Public License for more details.
  18 *
  19 * BSD LICENSE
  20 *
  21 * Copyright(c) 2017 Intel Deutschland GmbH
  22 * All rights reserved.
  23 *
  24 * Redistribution and use in source and binary forms, with or without
  25 * modification, are permitted provided that the following conditions
  26 * are met:
  27 *
  28 *  * Redistributions of source code must retain the above copyright
  29 *    notice, this list of conditions and the following disclaimer.
  30 *  * Redistributions in binary form must reproduce the above copyright
  31 *    notice, this list of conditions and the following disclaimer in
  32 *    the documentation and/or other materials provided with the
  33 *    distribution.
  34 *  * Neither the name Intel Corporation nor the names of its
  35 *    contributors may be used to endorse or promote products derived
  36 *    from this software without specific prior written permission.
  37 *
  38 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  39 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  40 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  41 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  42 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  43 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  44 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  45 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  46 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  47 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  48 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  49 *
  50 *****************************************************************************/
  51#include <linux/pm_runtime.h>
  52#include <net/tso.h>
 
  53
  54#include "iwl-debug.h"
  55#include "iwl-csr.h"
  56#include "iwl-io.h"
  57#include "internal.h"
  58#include "fw/api/tx.h"
  59
  60 /*
  61 * iwl_pcie_gen2_tx_stop - Stop all Tx DMA channels
  62 */
  63void iwl_pcie_gen2_tx_stop(struct iwl_trans *trans)
  64{
  65	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
  66	int txq_id;
  67
  68	/*
  69	 * This function can be called before the op_mode disabled the
  70	 * queues. This happens when we have an rfkill interrupt.
  71	 * Since we stop Tx altogether - mark the queues as stopped.
  72	 */
  73	memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped));
  74	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
  75
  76	/* Unmap DMA from host system and free skb's */
  77	for (txq_id = 0; txq_id < ARRAY_SIZE(trans_pcie->txq); txq_id++) {
  78		if (!trans_pcie->txq[txq_id])
  79			continue;
  80		iwl_pcie_gen2_txq_unmap(trans, txq_id);
  81	}
  82}
  83
  84/*
  85 * iwl_pcie_txq_update_byte_tbl - Set up entry in Tx byte-count array
  86 */
  87static void iwl_pcie_gen2_update_byte_tbl(struct iwl_txq *txq, u16 byte_cnt,
  88					  int num_tbs)
  89{
  90	struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr;
  91	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
  92	u8 filled_tfd_size, num_fetch_chunks;
  93	u16 len = byte_cnt;
  94	__le16 bc_ent;
  95
  96	len = DIV_ROUND_UP(len, 4);
  97
  98	if (WARN_ON(len > 0xFFF || idx >= txq->n_window))
  99		return;
 100
 101	filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) +
 102				   num_tbs * sizeof(struct iwl_tfh_tb);
 103	/*
 104	 * filled_tfd_size contains the number of filled bytes in the TFD.
 105	 * Dividing it by 64 will give the number of chunks to fetch
 106	 * to SRAM- 0 for one chunk, 1 for 2 and so on.
 107	 * If, for example, TFD contains only 3 TBs then 32 bytes
 108	 * of the TFD are used, and only one chunk of 64 bytes should
 109	 * be fetched
 110	 */
 111	num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1;
 112
 113	bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12));
 114	scd_bc_tbl->tfd_offset[idx] = bc_ent;
 115}
 116
 117/*
 118 * iwl_pcie_gen2_txq_inc_wr_ptr - Send new write index to hardware
 119 */
 120static void iwl_pcie_gen2_txq_inc_wr_ptr(struct iwl_trans *trans,
 121					 struct iwl_txq *txq)
 122{
 123	lockdep_assert_held(&txq->lock);
 124
 125	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr);
 126
 127	/*
 128	 * if not in power-save mode, uCode will never sleep when we're
 129	 * trying to tx (during RFKILL, we're not trying to tx).
 130	 */
 131	iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16));
 132}
 133
 134static u8 iwl_pcie_gen2_get_num_tbs(struct iwl_trans *trans,
 135				    struct iwl_tfh_tfd *tfd)
 136{
 137	return le16_to_cpu(tfd->num_tbs) & 0x1f;
 138}
 139
 140static void iwl_pcie_gen2_tfd_unmap(struct iwl_trans *trans,
 141				    struct iwl_cmd_meta *meta,
 142				    struct iwl_tfh_tfd *tfd)
 143{
 144	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 145	int i, num_tbs;
 146
 147	/* Sanity check on number of chunks */
 148	num_tbs = iwl_pcie_gen2_get_num_tbs(trans, tfd);
 149
 150	if (num_tbs > trans_pcie->max_tbs) {
 151		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
 152		return;
 153	}
 154
 155	/* first TB is never freed - it's the bidirectional DMA data */
 156	for (i = 1; i < num_tbs; i++) {
 157		if (meta->tbs & BIT(i))
 158			dma_unmap_page(trans->dev,
 159				       le64_to_cpu(tfd->tbs[i].addr),
 160				       le16_to_cpu(tfd->tbs[i].tb_len),
 161				       DMA_TO_DEVICE);
 162		else
 163			dma_unmap_single(trans->dev,
 164					 le64_to_cpu(tfd->tbs[i].addr),
 165					 le16_to_cpu(tfd->tbs[i].tb_len),
 166					 DMA_TO_DEVICE);
 167	}
 168
 169	tfd->num_tbs = 0;
 170}
 171
 172static void iwl_pcie_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
 173{
 174	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
 175	 * idx is bounded by n_window
 176	 */
 177	int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
 178
 179	lockdep_assert_held(&txq->lock);
 180
 181	iwl_pcie_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
 182				iwl_pcie_get_tfd(trans, txq, idx));
 183
 184	/* free SKB */
 185	if (txq->entries) {
 186		struct sk_buff *skb;
 187
 188		skb = txq->entries[idx].skb;
 189
 190		/* Can be called from irqs-disabled context
 191		 * If skb is not NULL, it means that the whole queue is being
 192		 * freed and that the queue is not empty - free the skb
 193		 */
 194		if (skb) {
 195			iwl_op_mode_free_skb(trans->op_mode, skb);
 196			txq->entries[idx].skb = NULL;
 197		}
 198	}
 199}
 200
 201static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
 202				struct iwl_tfh_tfd *tfd, dma_addr_t addr,
 203				u16 len)
 204{
 205	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 206	int idx = iwl_pcie_gen2_get_num_tbs(trans, tfd);
 207	struct iwl_tfh_tb *tb = &tfd->tbs[idx];
 208
 209	/* Each TFD can point to a maximum max_tbs Tx buffers */
 210	if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) {
 211		IWL_ERR(trans, "Error can not send more than %d chunks\n",
 212			trans_pcie->max_tbs);
 213		return -EINVAL;
 214	}
 215
 216	put_unaligned_le64(addr, &tb->addr);
 217	tb->tb_len = cpu_to_le16(len);
 218
 219	tfd->num_tbs = cpu_to_le16(idx + 1);
 220
 221	return idx;
 222}
 223
 224static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 225				     struct sk_buff *skb,
 226				     struct iwl_tfh_tfd *tfd, int start_len,
 227				     u8 hdr_len, struct iwl_device_cmd *dev_cmd)
 228{
 229#ifdef CONFIG_INET
 230	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 231	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
 232	struct ieee80211_hdr *hdr = (void *)skb->data;
 233	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
 234	unsigned int mss = skb_shinfo(skb)->gso_size;
 235	u16 length, iv_len, amsdu_pad;
 236	u8 *start_hdr;
 237	struct iwl_tso_hdr_page *hdr_page;
 238	struct page **page_ptr;
 239	struct tso_t tso;
 240
 241	/* if the packet is protected, then it must be CCMP or GCMP */
 242	iv_len = ieee80211_has_protected(hdr->frame_control) ?
 243		IEEE80211_CCMP_HDR_LEN : 0;
 244
 245	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
 246			     &dev_cmd->hdr, start_len, 0);
 247
 248	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
 249	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
 250	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len;
 251	amsdu_pad = 0;
 252
 253	/* total amount of header we may need for this A-MSDU */
 254	hdr_room = DIV_ROUND_UP(total_len, mss) *
 255		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len;
 256
 257	/* Our device supports 9 segments at most, it will fit in 1 page */
 258	hdr_page = get_page_hdr(trans, hdr_room);
 259	if (!hdr_page)
 260		return -ENOMEM;
 261
 262	get_page(hdr_page->page);
 263	start_hdr = hdr_page->pos;
 264	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
 265	*page_ptr = hdr_page->page;
 266	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
 267	hdr_page->pos += iv_len;
 268
 269	/*
 270	 * Pull the ieee80211 header + IV to be able to use TSO core,
 271	 * we will restore it for the tx_status flow.
 272	 */
 273	skb_pull(skb, hdr_len + iv_len);
 274
 275	/*
 276	 * Remove the length of all the headers that we don't actually
 277	 * have in the MPDU by themselves, but that we duplicate into
 278	 * all the different MSDUs inside the A-MSDU.
 279	 */
 280	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
 281
 282	tso_start(skb, &tso);
 283
 284	while (total_len) {
 285		/* this is the data left for this subframe */
 286		unsigned int data_left = min_t(unsigned int, mss, total_len);
 287		struct sk_buff *csum_skb = NULL;
 288		unsigned int tb_len;
 289		dma_addr_t tb_phys;
 290		u8 *subf_hdrs_start = hdr_page->pos;
 291
 292		total_len -= data_left;
 293
 294		memset(hdr_page->pos, 0, amsdu_pad);
 295		hdr_page->pos += amsdu_pad;
 296		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
 297				  data_left)) & 0x3;
 298		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
 299		hdr_page->pos += ETH_ALEN;
 300		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
 301		hdr_page->pos += ETH_ALEN;
 302
 303		length = snap_ip_tcp_hdrlen + data_left;
 304		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
 305		hdr_page->pos += sizeof(length);
 306
 307		/*
 308		 * This will copy the SNAP as well which will be considered
 309		 * as MAC header.
 310		 */
 311		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
 312
 313		hdr_page->pos += snap_ip_tcp_hdrlen;
 314
 315		tb_len = hdr_page->pos - start_hdr;
 316		tb_phys = dma_map_single(trans->dev, start_hdr,
 317					 tb_len, DMA_TO_DEVICE);
 318		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
 319			dev_kfree_skb(csum_skb);
 320			goto out_err;
 321		}
 322		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
 323		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr, tb_len);
 324		/* add this subframe's headers' length to the tx_cmd */
 325		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
 326
 327		/* prepare the start_hdr for the next subframe */
 328		start_hdr = hdr_page->pos;
 329
 330		/* put the payload */
 331		while (data_left) {
 332			tb_len = min_t(unsigned int, tso.size, data_left);
 333			tb_phys = dma_map_single(trans->dev, tso.data,
 334						 tb_len, DMA_TO_DEVICE);
 335			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
 336				dev_kfree_skb(csum_skb);
 337				goto out_err;
 338			}
 339			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
 340			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
 341						       tb_len);
 342
 343			data_left -= tb_len;
 344			tso_build_data(skb, &tso, tb_len);
 345		}
 346	}
 347
 348	/* re -add the WiFi header and IV */
 349	skb_push(skb, hdr_len + iv_len);
 350
 351	return 0;
 352
 353out_err:
 354#endif
 355	return -EINVAL;
 356}
 357
 358static
 359struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 360					    struct iwl_txq *txq,
 361					    struct iwl_device_cmd *dev_cmd,
 362					    struct sk_buff *skb,
 363					    struct iwl_cmd_meta *out_meta)
 364{
 365	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
 366	int idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
 367	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, idx);
 368	dma_addr_t tb_phys;
 369	bool amsdu;
 370	int i, len, tb1_len, tb2_len, hdr_len;
 371	void *tb1_addr;
 372
 373	memset(tfd, 0, sizeof(*tfd));
 374
 375	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
 376		(*ieee80211_get_qos_ctl(hdr) &
 377		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
 378
 379	tb_phys = iwl_pcie_get_first_tb_dma(txq, idx);
 380	/* The first TB points to bi-directional DMA data */
 381	if (!amsdu)
 382		memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr,
 383		       IWL_FIRST_TB_SIZE);
 384
 385	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
 386
 387	/* there must be data left over for TB1 or this code must be changed */
 388	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
 389
 390	/*
 391	 * The second TB (tb1) points to the remainder of the TX command
 392	 * and the 802.11 header - dword aligned size
 393	 * (This calculation modifies the TX command, so do it before the
 394	 * setup of the first TB)
 395	 */
 396	len = sizeof(struct iwl_tx_cmd_gen2) + sizeof(struct iwl_cmd_header) +
 397	      ieee80211_hdrlen(hdr->frame_control) - IWL_FIRST_TB_SIZE;
 398
 399	/* do not align A-MSDU to dword as the subframe header aligns it */
 400	if (amsdu)
 401		tb1_len = len;
 402	else
 403		tb1_len = ALIGN(len, 4);
 404
 405	/* map the data for TB1 */
 406	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
 407	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
 408	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
 409		goto out_err;
 410	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
 411
 412	hdr_len = ieee80211_hdrlen(hdr->frame_control);
 413
 414	if (amsdu) {
 415		if (iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
 416					      tb1_len + IWL_FIRST_TB_SIZE,
 417					      hdr_len, dev_cmd))
 418			goto out_err;
 419
 420		/*
 421		 * building the A-MSDU might have changed this data, so memcpy
 422		 * it now
 423		 */
 424		memcpy(&txq->first_tb_bufs[idx], &dev_cmd->hdr,
 425		       IWL_FIRST_TB_SIZE);
 426		return tfd;
 427	}
 428
 429	/* set up TFD's third entry to point to remainder of skb's head */
 430	tb2_len = skb_headlen(skb) - hdr_len;
 431
 432	if (tb2_len > 0) {
 433		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
 434					 tb2_len, DMA_TO_DEVICE);
 435		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
 436			goto out_err;
 437		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
 438	}
 439
 440	/* set up the remaining entries to point to the data */
 441	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 442		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 443		int tb_idx;
 444
 445		if (!skb_frag_size(frag))
 446			continue;
 447
 448		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
 449					   skb_frag_size(frag), DMA_TO_DEVICE);
 450
 451		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
 452			goto out_err;
 453		tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
 454					      skb_frag_size(frag));
 455
 456		out_meta->tbs |= BIT(tb_idx);
 457	}
 458
 459	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
 460			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
 461	trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len);
 462
 463	return tfd;
 464
 465out_err:
 466	iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
 467	return NULL;
 468}
 469
 470int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
 471			   struct iwl_device_cmd *dev_cmd, int txq_id)
 472{
 473	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 474	struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
 475	struct iwl_cmd_meta *out_meta;
 476	struct iwl_txq *txq = trans_pcie->txq[txq_id];
 477	int idx;
 478	void *tfd;
 479
 480	if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used),
 481		      "TX on unused queue %d\n", txq_id))
 482		return -EINVAL;
 483
 484	if (skb_is_nonlinear(skb) &&
 485	    skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) &&
 486	    __skb_linearize(skb))
 487		return -ENOMEM;
 488
 489	spin_lock(&txq->lock);
 490
 491	idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
 492
 493	/* Set up driver data for this TFD */
 494	txq->entries[idx].skb = skb;
 495	txq->entries[idx].cmd = dev_cmd;
 496
 497	dev_cmd->hdr.sequence =
 498		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
 499			    INDEX_TO_SEQ(idx)));
 500
 501	/* Set up first empty entry in queue's array of Tx/cmd buffers */
 502	out_meta = &txq->entries[idx].meta;
 503	out_meta->flags = 0;
 504
 505	tfd = iwl_pcie_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
 506	if (!tfd) {
 507		spin_unlock(&txq->lock);
 508		return -1;
 509	}
 510
 511	/* Set up entry for this TFD in Tx byte-count array */
 512	iwl_pcie_gen2_update_byte_tbl(txq, le16_to_cpu(tx_cmd->len),
 513				      iwl_pcie_gen2_get_num_tbs(trans, tfd));
 514
 515	/* start timer if queue currently empty */
 516	if (txq->read_ptr == txq->write_ptr) {
 517		if (txq->wd_timeout)
 518			mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
 519		IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id);
 520		iwl_trans_ref(trans);
 521	}
 522
 523	/* Tell device the write index *just past* this latest filled TFD */
 524	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
 525	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
 526	if (iwl_queue_space(txq) < txq->high_mark)
 527		iwl_stop_queue(trans, txq);
 528
 529	/*
 530	 * At this point the frame is "transmitted" successfully
 531	 * and we will get a TX status notification eventually.
 532	 */
 533	spin_unlock(&txq->lock);
 534	return 0;
 535}
 536
 537/*************** HOST COMMAND QUEUE FUNCTIONS   *****/
 538
 539/*
 540 * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command
 541 * @priv: device private data point
 542 * @cmd: a pointer to the ucode command structure
 543 *
 544 * The function returns < 0 values to indicate the operation
 545 * failed. On success, it returns the index (>= 0) of command in the
 546 * command queue.
 547 */
 548static int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans,
 549				      struct iwl_host_cmd *cmd)
 550{
 551	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 552	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
 553	struct iwl_device_cmd *out_cmd;
 554	struct iwl_cmd_meta *out_meta;
 555	unsigned long flags;
 556	void *dup_buf = NULL;
 557	dma_addr_t phys_addr;
 558	int i, cmd_pos, idx = iwl_pcie_get_cmd_index(txq, txq->write_ptr);
 559	u16 copy_size, cmd_size, tb0_size;
 560	bool had_nocopy = false;
 561	u8 group_id = iwl_cmd_groupid(cmd->id);
 562	const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD];
 563	u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD];
 564	struct iwl_tfh_tfd *tfd = iwl_pcie_get_tfd(trans, txq, txq->write_ptr);
 
 565
 566	memset(tfd, 0, sizeof(*tfd));
 
 567
 568	copy_size = sizeof(struct iwl_cmd_header_wide);
 569	cmd_size = sizeof(struct iwl_cmd_header_wide);
 570
 571	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
 572		cmddata[i] = cmd->data[i];
 573		cmdlen[i] = cmd->len[i];
 574
 575		if (!cmd->len[i])
 576			continue;
 577
 578		/* need at least IWL_FIRST_TB_SIZE copied */
 579		if (copy_size < IWL_FIRST_TB_SIZE) {
 580			int copy = IWL_FIRST_TB_SIZE - copy_size;
 581
 582			if (copy > cmdlen[i])
 583				copy = cmdlen[i];
 584			cmdlen[i] -= copy;
 585			cmddata[i] += copy;
 586			copy_size += copy;
 587		}
 588
 589		if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) {
 590			had_nocopy = true;
 591			if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) {
 592				idx = -EINVAL;
 593				goto free_dup_buf;
 594			}
 595		} else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) {
 596			/*
 597			 * This is also a chunk that isn't copied
 598			 * to the static buffer so set had_nocopy.
 599			 */
 600			had_nocopy = true;
 601
 602			/* only allowed once */
 603			if (WARN_ON(dup_buf)) {
 604				idx = -EINVAL;
 605				goto free_dup_buf;
 606			}
 607
 608			dup_buf = kmemdup(cmddata[i], cmdlen[i],
 609					  GFP_ATOMIC);
 610			if (!dup_buf)
 611				return -ENOMEM;
 612		} else {
 613			/* NOCOPY must not be followed by normal! */
 614			if (WARN_ON(had_nocopy)) {
 615				idx = -EINVAL;
 616				goto free_dup_buf;
 617			}
 618			copy_size += cmdlen[i];
 619		}
 620		cmd_size += cmd->len[i];
 621	}
 622
 623	/*
 624	 * If any of the command structures end up being larger than the
 625	 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into
 626	 * separate TFDs, then we will need to increase the size of the buffers
 627	 */
 628	if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE,
 629		 "Command %s (%#x) is too large (%d bytes)\n",
 630		 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) {
 631		idx = -EINVAL;
 632		goto free_dup_buf;
 633	}
 634
 635	spin_lock_bh(&txq->lock);
 
 
 
 
 636
 637	if (iwl_queue_space(txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) {
 638		spin_unlock_bh(&txq->lock);
 639
 640		IWL_ERR(trans, "No space in command queue\n");
 641		iwl_op_mode_cmd_queue_full(trans->op_mode);
 642		idx = -ENOSPC;
 643		goto free_dup_buf;
 644	}
 645
 646	out_cmd = txq->entries[idx].cmd;
 647	out_meta = &txq->entries[idx].meta;
 648
 649	/* re-initialize to NULL */
 650	memset(out_meta, 0, sizeof(*out_meta));
 651	if (cmd->flags & CMD_WANT_SKB)
 652		out_meta->source = cmd;
 653
 654	/* set up the header */
 655	out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id);
 656	out_cmd->hdr_wide.group_id = group_id;
 657	out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id);
 658	out_cmd->hdr_wide.length =
 659		cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide));
 660	out_cmd->hdr_wide.reserved = 0;
 661	out_cmd->hdr_wide.sequence =
 662		cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) |
 663					 INDEX_TO_SEQ(txq->write_ptr));
 664
 665	cmd_pos = sizeof(struct iwl_cmd_header_wide);
 666	copy_size = sizeof(struct iwl_cmd_header_wide);
 667
 668	/* and copy the data that needs to be copied */
 669	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
 670		int copy;
 671
 672		if (!cmd->len[i])
 673			continue;
 674
 675		/* copy everything if not nocopy/dup */
 676		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
 677					   IWL_HCMD_DFL_DUP))) {
 678			copy = cmd->len[i];
 679
 680			memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
 681			cmd_pos += copy;
 682			copy_size += copy;
 683			continue;
 684		}
 685
 686		/*
 687		 * Otherwise we need at least IWL_FIRST_TB_SIZE copied
 688		 * in total (for bi-directional DMA), but copy up to what
 689		 * we can fit into the payload for debug dump purposes.
 690		 */
 691		copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]);
 692
 693		memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy);
 694		cmd_pos += copy;
 695
 696		/* However, treat copy_size the proper way, we need it below */
 697		if (copy_size < IWL_FIRST_TB_SIZE) {
 698			copy = IWL_FIRST_TB_SIZE - copy_size;
 699
 700			if (copy > cmd->len[i])
 701				copy = cmd->len[i];
 702			copy_size += copy;
 703		}
 704	}
 705
 706	IWL_DEBUG_HC(trans,
 707		     "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n",
 708		     iwl_get_cmd_string(trans, cmd->id), group_id,
 709		     out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence),
 710		     cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue);
 711
 712	/* start the TFD with the minimum copy bytes */
 713	tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE);
 714	memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size);
 715	iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx),
 716			     tb0_size);
 717
 718	/* map first command fragment, if any remains */
 719	if (copy_size > tb0_size) {
 720		phys_addr = dma_map_single(trans->dev,
 721					   ((u8 *)&out_cmd->hdr) + tb0_size,
 722					   copy_size - tb0_size,
 723					   DMA_TO_DEVICE);
 724		if (dma_mapping_error(trans->dev, phys_addr)) {
 725			idx = -ENOMEM;
 726			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
 727			goto out;
 728		}
 729		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr,
 730				     copy_size - tb0_size);
 731	}
 732
 733	/* map the remaining (adjusted) nocopy/dup fragments */
 734	for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) {
 735		const void *data = cmddata[i];
 736
 737		if (!cmdlen[i])
 738			continue;
 739		if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY |
 740					   IWL_HCMD_DFL_DUP)))
 741			continue;
 742		if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP)
 743			data = dup_buf;
 744		phys_addr = dma_map_single(trans->dev, (void *)data,
 745					   cmdlen[i], DMA_TO_DEVICE);
 746		if (dma_mapping_error(trans->dev, phys_addr)) {
 747			idx = -ENOMEM;
 748			iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd);
 749			goto out;
 750		}
 751		iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]);
 752	}
 753
 754	BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE);
 755	out_meta->flags = cmd->flags;
 756	if (WARN_ON_ONCE(txq->entries[idx].free_buf))
 757		kzfree(txq->entries[idx].free_buf);
 758	txq->entries[idx].free_buf = dup_buf;
 759
 760	trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide);
 761
 762	/* start timer if queue currently empty */
 763	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
 764		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
 765
 766	spin_lock_irqsave(&trans_pcie->reg_lock, flags);
 767	if (!(cmd->flags & CMD_SEND_IN_IDLE) &&
 768	    !trans_pcie->ref_cmd_in_flight) {
 769		trans_pcie->ref_cmd_in_flight = true;
 770		IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n");
 771		iwl_trans_ref(trans);
 772	}
 773	/* Increment and update queue's write index */
 774	txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr);
 775	iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq);
 776	spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
 777
 778out:
 779	spin_unlock_bh(&txq->lock);
 780free_dup_buf:
 781	if (idx < 0)
 782		kfree(dup_buf);
 783	return idx;
 784}
 785
 786#define HOST_COMPLETE_TIMEOUT	(2 * HZ)
 787
 788static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans,
 789					struct iwl_host_cmd *cmd)
 790{
 791	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 792	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
 793	struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue];
 794	int cmd_idx;
 795	int ret;
 796
 797	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
 798
 799	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
 800				  &trans->status),
 801		 "Command %s: a command is already active!\n", cmd_str))
 802		return -EIO;
 803
 804	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
 805
 806	if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) {
 807		ret = wait_event_timeout(trans_pcie->d0i3_waitq,
 808				 pm_runtime_active(&trans_pcie->pci_dev->dev),
 809				 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT));
 810		if (!ret) {
 811			IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n");
 812			return -ETIMEDOUT;
 813		}
 814	}
 815
 816	cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
 817	if (cmd_idx < 0) {
 818		ret = cmd_idx;
 819		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
 820		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
 821			cmd_str, ret);
 822		return ret;
 823	}
 824
 825	ret = wait_event_timeout(trans_pcie->wait_command_queue,
 826				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
 827					   &trans->status),
 828				 HOST_COMPLETE_TIMEOUT);
 829	if (!ret) {
 830		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
 831			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
 832
 833		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
 834			txq->read_ptr, txq->write_ptr);
 835
 836		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
 837		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
 838			       cmd_str);
 839		ret = -ETIMEDOUT;
 840
 841		iwl_force_nmi(trans);
 842		iwl_trans_fw_error(trans);
 843
 844		goto cancel;
 845	}
 846
 847	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
 848		IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
 849		dump_stack();
 850		ret = -EIO;
 851		goto cancel;
 852	}
 853
 854	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
 855	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
 856		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
 857		ret = -ERFKILL;
 858		goto cancel;
 859	}
 860
 861	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
 862		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
 863		ret = -EIO;
 864		goto cancel;
 865	}
 866
 867	return 0;
 868
 869cancel:
 870	if (cmd->flags & CMD_WANT_SKB) {
 871		/*
 872		 * Cancel the CMD_WANT_SKB flag for the cmd in the
 873		 * TX cmd queue. Otherwise in case the cmd comes
 874		 * in later, it will possibly set an invalid
 875		 * address (cmd->meta.source).
 876		 */
 877		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
 878	}
 879
 880	if (cmd->resp_pkt) {
 881		iwl_free_resp(cmd);
 882		cmd->resp_pkt = NULL;
 883	}
 884
 885	return ret;
 886}
 887
 888int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans,
 889				  struct iwl_host_cmd *cmd)
 890{
 891	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
 892	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
 893		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
 894				  cmd->id);
 895		return -ERFKILL;
 896	}
 897
 898	if (cmd->flags & CMD_ASYNC) {
 899		int ret;
 900
 901		/* An asynchronous command can not expect an SKB to be set. */
 902		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
 903			return -EINVAL;
 904
 905		ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd);
 906		if (ret < 0) {
 907			IWL_ERR(trans,
 908				"Error sending %s: enqueue_hcmd failed: %d\n",
 909				iwl_get_cmd_string(trans, cmd->id), ret);
 910			return ret;
 911		}
 912		return 0;
 913	}
 914
 915	return iwl_pcie_gen2_send_hcmd_sync(trans, cmd);
 916}
 917
 918/*
 919 * iwl_pcie_gen2_txq_unmap -  Unmap any remaining DMA mappings and free skb's
 920 */
 921void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id)
 922{
 923	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 924	struct iwl_txq *txq = trans_pcie->txq[txq_id];
 925
 926	spin_lock_bh(&txq->lock);
 927	while (txq->write_ptr != txq->read_ptr) {
 928		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
 929				   txq_id, txq->read_ptr);
 930
 931		if (txq_id != trans_pcie->cmd_queue) {
 932			int idx = iwl_pcie_get_cmd_index(txq, txq->read_ptr);
 933			struct sk_buff *skb = txq->entries[idx].skb;
 934
 935			if (WARN_ON_ONCE(!skb))
 936				continue;
 937
 938			iwl_pcie_free_tso_page(trans_pcie, skb);
 939		}
 940		iwl_pcie_gen2_free_tfd(trans, txq);
 941		txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr);
 942
 943		if (txq->read_ptr == txq->write_ptr) {
 944			unsigned long flags;
 945
 946			spin_lock_irqsave(&trans_pcie->reg_lock, flags);
 947			if (txq_id != trans_pcie->cmd_queue) {
 948				IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n",
 949					      txq->id);
 950				iwl_trans_unref(trans);
 951			} else if (trans_pcie->ref_cmd_in_flight) {
 952				trans_pcie->ref_cmd_in_flight = false;
 953				IWL_DEBUG_RPM(trans,
 954					      "clear ref_cmd_in_flight\n");
 955				iwl_trans_unref(trans);
 956			}
 957			spin_unlock_irqrestore(&trans_pcie->reg_lock, flags);
 958		}
 959	}
 960	spin_unlock_bh(&txq->lock);
 961
 962	/* just in case - this queue may have been stopped */
 963	iwl_wake_queue(trans, txq);
 964}
 965
 966static void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans,
 967					  struct iwl_txq *txq)
 968{
 969	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 970	struct device *dev = trans->dev;
 971
 972	/* De-alloc circular buffer of TFDs */
 973	if (txq->tfds) {
 974		dma_free_coherent(dev,
 975				  trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX,
 976				  txq->tfds, txq->dma_addr);
 977		dma_free_coherent(dev,
 978				  sizeof(*txq->first_tb_bufs) * txq->n_window,
 979				  txq->first_tb_bufs, txq->first_tb_dma);
 980	}
 981
 982	kfree(txq->entries);
 983	iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl);
 984	kfree(txq);
 985}
 986
 987/*
 988 * iwl_pcie_txq_free - Deallocate DMA queue.
 989 * @txq: Transmit queue to deallocate.
 990 *
 991 * Empty queue by removing and destroying all BD's.
 992 * Free all buffers.
 993 * 0-fill, but do not free "txq" descriptor structure.
 994 */
 995static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id)
 996{
 997	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 998	struct iwl_txq *txq = trans_pcie->txq[txq_id];
 999	int i;
1000
1001	if (WARN_ON(!txq))
1002		return;
1003
1004	iwl_pcie_gen2_txq_unmap(trans, txq_id);
1005
1006	/* De-alloc array of command/tx buffers */
1007	if (txq_id == trans_pcie->cmd_queue)
1008		for (i = 0; i < txq->n_window; i++) {
1009			kzfree(txq->entries[i].cmd);
1010			kzfree(txq->entries[i].free_buf);
1011		}
1012	del_timer_sync(&txq->stuck_timer);
1013
1014	iwl_pcie_gen2_txq_free_memory(trans, txq);
1015
1016	trans_pcie->txq[txq_id] = NULL;
1017
1018	clear_bit(txq_id, trans_pcie->queue_used);
1019}
1020
1021int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
1022				 struct iwl_tx_queue_cfg_cmd *cmd,
1023				 int cmd_id,
1024				 unsigned int timeout)
1025{
1026	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1027	struct iwl_tx_queue_cfg_rsp *rsp;
1028	struct iwl_txq *txq;
1029	struct iwl_host_cmd hcmd = {
1030		.id = cmd_id,
1031		.len = { sizeof(*cmd) },
1032		.data = { cmd, },
1033		.flags = CMD_WANT_SKB,
1034	};
1035	int ret, qid;
1036	u32 wr_ptr;
1037
1038	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1039	if (!txq)
1040		return -ENOMEM;
1041	ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl,
1042				     sizeof(struct iwlagn_scd_bc_tbl));
1043	if (ret) {
1044		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1045		kfree(txq);
1046		return -ENOMEM;
1047	}
1048
1049	ret = iwl_pcie_txq_alloc(trans, txq, TFD_TX_CMD_SLOTS, false);
1050	if (ret) {
1051		IWL_ERR(trans, "Tx queue alloc failed\n");
1052		goto error;
1053	}
1054	ret = iwl_pcie_txq_init(trans, txq, TFD_TX_CMD_SLOTS, false);
1055	if (ret) {
1056		IWL_ERR(trans, "Tx queue init failed\n");
1057		goto error;
1058	}
1059
1060	txq->wd_timeout = msecs_to_jiffies(timeout);
1061
1062	cmd->tfdq_addr = cpu_to_le64(txq->dma_addr);
1063	cmd->byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1064	cmd->cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(TFD_TX_CMD_SLOTS));
1065
1066	ret = iwl_trans_send_cmd(trans, &hcmd);
1067	if (ret)
1068		goto error;
1069
1070	if (WARN_ON(iwl_rx_packet_payload_len(hcmd.resp_pkt) != sizeof(*rsp))) {
1071		ret = -EINVAL;
1072		goto error_free_resp;
1073	}
1074
1075	rsp = (void *)hcmd.resp_pkt->data;
1076	qid = le16_to_cpu(rsp->queue_number);
1077	wr_ptr = le16_to_cpu(rsp->write_pointer);
1078
1079	if (qid >= ARRAY_SIZE(trans_pcie->txq)) {
1080		WARN_ONCE(1, "queue index %d unsupported", qid);
1081		ret = -EIO;
1082		goto error_free_resp;
1083	}
1084
1085	if (test_and_set_bit(qid, trans_pcie->queue_used)) {
1086		WARN_ONCE(1, "queue %d already used", qid);
1087		ret = -EIO;
1088		goto error_free_resp;
1089	}
1090
1091	txq->id = qid;
1092	trans_pcie->txq[qid] = txq;
1093	wr_ptr &= (TFD_QUEUE_SIZE_MAX - 1);
1094
1095	/* Place first TFD at index corresponding to start sequence number */
1096	txq->read_ptr = wr_ptr;
1097	txq->write_ptr = wr_ptr;
1098	iwl_write_direct32(trans, HBUS_TARG_WRPTR,
1099			   (txq->write_ptr) | (qid << 16));
1100	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1101
1102	iwl_free_resp(&hcmd);
1103	return qid;
1104
1105error_free_resp:
1106	iwl_free_resp(&hcmd);
1107error:
1108	iwl_pcie_gen2_txq_free_memory(trans, txq);
1109	return ret;
1110}
1111
1112void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue)
1113{
1114	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1115
1116	/*
1117	 * Upon HW Rfkill - we stop the device, and then stop the queues
1118	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1119	 * allow the op_mode to call txq_disable after it already called
1120	 * stop_device.
1121	 */
1122	if (!test_and_clear_bit(queue, trans_pcie->queue_used)) {
1123		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1124			  "queue %d not used", queue);
1125		return;
1126	}
1127
1128	iwl_pcie_gen2_txq_unmap(trans, queue);
1129
1130	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1131}
1132
1133void iwl_pcie_gen2_tx_free(struct iwl_trans *trans)
1134{
1135	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1136	int i;
1137
1138	memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used));
1139
1140	/* Free all TX queues */
1141	for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) {
1142		if (!trans_pcie->txq[i])
1143			continue;
1144
1145		iwl_pcie_gen2_txq_free(trans, i);
1146	}
1147}
1148
1149int iwl_pcie_gen2_tx_init(struct iwl_trans *trans)
1150{
1151	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
1152	struct iwl_txq *cmd_queue;
1153	int txq_id = trans_pcie->cmd_queue, ret;
1154
1155	iwl_pcie_set_tx_cmd_queue_size(trans);
1156
1157	/* alloc and init the command queue */
1158	if (!trans_pcie->txq[txq_id]) {
1159		cmd_queue = kzalloc(sizeof(*cmd_queue), GFP_KERNEL);
1160		if (!cmd_queue) {
1161			IWL_ERR(trans, "Not enough memory for command queue\n");
1162			return -ENOMEM;
1163		}
1164		trans_pcie->txq[txq_id] = cmd_queue;
1165		ret = iwl_pcie_txq_alloc(trans, cmd_queue,
1166					 trans_pcie->tx_cmd_queue_size, true);
1167		if (ret) {
1168			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1169			goto error;
1170		}
1171	} else {
1172		cmd_queue = trans_pcie->txq[txq_id];
1173	}
1174
1175	ret = iwl_pcie_txq_init(trans, cmd_queue,
1176				trans_pcie->tx_cmd_queue_size, true);
1177	if (ret) {
1178		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1179		goto error;
1180	}
1181	trans_pcie->txq[txq_id]->id = txq_id;
1182	set_bit(txq_id, trans_pcie->queue_used);
1183
1184	return 0;
1185
1186error:
1187	iwl_pcie_gen2_tx_free(trans);
1188	return ret;
1189}
1190