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 53 #include "iwl-debug.h" 54 #include "iwl-csr.h" 55 #include "iwl-io.h" 56 #include "internal.h" 57 #include "mvm/fw-api.h" 58 59 /* 60 * iwl_pcie_gen2_tx_stop - Stop all Tx DMA channels 61 */ 62 void iwl_pcie_gen2_tx_stop(struct iwl_trans *trans) 63 { 64 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 65 int txq_id; 66 67 /* 68 * This function can be called before the op_mode disabled the 69 * queues. This happens when we have an rfkill interrupt. 70 * Since we stop Tx altogether - mark the queues as stopped. 71 */ 72 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped)); 73 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used)); 74 75 /* Unmap DMA from host system and free skb's */ 76 for (txq_id = 0; txq_id < ARRAY_SIZE(trans_pcie->txq); txq_id++) { 77 if (!trans_pcie->txq[txq_id]) 78 continue; 79 iwl_pcie_gen2_txq_unmap(trans, txq_id); 80 } 81 } 82 83 /* 84 * iwl_pcie_txq_update_byte_tbl - Set up entry in Tx byte-count array 85 */ 86 static void iwl_pcie_gen2_update_byte_tbl(struct iwl_txq *txq, u16 byte_cnt, 87 int num_tbs) 88 { 89 struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr; 90 int write_ptr = txq->write_ptr; 91 u8 filled_tfd_size, num_fetch_chunks; 92 u16 len = byte_cnt; 93 __le16 bc_ent; 94 95 len = DIV_ROUND_UP(len, 4); 96 97 if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX)) 98 return; 99 100 filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) + 101 num_tbs * sizeof(struct iwl_tfh_tb); 102 /* 103 * filled_tfd_size contains the number of filled bytes in the TFD. 104 * Dividing it by 64 will give the number of chunks to fetch 105 * to SRAM- 0 for one chunk, 1 for 2 and so on. 106 * If, for example, TFD contains only 3 TBs then 32 bytes 107 * of the TFD are used, and only one chunk of 64 bytes should 108 * be fetched 109 */ 110 num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1; 111 112 bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12)); 113 scd_bc_tbl->tfd_offset[write_ptr] = bc_ent; 114 } 115 116 /* 117 * iwl_pcie_gen2_txq_inc_wr_ptr - Send new write index to hardware 118 */ 119 static void iwl_pcie_gen2_txq_inc_wr_ptr(struct iwl_trans *trans, 120 struct iwl_txq *txq) 121 { 122 lockdep_assert_held(&txq->lock); 123 124 IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr); 125 126 /* 127 * if not in power-save mode, uCode will never sleep when we're 128 * trying to tx (during RFKILL, we're not trying to tx). 129 */ 130 iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16)); 131 } 132 133 static u8 iwl_pcie_gen2_get_num_tbs(struct iwl_trans *trans, 134 struct iwl_tfh_tfd *tfd) 135 { 136 return le16_to_cpu(tfd->num_tbs) & 0x1f; 137 } 138 139 static void iwl_pcie_gen2_tfd_unmap(struct iwl_trans *trans, 140 struct iwl_cmd_meta *meta, 141 struct iwl_tfh_tfd *tfd) 142 { 143 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 144 int i, num_tbs; 145 146 /* Sanity check on number of chunks */ 147 num_tbs = iwl_pcie_gen2_get_num_tbs(trans, tfd); 148 149 if (num_tbs >= trans_pcie->max_tbs) { 150 IWL_ERR(trans, "Too many chunks: %i\n", num_tbs); 151 return; 152 } 153 154 /* first TB is never freed - it's the bidirectional DMA data */ 155 for (i = 1; i < num_tbs; i++) { 156 if (meta->tbs & BIT(i)) 157 dma_unmap_page(trans->dev, 158 le64_to_cpu(tfd->tbs[i].addr), 159 le16_to_cpu(tfd->tbs[i].tb_len), 160 DMA_TO_DEVICE); 161 else 162 dma_unmap_single(trans->dev, 163 le64_to_cpu(tfd->tbs[i].addr), 164 le16_to_cpu(tfd->tbs[i].tb_len), 165 DMA_TO_DEVICE); 166 } 167 168 tfd->num_tbs = 0; 169 } 170 171 static void iwl_pcie_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq) 172 { 173 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 174 175 /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and 176 * idx is bounded by n_window 177 */ 178 int rd_ptr = txq->read_ptr; 179 int idx = get_cmd_index(txq, rd_ptr); 180 181 lockdep_assert_held(&txq->lock); 182 183 /* We have only q->n_window txq->entries, but we use 184 * TFD_QUEUE_SIZE_MAX tfds 185 */ 186 iwl_pcie_gen2_tfd_unmap(trans, &txq->entries[idx].meta, 187 iwl_pcie_get_tfd(trans_pcie, txq, rd_ptr)); 188 189 /* free SKB */ 190 if (txq->entries) { 191 struct sk_buff *skb; 192 193 skb = txq->entries[idx].skb; 194 195 /* Can be called from irqs-disabled context 196 * If skb is not NULL, it means that the whole queue is being 197 * freed and that the queue is not empty - free the skb 198 */ 199 if (skb) { 200 iwl_op_mode_free_skb(trans->op_mode, skb); 201 txq->entries[idx].skb = NULL; 202 } 203 } 204 } 205 206 static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans, 207 struct iwl_tfh_tfd *tfd, dma_addr_t addr, 208 u16 len) 209 { 210 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 211 int idx = iwl_pcie_gen2_get_num_tbs(trans, tfd); 212 struct iwl_tfh_tb *tb = &tfd->tbs[idx]; 213 214 /* Each TFD can point to a maximum max_tbs Tx buffers */ 215 if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->max_tbs) { 216 IWL_ERR(trans, "Error can not send more than %d chunks\n", 217 trans_pcie->max_tbs); 218 return -EINVAL; 219 } 220 221 put_unaligned_le64(addr, &tb->addr); 222 tb->tb_len = cpu_to_le16(len); 223 224 tfd->num_tbs = cpu_to_le16(idx + 1); 225 226 return idx; 227 } 228 229 static 230 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans, 231 struct iwl_txq *txq, 232 struct iwl_device_cmd *dev_cmd, 233 struct sk_buff *skb, 234 struct iwl_cmd_meta *out_meta) 235 { 236 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 237 struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data; 238 struct iwl_tfh_tfd *tfd = 239 iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr); 240 dma_addr_t tb_phys; 241 int i, len, tb1_len, tb2_len, hdr_len; 242 void *tb1_addr; 243 244 memset(tfd, 0, sizeof(*tfd)); 245 246 tb_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr); 247 /* The first TB points to bi-directional DMA data */ 248 memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr, 249 IWL_FIRST_TB_SIZE); 250 251 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE); 252 253 /* there must be data left over for TB1 or this code must be changed */ 254 BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE); 255 256 /* 257 * The second TB (tb1) points to the remainder of the TX command 258 * and the 802.11 header - dword aligned size 259 * (This calculation modifies the TX command, so do it before the 260 * setup of the first TB) 261 */ 262 len = sizeof(struct iwl_tx_cmd_gen2) + sizeof(struct iwl_cmd_header) + 263 ieee80211_hdrlen(hdr->frame_control) - IWL_FIRST_TB_SIZE; 264 265 tb1_len = ALIGN(len, 4); 266 267 /* map the data for TB1 */ 268 tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE; 269 tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE); 270 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) 271 goto out_err; 272 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len); 273 274 /* set up TFD's third entry to point to remainder of skb's head */ 275 hdr_len = ieee80211_hdrlen(hdr->frame_control); 276 tb2_len = skb_headlen(skb) - hdr_len; 277 278 if (tb2_len > 0) { 279 tb_phys = dma_map_single(trans->dev, skb->data + hdr_len, 280 tb2_len, DMA_TO_DEVICE); 281 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) 282 goto out_err; 283 iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len); 284 } 285 286 /* set up the remaining entries to point to the data */ 287 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 288 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 289 int tb_idx; 290 291 if (!skb_frag_size(frag)) 292 continue; 293 294 tb_phys = skb_frag_dma_map(trans->dev, frag, 0, 295 skb_frag_size(frag), DMA_TO_DEVICE); 296 297 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) 298 goto out_err; 299 tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, 300 skb_frag_size(frag)); 301 302 out_meta->tbs |= BIT(tb_idx); 303 } 304 305 trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr, 306 IWL_FIRST_TB_SIZE + tb1_len, 307 skb->data + hdr_len, tb2_len); 308 trace_iwlwifi_dev_tx_data(trans->dev, skb, hdr_len, 309 skb->len - hdr_len); 310 311 return tfd; 312 313 out_err: 314 iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd); 315 return NULL; 316 } 317 318 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb, 319 struct iwl_device_cmd *dev_cmd, int txq_id) 320 { 321 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 322 struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload; 323 struct iwl_cmd_meta *out_meta; 324 struct iwl_txq *txq = trans_pcie->txq[txq_id]; 325 void *tfd; 326 327 if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used), 328 "TX on unused queue %d\n", txq_id)) 329 return -EINVAL; 330 331 if (skb_is_nonlinear(skb) && 332 skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS(trans_pcie) && 333 __skb_linearize(skb)) 334 return -ENOMEM; 335 336 spin_lock(&txq->lock); 337 338 /* Set up driver data for this TFD */ 339 txq->entries[txq->write_ptr].skb = skb; 340 txq->entries[txq->write_ptr].cmd = dev_cmd; 341 342 dev_cmd->hdr.sequence = 343 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) | 344 INDEX_TO_SEQ(txq->write_ptr))); 345 346 /* Set up first empty entry in queue's array of Tx/cmd buffers */ 347 out_meta = &txq->entries[txq->write_ptr].meta; 348 out_meta->flags = 0; 349 350 tfd = iwl_pcie_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta); 351 if (!tfd) { 352 spin_unlock(&txq->lock); 353 return -1; 354 } 355 356 /* Set up entry for this TFD in Tx byte-count array */ 357 iwl_pcie_gen2_update_byte_tbl(txq, le16_to_cpu(tx_cmd->len), 358 iwl_pcie_gen2_get_num_tbs(trans, tfd)); 359 360 /* start timer if queue currently empty */ 361 if (txq->read_ptr == txq->write_ptr) { 362 if (txq->wd_timeout) 363 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); 364 IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", txq->id); 365 iwl_trans_ref(trans); 366 } 367 368 /* Tell device the write index *just past* this latest filled TFD */ 369 txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr); 370 iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq); 371 if (iwl_queue_space(txq) < txq->high_mark) 372 iwl_stop_queue(trans, txq); 373 374 /* 375 * At this point the frame is "transmitted" successfully 376 * and we will get a TX status notification eventually. 377 */ 378 spin_unlock(&txq->lock); 379 return 0; 380 } 381 382 /*************** HOST COMMAND QUEUE FUNCTIONS *****/ 383 384 /* 385 * iwl_pcie_gen2_enqueue_hcmd - enqueue a uCode command 386 * @priv: device private data point 387 * @cmd: a pointer to the ucode command structure 388 * 389 * The function returns < 0 values to indicate the operation 390 * failed. On success, it returns the index (>= 0) of command in the 391 * command queue. 392 */ 393 static int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans, 394 struct iwl_host_cmd *cmd) 395 { 396 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 397 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue]; 398 struct iwl_device_cmd *out_cmd; 399 struct iwl_cmd_meta *out_meta; 400 unsigned long flags; 401 void *dup_buf = NULL; 402 dma_addr_t phys_addr; 403 int idx, i, cmd_pos; 404 u16 copy_size, cmd_size, tb0_size; 405 bool had_nocopy = false; 406 u8 group_id = iwl_cmd_groupid(cmd->id); 407 const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD]; 408 u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD]; 409 struct iwl_tfh_tfd *tfd = 410 iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr); 411 412 memset(tfd, 0, sizeof(*tfd)); 413 414 copy_size = sizeof(struct iwl_cmd_header_wide); 415 cmd_size = sizeof(struct iwl_cmd_header_wide); 416 417 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 418 cmddata[i] = cmd->data[i]; 419 cmdlen[i] = cmd->len[i]; 420 421 if (!cmd->len[i]) 422 continue; 423 424 /* need at least IWL_FIRST_TB_SIZE copied */ 425 if (copy_size < IWL_FIRST_TB_SIZE) { 426 int copy = IWL_FIRST_TB_SIZE - copy_size; 427 428 if (copy > cmdlen[i]) 429 copy = cmdlen[i]; 430 cmdlen[i] -= copy; 431 cmddata[i] += copy; 432 copy_size += copy; 433 } 434 435 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) { 436 had_nocopy = true; 437 if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) { 438 idx = -EINVAL; 439 goto free_dup_buf; 440 } 441 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) { 442 /* 443 * This is also a chunk that isn't copied 444 * to the static buffer so set had_nocopy. 445 */ 446 had_nocopy = true; 447 448 /* only allowed once */ 449 if (WARN_ON(dup_buf)) { 450 idx = -EINVAL; 451 goto free_dup_buf; 452 } 453 454 dup_buf = kmemdup(cmddata[i], cmdlen[i], 455 GFP_ATOMIC); 456 if (!dup_buf) 457 return -ENOMEM; 458 } else { 459 /* NOCOPY must not be followed by normal! */ 460 if (WARN_ON(had_nocopy)) { 461 idx = -EINVAL; 462 goto free_dup_buf; 463 } 464 copy_size += cmdlen[i]; 465 } 466 cmd_size += cmd->len[i]; 467 } 468 469 /* 470 * If any of the command structures end up being larger than the 471 * TFD_MAX_PAYLOAD_SIZE and they aren't dynamically allocated into 472 * separate TFDs, then we will need to increase the size of the buffers 473 */ 474 if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE, 475 "Command %s (%#x) is too large (%d bytes)\n", 476 iwl_get_cmd_string(trans, cmd->id), cmd->id, copy_size)) { 477 idx = -EINVAL; 478 goto free_dup_buf; 479 } 480 481 spin_lock_bh(&txq->lock); 482 483 if (iwl_queue_space(txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) { 484 spin_unlock_bh(&txq->lock); 485 486 IWL_ERR(trans, "No space in command queue\n"); 487 iwl_op_mode_cmd_queue_full(trans->op_mode); 488 idx = -ENOSPC; 489 goto free_dup_buf; 490 } 491 492 idx = get_cmd_index(txq, txq->write_ptr); 493 out_cmd = txq->entries[idx].cmd; 494 out_meta = &txq->entries[idx].meta; 495 496 /* re-initialize to NULL */ 497 memset(out_meta, 0, sizeof(*out_meta)); 498 if (cmd->flags & CMD_WANT_SKB) 499 out_meta->source = cmd; 500 501 /* set up the header */ 502 out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id); 503 out_cmd->hdr_wide.group_id = group_id; 504 out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id); 505 out_cmd->hdr_wide.length = 506 cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide)); 507 out_cmd->hdr_wide.reserved = 0; 508 out_cmd->hdr_wide.sequence = 509 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) | 510 INDEX_TO_SEQ(txq->write_ptr)); 511 512 cmd_pos = sizeof(struct iwl_cmd_header_wide); 513 copy_size = sizeof(struct iwl_cmd_header_wide); 514 515 /* and copy the data that needs to be copied */ 516 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 517 int copy; 518 519 if (!cmd->len[i]) 520 continue; 521 522 /* copy everything if not nocopy/dup */ 523 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY | 524 IWL_HCMD_DFL_DUP))) { 525 copy = cmd->len[i]; 526 527 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy); 528 cmd_pos += copy; 529 copy_size += copy; 530 continue; 531 } 532 533 /* 534 * Otherwise we need at least IWL_FIRST_TB_SIZE copied 535 * in total (for bi-directional DMA), but copy up to what 536 * we can fit into the payload for debug dump purposes. 537 */ 538 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]); 539 540 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy); 541 cmd_pos += copy; 542 543 /* However, treat copy_size the proper way, we need it below */ 544 if (copy_size < IWL_FIRST_TB_SIZE) { 545 copy = IWL_FIRST_TB_SIZE - copy_size; 546 547 if (copy > cmd->len[i]) 548 copy = cmd->len[i]; 549 copy_size += copy; 550 } 551 } 552 553 IWL_DEBUG_HC(trans, 554 "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n", 555 iwl_get_cmd_string(trans, cmd->id), group_id, 556 out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence), 557 cmd_size, txq->write_ptr, idx, trans_pcie->cmd_queue); 558 559 /* start the TFD with the minimum copy bytes */ 560 tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE); 561 memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size); 562 iwl_pcie_gen2_set_tb(trans, tfd, iwl_pcie_get_first_tb_dma(txq, idx), 563 tb0_size); 564 565 /* map first command fragment, if any remains */ 566 if (copy_size > tb0_size) { 567 phys_addr = dma_map_single(trans->dev, 568 ((u8 *)&out_cmd->hdr) + tb0_size, 569 copy_size - tb0_size, 570 DMA_TO_DEVICE); 571 if (dma_mapping_error(trans->dev, phys_addr)) { 572 idx = -ENOMEM; 573 iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd); 574 goto out; 575 } 576 iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, 577 copy_size - tb0_size); 578 } 579 580 /* map the remaining (adjusted) nocopy/dup fragments */ 581 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 582 const void *data = cmddata[i]; 583 584 if (!cmdlen[i]) 585 continue; 586 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY | 587 IWL_HCMD_DFL_DUP))) 588 continue; 589 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) 590 data = dup_buf; 591 phys_addr = dma_map_single(trans->dev, (void *)data, 592 cmdlen[i], DMA_TO_DEVICE); 593 if (dma_mapping_error(trans->dev, phys_addr)) { 594 idx = -ENOMEM; 595 iwl_pcie_gen2_tfd_unmap(trans, out_meta, tfd); 596 goto out; 597 } 598 iwl_pcie_gen2_set_tb(trans, tfd, phys_addr, cmdlen[i]); 599 } 600 601 BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE); 602 out_meta->flags = cmd->flags; 603 if (WARN_ON_ONCE(txq->entries[idx].free_buf)) 604 kzfree(txq->entries[idx].free_buf); 605 txq->entries[idx].free_buf = dup_buf; 606 607 trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide); 608 609 /* start timer if queue currently empty */ 610 if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) 611 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); 612 613 spin_lock_irqsave(&trans_pcie->reg_lock, flags); 614 if (!(cmd->flags & CMD_SEND_IN_IDLE) && 615 !trans_pcie->ref_cmd_in_flight) { 616 trans_pcie->ref_cmd_in_flight = true; 617 IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n"); 618 iwl_trans_ref(trans); 619 } 620 /* Increment and update queue's write index */ 621 txq->write_ptr = iwl_queue_inc_wrap(txq->write_ptr); 622 iwl_pcie_gen2_txq_inc_wr_ptr(trans, txq); 623 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags); 624 625 out: 626 spin_unlock_bh(&txq->lock); 627 free_dup_buf: 628 if (idx < 0) 629 kfree(dup_buf); 630 return idx; 631 } 632 633 #define HOST_COMPLETE_TIMEOUT (2 * HZ) 634 635 static int iwl_pcie_gen2_send_hcmd_sync(struct iwl_trans *trans, 636 struct iwl_host_cmd *cmd) 637 { 638 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 639 const char *cmd_str = iwl_get_cmd_string(trans, cmd->id); 640 struct iwl_txq *txq = trans_pcie->txq[trans_pcie->cmd_queue]; 641 int cmd_idx; 642 int ret; 643 644 IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str); 645 646 if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE, 647 &trans->status), 648 "Command %s: a command is already active!\n", cmd_str)) 649 return -EIO; 650 651 IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str); 652 653 if (pm_runtime_suspended(&trans_pcie->pci_dev->dev)) { 654 ret = wait_event_timeout(trans_pcie->d0i3_waitq, 655 pm_runtime_active(&trans_pcie->pci_dev->dev), 656 msecs_to_jiffies(IWL_TRANS_IDLE_TIMEOUT)); 657 if (!ret) { 658 IWL_ERR(trans, "Timeout exiting D0i3 before hcmd\n"); 659 return -ETIMEDOUT; 660 } 661 } 662 663 cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd); 664 if (cmd_idx < 0) { 665 ret = cmd_idx; 666 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 667 IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n", 668 cmd_str, ret); 669 return ret; 670 } 671 672 ret = wait_event_timeout(trans_pcie->wait_command_queue, 673 !test_bit(STATUS_SYNC_HCMD_ACTIVE, 674 &trans->status), 675 HOST_COMPLETE_TIMEOUT); 676 if (!ret) { 677 IWL_ERR(trans, "Error sending %s: time out after %dms.\n", 678 cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT)); 679 680 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n", 681 txq->read_ptr, txq->write_ptr); 682 683 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 684 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n", 685 cmd_str); 686 ret = -ETIMEDOUT; 687 688 iwl_force_nmi(trans); 689 iwl_trans_fw_error(trans); 690 691 goto cancel; 692 } 693 694 if (test_bit(STATUS_FW_ERROR, &trans->status)) { 695 IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str); 696 dump_stack(); 697 ret = -EIO; 698 goto cancel; 699 } 700 701 if (!(cmd->flags & CMD_SEND_IN_RFKILL) && 702 test_bit(STATUS_RFKILL, &trans->status)) { 703 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n"); 704 ret = -ERFKILL; 705 goto cancel; 706 } 707 708 if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) { 709 IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str); 710 ret = -EIO; 711 goto cancel; 712 } 713 714 return 0; 715 716 cancel: 717 if (cmd->flags & CMD_WANT_SKB) { 718 /* 719 * Cancel the CMD_WANT_SKB flag for the cmd in the 720 * TX cmd queue. Otherwise in case the cmd comes 721 * in later, it will possibly set an invalid 722 * address (cmd->meta.source). 723 */ 724 txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB; 725 } 726 727 if (cmd->resp_pkt) { 728 iwl_free_resp(cmd); 729 cmd->resp_pkt = NULL; 730 } 731 732 return ret; 733 } 734 735 int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans, 736 struct iwl_host_cmd *cmd) 737 { 738 if (!(cmd->flags & CMD_SEND_IN_RFKILL) && 739 test_bit(STATUS_RFKILL, &trans->status)) { 740 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n", 741 cmd->id); 742 return -ERFKILL; 743 } 744 745 if (cmd->flags & CMD_ASYNC) { 746 int ret; 747 748 /* An asynchronous command can not expect an SKB to be set. */ 749 if (WARN_ON(cmd->flags & CMD_WANT_SKB)) 750 return -EINVAL; 751 752 ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd); 753 if (ret < 0) { 754 IWL_ERR(trans, 755 "Error sending %s: enqueue_hcmd failed: %d\n", 756 iwl_get_cmd_string(trans, cmd->id), ret); 757 return ret; 758 } 759 return 0; 760 } 761 762 return iwl_pcie_gen2_send_hcmd_sync(trans, cmd); 763 } 764 765 /* 766 * iwl_pcie_gen2_txq_unmap - Unmap any remaining DMA mappings and free skb's 767 */ 768 void iwl_pcie_gen2_txq_unmap(struct iwl_trans *trans, int txq_id) 769 { 770 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 771 struct iwl_txq *txq = trans_pcie->txq[txq_id]; 772 773 spin_lock_bh(&txq->lock); 774 while (txq->write_ptr != txq->read_ptr) { 775 IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n", 776 txq_id, txq->read_ptr); 777 778 iwl_pcie_gen2_free_tfd(trans, txq); 779 txq->read_ptr = iwl_queue_inc_wrap(txq->read_ptr); 780 781 if (txq->read_ptr == txq->write_ptr) { 782 unsigned long flags; 783 784 spin_lock_irqsave(&trans_pcie->reg_lock, flags); 785 if (txq_id != trans_pcie->cmd_queue) { 786 IWL_DEBUG_RPM(trans, "Q %d - last tx freed\n", 787 txq->id); 788 iwl_trans_unref(trans); 789 } else if (trans_pcie->ref_cmd_in_flight) { 790 trans_pcie->ref_cmd_in_flight = false; 791 IWL_DEBUG_RPM(trans, 792 "clear ref_cmd_in_flight\n"); 793 iwl_trans_unref(trans); 794 } 795 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags); 796 } 797 } 798 spin_unlock_bh(&txq->lock); 799 800 /* just in case - this queue may have been stopped */ 801 iwl_wake_queue(trans, txq); 802 } 803 804 static void iwl_pcie_gen2_txq_free_memory(struct iwl_trans *trans, 805 struct iwl_txq *txq) 806 { 807 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 808 struct device *dev = trans->dev; 809 810 /* De-alloc circular buffer of TFDs */ 811 if (txq->tfds) { 812 dma_free_coherent(dev, 813 trans_pcie->tfd_size * TFD_QUEUE_SIZE_MAX, 814 txq->tfds, txq->dma_addr); 815 dma_free_coherent(dev, 816 sizeof(*txq->first_tb_bufs) * txq->n_window, 817 txq->first_tb_bufs, txq->first_tb_dma); 818 } 819 820 kfree(txq->entries); 821 iwl_pcie_free_dma_ptr(trans, &txq->bc_tbl); 822 kfree(txq); 823 } 824 825 /* 826 * iwl_pcie_txq_free - Deallocate DMA queue. 827 * @txq: Transmit queue to deallocate. 828 * 829 * Empty queue by removing and destroying all BD's. 830 * Free all buffers. 831 * 0-fill, but do not free "txq" descriptor structure. 832 */ 833 static void iwl_pcie_gen2_txq_free(struct iwl_trans *trans, int txq_id) 834 { 835 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 836 struct iwl_txq *txq = trans_pcie->txq[txq_id]; 837 int i; 838 839 if (WARN_ON(!txq)) 840 return; 841 842 iwl_pcie_gen2_txq_unmap(trans, txq_id); 843 844 /* De-alloc array of command/tx buffers */ 845 if (txq_id == trans_pcie->cmd_queue) 846 for (i = 0; i < txq->n_window; i++) { 847 kzfree(txq->entries[i].cmd); 848 kzfree(txq->entries[i].free_buf); 849 } 850 del_timer_sync(&txq->stuck_timer); 851 852 iwl_pcie_gen2_txq_free_memory(trans, txq); 853 854 trans_pcie->txq[txq_id] = NULL; 855 856 clear_bit(txq_id, trans_pcie->queue_used); 857 } 858 859 int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans, 860 struct iwl_tx_queue_cfg_cmd *cmd, 861 int cmd_id, 862 unsigned int timeout) 863 { 864 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 865 struct iwl_tx_queue_cfg_rsp *rsp; 866 struct iwl_txq *txq; 867 struct iwl_host_cmd hcmd = { 868 .id = cmd_id, 869 .len = { sizeof(*cmd) }, 870 .data = { cmd, }, 871 .flags = CMD_WANT_SKB, 872 }; 873 int ret, qid; 874 875 txq = kzalloc(sizeof(*txq), GFP_KERNEL); 876 if (!txq) 877 return -ENOMEM; 878 ret = iwl_pcie_alloc_dma_ptr(trans, &txq->bc_tbl, 879 sizeof(struct iwlagn_scd_bc_tbl)); 880 if (ret) { 881 IWL_ERR(trans, "Scheduler BC Table allocation failed\n"); 882 kfree(txq); 883 return -ENOMEM; 884 } 885 886 ret = iwl_pcie_txq_alloc(trans, txq, TFD_TX_CMD_SLOTS, false); 887 if (ret) { 888 IWL_ERR(trans, "Tx queue alloc failed\n"); 889 goto error; 890 } 891 ret = iwl_pcie_txq_init(trans, txq, TFD_TX_CMD_SLOTS, false); 892 if (ret) { 893 IWL_ERR(trans, "Tx queue init failed\n"); 894 goto error; 895 } 896 897 txq->wd_timeout = msecs_to_jiffies(timeout); 898 899 cmd->tfdq_addr = cpu_to_le64(txq->dma_addr); 900 cmd->byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma); 901 cmd->cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(TFD_QUEUE_SIZE_MAX)); 902 903 ret = iwl_trans_send_cmd(trans, &hcmd); 904 if (ret) 905 goto error; 906 907 if (WARN_ON(iwl_rx_packet_payload_len(hcmd.resp_pkt) != sizeof(*rsp))) { 908 ret = -EINVAL; 909 goto error_free_resp; 910 } 911 912 rsp = (void *)hcmd.resp_pkt->data; 913 qid = le16_to_cpu(rsp->queue_number); 914 915 if (qid > ARRAY_SIZE(trans_pcie->txq)) { 916 WARN_ONCE(1, "queue index %d unsupported", qid); 917 ret = -EIO; 918 goto error_free_resp; 919 } 920 921 if (test_and_set_bit(qid, trans_pcie->queue_used)) { 922 WARN_ONCE(1, "queue %d already used", qid); 923 ret = -EIO; 924 goto error_free_resp; 925 } 926 927 txq->id = qid; 928 trans_pcie->txq[qid] = txq; 929 930 /* Place first TFD at index corresponding to start sequence number */ 931 txq->read_ptr = le16_to_cpu(rsp->write_pointer); 932 txq->write_ptr = le16_to_cpu(rsp->write_pointer); 933 iwl_write_direct32(trans, HBUS_TARG_WRPTR, 934 (txq->write_ptr) | (qid << 16)); 935 IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid); 936 937 iwl_free_resp(&hcmd); 938 return qid; 939 940 error_free_resp: 941 iwl_free_resp(&hcmd); 942 error: 943 iwl_pcie_gen2_txq_free_memory(trans, txq); 944 return ret; 945 } 946 947 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue) 948 { 949 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 950 951 /* 952 * Upon HW Rfkill - we stop the device, and then stop the queues 953 * in the op_mode. Just for the sake of the simplicity of the op_mode, 954 * allow the op_mode to call txq_disable after it already called 955 * stop_device. 956 */ 957 if (!test_and_clear_bit(queue, trans_pcie->queue_used)) { 958 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status), 959 "queue %d not used", queue); 960 return; 961 } 962 963 iwl_pcie_gen2_txq_unmap(trans, queue); 964 965 IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue); 966 } 967 968 void iwl_pcie_gen2_tx_free(struct iwl_trans *trans) 969 { 970 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 971 int i; 972 973 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used)); 974 975 /* Free all TX queues */ 976 for (i = 0; i < ARRAY_SIZE(trans_pcie->txq); i++) { 977 if (!trans_pcie->txq[i]) 978 continue; 979 980 iwl_pcie_gen2_txq_free(trans, i); 981 } 982 } 983 984 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans) 985 { 986 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 987 struct iwl_txq *cmd_queue; 988 int txq_id = trans_pcie->cmd_queue, ret; 989 990 /* alloc and init the command queue */ 991 if (!trans_pcie->txq[txq_id]) { 992 cmd_queue = kzalloc(sizeof(*cmd_queue), GFP_KERNEL); 993 if (!cmd_queue) { 994 IWL_ERR(trans, "Not enough memory for command queue\n"); 995 return -ENOMEM; 996 } 997 trans_pcie->txq[txq_id] = cmd_queue; 998 ret = iwl_pcie_txq_alloc(trans, cmd_queue, TFD_CMD_SLOTS, true); 999 if (ret) { 1000 IWL_ERR(trans, "Tx %d queue init failed\n", txq_id); 1001 goto error; 1002 } 1003 } else { 1004 cmd_queue = trans_pcie->txq[txq_id]; 1005 } 1006 1007 ret = iwl_pcie_txq_init(trans, cmd_queue, TFD_CMD_SLOTS, true); 1008 if (ret) { 1009 IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id); 1010 goto error; 1011 } 1012 trans_pcie->txq[txq_id]->id = txq_id; 1013 set_bit(txq_id, trans_pcie->queue_used); 1014 1015 return 0; 1016 1017 error: 1018 iwl_pcie_gen2_tx_free(trans); 1019 return ret; 1020 } 1021 1022