1 /****************************************************************************** 2 * 3 * Copyright(c) 2003 - 2014 Intel Corporation. All rights reserved. 4 * Copyright(c) 2013 - 2015 Intel Mobile Communications GmbH 5 * Copyright(c) 2016 Intel Deutschland GmbH 6 * 7 * Portions of this file are derived from the ipw3945 project, as well 8 * as portions of the ieee80211 subsystem header files. 9 * 10 * This program is free software; you can redistribute it and/or modify it 11 * under the terms of version 2 of the GNU General Public License as 12 * published by the Free Software Foundation. 13 * 14 * This program is distributed in the hope that it will be useful, but WITHOUT 15 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 16 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 17 * more details. 18 * 19 * You should have received a copy of the GNU General Public License along with 20 * this program; if not, write to the Free Software Foundation, Inc., 21 * 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA 22 * 23 * The full GNU General Public License is included in this distribution in the 24 * file called LICENSE. 25 * 26 * Contact Information: 27 * Intel Linux Wireless <linuxwifi@intel.com> 28 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 29 * 30 *****************************************************************************/ 31 #include <linux/etherdevice.h> 32 #include <linux/ieee80211.h> 33 #include <linux/slab.h> 34 #include <linux/sched.h> 35 #include <net/ip6_checksum.h> 36 #include <net/tso.h> 37 38 #include "iwl-debug.h" 39 #include "iwl-csr.h" 40 #include "iwl-prph.h" 41 #include "iwl-io.h" 42 #include "iwl-scd.h" 43 #include "iwl-op-mode.h" 44 #include "internal.h" 45 /* FIXME: need to abstract out TX command (once we know what it looks like) */ 46 #include "dvm/commands.h" 47 48 #define IWL_TX_CRC_SIZE 4 49 #define IWL_TX_DELIMITER_SIZE 4 50 51 /*************** DMA-QUEUE-GENERAL-FUNCTIONS ***** 52 * DMA services 53 * 54 * Theory of operation 55 * 56 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer 57 * of buffer descriptors, each of which points to one or more data buffers for 58 * the device to read from or fill. Driver and device exchange status of each 59 * queue via "read" and "write" pointers. Driver keeps minimum of 2 empty 60 * entries in each circular buffer, to protect against confusing empty and full 61 * queue states. 62 * 63 * The device reads or writes the data in the queues via the device's several 64 * DMA/FIFO channels. Each queue is mapped to a single DMA channel. 65 * 66 * For Tx queue, there are low mark and high mark limits. If, after queuing 67 * the packet for Tx, free space become < low mark, Tx queue stopped. When 68 * reclaiming packets (on 'tx done IRQ), if free space become > high mark, 69 * Tx queue resumed. 70 * 71 ***************************************************/ 72 static int iwl_queue_space(const struct iwl_queue *q) 73 { 74 unsigned int max; 75 unsigned int used; 76 77 /* 78 * To avoid ambiguity between empty and completely full queues, there 79 * should always be less than TFD_QUEUE_SIZE_MAX elements in the queue. 80 * If q->n_window is smaller than TFD_QUEUE_SIZE_MAX, there is no need 81 * to reserve any queue entries for this purpose. 82 */ 83 if (q->n_window < TFD_QUEUE_SIZE_MAX) 84 max = q->n_window; 85 else 86 max = TFD_QUEUE_SIZE_MAX - 1; 87 88 /* 89 * TFD_QUEUE_SIZE_MAX is a power of 2, so the following is equivalent to 90 * modulo by TFD_QUEUE_SIZE_MAX and is well defined. 91 */ 92 used = (q->write_ptr - q->read_ptr) & (TFD_QUEUE_SIZE_MAX - 1); 93 94 if (WARN_ON(used > max)) 95 return 0; 96 97 return max - used; 98 } 99 100 /* 101 * iwl_queue_init - Initialize queue's high/low-water and read/write indexes 102 */ 103 static int iwl_queue_init(struct iwl_queue *q, int slots_num, u32 id) 104 { 105 q->n_window = slots_num; 106 q->id = id; 107 108 /* slots_num must be power-of-two size, otherwise 109 * get_cmd_index is broken. */ 110 if (WARN_ON(!is_power_of_2(slots_num))) 111 return -EINVAL; 112 113 q->low_mark = q->n_window / 4; 114 if (q->low_mark < 4) 115 q->low_mark = 4; 116 117 q->high_mark = q->n_window / 8; 118 if (q->high_mark < 2) 119 q->high_mark = 2; 120 121 q->write_ptr = 0; 122 q->read_ptr = 0; 123 124 return 0; 125 } 126 127 static int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans, 128 struct iwl_dma_ptr *ptr, size_t size) 129 { 130 if (WARN_ON(ptr->addr)) 131 return -EINVAL; 132 133 ptr->addr = dma_alloc_coherent(trans->dev, size, 134 &ptr->dma, GFP_KERNEL); 135 if (!ptr->addr) 136 return -ENOMEM; 137 ptr->size = size; 138 return 0; 139 } 140 141 static void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, 142 struct iwl_dma_ptr *ptr) 143 { 144 if (unlikely(!ptr->addr)) 145 return; 146 147 dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma); 148 memset(ptr, 0, sizeof(*ptr)); 149 } 150 151 static void iwl_pcie_txq_stuck_timer(unsigned long data) 152 { 153 struct iwl_txq *txq = (void *)data; 154 struct iwl_trans_pcie *trans_pcie = txq->trans_pcie; 155 struct iwl_trans *trans = iwl_trans_pcie_get_trans(trans_pcie); 156 u32 scd_sram_addr = trans_pcie->scd_base_addr + 157 SCD_TX_STTS_QUEUE_OFFSET(txq->q.id); 158 u8 buf[16]; 159 int i; 160 161 spin_lock(&txq->lock); 162 /* check if triggered erroneously */ 163 if (txq->q.read_ptr == txq->q.write_ptr) { 164 spin_unlock(&txq->lock); 165 return; 166 } 167 spin_unlock(&txq->lock); 168 169 IWL_ERR(trans, "Queue %d stuck for %u ms.\n", txq->q.id, 170 jiffies_to_msecs(txq->wd_timeout)); 171 IWL_ERR(trans, "Current SW read_ptr %d write_ptr %d\n", 172 txq->q.read_ptr, txq->q.write_ptr); 173 174 iwl_trans_read_mem_bytes(trans, scd_sram_addr, buf, sizeof(buf)); 175 176 iwl_print_hex_error(trans, buf, sizeof(buf)); 177 178 for (i = 0; i < FH_TCSR_CHNL_NUM; i++) 179 IWL_ERR(trans, "FH TRBs(%d) = 0x%08x\n", i, 180 iwl_read_direct32(trans, FH_TX_TRB_REG(i))); 181 182 for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) { 183 u32 status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(i)); 184 u8 fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7; 185 bool active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE)); 186 u32 tbl_dw = 187 iwl_trans_read_mem32(trans, 188 trans_pcie->scd_base_addr + 189 SCD_TRANS_TBL_OFFSET_QUEUE(i)); 190 191 if (i & 0x1) 192 tbl_dw = (tbl_dw & 0xFFFF0000) >> 16; 193 else 194 tbl_dw = tbl_dw & 0x0000FFFF; 195 196 IWL_ERR(trans, 197 "Q %d is %sactive and mapped to fifo %d ra_tid 0x%04x [%d,%d]\n", 198 i, active ? "" : "in", fifo, tbl_dw, 199 iwl_read_prph(trans, SCD_QUEUE_RDPTR(i)) & 200 (TFD_QUEUE_SIZE_MAX - 1), 201 iwl_read_prph(trans, SCD_QUEUE_WRPTR(i))); 202 } 203 204 iwl_force_nmi(trans); 205 } 206 207 /* 208 * iwl_pcie_txq_update_byte_cnt_tbl - Set up entry in Tx byte-count array 209 */ 210 static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans, 211 struct iwl_txq *txq, u16 byte_cnt) 212 { 213 struct iwlagn_scd_bc_tbl *scd_bc_tbl; 214 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 215 int write_ptr = txq->q.write_ptr; 216 int txq_id = txq->q.id; 217 u8 sec_ctl = 0; 218 u8 sta_id = 0; 219 u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE; 220 __le16 bc_ent; 221 struct iwl_tx_cmd *tx_cmd = 222 (void *) txq->entries[txq->q.write_ptr].cmd->payload; 223 224 scd_bc_tbl = trans_pcie->scd_bc_tbls.addr; 225 226 sta_id = tx_cmd->sta_id; 227 sec_ctl = tx_cmd->sec_ctl; 228 229 switch (sec_ctl & TX_CMD_SEC_MSK) { 230 case TX_CMD_SEC_CCM: 231 len += IEEE80211_CCMP_MIC_LEN; 232 break; 233 case TX_CMD_SEC_TKIP: 234 len += IEEE80211_TKIP_ICV_LEN; 235 break; 236 case TX_CMD_SEC_WEP: 237 len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN; 238 break; 239 } 240 241 if (trans_pcie->bc_table_dword) 242 len = DIV_ROUND_UP(len, 4); 243 244 if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX)) 245 return; 246 247 bc_ent = cpu_to_le16(len | (sta_id << 12)); 248 249 scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent; 250 251 if (write_ptr < TFD_QUEUE_SIZE_BC_DUP) 252 scd_bc_tbl[txq_id]. 253 tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = bc_ent; 254 } 255 256 static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans, 257 struct iwl_txq *txq) 258 { 259 struct iwl_trans_pcie *trans_pcie = 260 IWL_TRANS_GET_PCIE_TRANS(trans); 261 struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->scd_bc_tbls.addr; 262 int txq_id = txq->q.id; 263 int read_ptr = txq->q.read_ptr; 264 u8 sta_id = 0; 265 __le16 bc_ent; 266 struct iwl_tx_cmd *tx_cmd = 267 (void *)txq->entries[txq->q.read_ptr].cmd->payload; 268 269 WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX); 270 271 if (txq_id != trans_pcie->cmd_queue) 272 sta_id = tx_cmd->sta_id; 273 274 bc_ent = cpu_to_le16(1 | (sta_id << 12)); 275 scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent; 276 277 if (read_ptr < TFD_QUEUE_SIZE_BC_DUP) 278 scd_bc_tbl[txq_id]. 279 tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = bc_ent; 280 } 281 282 /* 283 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware 284 */ 285 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, 286 struct iwl_txq *txq) 287 { 288 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 289 u32 reg = 0; 290 int txq_id = txq->q.id; 291 292 lockdep_assert_held(&txq->lock); 293 294 /* 295 * explicitly wake up the NIC if: 296 * 1. shadow registers aren't enabled 297 * 2. NIC is woken up for CMD regardless of shadow outside this function 298 * 3. there is a chance that the NIC is asleep 299 */ 300 if (!trans->cfg->base_params->shadow_reg_enable && 301 txq_id != trans_pcie->cmd_queue && 302 test_bit(STATUS_TPOWER_PMI, &trans->status)) { 303 /* 304 * wake up nic if it's powered down ... 305 * uCode will wake up, and interrupt us again, so next 306 * time we'll skip this part. 307 */ 308 reg = iwl_read32(trans, CSR_UCODE_DRV_GP1); 309 310 if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) { 311 IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n", 312 txq_id, reg); 313 iwl_set_bit(trans, CSR_GP_CNTRL, 314 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); 315 txq->need_update = true; 316 return; 317 } 318 } 319 320 /* 321 * if not in power-save mode, uCode will never sleep when we're 322 * trying to tx (during RFKILL, we're not trying to tx). 323 */ 324 IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->q.write_ptr); 325 if (!txq->block) 326 iwl_write32(trans, HBUS_TARG_WRPTR, 327 txq->q.write_ptr | (txq_id << 8)); 328 } 329 330 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans) 331 { 332 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 333 int i; 334 335 for (i = 0; i < trans->cfg->base_params->num_of_queues; i++) { 336 struct iwl_txq *txq = &trans_pcie->txq[i]; 337 338 spin_lock_bh(&txq->lock); 339 if (trans_pcie->txq[i].need_update) { 340 iwl_pcie_txq_inc_wr_ptr(trans, txq); 341 trans_pcie->txq[i].need_update = false; 342 } 343 spin_unlock_bh(&txq->lock); 344 } 345 } 346 347 static inline dma_addr_t iwl_pcie_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx) 348 { 349 struct iwl_tfd_tb *tb = &tfd->tbs[idx]; 350 351 dma_addr_t addr = get_unaligned_le32(&tb->lo); 352 if (sizeof(dma_addr_t) > sizeof(u32)) 353 addr |= 354 ((dma_addr_t)(le16_to_cpu(tb->hi_n_len) & 0xF) << 16) << 16; 355 356 return addr; 357 } 358 359 static inline void iwl_pcie_tfd_set_tb(struct iwl_tfd *tfd, u8 idx, 360 dma_addr_t addr, u16 len) 361 { 362 struct iwl_tfd_tb *tb = &tfd->tbs[idx]; 363 u16 hi_n_len = len << 4; 364 365 put_unaligned_le32(addr, &tb->lo); 366 if (sizeof(dma_addr_t) > sizeof(u32)) 367 hi_n_len |= ((addr >> 16) >> 16) & 0xF; 368 369 tb->hi_n_len = cpu_to_le16(hi_n_len); 370 371 tfd->num_tbs = idx + 1; 372 } 373 374 static inline u8 iwl_pcie_tfd_get_num_tbs(struct iwl_tfd *tfd) 375 { 376 return tfd->num_tbs & 0x1f; 377 } 378 379 static void iwl_pcie_tfd_unmap(struct iwl_trans *trans, 380 struct iwl_cmd_meta *meta, 381 struct iwl_tfd *tfd) 382 { 383 int i; 384 int num_tbs; 385 386 /* Sanity check on number of chunks */ 387 num_tbs = iwl_pcie_tfd_get_num_tbs(tfd); 388 389 if (num_tbs >= IWL_NUM_OF_TBS) { 390 IWL_ERR(trans, "Too many chunks: %i\n", num_tbs); 391 /* @todo issue fatal error, it is quite serious situation */ 392 return; 393 } 394 395 /* first TB is never freed - it's the scratchbuf data */ 396 397 for (i = 1; i < num_tbs; i++) { 398 if (meta->flags & BIT(i + CMD_TB_BITMAP_POS)) 399 dma_unmap_page(trans->dev, 400 iwl_pcie_tfd_tb_get_addr(tfd, i), 401 iwl_pcie_tfd_tb_get_len(tfd, i), 402 DMA_TO_DEVICE); 403 else 404 dma_unmap_single(trans->dev, 405 iwl_pcie_tfd_tb_get_addr(tfd, i), 406 iwl_pcie_tfd_tb_get_len(tfd, i), 407 DMA_TO_DEVICE); 408 } 409 tfd->num_tbs = 0; 410 } 411 412 /* 413 * iwl_pcie_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr] 414 * @trans - transport private data 415 * @txq - tx queue 416 * @dma_dir - the direction of the DMA mapping 417 * 418 * Does NOT advance any TFD circular buffer read/write indexes 419 * Does NOT free the TFD itself (which is within circular buffer) 420 */ 421 static void iwl_pcie_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq) 422 { 423 struct iwl_tfd *tfd_tmp = txq->tfds; 424 425 /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and 426 * idx is bounded by n_window 427 */ 428 int rd_ptr = txq->q.read_ptr; 429 int idx = get_cmd_index(&txq->q, rd_ptr); 430 431 lockdep_assert_held(&txq->lock); 432 433 /* We have only q->n_window txq->entries, but we use 434 * TFD_QUEUE_SIZE_MAX tfds 435 */ 436 iwl_pcie_tfd_unmap(trans, &txq->entries[idx].meta, &tfd_tmp[rd_ptr]); 437 438 /* free SKB */ 439 if (txq->entries) { 440 struct sk_buff *skb; 441 442 skb = txq->entries[idx].skb; 443 444 /* Can be called from irqs-disabled context 445 * If skb is not NULL, it means that the whole queue is being 446 * freed and that the queue is not empty - free the skb 447 */ 448 if (skb) { 449 iwl_op_mode_free_skb(trans->op_mode, skb); 450 txq->entries[idx].skb = NULL; 451 } 452 } 453 } 454 455 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq, 456 dma_addr_t addr, u16 len, bool reset) 457 { 458 struct iwl_queue *q; 459 struct iwl_tfd *tfd, *tfd_tmp; 460 u32 num_tbs; 461 462 q = &txq->q; 463 tfd_tmp = txq->tfds; 464 tfd = &tfd_tmp[q->write_ptr]; 465 466 if (reset) 467 memset(tfd, 0, sizeof(*tfd)); 468 469 num_tbs = iwl_pcie_tfd_get_num_tbs(tfd); 470 471 /* Each TFD can point to a maximum 20 Tx buffers */ 472 if (num_tbs >= IWL_NUM_OF_TBS) { 473 IWL_ERR(trans, "Error can not send more than %d chunks\n", 474 IWL_NUM_OF_TBS); 475 return -EINVAL; 476 } 477 478 if (WARN(addr & ~IWL_TX_DMA_MASK, 479 "Unaligned address = %llx\n", (unsigned long long)addr)) 480 return -EINVAL; 481 482 iwl_pcie_tfd_set_tb(tfd, num_tbs, addr, len); 483 484 return num_tbs; 485 } 486 487 static int iwl_pcie_txq_alloc(struct iwl_trans *trans, 488 struct iwl_txq *txq, int slots_num, 489 u32 txq_id) 490 { 491 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 492 size_t tfd_sz = sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX; 493 size_t scratchbuf_sz; 494 int i; 495 496 if (WARN_ON(txq->entries || txq->tfds)) 497 return -EINVAL; 498 499 setup_timer(&txq->stuck_timer, iwl_pcie_txq_stuck_timer, 500 (unsigned long)txq); 501 txq->trans_pcie = trans_pcie; 502 503 txq->q.n_window = slots_num; 504 505 txq->entries = kcalloc(slots_num, 506 sizeof(struct iwl_pcie_txq_entry), 507 GFP_KERNEL); 508 509 if (!txq->entries) 510 goto error; 511 512 if (txq_id == trans_pcie->cmd_queue) 513 for (i = 0; i < slots_num; i++) { 514 txq->entries[i].cmd = 515 kmalloc(sizeof(struct iwl_device_cmd), 516 GFP_KERNEL); 517 if (!txq->entries[i].cmd) 518 goto error; 519 } 520 521 /* Circular buffer of transmit frame descriptors (TFDs), 522 * shared with device */ 523 txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz, 524 &txq->q.dma_addr, GFP_KERNEL); 525 if (!txq->tfds) 526 goto error; 527 528 BUILD_BUG_ON(IWL_HCMD_SCRATCHBUF_SIZE != sizeof(*txq->scratchbufs)); 529 BUILD_BUG_ON(offsetof(struct iwl_pcie_txq_scratch_buf, scratch) != 530 sizeof(struct iwl_cmd_header) + 531 offsetof(struct iwl_tx_cmd, scratch)); 532 533 scratchbuf_sz = sizeof(*txq->scratchbufs) * slots_num; 534 535 txq->scratchbufs = dma_alloc_coherent(trans->dev, scratchbuf_sz, 536 &txq->scratchbufs_dma, 537 GFP_KERNEL); 538 if (!txq->scratchbufs) 539 goto err_free_tfds; 540 541 txq->q.id = txq_id; 542 543 return 0; 544 err_free_tfds: 545 dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->q.dma_addr); 546 error: 547 if (txq->entries && txq_id == trans_pcie->cmd_queue) 548 for (i = 0; i < slots_num; i++) 549 kfree(txq->entries[i].cmd); 550 kfree(txq->entries); 551 txq->entries = NULL; 552 553 return -ENOMEM; 554 555 } 556 557 static int iwl_pcie_txq_init(struct iwl_trans *trans, struct iwl_txq *txq, 558 int slots_num, u32 txq_id) 559 { 560 int ret; 561 562 txq->need_update = false; 563 564 /* TFD_QUEUE_SIZE_MAX must be power-of-two size, otherwise 565 * iwl_queue_inc_wrap and iwl_queue_dec_wrap are broken. */ 566 BUILD_BUG_ON(TFD_QUEUE_SIZE_MAX & (TFD_QUEUE_SIZE_MAX - 1)); 567 568 /* Initialize queue's high/low-water marks, and head/tail indexes */ 569 ret = iwl_queue_init(&txq->q, slots_num, txq_id); 570 if (ret) 571 return ret; 572 573 spin_lock_init(&txq->lock); 574 __skb_queue_head_init(&txq->overflow_q); 575 576 /* 577 * Tell nic where to find circular buffer of Tx Frame Descriptors for 578 * given Tx queue, and enable the DMA channel used for that queue. 579 * Circular buffer (TFD queue in DRAM) physical base address */ 580 iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id), 581 txq->q.dma_addr >> 8); 582 583 return 0; 584 } 585 586 static void iwl_pcie_free_tso_page(struct sk_buff *skb) 587 { 588 struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); 589 590 if (info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA]) { 591 struct page *page = 592 info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA]; 593 594 __free_page(page); 595 info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA] = NULL; 596 } 597 } 598 599 /* 600 * iwl_pcie_txq_unmap - Unmap any remaining DMA mappings and free skb's 601 */ 602 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id) 603 { 604 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 605 struct iwl_txq *txq = &trans_pcie->txq[txq_id]; 606 struct iwl_queue *q = &txq->q; 607 608 spin_lock_bh(&txq->lock); 609 while (q->write_ptr != q->read_ptr) { 610 IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n", 611 txq_id, q->read_ptr); 612 613 if (txq_id != trans_pcie->cmd_queue) { 614 struct sk_buff *skb = txq->entries[q->read_ptr].skb; 615 616 if (WARN_ON_ONCE(!skb)) 617 continue; 618 619 iwl_pcie_free_tso_page(skb); 620 } 621 iwl_pcie_txq_free_tfd(trans, txq); 622 q->read_ptr = iwl_queue_inc_wrap(q->read_ptr); 623 } 624 txq->active = false; 625 626 while (!skb_queue_empty(&txq->overflow_q)) { 627 struct sk_buff *skb = __skb_dequeue(&txq->overflow_q); 628 629 iwl_op_mode_free_skb(trans->op_mode, skb); 630 } 631 632 spin_unlock_bh(&txq->lock); 633 634 /* just in case - this queue may have been stopped */ 635 iwl_wake_queue(trans, txq); 636 } 637 638 /* 639 * iwl_pcie_txq_free - Deallocate DMA queue. 640 * @txq: Transmit queue to deallocate. 641 * 642 * Empty queue by removing and destroying all BD's. 643 * Free all buffers. 644 * 0-fill, but do not free "txq" descriptor structure. 645 */ 646 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id) 647 { 648 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 649 struct iwl_txq *txq = &trans_pcie->txq[txq_id]; 650 struct device *dev = trans->dev; 651 int i; 652 653 if (WARN_ON(!txq)) 654 return; 655 656 iwl_pcie_txq_unmap(trans, txq_id); 657 658 /* De-alloc array of command/tx buffers */ 659 if (txq_id == trans_pcie->cmd_queue) 660 for (i = 0; i < txq->q.n_window; i++) { 661 kzfree(txq->entries[i].cmd); 662 kzfree(txq->entries[i].free_buf); 663 } 664 665 /* De-alloc circular buffer of TFDs */ 666 if (txq->tfds) { 667 dma_free_coherent(dev, 668 sizeof(struct iwl_tfd) * TFD_QUEUE_SIZE_MAX, 669 txq->tfds, txq->q.dma_addr); 670 txq->q.dma_addr = 0; 671 txq->tfds = NULL; 672 673 dma_free_coherent(dev, 674 sizeof(*txq->scratchbufs) * txq->q.n_window, 675 txq->scratchbufs, txq->scratchbufs_dma); 676 } 677 678 kfree(txq->entries); 679 txq->entries = NULL; 680 681 del_timer_sync(&txq->stuck_timer); 682 683 /* 0-fill queue descriptor structure */ 684 memset(txq, 0, sizeof(*txq)); 685 } 686 687 void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr) 688 { 689 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 690 int nq = trans->cfg->base_params->num_of_queues; 691 int chan; 692 u32 reg_val; 693 int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) - 694 SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32); 695 696 /* make sure all queue are not stopped/used */ 697 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped)); 698 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used)); 699 700 trans_pcie->scd_base_addr = 701 iwl_read_prph(trans, SCD_SRAM_BASE_ADDR); 702 703 WARN_ON(scd_base_addr != 0 && 704 scd_base_addr != trans_pcie->scd_base_addr); 705 706 /* reset context data, TX status and translation data */ 707 iwl_trans_write_mem(trans, trans_pcie->scd_base_addr + 708 SCD_CONTEXT_MEM_LOWER_BOUND, 709 NULL, clear_dwords); 710 711 iwl_write_prph(trans, SCD_DRAM_BASE_ADDR, 712 trans_pcie->scd_bc_tbls.dma >> 10); 713 714 /* The chain extension of the SCD doesn't work well. This feature is 715 * enabled by default by the HW, so we need to disable it manually. 716 */ 717 if (trans->cfg->base_params->scd_chain_ext_wa) 718 iwl_write_prph(trans, SCD_CHAINEXT_EN, 0); 719 720 iwl_trans_ac_txq_enable(trans, trans_pcie->cmd_queue, 721 trans_pcie->cmd_fifo, 722 trans_pcie->cmd_q_wdg_timeout); 723 724 /* Activate all Tx DMA/FIFO channels */ 725 iwl_scd_activate_fifos(trans); 726 727 /* Enable DMA channel */ 728 for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++) 729 iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan), 730 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE | 731 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE); 732 733 /* Update FH chicken bits */ 734 reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG); 735 iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG, 736 reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN); 737 738 /* Enable L1-Active */ 739 if (trans->cfg->device_family != IWL_DEVICE_FAMILY_8000) 740 iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG, 741 APMG_PCIDEV_STT_VAL_L1_ACT_DIS); 742 } 743 744 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans) 745 { 746 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 747 int txq_id; 748 749 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues; 750 txq_id++) { 751 struct iwl_txq *txq = &trans_pcie->txq[txq_id]; 752 753 iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(txq_id), 754 txq->q.dma_addr >> 8); 755 iwl_pcie_txq_unmap(trans, txq_id); 756 txq->q.read_ptr = 0; 757 txq->q.write_ptr = 0; 758 } 759 760 /* Tell NIC where to find the "keep warm" buffer */ 761 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG, 762 trans_pcie->kw.dma >> 4); 763 764 /* 765 * Send 0 as the scd_base_addr since the device may have be reset 766 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will 767 * contain garbage. 768 */ 769 iwl_pcie_tx_start(trans, 0); 770 } 771 772 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans) 773 { 774 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 775 unsigned long flags; 776 int ch, ret; 777 u32 mask = 0; 778 779 spin_lock(&trans_pcie->irq_lock); 780 781 if (!iwl_trans_grab_nic_access(trans, &flags)) 782 goto out; 783 784 /* Stop each Tx DMA channel */ 785 for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) { 786 iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0); 787 mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch); 788 } 789 790 /* Wait for DMA channels to be idle */ 791 ret = iwl_poll_bit(trans, FH_TSSR_TX_STATUS_REG, mask, mask, 5000); 792 if (ret < 0) 793 IWL_ERR(trans, 794 "Failing on timeout while stopping DMA channel %d [0x%08x]\n", 795 ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG)); 796 797 iwl_trans_release_nic_access(trans, &flags); 798 799 out: 800 spin_unlock(&trans_pcie->irq_lock); 801 } 802 803 /* 804 * iwl_pcie_tx_stop - Stop all Tx DMA channels 805 */ 806 int iwl_pcie_tx_stop(struct iwl_trans *trans) 807 { 808 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 809 int txq_id; 810 811 /* Turn off all Tx DMA fifos */ 812 iwl_scd_deactivate_fifos(trans); 813 814 /* Turn off all Tx DMA channels */ 815 iwl_pcie_tx_stop_fh(trans); 816 817 /* 818 * This function can be called before the op_mode disabled the 819 * queues. This happens when we have an rfkill interrupt. 820 * Since we stop Tx altogether - mark the queues as stopped. 821 */ 822 memset(trans_pcie->queue_stopped, 0, sizeof(trans_pcie->queue_stopped)); 823 memset(trans_pcie->queue_used, 0, sizeof(trans_pcie->queue_used)); 824 825 /* This can happen: start_hw, stop_device */ 826 if (!trans_pcie->txq) 827 return 0; 828 829 /* Unmap DMA from host system and free skb's */ 830 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues; 831 txq_id++) 832 iwl_pcie_txq_unmap(trans, txq_id); 833 834 return 0; 835 } 836 837 /* 838 * iwl_trans_tx_free - Free TXQ Context 839 * 840 * Destroy all TX DMA queues and structures 841 */ 842 void iwl_pcie_tx_free(struct iwl_trans *trans) 843 { 844 int txq_id; 845 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 846 847 /* Tx queues */ 848 if (trans_pcie->txq) { 849 for (txq_id = 0; 850 txq_id < trans->cfg->base_params->num_of_queues; txq_id++) 851 iwl_pcie_txq_free(trans, txq_id); 852 } 853 854 kfree(trans_pcie->txq); 855 trans_pcie->txq = NULL; 856 857 iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw); 858 859 iwl_pcie_free_dma_ptr(trans, &trans_pcie->scd_bc_tbls); 860 } 861 862 /* 863 * iwl_pcie_tx_alloc - allocate TX context 864 * Allocate all Tx DMA structures and initialize them 865 */ 866 static int iwl_pcie_tx_alloc(struct iwl_trans *trans) 867 { 868 int ret; 869 int txq_id, slots_num; 870 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 871 872 u16 scd_bc_tbls_size = trans->cfg->base_params->num_of_queues * 873 sizeof(struct iwlagn_scd_bc_tbl); 874 875 /*It is not allowed to alloc twice, so warn when this happens. 876 * We cannot rely on the previous allocation, so free and fail */ 877 if (WARN_ON(trans_pcie->txq)) { 878 ret = -EINVAL; 879 goto error; 880 } 881 882 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->scd_bc_tbls, 883 scd_bc_tbls_size); 884 if (ret) { 885 IWL_ERR(trans, "Scheduler BC Table allocation failed\n"); 886 goto error; 887 } 888 889 /* Alloc keep-warm buffer */ 890 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE); 891 if (ret) { 892 IWL_ERR(trans, "Keep Warm allocation failed\n"); 893 goto error; 894 } 895 896 trans_pcie->txq = kcalloc(trans->cfg->base_params->num_of_queues, 897 sizeof(struct iwl_txq), GFP_KERNEL); 898 if (!trans_pcie->txq) { 899 IWL_ERR(trans, "Not enough memory for txq\n"); 900 ret = -ENOMEM; 901 goto error; 902 } 903 904 /* Alloc and init all Tx queues, including the command queue (#4/#9) */ 905 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues; 906 txq_id++) { 907 slots_num = (txq_id == trans_pcie->cmd_queue) ? 908 TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS; 909 ret = iwl_pcie_txq_alloc(trans, &trans_pcie->txq[txq_id], 910 slots_num, txq_id); 911 if (ret) { 912 IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id); 913 goto error; 914 } 915 } 916 917 return 0; 918 919 error: 920 iwl_pcie_tx_free(trans); 921 922 return ret; 923 } 924 int iwl_pcie_tx_init(struct iwl_trans *trans) 925 { 926 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 927 int ret; 928 int txq_id, slots_num; 929 bool alloc = false; 930 931 if (!trans_pcie->txq) { 932 ret = iwl_pcie_tx_alloc(trans); 933 if (ret) 934 goto error; 935 alloc = true; 936 } 937 938 spin_lock(&trans_pcie->irq_lock); 939 940 /* Turn off all Tx DMA fifos */ 941 iwl_scd_deactivate_fifos(trans); 942 943 /* Tell NIC where to find the "keep warm" buffer */ 944 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG, 945 trans_pcie->kw.dma >> 4); 946 947 spin_unlock(&trans_pcie->irq_lock); 948 949 /* Alloc and init all Tx queues, including the command queue (#4/#9) */ 950 for (txq_id = 0; txq_id < trans->cfg->base_params->num_of_queues; 951 txq_id++) { 952 slots_num = (txq_id == trans_pcie->cmd_queue) ? 953 TFD_CMD_SLOTS : TFD_TX_CMD_SLOTS; 954 ret = iwl_pcie_txq_init(trans, &trans_pcie->txq[txq_id], 955 slots_num, txq_id); 956 if (ret) { 957 IWL_ERR(trans, "Tx %d queue init failed\n", txq_id); 958 goto error; 959 } 960 } 961 962 iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE); 963 if (trans->cfg->base_params->num_of_queues > 20) 964 iwl_set_bits_prph(trans, SCD_GP_CTRL, 965 SCD_GP_CTRL_ENABLE_31_QUEUES); 966 967 return 0; 968 error: 969 /*Upon error, free only if we allocated something */ 970 if (alloc) 971 iwl_pcie_tx_free(trans); 972 return ret; 973 } 974 975 static inline void iwl_pcie_txq_progress(struct iwl_txq *txq) 976 { 977 lockdep_assert_held(&txq->lock); 978 979 if (!txq->wd_timeout) 980 return; 981 982 /* 983 * station is asleep and we send data - that must 984 * be uAPSD or PS-Poll. Don't rearm the timer. 985 */ 986 if (txq->frozen) 987 return; 988 989 /* 990 * if empty delete timer, otherwise move timer forward 991 * since we're making progress on this queue 992 */ 993 if (txq->q.read_ptr == txq->q.write_ptr) 994 del_timer(&txq->stuck_timer); 995 else 996 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); 997 } 998 999 /* Frees buffers until index _not_ inclusive */ 1000 void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn, 1001 struct sk_buff_head *skbs) 1002 { 1003 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1004 struct iwl_txq *txq = &trans_pcie->txq[txq_id]; 1005 int tfd_num = ssn & (TFD_QUEUE_SIZE_MAX - 1); 1006 struct iwl_queue *q = &txq->q; 1007 int last_to_free; 1008 1009 /* This function is not meant to release cmd queue*/ 1010 if (WARN_ON(txq_id == trans_pcie->cmd_queue)) 1011 return; 1012 1013 spin_lock_bh(&txq->lock); 1014 1015 if (!txq->active) { 1016 IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n", 1017 txq_id, ssn); 1018 goto out; 1019 } 1020 1021 if (txq->q.read_ptr == tfd_num) 1022 goto out; 1023 1024 IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n", 1025 txq_id, txq->q.read_ptr, tfd_num, ssn); 1026 1027 /*Since we free until index _not_ inclusive, the one before index is 1028 * the last we will free. This one must be used */ 1029 last_to_free = iwl_queue_dec_wrap(tfd_num); 1030 1031 if (!iwl_queue_used(q, last_to_free)) { 1032 IWL_ERR(trans, 1033 "%s: Read index for DMA queue txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n", 1034 __func__, txq_id, last_to_free, TFD_QUEUE_SIZE_MAX, 1035 q->write_ptr, q->read_ptr); 1036 goto out; 1037 } 1038 1039 if (WARN_ON(!skb_queue_empty(skbs))) 1040 goto out; 1041 1042 for (; 1043 q->read_ptr != tfd_num; 1044 q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) { 1045 struct sk_buff *skb = txq->entries[txq->q.read_ptr].skb; 1046 1047 if (WARN_ON_ONCE(!skb)) 1048 continue; 1049 1050 iwl_pcie_free_tso_page(skb); 1051 1052 __skb_queue_tail(skbs, skb); 1053 1054 txq->entries[txq->q.read_ptr].skb = NULL; 1055 1056 iwl_pcie_txq_inval_byte_cnt_tbl(trans, txq); 1057 1058 iwl_pcie_txq_free_tfd(trans, txq); 1059 } 1060 1061 iwl_pcie_txq_progress(txq); 1062 1063 if (iwl_queue_space(&txq->q) > txq->q.low_mark && 1064 test_bit(txq_id, trans_pcie->queue_stopped)) { 1065 struct sk_buff_head overflow_skbs; 1066 1067 __skb_queue_head_init(&overflow_skbs); 1068 skb_queue_splice_init(&txq->overflow_q, &overflow_skbs); 1069 1070 /* 1071 * This is tricky: we are in reclaim path which is non 1072 * re-entrant, so noone will try to take the access the 1073 * txq data from that path. We stopped tx, so we can't 1074 * have tx as well. Bottom line, we can unlock and re-lock 1075 * later. 1076 */ 1077 spin_unlock_bh(&txq->lock); 1078 1079 while (!skb_queue_empty(&overflow_skbs)) { 1080 struct sk_buff *skb = __skb_dequeue(&overflow_skbs); 1081 struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); 1082 u8 dev_cmd_idx = IWL_TRANS_FIRST_DRIVER_DATA + 1; 1083 struct iwl_device_cmd *dev_cmd = 1084 info->driver_data[dev_cmd_idx]; 1085 1086 /* 1087 * Note that we can very well be overflowing again. 1088 * In that case, iwl_queue_space will be small again 1089 * and we won't wake mac80211's queue. 1090 */ 1091 iwl_trans_pcie_tx(trans, skb, dev_cmd, txq_id); 1092 } 1093 spin_lock_bh(&txq->lock); 1094 1095 if (iwl_queue_space(&txq->q) > txq->q.low_mark) 1096 iwl_wake_queue(trans, txq); 1097 } 1098 1099 if (q->read_ptr == q->write_ptr) { 1100 IWL_DEBUG_RPM(trans, "Q %d - last tx reclaimed\n", q->id); 1101 iwl_trans_pcie_unref(trans); 1102 } 1103 1104 out: 1105 spin_unlock_bh(&txq->lock); 1106 } 1107 1108 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans, 1109 const struct iwl_host_cmd *cmd) 1110 { 1111 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1112 int ret; 1113 1114 lockdep_assert_held(&trans_pcie->reg_lock); 1115 1116 if (!(cmd->flags & CMD_SEND_IN_IDLE) && 1117 !trans_pcie->ref_cmd_in_flight) { 1118 trans_pcie->ref_cmd_in_flight = true; 1119 IWL_DEBUG_RPM(trans, "set ref_cmd_in_flight - ref\n"); 1120 iwl_trans_pcie_ref(trans); 1121 } 1122 1123 /* 1124 * wake up the NIC to make sure that the firmware will see the host 1125 * command - we will let the NIC sleep once all the host commands 1126 * returned. This needs to be done only on NICs that have 1127 * apmg_wake_up_wa set. 1128 */ 1129 if (trans->cfg->base_params->apmg_wake_up_wa && 1130 !trans_pcie->cmd_hold_nic_awake) { 1131 __iwl_trans_pcie_set_bit(trans, CSR_GP_CNTRL, 1132 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); 1133 1134 ret = iwl_poll_bit(trans, CSR_GP_CNTRL, 1135 CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN, 1136 (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY | 1137 CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP), 1138 15000); 1139 if (ret < 0) { 1140 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL, 1141 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); 1142 IWL_ERR(trans, "Failed to wake NIC for hcmd\n"); 1143 return -EIO; 1144 } 1145 trans_pcie->cmd_hold_nic_awake = true; 1146 } 1147 1148 return 0; 1149 } 1150 1151 static int iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans) 1152 { 1153 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1154 1155 lockdep_assert_held(&trans_pcie->reg_lock); 1156 1157 if (trans_pcie->ref_cmd_in_flight) { 1158 trans_pcie->ref_cmd_in_flight = false; 1159 IWL_DEBUG_RPM(trans, "clear ref_cmd_in_flight - unref\n"); 1160 iwl_trans_pcie_unref(trans); 1161 } 1162 1163 if (trans->cfg->base_params->apmg_wake_up_wa) { 1164 if (WARN_ON(!trans_pcie->cmd_hold_nic_awake)) 1165 return 0; 1166 1167 trans_pcie->cmd_hold_nic_awake = false; 1168 __iwl_trans_pcie_clear_bit(trans, CSR_GP_CNTRL, 1169 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); 1170 } 1171 return 0; 1172 } 1173 1174 /* 1175 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd 1176 * 1177 * When FW advances 'R' index, all entries between old and new 'R' index 1178 * need to be reclaimed. As result, some free space forms. If there is 1179 * enough free space (> low mark), wake the stack that feeds us. 1180 */ 1181 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx) 1182 { 1183 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1184 struct iwl_txq *txq = &trans_pcie->txq[txq_id]; 1185 struct iwl_queue *q = &txq->q; 1186 unsigned long flags; 1187 int nfreed = 0; 1188 1189 lockdep_assert_held(&txq->lock); 1190 1191 if ((idx >= TFD_QUEUE_SIZE_MAX) || (!iwl_queue_used(q, idx))) { 1192 IWL_ERR(trans, 1193 "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n", 1194 __func__, txq_id, idx, TFD_QUEUE_SIZE_MAX, 1195 q->write_ptr, q->read_ptr); 1196 return; 1197 } 1198 1199 for (idx = iwl_queue_inc_wrap(idx); q->read_ptr != idx; 1200 q->read_ptr = iwl_queue_inc_wrap(q->read_ptr)) { 1201 1202 if (nfreed++ > 0) { 1203 IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n", 1204 idx, q->write_ptr, q->read_ptr); 1205 iwl_force_nmi(trans); 1206 } 1207 } 1208 1209 if (q->read_ptr == q->write_ptr) { 1210 spin_lock_irqsave(&trans_pcie->reg_lock, flags); 1211 iwl_pcie_clear_cmd_in_flight(trans); 1212 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags); 1213 } 1214 1215 iwl_pcie_txq_progress(txq); 1216 } 1217 1218 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid, 1219 u16 txq_id) 1220 { 1221 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1222 u32 tbl_dw_addr; 1223 u32 tbl_dw; 1224 u16 scd_q2ratid; 1225 1226 scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK; 1227 1228 tbl_dw_addr = trans_pcie->scd_base_addr + 1229 SCD_TRANS_TBL_OFFSET_QUEUE(txq_id); 1230 1231 tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr); 1232 1233 if (txq_id & 0x1) 1234 tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF); 1235 else 1236 tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000); 1237 1238 iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw); 1239 1240 return 0; 1241 } 1242 1243 /* Receiver address (actually, Rx station's index into station table), 1244 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */ 1245 #define BUILD_RAxTID(sta_id, tid) (((sta_id) << 4) + (tid)) 1246 1247 void iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, 1248 const struct iwl_trans_txq_scd_cfg *cfg, 1249 unsigned int wdg_timeout) 1250 { 1251 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1252 struct iwl_txq *txq = &trans_pcie->txq[txq_id]; 1253 int fifo = -1; 1254 1255 if (test_and_set_bit(txq_id, trans_pcie->queue_used)) 1256 WARN_ONCE(1, "queue %d already used - expect issues", txq_id); 1257 1258 txq->wd_timeout = msecs_to_jiffies(wdg_timeout); 1259 1260 if (cfg) { 1261 fifo = cfg->fifo; 1262 1263 /* Disable the scheduler prior configuring the cmd queue */ 1264 if (txq_id == trans_pcie->cmd_queue && 1265 trans_pcie->scd_set_active) 1266 iwl_scd_enable_set_active(trans, 0); 1267 1268 /* Stop this Tx queue before configuring it */ 1269 iwl_scd_txq_set_inactive(trans, txq_id); 1270 1271 /* Set this queue as a chain-building queue unless it is CMD */ 1272 if (txq_id != trans_pcie->cmd_queue) 1273 iwl_scd_txq_set_chain(trans, txq_id); 1274 1275 if (cfg->aggregate) { 1276 u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid); 1277 1278 /* Map receiver-address / traffic-ID to this queue */ 1279 iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id); 1280 1281 /* enable aggregations for the queue */ 1282 iwl_scd_txq_enable_agg(trans, txq_id); 1283 txq->ampdu = true; 1284 } else { 1285 /* 1286 * disable aggregations for the queue, this will also 1287 * make the ra_tid mapping configuration irrelevant 1288 * since it is now a non-AGG queue. 1289 */ 1290 iwl_scd_txq_disable_agg(trans, txq_id); 1291 1292 ssn = txq->q.read_ptr; 1293 } 1294 } 1295 1296 /* Place first TFD at index corresponding to start sequence number. 1297 * Assumes that ssn_idx is valid (!= 0xFFF) */ 1298 txq->q.read_ptr = (ssn & 0xff); 1299 txq->q.write_ptr = (ssn & 0xff); 1300 iwl_write_direct32(trans, HBUS_TARG_WRPTR, 1301 (ssn & 0xff) | (txq_id << 8)); 1302 1303 if (cfg) { 1304 u8 frame_limit = cfg->frame_limit; 1305 1306 iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn); 1307 1308 /* Set up Tx window size and frame limit for this queue */ 1309 iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr + 1310 SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0); 1311 iwl_trans_write_mem32(trans, 1312 trans_pcie->scd_base_addr + 1313 SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32), 1314 ((frame_limit << SCD_QUEUE_CTX_REG2_WIN_SIZE_POS) & 1315 SCD_QUEUE_CTX_REG2_WIN_SIZE_MSK) | 1316 ((frame_limit << SCD_QUEUE_CTX_REG2_FRAME_LIMIT_POS) & 1317 SCD_QUEUE_CTX_REG2_FRAME_LIMIT_MSK)); 1318 1319 /* Set up status area in SRAM, map to Tx DMA/FIFO, activate */ 1320 iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id), 1321 (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) | 1322 (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) | 1323 (1 << SCD_QUEUE_STTS_REG_POS_WSL) | 1324 SCD_QUEUE_STTS_REG_MSK); 1325 1326 /* enable the scheduler for this queue (only) */ 1327 if (txq_id == trans_pcie->cmd_queue && 1328 trans_pcie->scd_set_active) 1329 iwl_scd_enable_set_active(trans, BIT(txq_id)); 1330 1331 IWL_DEBUG_TX_QUEUES(trans, 1332 "Activate queue %d on FIFO %d WrPtr: %d\n", 1333 txq_id, fifo, ssn & 0xff); 1334 } else { 1335 IWL_DEBUG_TX_QUEUES(trans, 1336 "Activate queue %d WrPtr: %d\n", 1337 txq_id, ssn & 0xff); 1338 } 1339 1340 txq->active = true; 1341 } 1342 1343 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id, 1344 bool configure_scd) 1345 { 1346 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1347 u32 stts_addr = trans_pcie->scd_base_addr + 1348 SCD_TX_STTS_QUEUE_OFFSET(txq_id); 1349 static const u32 zero_val[4] = {}; 1350 1351 trans_pcie->txq[txq_id].frozen_expiry_remainder = 0; 1352 trans_pcie->txq[txq_id].frozen = false; 1353 1354 /* 1355 * Upon HW Rfkill - we stop the device, and then stop the queues 1356 * in the op_mode. Just for the sake of the simplicity of the op_mode, 1357 * allow the op_mode to call txq_disable after it already called 1358 * stop_device. 1359 */ 1360 if (!test_and_clear_bit(txq_id, trans_pcie->queue_used)) { 1361 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status), 1362 "queue %d not used", txq_id); 1363 return; 1364 } 1365 1366 if (configure_scd) { 1367 iwl_scd_txq_set_inactive(trans, txq_id); 1368 1369 iwl_trans_write_mem(trans, stts_addr, (void *)zero_val, 1370 ARRAY_SIZE(zero_val)); 1371 } 1372 1373 iwl_pcie_txq_unmap(trans, txq_id); 1374 trans_pcie->txq[txq_id].ampdu = false; 1375 1376 IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id); 1377 } 1378 1379 /*************** HOST COMMAND QUEUE FUNCTIONS *****/ 1380 1381 /* 1382 * iwl_pcie_enqueue_hcmd - enqueue a uCode command 1383 * @priv: device private data point 1384 * @cmd: a pointer to the ucode command structure 1385 * 1386 * The function returns < 0 values to indicate the operation 1387 * failed. On success, it returns the index (>= 0) of command in the 1388 * command queue. 1389 */ 1390 static int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, 1391 struct iwl_host_cmd *cmd) 1392 { 1393 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1394 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue]; 1395 struct iwl_queue *q = &txq->q; 1396 struct iwl_device_cmd *out_cmd; 1397 struct iwl_cmd_meta *out_meta; 1398 unsigned long flags; 1399 void *dup_buf = NULL; 1400 dma_addr_t phys_addr; 1401 int idx; 1402 u16 copy_size, cmd_size, scratch_size; 1403 bool had_nocopy = false; 1404 u8 group_id = iwl_cmd_groupid(cmd->id); 1405 int i, ret; 1406 u32 cmd_pos; 1407 const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD]; 1408 u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD]; 1409 1410 if (WARN(!trans_pcie->wide_cmd_header && 1411 group_id > IWL_ALWAYS_LONG_GROUP, 1412 "unsupported wide command %#x\n", cmd->id)) 1413 return -EINVAL; 1414 1415 if (group_id != 0) { 1416 copy_size = sizeof(struct iwl_cmd_header_wide); 1417 cmd_size = sizeof(struct iwl_cmd_header_wide); 1418 } else { 1419 copy_size = sizeof(struct iwl_cmd_header); 1420 cmd_size = sizeof(struct iwl_cmd_header); 1421 } 1422 1423 /* need one for the header if the first is NOCOPY */ 1424 BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1); 1425 1426 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 1427 cmddata[i] = cmd->data[i]; 1428 cmdlen[i] = cmd->len[i]; 1429 1430 if (!cmd->len[i]) 1431 continue; 1432 1433 /* need at least IWL_HCMD_SCRATCHBUF_SIZE copied */ 1434 if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) { 1435 int copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size; 1436 1437 if (copy > cmdlen[i]) 1438 copy = cmdlen[i]; 1439 cmdlen[i] -= copy; 1440 cmddata[i] += copy; 1441 copy_size += copy; 1442 } 1443 1444 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) { 1445 had_nocopy = true; 1446 if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) { 1447 idx = -EINVAL; 1448 goto free_dup_buf; 1449 } 1450 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) { 1451 /* 1452 * This is also a chunk that isn't copied 1453 * to the static buffer so set had_nocopy. 1454 */ 1455 had_nocopy = true; 1456 1457 /* only allowed once */ 1458 if (WARN_ON(dup_buf)) { 1459 idx = -EINVAL; 1460 goto free_dup_buf; 1461 } 1462 1463 dup_buf = kmemdup(cmddata[i], cmdlen[i], 1464 GFP_ATOMIC); 1465 if (!dup_buf) 1466 return -ENOMEM; 1467 } else { 1468 /* NOCOPY must not be followed by normal! */ 1469 if (WARN_ON(had_nocopy)) { 1470 idx = -EINVAL; 1471 goto free_dup_buf; 1472 } 1473 copy_size += cmdlen[i]; 1474 } 1475 cmd_size += cmd->len[i]; 1476 } 1477 1478 /* 1479 * If any of the command structures end up being larger than 1480 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically 1481 * allocated into separate TFDs, then we will need to 1482 * increase the size of the buffers. 1483 */ 1484 if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE, 1485 "Command %s (%#x) is too large (%d bytes)\n", 1486 iwl_get_cmd_string(trans, cmd->id), 1487 cmd->id, copy_size)) { 1488 idx = -EINVAL; 1489 goto free_dup_buf; 1490 } 1491 1492 spin_lock_bh(&txq->lock); 1493 1494 if (iwl_queue_space(q) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) { 1495 spin_unlock_bh(&txq->lock); 1496 1497 IWL_ERR(trans, "No space in command queue\n"); 1498 iwl_op_mode_cmd_queue_full(trans->op_mode); 1499 idx = -ENOSPC; 1500 goto free_dup_buf; 1501 } 1502 1503 idx = get_cmd_index(q, q->write_ptr); 1504 out_cmd = txq->entries[idx].cmd; 1505 out_meta = &txq->entries[idx].meta; 1506 1507 memset(out_meta, 0, sizeof(*out_meta)); /* re-initialize to NULL */ 1508 if (cmd->flags & CMD_WANT_SKB) 1509 out_meta->source = cmd; 1510 1511 /* set up the header */ 1512 if (group_id != 0) { 1513 out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id); 1514 out_cmd->hdr_wide.group_id = group_id; 1515 out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id); 1516 out_cmd->hdr_wide.length = 1517 cpu_to_le16(cmd_size - 1518 sizeof(struct iwl_cmd_header_wide)); 1519 out_cmd->hdr_wide.reserved = 0; 1520 out_cmd->hdr_wide.sequence = 1521 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) | 1522 INDEX_TO_SEQ(q->write_ptr)); 1523 1524 cmd_pos = sizeof(struct iwl_cmd_header_wide); 1525 copy_size = sizeof(struct iwl_cmd_header_wide); 1526 } else { 1527 out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id); 1528 out_cmd->hdr.sequence = 1529 cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->cmd_queue) | 1530 INDEX_TO_SEQ(q->write_ptr)); 1531 out_cmd->hdr.group_id = 0; 1532 1533 cmd_pos = sizeof(struct iwl_cmd_header); 1534 copy_size = sizeof(struct iwl_cmd_header); 1535 } 1536 1537 /* and copy the data that needs to be copied */ 1538 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 1539 int copy; 1540 1541 if (!cmd->len[i]) 1542 continue; 1543 1544 /* copy everything if not nocopy/dup */ 1545 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY | 1546 IWL_HCMD_DFL_DUP))) { 1547 copy = cmd->len[i]; 1548 1549 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy); 1550 cmd_pos += copy; 1551 copy_size += copy; 1552 continue; 1553 } 1554 1555 /* 1556 * Otherwise we need at least IWL_HCMD_SCRATCHBUF_SIZE copied 1557 * in total (for the scratchbuf handling), but copy up to what 1558 * we can fit into the payload for debug dump purposes. 1559 */ 1560 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]); 1561 1562 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy); 1563 cmd_pos += copy; 1564 1565 /* However, treat copy_size the proper way, we need it below */ 1566 if (copy_size < IWL_HCMD_SCRATCHBUF_SIZE) { 1567 copy = IWL_HCMD_SCRATCHBUF_SIZE - copy_size; 1568 1569 if (copy > cmd->len[i]) 1570 copy = cmd->len[i]; 1571 copy_size += copy; 1572 } 1573 } 1574 1575 IWL_DEBUG_HC(trans, 1576 "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n", 1577 iwl_get_cmd_string(trans, cmd->id), 1578 group_id, out_cmd->hdr.cmd, 1579 le16_to_cpu(out_cmd->hdr.sequence), 1580 cmd_size, q->write_ptr, idx, trans_pcie->cmd_queue); 1581 1582 /* start the TFD with the scratchbuf */ 1583 scratch_size = min_t(int, copy_size, IWL_HCMD_SCRATCHBUF_SIZE); 1584 memcpy(&txq->scratchbufs[q->write_ptr], &out_cmd->hdr, scratch_size); 1585 iwl_pcie_txq_build_tfd(trans, txq, 1586 iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr), 1587 scratch_size, true); 1588 1589 /* map first command fragment, if any remains */ 1590 if (copy_size > scratch_size) { 1591 phys_addr = dma_map_single(trans->dev, 1592 ((u8 *)&out_cmd->hdr) + scratch_size, 1593 copy_size - scratch_size, 1594 DMA_TO_DEVICE); 1595 if (dma_mapping_error(trans->dev, phys_addr)) { 1596 iwl_pcie_tfd_unmap(trans, out_meta, 1597 &txq->tfds[q->write_ptr]); 1598 idx = -ENOMEM; 1599 goto out; 1600 } 1601 1602 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, 1603 copy_size - scratch_size, false); 1604 } 1605 1606 /* map the remaining (adjusted) nocopy/dup fragments */ 1607 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 1608 const void *data = cmddata[i]; 1609 1610 if (!cmdlen[i]) 1611 continue; 1612 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY | 1613 IWL_HCMD_DFL_DUP))) 1614 continue; 1615 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) 1616 data = dup_buf; 1617 phys_addr = dma_map_single(trans->dev, (void *)data, 1618 cmdlen[i], DMA_TO_DEVICE); 1619 if (dma_mapping_error(trans->dev, phys_addr)) { 1620 iwl_pcie_tfd_unmap(trans, out_meta, 1621 &txq->tfds[q->write_ptr]); 1622 idx = -ENOMEM; 1623 goto out; 1624 } 1625 1626 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false); 1627 } 1628 1629 BUILD_BUG_ON(IWL_NUM_OF_TBS + CMD_TB_BITMAP_POS > 1630 sizeof(out_meta->flags) * BITS_PER_BYTE); 1631 out_meta->flags = cmd->flags; 1632 if (WARN_ON_ONCE(txq->entries[idx].free_buf)) 1633 kzfree(txq->entries[idx].free_buf); 1634 txq->entries[idx].free_buf = dup_buf; 1635 1636 trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide); 1637 1638 /* start timer if queue currently empty */ 1639 if (q->read_ptr == q->write_ptr && txq->wd_timeout) 1640 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); 1641 1642 spin_lock_irqsave(&trans_pcie->reg_lock, flags); 1643 ret = iwl_pcie_set_cmd_in_flight(trans, cmd); 1644 if (ret < 0) { 1645 idx = ret; 1646 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags); 1647 goto out; 1648 } 1649 1650 /* Increment and update queue's write index */ 1651 q->write_ptr = iwl_queue_inc_wrap(q->write_ptr); 1652 iwl_pcie_txq_inc_wr_ptr(trans, txq); 1653 1654 spin_unlock_irqrestore(&trans_pcie->reg_lock, flags); 1655 1656 out: 1657 spin_unlock_bh(&txq->lock); 1658 free_dup_buf: 1659 if (idx < 0) 1660 kfree(dup_buf); 1661 return idx; 1662 } 1663 1664 /* 1665 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them 1666 * @rxb: Rx buffer to reclaim 1667 */ 1668 void iwl_pcie_hcmd_complete(struct iwl_trans *trans, 1669 struct iwl_rx_cmd_buffer *rxb) 1670 { 1671 struct iwl_rx_packet *pkt = rxb_addr(rxb); 1672 u16 sequence = le16_to_cpu(pkt->hdr.sequence); 1673 u8 group_id = iwl_cmd_groupid(pkt->hdr.group_id); 1674 u32 cmd_id; 1675 int txq_id = SEQ_TO_QUEUE(sequence); 1676 int index = SEQ_TO_INDEX(sequence); 1677 int cmd_index; 1678 struct iwl_device_cmd *cmd; 1679 struct iwl_cmd_meta *meta; 1680 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1681 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue]; 1682 1683 /* If a Tx command is being handled and it isn't in the actual 1684 * command queue then there a command routing bug has been introduced 1685 * in the queue management code. */ 1686 if (WARN(txq_id != trans_pcie->cmd_queue, 1687 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n", 1688 txq_id, trans_pcie->cmd_queue, sequence, 1689 trans_pcie->txq[trans_pcie->cmd_queue].q.read_ptr, 1690 trans_pcie->txq[trans_pcie->cmd_queue].q.write_ptr)) { 1691 iwl_print_hex_error(trans, pkt, 32); 1692 return; 1693 } 1694 1695 spin_lock_bh(&txq->lock); 1696 1697 cmd_index = get_cmd_index(&txq->q, index); 1698 cmd = txq->entries[cmd_index].cmd; 1699 meta = &txq->entries[cmd_index].meta; 1700 cmd_id = iwl_cmd_id(cmd->hdr.cmd, group_id, 0); 1701 1702 iwl_pcie_tfd_unmap(trans, meta, &txq->tfds[index]); 1703 1704 /* Input error checking is done when commands are added to queue. */ 1705 if (meta->flags & CMD_WANT_SKB) { 1706 struct page *p = rxb_steal_page(rxb); 1707 1708 meta->source->resp_pkt = pkt; 1709 meta->source->_rx_page_addr = (unsigned long)page_address(p); 1710 meta->source->_rx_page_order = trans_pcie->rx_page_order; 1711 } 1712 1713 if (meta->flags & CMD_WANT_ASYNC_CALLBACK) 1714 iwl_op_mode_async_cb(trans->op_mode, cmd); 1715 1716 iwl_pcie_cmdq_reclaim(trans, txq_id, index); 1717 1718 if (!(meta->flags & CMD_ASYNC)) { 1719 if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) { 1720 IWL_WARN(trans, 1721 "HCMD_ACTIVE already clear for command %s\n", 1722 iwl_get_cmd_string(trans, cmd_id)); 1723 } 1724 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 1725 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n", 1726 iwl_get_cmd_string(trans, cmd_id)); 1727 wake_up(&trans_pcie->wait_command_queue); 1728 } 1729 1730 if (meta->flags & CMD_MAKE_TRANS_IDLE) { 1731 IWL_DEBUG_INFO(trans, "complete %s - mark trans as idle\n", 1732 iwl_get_cmd_string(trans, cmd->hdr.cmd)); 1733 set_bit(STATUS_TRANS_IDLE, &trans->status); 1734 wake_up(&trans_pcie->d0i3_waitq); 1735 } 1736 1737 if (meta->flags & CMD_WAKE_UP_TRANS) { 1738 IWL_DEBUG_INFO(trans, "complete %s - clear trans idle flag\n", 1739 iwl_get_cmd_string(trans, cmd->hdr.cmd)); 1740 clear_bit(STATUS_TRANS_IDLE, &trans->status); 1741 wake_up(&trans_pcie->d0i3_waitq); 1742 } 1743 1744 meta->flags = 0; 1745 1746 spin_unlock_bh(&txq->lock); 1747 } 1748 1749 #define HOST_COMPLETE_TIMEOUT (2 * HZ) 1750 1751 static int iwl_pcie_send_hcmd_async(struct iwl_trans *trans, 1752 struct iwl_host_cmd *cmd) 1753 { 1754 int ret; 1755 1756 /* An asynchronous command can not expect an SKB to be set. */ 1757 if (WARN_ON(cmd->flags & CMD_WANT_SKB)) 1758 return -EINVAL; 1759 1760 ret = iwl_pcie_enqueue_hcmd(trans, cmd); 1761 if (ret < 0) { 1762 IWL_ERR(trans, 1763 "Error sending %s: enqueue_hcmd failed: %d\n", 1764 iwl_get_cmd_string(trans, cmd->id), ret); 1765 return ret; 1766 } 1767 return 0; 1768 } 1769 1770 static int iwl_pcie_send_hcmd_sync(struct iwl_trans *trans, 1771 struct iwl_host_cmd *cmd) 1772 { 1773 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1774 int cmd_idx; 1775 int ret; 1776 1777 IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", 1778 iwl_get_cmd_string(trans, cmd->id)); 1779 1780 if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE, 1781 &trans->status), 1782 "Command %s: a command is already active!\n", 1783 iwl_get_cmd_string(trans, cmd->id))) 1784 return -EIO; 1785 1786 IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", 1787 iwl_get_cmd_string(trans, cmd->id)); 1788 1789 cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd); 1790 if (cmd_idx < 0) { 1791 ret = cmd_idx; 1792 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 1793 IWL_ERR(trans, 1794 "Error sending %s: enqueue_hcmd failed: %d\n", 1795 iwl_get_cmd_string(trans, cmd->id), ret); 1796 return ret; 1797 } 1798 1799 ret = wait_event_timeout(trans_pcie->wait_command_queue, 1800 !test_bit(STATUS_SYNC_HCMD_ACTIVE, 1801 &trans->status), 1802 HOST_COMPLETE_TIMEOUT); 1803 if (!ret) { 1804 struct iwl_txq *txq = &trans_pcie->txq[trans_pcie->cmd_queue]; 1805 struct iwl_queue *q = &txq->q; 1806 1807 IWL_ERR(trans, "Error sending %s: time out after %dms.\n", 1808 iwl_get_cmd_string(trans, cmd->id), 1809 jiffies_to_msecs(HOST_COMPLETE_TIMEOUT)); 1810 1811 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n", 1812 q->read_ptr, q->write_ptr); 1813 1814 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 1815 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n", 1816 iwl_get_cmd_string(trans, cmd->id)); 1817 ret = -ETIMEDOUT; 1818 1819 iwl_force_nmi(trans); 1820 iwl_trans_fw_error(trans); 1821 1822 goto cancel; 1823 } 1824 1825 if (test_bit(STATUS_FW_ERROR, &trans->status)) { 1826 IWL_ERR(trans, "FW error in SYNC CMD %s\n", 1827 iwl_get_cmd_string(trans, cmd->id)); 1828 dump_stack(); 1829 ret = -EIO; 1830 goto cancel; 1831 } 1832 1833 if (!(cmd->flags & CMD_SEND_IN_RFKILL) && 1834 test_bit(STATUS_RFKILL, &trans->status)) { 1835 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n"); 1836 ret = -ERFKILL; 1837 goto cancel; 1838 } 1839 1840 if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) { 1841 IWL_ERR(trans, "Error: Response NULL in '%s'\n", 1842 iwl_get_cmd_string(trans, cmd->id)); 1843 ret = -EIO; 1844 goto cancel; 1845 } 1846 1847 return 0; 1848 1849 cancel: 1850 if (cmd->flags & CMD_WANT_SKB) { 1851 /* 1852 * Cancel the CMD_WANT_SKB flag for the cmd in the 1853 * TX cmd queue. Otherwise in case the cmd comes 1854 * in later, it will possibly set an invalid 1855 * address (cmd->meta.source). 1856 */ 1857 trans_pcie->txq[trans_pcie->cmd_queue]. 1858 entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB; 1859 } 1860 1861 if (cmd->resp_pkt) { 1862 iwl_free_resp(cmd); 1863 cmd->resp_pkt = NULL; 1864 } 1865 1866 return ret; 1867 } 1868 1869 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd) 1870 { 1871 if (!(cmd->flags & CMD_SEND_IN_RFKILL) && 1872 test_bit(STATUS_RFKILL, &trans->status)) { 1873 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n", 1874 cmd->id); 1875 return -ERFKILL; 1876 } 1877 1878 if (cmd->flags & CMD_ASYNC) 1879 return iwl_pcie_send_hcmd_async(trans, cmd); 1880 1881 /* We still can fail on RFKILL that can be asserted while we wait */ 1882 return iwl_pcie_send_hcmd_sync(trans, cmd); 1883 } 1884 1885 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb, 1886 struct iwl_txq *txq, u8 hdr_len, 1887 struct iwl_cmd_meta *out_meta, 1888 struct iwl_device_cmd *dev_cmd, u16 tb1_len) 1889 { 1890 struct iwl_queue *q = &txq->q; 1891 u16 tb2_len; 1892 int i; 1893 1894 /* 1895 * Set up TFD's third entry to point directly to remainder 1896 * of skb's head, if any 1897 */ 1898 tb2_len = skb_headlen(skb) - hdr_len; 1899 1900 if (tb2_len > 0) { 1901 dma_addr_t tb2_phys = dma_map_single(trans->dev, 1902 skb->data + hdr_len, 1903 tb2_len, DMA_TO_DEVICE); 1904 if (unlikely(dma_mapping_error(trans->dev, tb2_phys))) { 1905 iwl_pcie_tfd_unmap(trans, out_meta, 1906 &txq->tfds[q->write_ptr]); 1907 return -EINVAL; 1908 } 1909 iwl_pcie_txq_build_tfd(trans, txq, tb2_phys, tb2_len, false); 1910 } 1911 1912 /* set up the remaining entries to point to the data */ 1913 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1914 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1915 dma_addr_t tb_phys; 1916 int tb_idx; 1917 1918 if (!skb_frag_size(frag)) 1919 continue; 1920 1921 tb_phys = skb_frag_dma_map(trans->dev, frag, 0, 1922 skb_frag_size(frag), DMA_TO_DEVICE); 1923 1924 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) { 1925 iwl_pcie_tfd_unmap(trans, out_meta, 1926 &txq->tfds[q->write_ptr]); 1927 return -EINVAL; 1928 } 1929 tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys, 1930 skb_frag_size(frag), false); 1931 1932 out_meta->flags |= BIT(tb_idx + CMD_TB_BITMAP_POS); 1933 } 1934 1935 trace_iwlwifi_dev_tx(trans->dev, skb, 1936 &txq->tfds[txq->q.write_ptr], 1937 sizeof(struct iwl_tfd), 1938 &dev_cmd->hdr, IWL_HCMD_SCRATCHBUF_SIZE + tb1_len, 1939 skb->data + hdr_len, tb2_len); 1940 trace_iwlwifi_dev_tx_data(trans->dev, skb, 1941 hdr_len, skb->len - hdr_len); 1942 return 0; 1943 } 1944 1945 #ifdef CONFIG_INET 1946 static struct iwl_tso_hdr_page * 1947 get_page_hdr(struct iwl_trans *trans, size_t len) 1948 { 1949 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1950 struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->tso_hdr_page); 1951 1952 if (!p->page) 1953 goto alloc; 1954 1955 /* enough room on this page */ 1956 if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE) 1957 return p; 1958 1959 /* We don't have enough room on this page, get a new one. */ 1960 __free_page(p->page); 1961 1962 alloc: 1963 p->page = alloc_page(GFP_ATOMIC); 1964 if (!p->page) 1965 return NULL; 1966 p->pos = page_address(p->page); 1967 return p; 1968 } 1969 1970 static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph, 1971 bool ipv6, unsigned int len) 1972 { 1973 if (ipv6) { 1974 struct ipv6hdr *iphv6 = iph; 1975 1976 tcph->check = ~csum_ipv6_magic(&iphv6->saddr, &iphv6->daddr, 1977 len + tcph->doff * 4, 1978 IPPROTO_TCP, 0); 1979 } else { 1980 struct iphdr *iphv4 = iph; 1981 1982 ip_send_check(iphv4); 1983 tcph->check = ~csum_tcpudp_magic(iphv4->saddr, iphv4->daddr, 1984 len + tcph->doff * 4, 1985 IPPROTO_TCP, 0); 1986 } 1987 } 1988 1989 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, 1990 struct iwl_txq *txq, u8 hdr_len, 1991 struct iwl_cmd_meta *out_meta, 1992 struct iwl_device_cmd *dev_cmd, u16 tb1_len) 1993 { 1994 struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); 1995 struct iwl_trans_pcie *trans_pcie = txq->trans_pcie; 1996 struct ieee80211_hdr *hdr = (void *)skb->data; 1997 unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room; 1998 unsigned int mss = skb_shinfo(skb)->gso_size; 1999 struct iwl_queue *q = &txq->q; 2000 u16 length, iv_len, amsdu_pad; 2001 u8 *start_hdr; 2002 struct iwl_tso_hdr_page *hdr_page; 2003 int ret; 2004 struct tso_t tso; 2005 2006 /* if the packet is protected, then it must be CCMP or GCMP */ 2007 BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN); 2008 iv_len = ieee80211_has_protected(hdr->frame_control) ? 2009 IEEE80211_CCMP_HDR_LEN : 0; 2010 2011 trace_iwlwifi_dev_tx(trans->dev, skb, 2012 &txq->tfds[txq->q.write_ptr], 2013 sizeof(struct iwl_tfd), 2014 &dev_cmd->hdr, IWL_HCMD_SCRATCHBUF_SIZE + tb1_len, 2015 NULL, 0); 2016 2017 ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb); 2018 snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb); 2019 total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len; 2020 amsdu_pad = 0; 2021 2022 /* total amount of header we may need for this A-MSDU */ 2023 hdr_room = DIV_ROUND_UP(total_len, mss) * 2024 (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len; 2025 2026 /* Our device supports 9 segments at most, it will fit in 1 page */ 2027 hdr_page = get_page_hdr(trans, hdr_room); 2028 if (!hdr_page) 2029 return -ENOMEM; 2030 2031 get_page(hdr_page->page); 2032 start_hdr = hdr_page->pos; 2033 info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA] = hdr_page->page; 2034 memcpy(hdr_page->pos, skb->data + hdr_len, iv_len); 2035 hdr_page->pos += iv_len; 2036 2037 /* 2038 * Pull the ieee80211 header + IV to be able to use TSO core, 2039 * we will restore it for the tx_status flow. 2040 */ 2041 skb_pull(skb, hdr_len + iv_len); 2042 2043 tso_start(skb, &tso); 2044 2045 while (total_len) { 2046 /* this is the data left for this subframe */ 2047 unsigned int data_left = 2048 min_t(unsigned int, mss, total_len); 2049 struct sk_buff *csum_skb = NULL; 2050 unsigned int hdr_tb_len; 2051 dma_addr_t hdr_tb_phys; 2052 struct tcphdr *tcph; 2053 u8 *iph; 2054 2055 total_len -= data_left; 2056 2057 memset(hdr_page->pos, 0, amsdu_pad); 2058 hdr_page->pos += amsdu_pad; 2059 amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen + 2060 data_left)) & 0x3; 2061 ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr)); 2062 hdr_page->pos += ETH_ALEN; 2063 ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr)); 2064 hdr_page->pos += ETH_ALEN; 2065 2066 length = snap_ip_tcp_hdrlen + data_left; 2067 *((__be16 *)hdr_page->pos) = cpu_to_be16(length); 2068 hdr_page->pos += sizeof(length); 2069 2070 /* 2071 * This will copy the SNAP as well which will be considered 2072 * as MAC header. 2073 */ 2074 tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len); 2075 iph = hdr_page->pos + 8; 2076 tcph = (void *)(iph + ip_hdrlen); 2077 2078 /* For testing on current hardware only */ 2079 if (trans_pcie->sw_csum_tx) { 2080 csum_skb = alloc_skb(data_left + tcp_hdrlen(skb), 2081 GFP_ATOMIC); 2082 if (!csum_skb) { 2083 ret = -ENOMEM; 2084 goto out_unmap; 2085 } 2086 2087 iwl_compute_pseudo_hdr_csum(iph, tcph, 2088 skb->protocol == 2089 htons(ETH_P_IPV6), 2090 data_left); 2091 2092 memcpy(skb_put(csum_skb, tcp_hdrlen(skb)), 2093 tcph, tcp_hdrlen(skb)); 2094 skb_set_transport_header(csum_skb, 0); 2095 csum_skb->csum_start = 2096 (unsigned char *)tcp_hdr(csum_skb) - 2097 csum_skb->head; 2098 } 2099 2100 hdr_page->pos += snap_ip_tcp_hdrlen; 2101 2102 hdr_tb_len = hdr_page->pos - start_hdr; 2103 hdr_tb_phys = dma_map_single(trans->dev, start_hdr, 2104 hdr_tb_len, DMA_TO_DEVICE); 2105 if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) { 2106 dev_kfree_skb(csum_skb); 2107 ret = -EINVAL; 2108 goto out_unmap; 2109 } 2110 iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys, 2111 hdr_tb_len, false); 2112 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_hdr, 2113 hdr_tb_len); 2114 2115 /* prepare the start_hdr for the next subframe */ 2116 start_hdr = hdr_page->pos; 2117 2118 /* put the payload */ 2119 while (data_left) { 2120 unsigned int size = min_t(unsigned int, tso.size, 2121 data_left); 2122 dma_addr_t tb_phys; 2123 2124 if (trans_pcie->sw_csum_tx) 2125 memcpy(skb_put(csum_skb, size), tso.data, size); 2126 2127 tb_phys = dma_map_single(trans->dev, tso.data, 2128 size, DMA_TO_DEVICE); 2129 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) { 2130 dev_kfree_skb(csum_skb); 2131 ret = -EINVAL; 2132 goto out_unmap; 2133 } 2134 2135 iwl_pcie_txq_build_tfd(trans, txq, tb_phys, 2136 size, false); 2137 trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data, 2138 size); 2139 2140 data_left -= size; 2141 tso_build_data(skb, &tso, size); 2142 } 2143 2144 /* For testing on early hardware only */ 2145 if (trans_pcie->sw_csum_tx) { 2146 __wsum csum; 2147 2148 csum = skb_checksum(csum_skb, 2149 skb_checksum_start_offset(csum_skb), 2150 csum_skb->len - 2151 skb_checksum_start_offset(csum_skb), 2152 0); 2153 dev_kfree_skb(csum_skb); 2154 dma_sync_single_for_cpu(trans->dev, hdr_tb_phys, 2155 hdr_tb_len, DMA_TO_DEVICE); 2156 tcph->check = csum_fold(csum); 2157 dma_sync_single_for_device(trans->dev, hdr_tb_phys, 2158 hdr_tb_len, DMA_TO_DEVICE); 2159 } 2160 } 2161 2162 /* re -add the WiFi header and IV */ 2163 skb_push(skb, hdr_len + iv_len); 2164 2165 return 0; 2166 2167 out_unmap: 2168 iwl_pcie_tfd_unmap(trans, out_meta, &txq->tfds[q->write_ptr]); 2169 return ret; 2170 } 2171 #else /* CONFIG_INET */ 2172 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, 2173 struct iwl_txq *txq, u8 hdr_len, 2174 struct iwl_cmd_meta *out_meta, 2175 struct iwl_device_cmd *dev_cmd, u16 tb1_len) 2176 { 2177 /* No A-MSDU without CONFIG_INET */ 2178 WARN_ON(1); 2179 2180 return -1; 2181 } 2182 #endif /* CONFIG_INET */ 2183 2184 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, 2185 struct iwl_device_cmd *dev_cmd, int txq_id) 2186 { 2187 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2188 struct ieee80211_hdr *hdr; 2189 struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload; 2190 struct iwl_cmd_meta *out_meta; 2191 struct iwl_txq *txq; 2192 struct iwl_queue *q; 2193 dma_addr_t tb0_phys, tb1_phys, scratch_phys; 2194 void *tb1_addr; 2195 u16 len, tb1_len; 2196 bool wait_write_ptr; 2197 __le16 fc; 2198 u8 hdr_len; 2199 u16 wifi_seq; 2200 2201 txq = &trans_pcie->txq[txq_id]; 2202 q = &txq->q; 2203 2204 if (WARN_ONCE(!test_bit(txq_id, trans_pcie->queue_used), 2205 "TX on unused queue %d\n", txq_id)) 2206 return -EINVAL; 2207 2208 if (unlikely(trans_pcie->sw_csum_tx && 2209 skb->ip_summed == CHECKSUM_PARTIAL)) { 2210 int offs = skb_checksum_start_offset(skb); 2211 int csum_offs = offs + skb->csum_offset; 2212 __wsum csum; 2213 2214 if (skb_ensure_writable(skb, csum_offs + sizeof(__sum16))) 2215 return -1; 2216 2217 csum = skb_checksum(skb, offs, skb->len - offs, 0); 2218 *(__sum16 *)(skb->data + csum_offs) = csum_fold(csum); 2219 2220 skb->ip_summed = CHECKSUM_UNNECESSARY; 2221 } 2222 2223 if (skb_is_nonlinear(skb) && 2224 skb_shinfo(skb)->nr_frags > IWL_PCIE_MAX_FRAGS && 2225 __skb_linearize(skb)) 2226 return -ENOMEM; 2227 2228 /* mac80211 always puts the full header into the SKB's head, 2229 * so there's no need to check if it's readable there 2230 */ 2231 hdr = (struct ieee80211_hdr *)skb->data; 2232 fc = hdr->frame_control; 2233 hdr_len = ieee80211_hdrlen(fc); 2234 2235 spin_lock(&txq->lock); 2236 2237 if (iwl_queue_space(q) < q->high_mark) { 2238 iwl_stop_queue(trans, txq); 2239 2240 /* don't put the packet on the ring, if there is no room */ 2241 if (unlikely(iwl_queue_space(q) < 3)) { 2242 struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); 2243 2244 info->driver_data[IWL_TRANS_FIRST_DRIVER_DATA + 1] = 2245 dev_cmd; 2246 __skb_queue_tail(&txq->overflow_q, skb); 2247 2248 spin_unlock(&txq->lock); 2249 return 0; 2250 } 2251 } 2252 2253 /* In AGG mode, the index in the ring must correspond to the WiFi 2254 * sequence number. This is a HW requirements to help the SCD to parse 2255 * the BA. 2256 * Check here that the packets are in the right place on the ring. 2257 */ 2258 wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl)); 2259 WARN_ONCE(txq->ampdu && 2260 (wifi_seq & 0xff) != q->write_ptr, 2261 "Q: %d WiFi Seq %d tfdNum %d", 2262 txq_id, wifi_seq, q->write_ptr); 2263 2264 /* Set up driver data for this TFD */ 2265 txq->entries[q->write_ptr].skb = skb; 2266 txq->entries[q->write_ptr].cmd = dev_cmd; 2267 2268 dev_cmd->hdr.sequence = 2269 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) | 2270 INDEX_TO_SEQ(q->write_ptr))); 2271 2272 tb0_phys = iwl_pcie_get_scratchbuf_dma(txq, q->write_ptr); 2273 scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) + 2274 offsetof(struct iwl_tx_cmd, scratch); 2275 2276 tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys); 2277 tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys); 2278 2279 /* Set up first empty entry in queue's array of Tx/cmd buffers */ 2280 out_meta = &txq->entries[q->write_ptr].meta; 2281 out_meta->flags = 0; 2282 2283 /* 2284 * The second TB (tb1) points to the remainder of the TX command 2285 * and the 802.11 header - dword aligned size 2286 * (This calculation modifies the TX command, so do it before the 2287 * setup of the first TB) 2288 */ 2289 len = sizeof(struct iwl_tx_cmd) + sizeof(struct iwl_cmd_header) + 2290 hdr_len - IWL_HCMD_SCRATCHBUF_SIZE; 2291 tb1_len = ALIGN(len, 4); 2292 2293 /* Tell NIC about any 2-byte padding after MAC header */ 2294 if (tb1_len != len) 2295 tx_cmd->tx_flags |= TX_CMD_FLG_MH_PAD_MSK; 2296 2297 /* The first TB points to the scratchbuf data - min_copy bytes */ 2298 memcpy(&txq->scratchbufs[q->write_ptr], &dev_cmd->hdr, 2299 IWL_HCMD_SCRATCHBUF_SIZE); 2300 iwl_pcie_txq_build_tfd(trans, txq, tb0_phys, 2301 IWL_HCMD_SCRATCHBUF_SIZE, true); 2302 2303 /* there must be data left over for TB1 or this code must be changed */ 2304 BUILD_BUG_ON(sizeof(struct iwl_tx_cmd) < IWL_HCMD_SCRATCHBUF_SIZE); 2305 2306 /* map the data for TB1 */ 2307 tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_HCMD_SCRATCHBUF_SIZE; 2308 tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE); 2309 if (unlikely(dma_mapping_error(trans->dev, tb1_phys))) 2310 goto out_err; 2311 iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false); 2312 2313 if (ieee80211_is_data_qos(fc) && 2314 (*ieee80211_get_qos_ctl(hdr) & IEEE80211_QOS_CTL_A_MSDU_PRESENT)) { 2315 if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len, 2316 out_meta, dev_cmd, 2317 tb1_len))) 2318 goto out_err; 2319 } else if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len, 2320 out_meta, dev_cmd, tb1_len))) { 2321 goto out_err; 2322 } 2323 2324 /* Set up entry for this TFD in Tx byte-count array */ 2325 iwl_pcie_txq_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len)); 2326 2327 wait_write_ptr = ieee80211_has_morefrags(fc); 2328 2329 /* start timer if queue currently empty */ 2330 if (q->read_ptr == q->write_ptr) { 2331 if (txq->wd_timeout) { 2332 /* 2333 * If the TXQ is active, then set the timer, if not, 2334 * set the timer in remainder so that the timer will 2335 * be armed with the right value when the station will 2336 * wake up. 2337 */ 2338 if (!txq->frozen) 2339 mod_timer(&txq->stuck_timer, 2340 jiffies + txq->wd_timeout); 2341 else 2342 txq->frozen_expiry_remainder = txq->wd_timeout; 2343 } 2344 IWL_DEBUG_RPM(trans, "Q: %d first tx - take ref\n", q->id); 2345 iwl_trans_pcie_ref(trans); 2346 } 2347 2348 /* Tell device the write index *just past* this latest filled TFD */ 2349 q->write_ptr = iwl_queue_inc_wrap(q->write_ptr); 2350 if (!wait_write_ptr) 2351 iwl_pcie_txq_inc_wr_ptr(trans, txq); 2352 2353 /* 2354 * At this point the frame is "transmitted" successfully 2355 * and we will get a TX status notification eventually. 2356 */ 2357 spin_unlock(&txq->lock); 2358 return 0; 2359 out_err: 2360 spin_unlock(&txq->lock); 2361 return -1; 2362 } 2363