1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause 2 /* 3 * Copyright (C) 2003-2014, 2018-2021, 2023-2025 Intel Corporation 4 * Copyright (C) 2013-2015 Intel Mobile Communications GmbH 5 * Copyright (C) 2016-2017 Intel Deutschland GmbH 6 */ 7 #include <linux/etherdevice.h> 8 #include <linux/ieee80211.h> 9 #include <linux/dmapool.h> 10 #include <linux/slab.h> 11 #include <linux/sched.h> 12 #include <linux/tcp.h> 13 #include <net/ip6_checksum.h> 14 #include <net/tso.h> 15 16 #include "fw/api/commands.h" 17 #include "fw/api/datapath.h" 18 #include "fw/api/debug.h" 19 #include "iwl-fh.h" 20 #include "iwl-debug.h" 21 #include "iwl-csr.h" 22 #include "iwl-prph.h" 23 #include "iwl-io.h" 24 #include "iwl-scd.h" 25 #include "iwl-op-mode.h" 26 #include "internal.h" 27 #include "fw/api/tx.h" 28 #include "fw/dbg.h" 29 #include "pcie/utils.h" 30 31 /*************** DMA-QUEUE-GENERAL-FUNCTIONS ***** 32 * DMA services 33 * 34 * Theory of operation 35 * 36 * A Tx or Rx queue resides in host DRAM, and is comprised of a circular buffer 37 * of buffer descriptors, each of which points to one or more data buffers for 38 * the device to read from or fill. Driver and device exchange status of each 39 * queue via "read" and "write" pointers. Driver keeps minimum of 2 empty 40 * entries in each circular buffer, to protect against confusing empty and full 41 * queue states. 42 * 43 * The device reads or writes the data in the queues via the device's several 44 * DMA/FIFO channels. Each queue is mapped to a single DMA channel. 45 * 46 * For Tx queue, there are low mark and high mark limits. If, after queuing 47 * the packet for Tx, free space become < low mark, Tx queue stopped. When 48 * reclaiming packets (on 'tx done IRQ), if free space become > high mark, 49 * Tx queue resumed. 50 * 51 ***************************************************/ 52 53 54 int iwl_pcie_alloc_dma_ptr(struct iwl_trans *trans, 55 struct iwl_dma_ptr *ptr, size_t size) 56 { 57 if (WARN_ON(ptr->addr)) 58 return -EINVAL; 59 60 ptr->addr = dma_alloc_coherent(trans->dev, size, 61 &ptr->dma, GFP_KERNEL); 62 if (!ptr->addr) 63 return -ENOMEM; 64 ptr->size = size; 65 return 0; 66 } 67 68 void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr) 69 { 70 if (unlikely(!ptr->addr)) 71 return; 72 73 dma_free_coherent(trans->dev, ptr->size, ptr->addr, ptr->dma); 74 memset(ptr, 0, sizeof(*ptr)); 75 } 76 77 /* 78 * iwl_pcie_txq_inc_wr_ptr - Send new write index to hardware 79 */ 80 static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, 81 struct iwl_txq *txq) 82 { 83 u32 reg = 0; 84 int txq_id = txq->id; 85 86 lockdep_assert_held(&txq->lock); 87 88 /* 89 * explicitly wake up the NIC if: 90 * 1. shadow registers aren't enabled 91 * 2. NIC is woken up for CMD regardless of shadow outside this function 92 * 3. there is a chance that the NIC is asleep 93 */ 94 if (!trans->mac_cfg->base->shadow_reg_enable && 95 txq_id != trans->conf.cmd_queue && 96 test_bit(STATUS_TPOWER_PMI, &trans->status)) { 97 /* 98 * wake up nic if it's powered down ... 99 * uCode will wake up, and interrupt us again, so next 100 * time we'll skip this part. 101 */ 102 reg = iwl_read32(trans, CSR_UCODE_DRV_GP1); 103 104 if (reg & CSR_UCODE_DRV_GP1_BIT_MAC_SLEEP) { 105 IWL_DEBUG_INFO(trans, "Tx queue %d requesting wakeup, GP1 = 0x%x\n", 106 txq_id, reg); 107 iwl_set_bit(trans, CSR_GP_CNTRL, 108 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); 109 txq->need_update = true; 110 return; 111 } 112 } 113 114 /* 115 * if not in power-save mode, uCode will never sleep when we're 116 * trying to tx (during RFKILL, we're not trying to tx). 117 */ 118 IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq_id, txq->write_ptr); 119 if (!txq->block) 120 iwl_write32(trans, HBUS_TARG_WRPTR, 121 txq->write_ptr | (txq_id << 8)); 122 } 123 124 void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans) 125 { 126 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 127 int i; 128 129 for (i = 0; i < trans->mac_cfg->base->num_of_queues; i++) { 130 struct iwl_txq *txq = trans_pcie->txqs.txq[i]; 131 132 if (!test_bit(i, trans_pcie->txqs.queue_used)) 133 continue; 134 135 spin_lock_bh(&txq->lock); 136 if (txq->need_update) { 137 iwl_pcie_txq_inc_wr_ptr(trans, txq); 138 txq->need_update = false; 139 } 140 spin_unlock_bh(&txq->lock); 141 } 142 } 143 144 static inline void iwl_pcie_gen1_tfd_set_tb(struct iwl_tfd *tfd, 145 u8 idx, dma_addr_t addr, u16 len) 146 { 147 struct iwl_tfd_tb *tb = &tfd->tbs[idx]; 148 u16 hi_n_len = len << 4; 149 150 put_unaligned_le32(addr, &tb->lo); 151 hi_n_len |= iwl_get_dma_hi_addr(addr); 152 153 tb->hi_n_len = cpu_to_le16(hi_n_len); 154 155 tfd->num_tbs = idx + 1; 156 } 157 158 static inline u8 iwl_txq_gen1_tfd_get_num_tbs(struct iwl_tfd *tfd) 159 { 160 return tfd->num_tbs & 0x1f; 161 } 162 163 static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq, 164 dma_addr_t addr, u16 len, bool reset) 165 { 166 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 167 void *tfd; 168 u32 num_tbs; 169 170 tfd = (u8 *)txq->tfds + trans_pcie->txqs.tfd.size * txq->write_ptr; 171 172 if (reset) 173 memset(tfd, 0, trans_pcie->txqs.tfd.size); 174 175 num_tbs = iwl_txq_gen1_tfd_get_num_tbs(tfd); 176 177 /* Each TFD can point to a maximum max_tbs Tx buffers */ 178 if (num_tbs >= trans_pcie->txqs.tfd.max_tbs) { 179 IWL_ERR(trans, "Error can not send more than %d chunks\n", 180 trans_pcie->txqs.tfd.max_tbs); 181 return -EINVAL; 182 } 183 184 if (WARN(addr & ~IWL_TX_DMA_MASK, 185 "Unaligned address = %llx\n", (unsigned long long)addr)) 186 return -EINVAL; 187 188 iwl_pcie_gen1_tfd_set_tb(tfd, num_tbs, addr, len); 189 190 return num_tbs; 191 } 192 193 static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans) 194 { 195 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 196 197 if (!trans->mac_cfg->base->apmg_wake_up_wa) 198 return; 199 200 spin_lock(&trans_pcie->reg_lock); 201 202 if (WARN_ON(!trans_pcie->cmd_hold_nic_awake)) { 203 spin_unlock(&trans_pcie->reg_lock); 204 return; 205 } 206 207 trans_pcie->cmd_hold_nic_awake = false; 208 iwl_trans_clear_bit(trans, CSR_GP_CNTRL, 209 CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); 210 spin_unlock(&trans_pcie->reg_lock); 211 } 212 213 static void iwl_pcie_free_and_unmap_tso_page(struct iwl_trans *trans, 214 struct page *page) 215 { 216 struct iwl_tso_page_info *info = IWL_TSO_PAGE_INFO(page_address(page)); 217 218 /* Decrease internal use count and unmap/free page if needed */ 219 if (refcount_dec_and_test(&info->use_count)) { 220 dma_unmap_page(trans->dev, info->dma_addr, PAGE_SIZE, 221 DMA_TO_DEVICE); 222 223 __free_page(page); 224 } 225 } 226 227 void iwl_pcie_free_tso_pages(struct iwl_trans *trans, struct sk_buff *skb, 228 struct iwl_cmd_meta *cmd_meta) 229 { 230 struct page **page_ptr; 231 struct page *next; 232 233 page_ptr = (void *)((u8 *)skb->cb + trans->conf.cb_data_offs); 234 next = *page_ptr; 235 *page_ptr = NULL; 236 237 while (next) { 238 struct iwl_tso_page_info *info; 239 struct page *tmp = next; 240 241 info = IWL_TSO_PAGE_INFO(page_address(next)); 242 next = info->next; 243 244 /* Unmap the scatter gather list that is on the last page */ 245 if (!next && cmd_meta->sg_offset) { 246 struct sg_table *sgt; 247 248 sgt = (void *)((u8 *)page_address(tmp) + 249 cmd_meta->sg_offset); 250 251 dma_unmap_sgtable(trans->dev, sgt, DMA_TO_DEVICE, 0); 252 } 253 254 iwl_pcie_free_and_unmap_tso_page(trans, tmp); 255 } 256 } 257 258 static inline dma_addr_t 259 iwl_txq_gen1_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx) 260 { 261 struct iwl_tfd_tb *tb = &tfd->tbs[idx]; 262 dma_addr_t addr; 263 dma_addr_t hi_len; 264 265 addr = get_unaligned_le32(&tb->lo); 266 267 if (sizeof(dma_addr_t) <= sizeof(u32)) 268 return addr; 269 270 hi_len = le16_to_cpu(tb->hi_n_len) & 0xF; 271 272 /* 273 * shift by 16 twice to avoid warnings on 32-bit 274 * (where this code never runs anyway due to the 275 * if statement above) 276 */ 277 return addr | ((hi_len << 16) << 16); 278 } 279 280 static void iwl_txq_set_tfd_invalid_gen1(struct iwl_trans *trans, 281 struct iwl_tfd *tfd) 282 { 283 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 284 285 tfd->num_tbs = 0; 286 287 iwl_pcie_gen1_tfd_set_tb(tfd, 0, trans_pcie->invalid_tx_cmd.dma, 288 trans_pcie->invalid_tx_cmd.size); 289 } 290 291 static void iwl_txq_gen1_tfd_unmap(struct iwl_trans *trans, 292 struct iwl_cmd_meta *meta, 293 struct iwl_txq *txq, int index) 294 { 295 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 296 int i, num_tbs; 297 struct iwl_tfd *tfd = iwl_txq_get_tfd(trans, txq, index); 298 299 /* Sanity check on number of chunks */ 300 num_tbs = iwl_txq_gen1_tfd_get_num_tbs(tfd); 301 302 if (num_tbs > trans_pcie->txqs.tfd.max_tbs) { 303 IWL_ERR(trans, "Too many chunks: %i\n", num_tbs); 304 /* @todo issue fatal error, it is quite serious situation */ 305 return; 306 } 307 308 /* TB1 is mapped directly, the rest is the TSO page and SG list. */ 309 if (meta->sg_offset) 310 num_tbs = 2; 311 312 /* first TB is never freed - it's the bidirectional DMA data */ 313 314 for (i = 1; i < num_tbs; i++) { 315 if (meta->tbs & BIT(i)) 316 dma_unmap_page(trans->dev, 317 iwl_txq_gen1_tfd_tb_get_addr(tfd, i), 318 iwl_txq_gen1_tfd_tb_get_len(trans, 319 tfd, i), 320 DMA_TO_DEVICE); 321 else 322 dma_unmap_single(trans->dev, 323 iwl_txq_gen1_tfd_tb_get_addr(tfd, i), 324 iwl_txq_gen1_tfd_tb_get_len(trans, 325 tfd, i), 326 DMA_TO_DEVICE); 327 } 328 329 meta->tbs = 0; 330 331 iwl_txq_set_tfd_invalid_gen1(trans, tfd); 332 } 333 334 /** 335 * iwl_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr] 336 * @trans: transport private data 337 * @txq: tx queue 338 * @read_ptr: the TXQ read_ptr to free 339 * 340 * Does NOT advance any TFD circular buffer read/write indexes 341 * Does NOT free the TFD itself (which is within circular buffer) 342 */ 343 static void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq, 344 int read_ptr) 345 { 346 /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and 347 * idx is bounded by n_window 348 */ 349 int idx = iwl_txq_get_cmd_index(txq, read_ptr); 350 struct sk_buff *skb; 351 352 lockdep_assert_held(&txq->reclaim_lock); 353 354 if (!txq->entries) 355 return; 356 357 /* We have only q->n_window txq->entries, but we use 358 * TFD_QUEUE_SIZE_MAX tfds 359 */ 360 if (trans->mac_cfg->gen2) 361 iwl_txq_gen2_tfd_unmap(trans, &txq->entries[idx].meta, 362 iwl_txq_get_tfd(trans, txq, read_ptr)); 363 else 364 iwl_txq_gen1_tfd_unmap(trans, &txq->entries[idx].meta, 365 txq, read_ptr); 366 367 /* free SKB */ 368 skb = txq->entries[idx].skb; 369 370 /* Can be called from irqs-disabled context 371 * If skb is not NULL, it means that the whole queue is being 372 * freed and that the queue is not empty - free the skb 373 */ 374 if (skb) { 375 iwl_op_mode_free_skb(trans->op_mode, skb); 376 txq->entries[idx].skb = NULL; 377 } 378 } 379 380 /* 381 * iwl_pcie_txq_unmap - Unmap any remaining DMA mappings and free skb's 382 */ 383 static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id) 384 { 385 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 386 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 387 388 if (!txq) { 389 IWL_ERR(trans, "Trying to free a queue that wasn't allocated?\n"); 390 return; 391 } 392 393 spin_lock_bh(&txq->reclaim_lock); 394 spin_lock(&txq->lock); 395 while (txq->write_ptr != txq->read_ptr) { 396 IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n", 397 txq_id, txq->read_ptr); 398 399 if (txq_id != trans->conf.cmd_queue) { 400 struct sk_buff *skb = txq->entries[txq->read_ptr].skb; 401 struct iwl_cmd_meta *cmd_meta = 402 &txq->entries[txq->read_ptr].meta; 403 404 if (WARN_ON_ONCE(!skb)) 405 continue; 406 407 iwl_pcie_free_tso_pages(trans, skb, cmd_meta); 408 } 409 iwl_txq_free_tfd(trans, txq, txq->read_ptr); 410 txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr); 411 412 if (txq->read_ptr == txq->write_ptr && 413 txq_id == trans->conf.cmd_queue) 414 iwl_pcie_clear_cmd_in_flight(trans); 415 } 416 417 while (!skb_queue_empty(&txq->overflow_q)) { 418 struct sk_buff *skb = __skb_dequeue(&txq->overflow_q); 419 420 iwl_op_mode_free_skb(trans->op_mode, skb); 421 } 422 423 spin_unlock(&txq->lock); 424 spin_unlock_bh(&txq->reclaim_lock); 425 426 /* just in case - this queue may have been stopped */ 427 iwl_trans_pcie_wake_queue(trans, txq); 428 } 429 430 /* 431 * iwl_pcie_txq_free - Deallocate DMA queue. 432 * @txq: Transmit queue to deallocate. 433 * 434 * Empty queue by removing and destroying all BD's. 435 * Free all buffers. 436 * 0-fill, but do not free "txq" descriptor structure. 437 */ 438 static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id) 439 { 440 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 441 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 442 struct device *dev = trans->dev; 443 int i; 444 445 if (WARN_ON(!txq)) 446 return; 447 448 iwl_pcie_txq_unmap(trans, txq_id); 449 450 /* De-alloc array of command/tx buffers */ 451 if (txq_id == trans->conf.cmd_queue) 452 for (i = 0; i < txq->n_window; i++) { 453 kfree_sensitive(txq->entries[i].cmd); 454 kfree_sensitive(txq->entries[i].free_buf); 455 } 456 457 /* De-alloc circular buffer of TFDs */ 458 if (txq->tfds) { 459 dma_free_coherent(dev, 460 trans_pcie->txqs.tfd.size * 461 trans->mac_cfg->base->max_tfd_queue_size, 462 txq->tfds, txq->dma_addr); 463 txq->dma_addr = 0; 464 txq->tfds = NULL; 465 466 dma_free_coherent(dev, 467 sizeof(*txq->first_tb_bufs) * txq->n_window, 468 txq->first_tb_bufs, txq->first_tb_dma); 469 } 470 471 kfree(txq->entries); 472 txq->entries = NULL; 473 474 timer_delete_sync(&txq->stuck_timer); 475 476 /* 0-fill queue descriptor structure */ 477 memset(txq, 0, sizeof(*txq)); 478 } 479 480 void iwl_pcie_tx_start(struct iwl_trans *trans) 481 { 482 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 483 int nq = trans->mac_cfg->base->num_of_queues; 484 int chan; 485 u32 reg_val; 486 int clear_dwords = (SCD_TRANS_TBL_OFFSET_QUEUE(nq) - 487 SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32); 488 489 /* make sure all queue are not stopped/used */ 490 memset(trans_pcie->txqs.queue_stopped, 0, 491 sizeof(trans_pcie->txqs.queue_stopped)); 492 memset(trans_pcie->txqs.queue_used, 0, 493 sizeof(trans_pcie->txqs.queue_used)); 494 495 trans_pcie->scd_base_addr = 496 iwl_read_prph(trans, SCD_SRAM_BASE_ADDR); 497 498 /* reset context data, TX status and translation data */ 499 iwl_trans_write_mem(trans, trans_pcie->scd_base_addr + 500 SCD_CONTEXT_MEM_LOWER_BOUND, 501 NULL, clear_dwords); 502 503 iwl_write_prph(trans, SCD_DRAM_BASE_ADDR, 504 trans_pcie->txqs.scd_bc_tbls.dma >> 10); 505 506 /* The chain extension of the SCD doesn't work well. This feature is 507 * enabled by default by the HW, so we need to disable it manually. 508 */ 509 if (trans->mac_cfg->base->scd_chain_ext_wa) 510 iwl_write_prph(trans, SCD_CHAINEXT_EN, 0); 511 512 iwl_trans_ac_txq_enable(trans, trans->conf.cmd_queue, 513 trans->conf.cmd_fifo, 514 IWL_DEF_WD_TIMEOUT); 515 516 /* Activate all Tx DMA/FIFO channels */ 517 iwl_scd_activate_fifos(trans); 518 519 /* Enable DMA channel */ 520 for (chan = 0; chan < FH_TCSR_CHNL_NUM; chan++) 521 iwl_write_direct32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(chan), 522 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CHNL_ENABLE | 523 FH_TCSR_TX_CONFIG_REG_VAL_DMA_CREDIT_ENABLE); 524 525 /* Update FH chicken bits */ 526 reg_val = iwl_read_direct32(trans, FH_TX_CHICKEN_BITS_REG); 527 iwl_write_direct32(trans, FH_TX_CHICKEN_BITS_REG, 528 reg_val | FH_TX_CHICKEN_BITS_SCD_AUTO_RETRY_EN); 529 530 /* Enable L1-Active */ 531 if (trans->mac_cfg->device_family < IWL_DEVICE_FAMILY_8000) 532 iwl_clear_bits_prph(trans, APMG_PCIDEV_STT_REG, 533 APMG_PCIDEV_STT_VAL_L1_ACT_DIS); 534 } 535 536 void iwl_trans_pcie_tx_reset(struct iwl_trans *trans) 537 { 538 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 539 int txq_id; 540 541 /* 542 * we should never get here in gen2 trans mode return early to avoid 543 * having invalid accesses 544 */ 545 if (WARN_ON_ONCE(trans->mac_cfg->gen2)) 546 return; 547 548 for (txq_id = 0; txq_id < trans->mac_cfg->base->num_of_queues; 549 txq_id++) { 550 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 551 if (trans->mac_cfg->gen2) 552 iwl_write_direct64(trans, 553 FH_MEM_CBBC_QUEUE(trans, txq_id), 554 txq->dma_addr); 555 else 556 iwl_write_direct32(trans, 557 FH_MEM_CBBC_QUEUE(trans, txq_id), 558 txq->dma_addr >> 8); 559 iwl_pcie_txq_unmap(trans, txq_id); 560 txq->read_ptr = 0; 561 txq->write_ptr = 0; 562 } 563 564 /* Tell NIC where to find the "keep warm" buffer */ 565 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG, 566 trans_pcie->kw.dma >> 4); 567 568 /* 569 * Send 0 as the scd_base_addr since the device may have be reset 570 * while we were in WoWLAN in which case SCD_SRAM_BASE_ADDR will 571 * contain garbage. 572 */ 573 iwl_pcie_tx_start(trans); 574 } 575 576 static void iwl_pcie_tx_stop_fh(struct iwl_trans *trans) 577 { 578 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 579 int ch, ret; 580 u32 mask = 0; 581 582 spin_lock_bh(&trans_pcie->irq_lock); 583 584 if (!iwl_trans_grab_nic_access(trans)) 585 goto out; 586 587 /* Stop each Tx DMA channel */ 588 for (ch = 0; ch < FH_TCSR_CHNL_NUM; ch++) { 589 iwl_write32(trans, FH_TCSR_CHNL_TX_CONFIG_REG(ch), 0x0); 590 mask |= FH_TSSR_TX_STATUS_REG_MSK_CHNL_IDLE(ch); 591 } 592 593 /* Wait for DMA channels to be idle */ 594 ret = iwl_poll_bits(trans, FH_TSSR_TX_STATUS_REG, mask, 5000); 595 if (ret) 596 IWL_ERR(trans, 597 "Failing on timeout while stopping DMA channel %d [0x%08x]\n", 598 ch, iwl_read32(trans, FH_TSSR_TX_STATUS_REG)); 599 600 iwl_trans_release_nic_access(trans); 601 602 out: 603 spin_unlock_bh(&trans_pcie->irq_lock); 604 } 605 606 /* 607 * iwl_pcie_tx_stop - Stop all Tx DMA channels 608 */ 609 int iwl_pcie_tx_stop(struct iwl_trans *trans) 610 { 611 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 612 int txq_id; 613 614 /* Turn off all Tx DMA fifos */ 615 iwl_scd_deactivate_fifos(trans); 616 617 /* Turn off all Tx DMA channels */ 618 iwl_pcie_tx_stop_fh(trans); 619 620 /* 621 * This function can be called before the op_mode disabled the 622 * queues. This happens when we have an rfkill interrupt. 623 * Since we stop Tx altogether - mark the queues as stopped. 624 */ 625 memset(trans_pcie->txqs.queue_stopped, 0, 626 sizeof(trans_pcie->txqs.queue_stopped)); 627 memset(trans_pcie->txqs.queue_used, 0, 628 sizeof(trans_pcie->txqs.queue_used)); 629 630 /* This can happen: start_hw, stop_device */ 631 if (!trans_pcie->txq_memory) 632 return 0; 633 634 /* Unmap DMA from host system and free skb's */ 635 for (txq_id = 0; txq_id < trans->mac_cfg->base->num_of_queues; 636 txq_id++) 637 iwl_pcie_txq_unmap(trans, txq_id); 638 639 return 0; 640 } 641 642 /* 643 * iwl_trans_tx_free - Free TXQ Context 644 * 645 * Destroy all TX DMA queues and structures 646 */ 647 void iwl_pcie_tx_free(struct iwl_trans *trans) 648 { 649 int txq_id; 650 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 651 652 memset(trans_pcie->txqs.queue_used, 0, 653 sizeof(trans_pcie->txqs.queue_used)); 654 655 /* Tx queues */ 656 if (trans_pcie->txq_memory) { 657 for (txq_id = 0; 658 txq_id < trans->mac_cfg->base->num_of_queues; 659 txq_id++) { 660 iwl_pcie_txq_free(trans, txq_id); 661 trans_pcie->txqs.txq[txq_id] = NULL; 662 } 663 } 664 665 kfree(trans_pcie->txq_memory); 666 trans_pcie->txq_memory = NULL; 667 668 iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw); 669 670 iwl_pcie_free_dma_ptr(trans, &trans_pcie->txqs.scd_bc_tbls); 671 } 672 673 void iwl_txq_log_scd_error(struct iwl_trans *trans, struct iwl_txq *txq) 674 { 675 u32 txq_id = txq->id; 676 u32 status; 677 bool active; 678 u8 fifo; 679 680 if (trans->mac_cfg->gen2) { 681 IWL_ERR(trans, "Queue %d is stuck %d %d\n", txq_id, 682 txq->read_ptr, txq->write_ptr); 683 /* TODO: access new SCD registers and dump them */ 684 return; 685 } 686 687 status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id)); 688 fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7; 689 active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE)); 690 691 IWL_ERR(trans, 692 "Queue %d is %sactive on fifo %d and stuck for %u ms. SW [%d, %d] HW [%d, %d] FH TRB=0x0%x\n", 693 txq_id, active ? "" : "in", fifo, 694 jiffies_to_msecs(txq->wd_timeout), 695 txq->read_ptr, txq->write_ptr, 696 iwl_read_prph(trans, SCD_QUEUE_RDPTR(txq_id)) & 697 (trans->mac_cfg->base->max_tfd_queue_size - 1), 698 iwl_read_prph(trans, SCD_QUEUE_WRPTR(txq_id)) & 699 (trans->mac_cfg->base->max_tfd_queue_size - 1), 700 iwl_read_direct32(trans, FH_TX_TRB_REG(fifo))); 701 } 702 703 static void iwl_txq_stuck_timer(struct timer_list *t) 704 { 705 struct iwl_txq *txq = timer_container_of(txq, t, stuck_timer); 706 struct iwl_trans *trans = txq->trans; 707 708 spin_lock(&txq->lock); 709 /* check if triggered erroneously */ 710 if (txq->read_ptr == txq->write_ptr) { 711 spin_unlock(&txq->lock); 712 return; 713 } 714 spin_unlock(&txq->lock); 715 716 iwl_txq_log_scd_error(trans, txq); 717 718 iwl_force_nmi(trans); 719 } 720 721 int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq, 722 int slots_num, bool cmd_queue) 723 { 724 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 725 size_t num_entries = trans->mac_cfg->gen2 ? 726 slots_num : trans->mac_cfg->base->max_tfd_queue_size; 727 size_t tfd_sz; 728 size_t tb0_buf_sz; 729 int i; 730 731 if (WARN_ONCE(slots_num <= 0, "Invalid slots num:%d\n", slots_num)) 732 return -EINVAL; 733 734 if (WARN_ON(txq->entries || txq->tfds)) 735 return -EINVAL; 736 737 tfd_sz = trans_pcie->txqs.tfd.size * num_entries; 738 739 timer_setup(&txq->stuck_timer, iwl_txq_stuck_timer, 0); 740 txq->trans = trans; 741 742 txq->n_window = slots_num; 743 744 txq->entries = kcalloc(slots_num, 745 sizeof(struct iwl_pcie_txq_entry), 746 GFP_KERNEL); 747 748 if (!txq->entries) 749 goto error; 750 751 if (cmd_queue) 752 for (i = 0; i < slots_num; i++) { 753 txq->entries[i].cmd = 754 kmalloc(sizeof(struct iwl_device_cmd), 755 GFP_KERNEL); 756 if (!txq->entries[i].cmd) 757 goto error; 758 } 759 760 /* Circular buffer of transmit frame descriptors (TFDs), 761 * shared with device 762 */ 763 txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz, 764 &txq->dma_addr, GFP_KERNEL); 765 if (!txq->tfds) 766 goto error; 767 768 BUILD_BUG_ON(sizeof(*txq->first_tb_bufs) != IWL_FIRST_TB_SIZE_ALIGN); 769 770 tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num; 771 772 txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz, 773 &txq->first_tb_dma, 774 GFP_KERNEL); 775 if (!txq->first_tb_bufs) 776 goto err_free_tfds; 777 778 for (i = 0; i < num_entries; i++) { 779 void *tfd = iwl_txq_get_tfd(trans, txq, i); 780 781 if (trans->mac_cfg->gen2) 782 iwl_txq_set_tfd_invalid_gen2(trans, tfd); 783 else 784 iwl_txq_set_tfd_invalid_gen1(trans, tfd); 785 } 786 787 return 0; 788 err_free_tfds: 789 dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr); 790 txq->tfds = NULL; 791 error: 792 if (txq->entries && cmd_queue) 793 for (i = 0; i < slots_num; i++) 794 kfree(txq->entries[i].cmd); 795 kfree(txq->entries); 796 txq->entries = NULL; 797 798 return -ENOMEM; 799 } 800 801 #define BC_TABLE_SIZE (sizeof(struct iwl_bc_tbl_entry) * TFD_QUEUE_BC_SIZE) 802 803 /* 804 * iwl_pcie_tx_alloc - allocate TX context 805 * Allocate all Tx DMA structures and initialize them 806 */ 807 static int iwl_pcie_tx_alloc(struct iwl_trans *trans) 808 { 809 int ret; 810 int txq_id, slots_num; 811 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 812 u16 bc_tbls_size = trans->mac_cfg->base->num_of_queues; 813 814 if (WARN_ON(trans->mac_cfg->device_family >= IWL_DEVICE_FAMILY_AX210)) 815 return -EINVAL; 816 817 bc_tbls_size *= BC_TABLE_SIZE; 818 819 /*It is not allowed to alloc twice, so warn when this happens. 820 * We cannot rely on the previous allocation, so free and fail */ 821 if (WARN_ON(trans_pcie->txq_memory)) { 822 ret = -EINVAL; 823 goto error; 824 } 825 826 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->txqs.scd_bc_tbls, 827 bc_tbls_size); 828 if (ret) { 829 IWL_ERR(trans, "Scheduler BC Table allocation failed\n"); 830 goto error; 831 } 832 833 /* Alloc keep-warm buffer */ 834 ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->kw, IWL_KW_SIZE); 835 if (ret) { 836 IWL_ERR(trans, "Keep Warm allocation failed\n"); 837 goto error; 838 } 839 840 trans_pcie->txq_memory = 841 kcalloc(trans->mac_cfg->base->num_of_queues, 842 sizeof(struct iwl_txq), GFP_KERNEL); 843 if (!trans_pcie->txq_memory) { 844 IWL_ERR(trans, "Not enough memory for txq\n"); 845 ret = -ENOMEM; 846 goto error; 847 } 848 849 /* Alloc and init all Tx queues, including the command queue (#4/#9) */ 850 for (txq_id = 0; txq_id < trans->mac_cfg->base->num_of_queues; 851 txq_id++) { 852 bool cmd_queue = (txq_id == trans->conf.cmd_queue); 853 854 if (cmd_queue) 855 slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE, 856 trans->mac_cfg->base->min_txq_size); 857 else 858 slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE, 859 trans->mac_cfg->base->min_ba_txq_size); 860 trans_pcie->txqs.txq[txq_id] = &trans_pcie->txq_memory[txq_id]; 861 ret = iwl_pcie_txq_alloc(trans, trans_pcie->txqs.txq[txq_id], 862 slots_num, cmd_queue); 863 if (ret) { 864 IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id); 865 goto error; 866 } 867 trans_pcie->txqs.txq[txq_id]->id = txq_id; 868 } 869 870 return 0; 871 872 error: 873 iwl_pcie_tx_free(trans); 874 875 return ret; 876 } 877 878 /* 879 * iwl_queue_init - Initialize queue's high/low-water and read/write indexes 880 */ 881 static int iwl_queue_init(struct iwl_txq *q, int slots_num) 882 { 883 q->n_window = slots_num; 884 885 /* slots_num must be power-of-two size, otherwise 886 * iwl_txq_get_cmd_index is broken. 887 */ 888 if (WARN_ON(!is_power_of_2(slots_num))) 889 return -EINVAL; 890 891 q->low_mark = q->n_window / 4; 892 if (q->low_mark < 4) 893 q->low_mark = 4; 894 895 q->high_mark = q->n_window / 8; 896 if (q->high_mark < 2) 897 q->high_mark = 2; 898 899 q->write_ptr = 0; 900 q->read_ptr = 0; 901 902 return 0; 903 } 904 905 int iwl_txq_init(struct iwl_trans *trans, struct iwl_txq *txq, 906 int slots_num, bool cmd_queue) 907 { 908 u32 tfd_queue_max_size = 909 trans->mac_cfg->base->max_tfd_queue_size; 910 int ret; 911 912 txq->need_update = false; 913 914 /* max_tfd_queue_size must be power-of-two size, otherwise 915 * iwl_txq_inc_wrap and iwl_txq_dec_wrap are broken. 916 */ 917 if (WARN_ONCE(tfd_queue_max_size & (tfd_queue_max_size - 1), 918 "Max tfd queue size must be a power of two, but is %d", 919 tfd_queue_max_size)) 920 return -EINVAL; 921 922 /* Initialize queue's high/low-water marks, and head/tail indexes */ 923 ret = iwl_queue_init(txq, slots_num); 924 if (ret) 925 return ret; 926 927 spin_lock_init(&txq->lock); 928 spin_lock_init(&txq->reclaim_lock); 929 930 if (cmd_queue) { 931 static struct lock_class_key iwl_txq_cmd_queue_lock_class; 932 933 lockdep_set_class(&txq->lock, &iwl_txq_cmd_queue_lock_class); 934 } 935 936 __skb_queue_head_init(&txq->overflow_q); 937 938 return 0; 939 } 940 941 int iwl_pcie_tx_init(struct iwl_trans *trans) 942 { 943 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 944 int ret; 945 int txq_id, slots_num; 946 bool alloc = false; 947 948 if (!trans_pcie->txq_memory) { 949 ret = iwl_pcie_tx_alloc(trans); 950 if (ret) 951 goto error; 952 alloc = true; 953 } 954 955 spin_lock_bh(&trans_pcie->irq_lock); 956 957 /* Turn off all Tx DMA fifos */ 958 iwl_scd_deactivate_fifos(trans); 959 960 /* Tell NIC where to find the "keep warm" buffer */ 961 iwl_write_direct32(trans, FH_KW_MEM_ADDR_REG, 962 trans_pcie->kw.dma >> 4); 963 964 spin_unlock_bh(&trans_pcie->irq_lock); 965 966 /* Alloc and init all Tx queues, including the command queue (#4/#9) */ 967 for (txq_id = 0; txq_id < trans->mac_cfg->base->num_of_queues; 968 txq_id++) { 969 bool cmd_queue = (txq_id == trans->conf.cmd_queue); 970 971 if (cmd_queue) 972 slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE, 973 trans->mac_cfg->base->min_txq_size); 974 else 975 slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE, 976 trans->mac_cfg->base->min_ba_txq_size); 977 ret = iwl_txq_init(trans, trans_pcie->txqs.txq[txq_id], slots_num, 978 cmd_queue); 979 if (ret) { 980 IWL_ERR(trans, "Tx %d queue init failed\n", txq_id); 981 goto error; 982 } 983 984 /* 985 * Tell nic where to find circular buffer of TFDs for a 986 * given Tx queue, and enable the DMA channel used for that 987 * queue. 988 * Circular buffer (TFD queue in DRAM) physical base address 989 */ 990 iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id), 991 trans_pcie->txqs.txq[txq_id]->dma_addr >> 8); 992 } 993 994 iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE); 995 if (trans->mac_cfg->base->num_of_queues > 20) 996 iwl_set_bits_prph(trans, SCD_GP_CTRL, 997 SCD_GP_CTRL_ENABLE_31_QUEUES); 998 999 return 0; 1000 error: 1001 /*Upon error, free only if we allocated something */ 1002 if (alloc) 1003 iwl_pcie_tx_free(trans); 1004 return ret; 1005 } 1006 1007 static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans, 1008 const struct iwl_host_cmd *cmd) 1009 { 1010 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1011 1012 /* Make sure the NIC is still alive in the bus */ 1013 if (test_bit(STATUS_TRANS_DEAD, &trans->status)) 1014 return -ENODEV; 1015 1016 if (!trans->mac_cfg->base->apmg_wake_up_wa) 1017 return 0; 1018 1019 /* 1020 * wake up the NIC to make sure that the firmware will see the host 1021 * command - we will let the NIC sleep once all the host commands 1022 * returned. This needs to be done only on NICs that have 1023 * apmg_wake_up_wa set (see above.) 1024 */ 1025 if (!_iwl_trans_pcie_grab_nic_access(trans, false)) 1026 return -EIO; 1027 1028 /* 1029 * In iwl_trans_grab_nic_access(), we've acquired the reg_lock. 1030 * There, we also returned immediately if cmd_hold_nic_awake is 1031 * already true, so it's OK to unconditionally set it to true. 1032 */ 1033 trans_pcie->cmd_hold_nic_awake = true; 1034 spin_unlock(&trans_pcie->reg_lock); 1035 1036 return 0; 1037 } 1038 1039 static void iwl_txq_progress(struct iwl_txq *txq) 1040 { 1041 lockdep_assert_held(&txq->lock); 1042 1043 if (!txq->wd_timeout) 1044 return; 1045 1046 /* 1047 * station is asleep and we send data - that must 1048 * be uAPSD or PS-Poll. Don't rearm the timer. 1049 */ 1050 if (txq->frozen) 1051 return; 1052 1053 /* 1054 * if empty delete timer, otherwise move timer forward 1055 * since we're making progress on this queue 1056 */ 1057 if (txq->read_ptr == txq->write_ptr) 1058 timer_delete(&txq->stuck_timer); 1059 else 1060 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); 1061 } 1062 1063 static inline bool iwl_txq_used(const struct iwl_txq *q, int i, 1064 int read_ptr, int write_ptr) 1065 { 1066 int index = iwl_txq_get_cmd_index(q, i); 1067 int r = iwl_txq_get_cmd_index(q, read_ptr); 1068 int w = iwl_txq_get_cmd_index(q, write_ptr); 1069 1070 return w >= r ? 1071 (index >= r && index < w) : 1072 !(index < r && index >= w); 1073 } 1074 1075 /* 1076 * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd 1077 * 1078 * When FW advances 'R' index, all entries between old and new 'R' index 1079 * need to be reclaimed. As result, some free space forms. If there is 1080 * enough free space (> low mark), wake the stack that feeds us. 1081 */ 1082 static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx) 1083 { 1084 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1085 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 1086 int nfreed = 0; 1087 u16 r; 1088 1089 lockdep_assert_held(&txq->lock); 1090 1091 idx = iwl_txq_get_cmd_index(txq, idx); 1092 r = iwl_txq_get_cmd_index(txq, txq->read_ptr); 1093 1094 if (idx >= trans->mac_cfg->base->max_tfd_queue_size || 1095 (!iwl_txq_used(txq, idx, txq->read_ptr, txq->write_ptr))) { 1096 WARN_ONCE(test_bit(txq_id, trans_pcie->txqs.queue_used), 1097 "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n", 1098 __func__, txq_id, idx, 1099 trans->mac_cfg->base->max_tfd_queue_size, 1100 txq->write_ptr, txq->read_ptr); 1101 return; 1102 } 1103 1104 for (idx = iwl_txq_inc_wrap(trans, idx); r != idx; 1105 r = iwl_txq_inc_wrap(trans, r)) { 1106 txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr); 1107 1108 if (nfreed++ > 0) { 1109 IWL_ERR(trans, "HCMD skipped: index (%d) %d %d\n", 1110 idx, txq->write_ptr, r); 1111 iwl_force_nmi(trans); 1112 } 1113 } 1114 1115 if (txq->read_ptr == txq->write_ptr) 1116 iwl_pcie_clear_cmd_in_flight(trans); 1117 1118 iwl_txq_progress(txq); 1119 } 1120 1121 static int iwl_pcie_txq_set_ratid_map(struct iwl_trans *trans, u16 ra_tid, 1122 u16 txq_id) 1123 { 1124 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1125 u32 tbl_dw_addr; 1126 u32 tbl_dw; 1127 u16 scd_q2ratid; 1128 1129 scd_q2ratid = ra_tid & SCD_QUEUE_RA_TID_MAP_RATID_MSK; 1130 1131 tbl_dw_addr = trans_pcie->scd_base_addr + 1132 SCD_TRANS_TBL_OFFSET_QUEUE(txq_id); 1133 1134 tbl_dw = iwl_trans_read_mem32(trans, tbl_dw_addr); 1135 1136 if (txq_id & 0x1) 1137 tbl_dw = (scd_q2ratid << 16) | (tbl_dw & 0x0000FFFF); 1138 else 1139 tbl_dw = scd_q2ratid | (tbl_dw & 0xFFFF0000); 1140 1141 iwl_trans_write_mem32(trans, tbl_dw_addr, tbl_dw); 1142 1143 return 0; 1144 } 1145 1146 /* Receiver address (actually, Rx station's index into station table), 1147 * combined with Traffic ID (QOS priority), in format used by Tx Scheduler */ 1148 #define BUILD_RAxTID(sta_id, tid) (((sta_id) << 4) + (tid)) 1149 1150 bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, 1151 const struct iwl_trans_txq_scd_cfg *cfg, 1152 unsigned int wdg_timeout) 1153 { 1154 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1155 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 1156 int fifo = -1; 1157 bool scd_bug = false; 1158 1159 if (test_and_set_bit(txq_id, trans_pcie->txqs.queue_used)) 1160 WARN_ONCE(1, "queue %d already used - expect issues", txq_id); 1161 1162 txq->wd_timeout = msecs_to_jiffies(wdg_timeout); 1163 1164 if (cfg) { 1165 fifo = cfg->fifo; 1166 1167 /* Disable the scheduler prior configuring the cmd queue */ 1168 if (txq_id == trans->conf.cmd_queue && 1169 trans->conf.scd_set_active) 1170 iwl_scd_enable_set_active(trans, 0); 1171 1172 /* Stop this Tx queue before configuring it */ 1173 iwl_scd_txq_set_inactive(trans, txq_id); 1174 1175 /* Set this queue as a chain-building queue unless it is CMD */ 1176 if (txq_id != trans->conf.cmd_queue) 1177 iwl_scd_txq_set_chain(trans, txq_id); 1178 1179 if (cfg->aggregate) { 1180 u16 ra_tid = BUILD_RAxTID(cfg->sta_id, cfg->tid); 1181 1182 /* Map receiver-address / traffic-ID to this queue */ 1183 iwl_pcie_txq_set_ratid_map(trans, ra_tid, txq_id); 1184 1185 /* enable aggregations for the queue */ 1186 iwl_scd_txq_enable_agg(trans, txq_id); 1187 txq->ampdu = true; 1188 } else { 1189 /* 1190 * disable aggregations for the queue, this will also 1191 * make the ra_tid mapping configuration irrelevant 1192 * since it is now a non-AGG queue. 1193 */ 1194 iwl_scd_txq_disable_agg(trans, txq_id); 1195 1196 ssn = txq->read_ptr; 1197 } 1198 } else { 1199 /* 1200 * If we need to move the SCD write pointer by steps of 1201 * 0x40, 0x80 or 0xc0, it gets stuck. Avoids this and let 1202 * the op_mode know by returning true later. 1203 * Do this only in case cfg is NULL since this trick can 1204 * be done only if we have DQA enabled which is true for mvm 1205 * only. And mvm never sets a cfg pointer. 1206 * This is really ugly, but this is the easiest way out for 1207 * this sad hardware issue. 1208 * This bug has been fixed on devices 9000 and up. 1209 */ 1210 scd_bug = !trans->mac_cfg->mq_rx_supported && 1211 !((ssn - txq->write_ptr) & 0x3f) && 1212 (ssn != txq->write_ptr); 1213 if (scd_bug) 1214 ssn++; 1215 } 1216 1217 /* Place first TFD at index corresponding to start sequence number. 1218 * Assumes that ssn_idx is valid (!= 0xFFF) */ 1219 txq->read_ptr = (ssn & 0xff); 1220 txq->write_ptr = (ssn & 0xff); 1221 iwl_write_direct32(trans, HBUS_TARG_WRPTR, 1222 (ssn & 0xff) | (txq_id << 8)); 1223 1224 if (cfg) { 1225 u8 frame_limit = cfg->frame_limit; 1226 1227 iwl_write_prph(trans, SCD_QUEUE_RDPTR(txq_id), ssn); 1228 1229 /* Set up Tx window size and frame limit for this queue */ 1230 iwl_trans_write_mem32(trans, trans_pcie->scd_base_addr + 1231 SCD_CONTEXT_QUEUE_OFFSET(txq_id), 0); 1232 iwl_trans_write_mem32(trans, 1233 trans_pcie->scd_base_addr + 1234 SCD_CONTEXT_QUEUE_OFFSET(txq_id) + sizeof(u32), 1235 SCD_QUEUE_CTX_REG2_VAL(WIN_SIZE, frame_limit) | 1236 SCD_QUEUE_CTX_REG2_VAL(FRAME_LIMIT, frame_limit)); 1237 1238 /* Set up status area in SRAM, map to Tx DMA/FIFO, activate */ 1239 iwl_write_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id), 1240 (1 << SCD_QUEUE_STTS_REG_POS_ACTIVE) | 1241 (cfg->fifo << SCD_QUEUE_STTS_REG_POS_TXF) | 1242 (1 << SCD_QUEUE_STTS_REG_POS_WSL) | 1243 SCD_QUEUE_STTS_REG_MSK); 1244 1245 /* enable the scheduler for this queue (only) */ 1246 if (txq_id == trans->conf.cmd_queue && 1247 trans->conf.scd_set_active) 1248 iwl_scd_enable_set_active(trans, BIT(txq_id)); 1249 1250 IWL_DEBUG_TX_QUEUES(trans, 1251 "Activate queue %d on FIFO %d WrPtr: %d\n", 1252 txq_id, fifo, ssn & 0xff); 1253 } else { 1254 IWL_DEBUG_TX_QUEUES(trans, 1255 "Activate queue %d WrPtr: %d\n", 1256 txq_id, ssn & 0xff); 1257 } 1258 1259 return scd_bug; 1260 } 1261 1262 void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id, 1263 bool shared_mode) 1264 { 1265 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1266 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 1267 1268 txq->ampdu = !shared_mode; 1269 } 1270 1271 void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id, 1272 bool configure_scd) 1273 { 1274 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1275 u32 stts_addr = trans_pcie->scd_base_addr + 1276 SCD_TX_STTS_QUEUE_OFFSET(txq_id); 1277 static const u32 zero_val[4] = {}; 1278 1279 trans_pcie->txqs.txq[txq_id]->frozen_expiry_remainder = 0; 1280 trans_pcie->txqs.txq[txq_id]->frozen = false; 1281 1282 /* 1283 * Upon HW Rfkill - we stop the device, and then stop the queues 1284 * in the op_mode. Just for the sake of the simplicity of the op_mode, 1285 * allow the op_mode to call txq_disable after it already called 1286 * stop_device. 1287 */ 1288 if (!test_and_clear_bit(txq_id, trans_pcie->txqs.queue_used)) { 1289 WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status), 1290 "queue %d not used", txq_id); 1291 return; 1292 } 1293 1294 if (configure_scd) { 1295 iwl_scd_txq_set_inactive(trans, txq_id); 1296 1297 iwl_trans_write_mem(trans, stts_addr, (const void *)zero_val, 1298 ARRAY_SIZE(zero_val)); 1299 } 1300 1301 iwl_pcie_txq_unmap(trans, txq_id); 1302 trans_pcie->txqs.txq[txq_id]->ampdu = false; 1303 1304 IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id); 1305 } 1306 1307 /*************** HOST COMMAND QUEUE FUNCTIONS *****/ 1308 1309 static void iwl_trans_pcie_block_txq_ptrs(struct iwl_trans *trans, bool block) 1310 { 1311 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1312 int i; 1313 1314 for (i = 0; i < trans->mac_cfg->base->num_of_queues; i++) { 1315 struct iwl_txq *txq = trans_pcie->txqs.txq[i]; 1316 1317 if (i == trans->conf.cmd_queue) 1318 continue; 1319 1320 /* we skip the command queue (obviously) so it's OK to nest */ 1321 spin_lock_nested(&txq->lock, 1); 1322 1323 if (!block && !(WARN_ON_ONCE(!txq->block))) { 1324 txq->block--; 1325 if (!txq->block) { 1326 iwl_write32(trans, HBUS_TARG_WRPTR, 1327 txq->write_ptr | (i << 8)); 1328 } 1329 } else if (block) { 1330 txq->block++; 1331 } 1332 1333 spin_unlock(&txq->lock); 1334 } 1335 } 1336 1337 /* 1338 * iwl_pcie_enqueue_hcmd - enqueue a uCode command 1339 * @priv: device private data point 1340 * @cmd: a pointer to the ucode command structure 1341 * 1342 * The function returns < 0 values to indicate the operation 1343 * failed. On success, it returns the index (>= 0) of command in the 1344 * command queue. 1345 */ 1346 int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, 1347 struct iwl_host_cmd *cmd) 1348 { 1349 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1350 struct iwl_txq *txq = trans_pcie->txqs.txq[trans->conf.cmd_queue]; 1351 struct iwl_device_cmd *out_cmd; 1352 struct iwl_cmd_meta *out_meta; 1353 void *dup_buf = NULL; 1354 dma_addr_t phys_addr; 1355 int idx; 1356 u16 copy_size, cmd_size, tb0_size; 1357 bool had_nocopy = false; 1358 u8 group_id = iwl_cmd_groupid(cmd->id); 1359 int i, ret; 1360 u32 cmd_pos; 1361 const u8 *cmddata[IWL_MAX_CMD_TBS_PER_TFD]; 1362 u16 cmdlen[IWL_MAX_CMD_TBS_PER_TFD]; 1363 unsigned long flags; 1364 1365 if (WARN(!trans->conf.wide_cmd_header && 1366 group_id > IWL_ALWAYS_LONG_GROUP, 1367 "unsupported wide command %#x\n", cmd->id)) 1368 return -EINVAL; 1369 1370 if (group_id != 0) { 1371 copy_size = sizeof(struct iwl_cmd_header_wide); 1372 cmd_size = sizeof(struct iwl_cmd_header_wide); 1373 } else { 1374 copy_size = sizeof(struct iwl_cmd_header); 1375 cmd_size = sizeof(struct iwl_cmd_header); 1376 } 1377 1378 /* need one for the header if the first is NOCOPY */ 1379 BUILD_BUG_ON(IWL_MAX_CMD_TBS_PER_TFD > IWL_NUM_OF_TBS - 1); 1380 1381 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 1382 cmddata[i] = cmd->data[i]; 1383 cmdlen[i] = cmd->len[i]; 1384 1385 if (!cmd->len[i]) 1386 continue; 1387 1388 /* need at least IWL_FIRST_TB_SIZE copied */ 1389 if (copy_size < IWL_FIRST_TB_SIZE) { 1390 int copy = IWL_FIRST_TB_SIZE - copy_size; 1391 1392 if (copy > cmdlen[i]) 1393 copy = cmdlen[i]; 1394 cmdlen[i] -= copy; 1395 cmddata[i] += copy; 1396 copy_size += copy; 1397 } 1398 1399 if (cmd->dataflags[i] & IWL_HCMD_DFL_NOCOPY) { 1400 had_nocopy = true; 1401 if (WARN_ON(cmd->dataflags[i] & IWL_HCMD_DFL_DUP)) { 1402 idx = -EINVAL; 1403 goto free_dup_buf; 1404 } 1405 } else if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) { 1406 /* 1407 * This is also a chunk that isn't copied 1408 * to the static buffer so set had_nocopy. 1409 */ 1410 had_nocopy = true; 1411 1412 /* only allowed once */ 1413 if (WARN_ON(dup_buf)) { 1414 idx = -EINVAL; 1415 goto free_dup_buf; 1416 } 1417 1418 dup_buf = kmemdup(cmddata[i], cmdlen[i], 1419 GFP_ATOMIC); 1420 if (!dup_buf) 1421 return -ENOMEM; 1422 } else { 1423 /* NOCOPY must not be followed by normal! */ 1424 if (WARN_ON(had_nocopy)) { 1425 idx = -EINVAL; 1426 goto free_dup_buf; 1427 } 1428 copy_size += cmdlen[i]; 1429 } 1430 cmd_size += cmd->len[i]; 1431 } 1432 1433 /* 1434 * If any of the command structures end up being larger than 1435 * the TFD_MAX_PAYLOAD_SIZE and they aren't dynamically 1436 * allocated into separate TFDs, then we will need to 1437 * increase the size of the buffers. 1438 */ 1439 if (WARN(copy_size > TFD_MAX_PAYLOAD_SIZE, 1440 "Command %s (%#x) is too large (%d bytes)\n", 1441 iwl_get_cmd_string(trans, cmd->id), 1442 cmd->id, copy_size)) { 1443 idx = -EINVAL; 1444 goto free_dup_buf; 1445 } 1446 1447 spin_lock_irqsave(&txq->lock, flags); 1448 1449 if (iwl_txq_space(trans, txq) < ((cmd->flags & CMD_ASYNC) ? 2 : 1)) { 1450 spin_unlock_irqrestore(&txq->lock, flags); 1451 1452 IWL_ERR(trans, "No space in command queue\n"); 1453 iwl_op_mode_nic_error(trans->op_mode, 1454 IWL_ERR_TYPE_CMD_QUEUE_FULL); 1455 iwl_trans_schedule_reset(trans, IWL_ERR_TYPE_CMD_QUEUE_FULL); 1456 idx = -ENOSPC; 1457 goto free_dup_buf; 1458 } 1459 1460 idx = iwl_txq_get_cmd_index(txq, txq->write_ptr); 1461 out_cmd = txq->entries[idx].cmd; 1462 out_meta = &txq->entries[idx].meta; 1463 1464 /* re-initialize, this also marks the SG list as unused */ 1465 memset(out_meta, 0, sizeof(*out_meta)); 1466 if (cmd->flags & CMD_WANT_SKB) 1467 out_meta->source = cmd; 1468 1469 /* set up the header */ 1470 if (group_id != 0) { 1471 out_cmd->hdr_wide.cmd = iwl_cmd_opcode(cmd->id); 1472 out_cmd->hdr_wide.group_id = group_id; 1473 out_cmd->hdr_wide.version = iwl_cmd_version(cmd->id); 1474 out_cmd->hdr_wide.length = 1475 cpu_to_le16(cmd_size - 1476 sizeof(struct iwl_cmd_header_wide)); 1477 out_cmd->hdr_wide.reserved = 0; 1478 out_cmd->hdr_wide.sequence = 1479 cpu_to_le16(QUEUE_TO_SEQ(trans->conf.cmd_queue) | 1480 INDEX_TO_SEQ(txq->write_ptr)); 1481 1482 cmd_pos = sizeof(struct iwl_cmd_header_wide); 1483 copy_size = sizeof(struct iwl_cmd_header_wide); 1484 } else { 1485 out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id); 1486 out_cmd->hdr.sequence = 1487 cpu_to_le16(QUEUE_TO_SEQ(trans->conf.cmd_queue) | 1488 INDEX_TO_SEQ(txq->write_ptr)); 1489 out_cmd->hdr.group_id = 0; 1490 1491 cmd_pos = sizeof(struct iwl_cmd_header); 1492 copy_size = sizeof(struct iwl_cmd_header); 1493 } 1494 1495 /* and copy the data that needs to be copied */ 1496 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 1497 int copy; 1498 1499 if (!cmd->len[i]) 1500 continue; 1501 1502 /* copy everything if not nocopy/dup */ 1503 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY | 1504 IWL_HCMD_DFL_DUP))) { 1505 copy = cmd->len[i]; 1506 1507 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy); 1508 cmd_pos += copy; 1509 copy_size += copy; 1510 continue; 1511 } 1512 1513 /* 1514 * Otherwise we need at least IWL_FIRST_TB_SIZE copied 1515 * in total (for bi-directional DMA), but copy up to what 1516 * we can fit into the payload for debug dump purposes. 1517 */ 1518 copy = min_t(int, TFD_MAX_PAYLOAD_SIZE - cmd_pos, cmd->len[i]); 1519 1520 memcpy((u8 *)out_cmd + cmd_pos, cmd->data[i], copy); 1521 cmd_pos += copy; 1522 1523 /* However, treat copy_size the proper way, we need it below */ 1524 if (copy_size < IWL_FIRST_TB_SIZE) { 1525 copy = IWL_FIRST_TB_SIZE - copy_size; 1526 1527 if (copy > cmd->len[i]) 1528 copy = cmd->len[i]; 1529 copy_size += copy; 1530 } 1531 } 1532 1533 IWL_DEBUG_HC(trans, 1534 "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n", 1535 iwl_get_cmd_string(trans, cmd->id), 1536 group_id, out_cmd->hdr.cmd, 1537 le16_to_cpu(out_cmd->hdr.sequence), 1538 cmd_size, txq->write_ptr, idx, trans->conf.cmd_queue); 1539 1540 /* start the TFD with the minimum copy bytes */ 1541 tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE); 1542 memcpy(&txq->first_tb_bufs[idx], &out_cmd->hdr, tb0_size); 1543 iwl_pcie_txq_build_tfd(trans, txq, 1544 iwl_txq_get_first_tb_dma(txq, idx), 1545 tb0_size, true); 1546 1547 /* map first command fragment, if any remains */ 1548 if (copy_size > tb0_size) { 1549 phys_addr = dma_map_single(trans->dev, 1550 ((u8 *)&out_cmd->hdr) + tb0_size, 1551 copy_size - tb0_size, 1552 DMA_TO_DEVICE); 1553 if (dma_mapping_error(trans->dev, phys_addr)) { 1554 iwl_txq_gen1_tfd_unmap(trans, out_meta, txq, 1555 txq->write_ptr); 1556 idx = -ENOMEM; 1557 goto out; 1558 } 1559 1560 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, 1561 copy_size - tb0_size, false); 1562 } 1563 1564 /* map the remaining (adjusted) nocopy/dup fragments */ 1565 for (i = 0; i < IWL_MAX_CMD_TBS_PER_TFD; i++) { 1566 void *data = (void *)(uintptr_t)cmddata[i]; 1567 1568 if (!cmdlen[i]) 1569 continue; 1570 if (!(cmd->dataflags[i] & (IWL_HCMD_DFL_NOCOPY | 1571 IWL_HCMD_DFL_DUP))) 1572 continue; 1573 if (cmd->dataflags[i] & IWL_HCMD_DFL_DUP) 1574 data = dup_buf; 1575 phys_addr = dma_map_single(trans->dev, data, 1576 cmdlen[i], DMA_TO_DEVICE); 1577 if (dma_mapping_error(trans->dev, phys_addr)) { 1578 iwl_txq_gen1_tfd_unmap(trans, out_meta, txq, 1579 txq->write_ptr); 1580 idx = -ENOMEM; 1581 goto out; 1582 } 1583 1584 iwl_pcie_txq_build_tfd(trans, txq, phys_addr, cmdlen[i], false); 1585 } 1586 1587 BUILD_BUG_ON(IWL_TFH_NUM_TBS > sizeof(out_meta->tbs) * BITS_PER_BYTE); 1588 out_meta->flags = cmd->flags; 1589 if (WARN_ON_ONCE(txq->entries[idx].free_buf)) 1590 kfree_sensitive(txq->entries[idx].free_buf); 1591 txq->entries[idx].free_buf = dup_buf; 1592 1593 trace_iwlwifi_dev_hcmd(trans->dev, cmd, cmd_size, &out_cmd->hdr_wide); 1594 1595 /* start timer if queue currently empty */ 1596 if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) 1597 mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); 1598 1599 ret = iwl_pcie_set_cmd_in_flight(trans, cmd); 1600 if (ret < 0) { 1601 idx = ret; 1602 goto out; 1603 } 1604 1605 if (cmd->flags & CMD_BLOCK_TXQS) 1606 iwl_trans_pcie_block_txq_ptrs(trans, true); 1607 1608 /* Increment and update queue's write index */ 1609 txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr); 1610 iwl_pcie_txq_inc_wr_ptr(trans, txq); 1611 1612 out: 1613 spin_unlock_irqrestore(&txq->lock, flags); 1614 free_dup_buf: 1615 if (idx < 0) 1616 kfree(dup_buf); 1617 return idx; 1618 } 1619 1620 /* 1621 * iwl_pcie_hcmd_complete - Pull unused buffers off the queue and reclaim them 1622 * @rxb: Rx buffer to reclaim 1623 */ 1624 void iwl_pcie_hcmd_complete(struct iwl_trans *trans, 1625 struct iwl_rx_cmd_buffer *rxb) 1626 { 1627 struct iwl_rx_packet *pkt = rxb_addr(rxb); 1628 u16 sequence = le16_to_cpu(pkt->hdr.sequence); 1629 u8 group_id; 1630 u32 cmd_id; 1631 int txq_id = SEQ_TO_QUEUE(sequence); 1632 int index = SEQ_TO_INDEX(sequence); 1633 int cmd_index; 1634 struct iwl_device_cmd *cmd; 1635 struct iwl_cmd_meta *meta; 1636 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1637 struct iwl_txq *txq = trans_pcie->txqs.txq[trans->conf.cmd_queue]; 1638 1639 /* If a Tx command is being handled and it isn't in the actual 1640 * command queue then there a command routing bug has been introduced 1641 * in the queue management code. */ 1642 if (IWL_FW_CHECK(trans, txq_id != trans->conf.cmd_queue, 1643 "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d pkt=%*phN\n", 1644 txq_id, trans->conf.cmd_queue, sequence, txq->read_ptr, 1645 txq->write_ptr, 32, pkt)) 1646 return; 1647 1648 spin_lock_bh(&txq->lock); 1649 1650 cmd_index = iwl_txq_get_cmd_index(txq, index); 1651 cmd = txq->entries[cmd_index].cmd; 1652 meta = &txq->entries[cmd_index].meta; 1653 group_id = cmd->hdr.group_id; 1654 cmd_id = WIDE_ID(group_id, cmd->hdr.cmd); 1655 1656 if (trans->mac_cfg->gen2) 1657 iwl_txq_gen2_tfd_unmap(trans, meta, 1658 iwl_txq_get_tfd(trans, txq, index)); 1659 else 1660 iwl_txq_gen1_tfd_unmap(trans, meta, txq, index); 1661 1662 /* Input error checking is done when commands are added to queue. */ 1663 if (meta->flags & CMD_WANT_SKB) { 1664 struct page *p = rxb_steal_page(rxb); 1665 1666 meta->source->resp_pkt = pkt; 1667 meta->source->_rx_page_addr = (unsigned long)page_address(p); 1668 meta->source->_rx_page_order = trans_pcie->rx_page_order; 1669 } 1670 1671 if (meta->flags & CMD_BLOCK_TXQS) 1672 iwl_trans_pcie_block_txq_ptrs(trans, false); 1673 1674 iwl_pcie_cmdq_reclaim(trans, txq_id, index); 1675 1676 if (!(meta->flags & CMD_ASYNC)) { 1677 if (!test_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status)) { 1678 IWL_WARN(trans, 1679 "HCMD_ACTIVE already clear for command %s\n", 1680 iwl_get_cmd_string(trans, cmd_id)); 1681 } 1682 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 1683 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n", 1684 iwl_get_cmd_string(trans, cmd_id)); 1685 wake_up(&trans_pcie->wait_command_queue); 1686 } 1687 1688 meta->flags = 0; 1689 1690 spin_unlock_bh(&txq->lock); 1691 } 1692 1693 static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb, 1694 struct iwl_txq *txq, u8 hdr_len, 1695 struct iwl_cmd_meta *out_meta) 1696 { 1697 u16 head_tb_len; 1698 int i; 1699 1700 /* 1701 * Set up TFD's third entry to point directly to remainder 1702 * of skb's head, if any 1703 */ 1704 head_tb_len = skb_headlen(skb) - hdr_len; 1705 1706 if (head_tb_len > 0) { 1707 dma_addr_t tb_phys = dma_map_single(trans->dev, 1708 skb->data + hdr_len, 1709 head_tb_len, DMA_TO_DEVICE); 1710 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) 1711 return -EINVAL; 1712 trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb->data + hdr_len, 1713 tb_phys, head_tb_len); 1714 iwl_pcie_txq_build_tfd(trans, txq, tb_phys, head_tb_len, false); 1715 } 1716 1717 /* set up the remaining entries to point to the data */ 1718 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1719 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1720 dma_addr_t tb_phys; 1721 int tb_idx; 1722 1723 if (!skb_frag_size(frag)) 1724 continue; 1725 1726 tb_phys = skb_frag_dma_map(trans->dev, frag, 0, 1727 skb_frag_size(frag), DMA_TO_DEVICE); 1728 1729 if (unlikely(dma_mapping_error(trans->dev, tb_phys))) 1730 return -EINVAL; 1731 trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb_frag_address(frag), 1732 tb_phys, skb_frag_size(frag)); 1733 tb_idx = iwl_pcie_txq_build_tfd(trans, txq, tb_phys, 1734 skb_frag_size(frag), false); 1735 if (tb_idx < 0) 1736 return tb_idx; 1737 1738 out_meta->tbs |= BIT(tb_idx); 1739 } 1740 1741 return 0; 1742 } 1743 1744 #ifdef CONFIG_INET 1745 static void *iwl_pcie_get_page_hdr(struct iwl_trans *trans, 1746 size_t len, struct sk_buff *skb) 1747 { 1748 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1749 struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->txqs.tso_hdr_page); 1750 struct iwl_tso_page_info *info; 1751 struct page **page_ptr; 1752 dma_addr_t phys; 1753 void *ret; 1754 1755 page_ptr = (void *)((u8 *)skb->cb + trans->conf.cb_data_offs); 1756 1757 if (WARN_ON(*page_ptr)) 1758 return NULL; 1759 1760 if (!p->page) 1761 goto alloc; 1762 1763 /* 1764 * Check if there's enough room on this page 1765 * 1766 * Note that we put a page chaining pointer *last* in the 1767 * page - we need it somewhere, and if it's there then we 1768 * avoid DMA mapping the last bits of the page which may 1769 * trigger the 32-bit boundary hardware bug. 1770 * 1771 * (see also get_workaround_page() in tx-gen2.c) 1772 */ 1773 if (((unsigned long)p->pos & ~PAGE_MASK) + len < IWL_TSO_PAGE_DATA_SIZE) { 1774 info = IWL_TSO_PAGE_INFO(page_address(p->page)); 1775 goto out; 1776 } 1777 1778 /* We don't have enough room on this page, get a new one. */ 1779 iwl_pcie_free_and_unmap_tso_page(trans, p->page); 1780 1781 alloc: 1782 p->page = alloc_page(GFP_ATOMIC); 1783 if (!p->page) 1784 return NULL; 1785 p->pos = page_address(p->page); 1786 1787 info = IWL_TSO_PAGE_INFO(page_address(p->page)); 1788 1789 /* set the chaining pointer to NULL */ 1790 info->next = NULL; 1791 1792 /* Create a DMA mapping for the page */ 1793 phys = dma_map_page_attrs(trans->dev, p->page, 0, PAGE_SIZE, 1794 DMA_TO_DEVICE, DMA_ATTR_SKIP_CPU_SYNC); 1795 if (unlikely(dma_mapping_error(trans->dev, phys))) { 1796 __free_page(p->page); 1797 p->page = NULL; 1798 1799 return NULL; 1800 } 1801 1802 /* Store physical address and set use count */ 1803 info->dma_addr = phys; 1804 refcount_set(&info->use_count, 1); 1805 out: 1806 *page_ptr = p->page; 1807 /* Return an internal reference for the caller */ 1808 refcount_inc(&info->use_count); 1809 ret = p->pos; 1810 p->pos += len; 1811 1812 return ret; 1813 } 1814 1815 /** 1816 * iwl_pcie_get_sgt_tb_phys - Find TB address in mapped SG list 1817 * @sgt: scatter gather table 1818 * @offset: Offset into the mapped memory (i.e. SKB payload data) 1819 * @len: Length of the area 1820 * 1821 * Find the DMA address that corresponds to the SKB payload data at the 1822 * position given by @offset. 1823 * 1824 * Returns: Address for TB entry 1825 */ 1826 dma_addr_t iwl_pcie_get_sgt_tb_phys(struct sg_table *sgt, unsigned int offset, 1827 unsigned int len) 1828 { 1829 struct scatterlist *sg; 1830 unsigned int sg_offset = 0; 1831 int i; 1832 1833 /* 1834 * Search the mapped DMA areas in the SG for the area that contains the 1835 * data at offset with the given length. 1836 */ 1837 for_each_sgtable_dma_sg(sgt, sg, i) { 1838 if (offset >= sg_offset && 1839 offset + len <= sg_offset + sg_dma_len(sg)) 1840 return sg_dma_address(sg) + offset - sg_offset; 1841 1842 sg_offset += sg_dma_len(sg); 1843 } 1844 1845 WARN_ON_ONCE(1); 1846 1847 return DMA_MAPPING_ERROR; 1848 } 1849 1850 /** 1851 * iwl_pcie_prep_tso - Prepare TSO page and SKB for sending 1852 * @trans: transport private data 1853 * @skb: the SKB to map 1854 * @cmd_meta: command meta to store the scatter list information for unmapping 1855 * @hdr: output argument for TSO headers 1856 * @hdr_room: requested length for TSO headers 1857 * @offset: offset into the data from which mapping should start 1858 * 1859 * Allocate space for a scatter gather list and TSO headers and map the SKB 1860 * using the scatter gather list. The SKB is unmapped again when the page is 1861 * free'ed again at the end of the operation. 1862 * 1863 * Returns: newly allocated and mapped scatter gather table with list 1864 */ 1865 struct sg_table *iwl_pcie_prep_tso(struct iwl_trans *trans, struct sk_buff *skb, 1866 struct iwl_cmd_meta *cmd_meta, 1867 u8 **hdr, unsigned int hdr_room, 1868 unsigned int offset) 1869 { 1870 struct sg_table *sgt; 1871 unsigned int n_segments = skb_shinfo(skb)->nr_frags + 1; 1872 int orig_nents; 1873 1874 if (WARN_ON_ONCE(skb_has_frag_list(skb))) 1875 return NULL; 1876 1877 *hdr = iwl_pcie_get_page_hdr(trans, 1878 hdr_room + __alignof__(struct sg_table) + 1879 sizeof(struct sg_table) + 1880 n_segments * sizeof(struct scatterlist), 1881 skb); 1882 if (!*hdr) 1883 return NULL; 1884 1885 sgt = (void *)PTR_ALIGN(*hdr + hdr_room, __alignof__(struct sg_table)); 1886 sgt->sgl = (void *)(sgt + 1); 1887 1888 sg_init_table(sgt->sgl, n_segments); 1889 1890 /* Only map the data, not the header (it is copied to the TSO page) */ 1891 orig_nents = skb_to_sgvec(skb, sgt->sgl, offset, skb->len - offset); 1892 if (WARN_ON_ONCE(orig_nents <= 0)) 1893 return NULL; 1894 1895 sgt->orig_nents = orig_nents; 1896 1897 /* And map the entire SKB */ 1898 if (dma_map_sgtable(trans->dev, sgt, DMA_TO_DEVICE, 0) < 0) 1899 return NULL; 1900 1901 /* Store non-zero (i.e. valid) offset for unmapping */ 1902 cmd_meta->sg_offset = (unsigned long) sgt & ~PAGE_MASK; 1903 1904 return sgt; 1905 } 1906 1907 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, 1908 struct iwl_txq *txq, u8 hdr_len, 1909 struct iwl_cmd_meta *out_meta, 1910 struct iwl_device_tx_cmd *dev_cmd, 1911 u16 tb1_len) 1912 { 1913 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 1914 struct iwl_tx_cmd_v6 *tx_cmd = (void *)dev_cmd->payload; 1915 struct ieee80211_hdr *hdr = (void *)skb->data; 1916 unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room; 1917 unsigned int mss = skb_shinfo(skb)->gso_size; 1918 unsigned int data_offset = 0; 1919 u16 length, iv_len, amsdu_pad; 1920 dma_addr_t start_hdr_phys; 1921 u8 *start_hdr, *pos_hdr; 1922 struct sg_table *sgt; 1923 struct tso_t tso; 1924 1925 /* if the packet is protected, then it must be CCMP or GCMP */ 1926 BUILD_BUG_ON(IEEE80211_CCMP_HDR_LEN != IEEE80211_GCMP_HDR_LEN); 1927 iv_len = ieee80211_has_protected(hdr->frame_control) ? 1928 IEEE80211_CCMP_HDR_LEN : 0; 1929 1930 trace_iwlwifi_dev_tx(trans->dev, skb, 1931 iwl_txq_get_tfd(trans, txq, txq->write_ptr), 1932 trans_pcie->txqs.tfd.size, 1933 &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0); 1934 1935 ip_hdrlen = skb_network_header_len(skb); 1936 snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb); 1937 total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_len; 1938 amsdu_pad = 0; 1939 1940 /* total amount of header we may need for this A-MSDU */ 1941 hdr_room = DIV_ROUND_UP(total_len, mss) * 1942 (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len; 1943 1944 /* Our device supports 9 segments at most, it will fit in 1 page */ 1945 sgt = iwl_pcie_prep_tso(trans, skb, out_meta, &start_hdr, hdr_room, 1946 snap_ip_tcp_hdrlen + hdr_len + iv_len); 1947 if (!sgt) 1948 return -ENOMEM; 1949 1950 start_hdr_phys = iwl_pcie_get_tso_page_phys(start_hdr); 1951 pos_hdr = start_hdr; 1952 memcpy(pos_hdr, skb->data + hdr_len, iv_len); 1953 pos_hdr += iv_len; 1954 1955 /* 1956 * Pull the ieee80211 header + IV to be able to use TSO core, 1957 * we will restore it for the tx_status flow. 1958 */ 1959 skb_pull(skb, hdr_len + iv_len); 1960 1961 /* 1962 * Remove the length of all the headers that we don't actually 1963 * have in the MPDU by themselves, but that we duplicate into 1964 * all the different MSDUs inside the A-MSDU. 1965 */ 1966 le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen); 1967 1968 tso_start(skb, &tso); 1969 1970 while (total_len) { 1971 /* this is the data left for this subframe */ 1972 unsigned int data_left = 1973 min_t(unsigned int, mss, total_len); 1974 unsigned int hdr_tb_len; 1975 dma_addr_t hdr_tb_phys; 1976 u8 *subf_hdrs_start = pos_hdr; 1977 1978 total_len -= data_left; 1979 1980 memset(pos_hdr, 0, amsdu_pad); 1981 pos_hdr += amsdu_pad; 1982 amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen + 1983 data_left)) & 0x3; 1984 ether_addr_copy(pos_hdr, ieee80211_get_DA(hdr)); 1985 pos_hdr += ETH_ALEN; 1986 ether_addr_copy(pos_hdr, ieee80211_get_SA(hdr)); 1987 pos_hdr += ETH_ALEN; 1988 1989 length = snap_ip_tcp_hdrlen + data_left; 1990 *((__be16 *)pos_hdr) = cpu_to_be16(length); 1991 pos_hdr += sizeof(length); 1992 1993 /* 1994 * This will copy the SNAP as well which will be considered 1995 * as MAC header. 1996 */ 1997 tso_build_hdr(skb, pos_hdr, &tso, data_left, !total_len); 1998 1999 pos_hdr += snap_ip_tcp_hdrlen; 2000 2001 hdr_tb_len = pos_hdr - start_hdr; 2002 hdr_tb_phys = iwl_pcie_get_tso_page_phys(start_hdr); 2003 2004 iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys, 2005 hdr_tb_len, false); 2006 trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr, 2007 hdr_tb_phys, hdr_tb_len); 2008 /* add this subframe's headers' length to the tx_cmd */ 2009 le16_add_cpu(&tx_cmd->len, pos_hdr - subf_hdrs_start); 2010 2011 /* prepare the start_hdr for the next subframe */ 2012 start_hdr = pos_hdr; 2013 2014 /* put the payload */ 2015 while (data_left) { 2016 unsigned int size = min_t(unsigned int, tso.size, 2017 data_left); 2018 dma_addr_t tb_phys; 2019 2020 tb_phys = iwl_pcie_get_sgt_tb_phys(sgt, data_offset, size); 2021 /* Not a real mapping error, use direct comparison */ 2022 if (unlikely(tb_phys == DMA_MAPPING_ERROR)) 2023 return -EINVAL; 2024 2025 iwl_pcie_txq_build_tfd(trans, txq, tb_phys, 2026 size, false); 2027 trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data, 2028 tb_phys, size); 2029 2030 data_left -= size; 2031 data_offset += size; 2032 tso_build_data(skb, &tso, size); 2033 } 2034 } 2035 2036 dma_sync_single_for_device(trans->dev, start_hdr_phys, hdr_room, 2037 DMA_TO_DEVICE); 2038 2039 /* re -add the WiFi header and IV */ 2040 skb_push(skb, hdr_len + iv_len); 2041 2042 return 0; 2043 } 2044 #else /* CONFIG_INET */ 2045 static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, 2046 struct iwl_txq *txq, u8 hdr_len, 2047 struct iwl_cmd_meta *out_meta, 2048 struct iwl_device_tx_cmd *dev_cmd, 2049 u16 tb1_len) 2050 { 2051 /* No A-MSDU without CONFIG_INET */ 2052 WARN_ON(1); 2053 2054 return -1; 2055 } 2056 #endif /* CONFIG_INET */ 2057 2058 #define IWL_TX_CRC_SIZE 4 2059 #define IWL_TX_DELIMITER_SIZE 4 2060 2061 /* 2062 * iwl_txq_gen1_update_byte_cnt_tbl - Set up entry in Tx byte-count array 2063 */ 2064 static void iwl_txq_gen1_update_byte_cnt_tbl(struct iwl_trans *trans, 2065 struct iwl_txq *txq, u16 byte_cnt, 2066 int num_tbs) 2067 { 2068 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2069 struct iwl_bc_tbl_entry *scd_bc_tbl; 2070 int write_ptr = txq->write_ptr; 2071 int txq_id = txq->id; 2072 u8 sec_ctl = 0; 2073 u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE; 2074 __le16 bc_ent; 2075 struct iwl_device_tx_cmd *dev_cmd = txq->entries[txq->write_ptr].cmd; 2076 struct iwl_tx_cmd_v6 *tx_cmd = (void *)dev_cmd->payload; 2077 u8 sta_id = tx_cmd->sta_id; 2078 2079 scd_bc_tbl = trans_pcie->txqs.scd_bc_tbls.addr; 2080 2081 sec_ctl = tx_cmd->sec_ctl; 2082 2083 switch (sec_ctl & TX_CMD_SEC_MSK) { 2084 case TX_CMD_SEC_CCM: 2085 len += IEEE80211_CCMP_MIC_LEN; 2086 break; 2087 case TX_CMD_SEC_TKIP: 2088 len += IEEE80211_TKIP_ICV_LEN; 2089 break; 2090 case TX_CMD_SEC_WEP: 2091 len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN; 2092 break; 2093 } 2094 2095 if (trans->mac_cfg->device_family < IWL_DEVICE_FAMILY_AX210) 2096 len = DIV_ROUND_UP(len, 4); 2097 2098 if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX)) 2099 return; 2100 2101 bc_ent = cpu_to_le16(len | (sta_id << 12)); 2102 2103 scd_bc_tbl[txq_id * BC_TABLE_SIZE + write_ptr].tfd_offset = bc_ent; 2104 2105 if (write_ptr < TFD_QUEUE_SIZE_BC_DUP) 2106 scd_bc_tbl[txq_id * BC_TABLE_SIZE + TFD_QUEUE_SIZE_MAX + write_ptr].tfd_offset = 2107 bc_ent; 2108 } 2109 2110 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, 2111 struct iwl_device_tx_cmd *dev_cmd, int txq_id) 2112 { 2113 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2114 struct ieee80211_hdr *hdr; 2115 struct iwl_tx_cmd_v6 *tx_cmd = (struct iwl_tx_cmd_v6 *)dev_cmd->payload; 2116 struct iwl_cmd_meta *out_meta; 2117 struct iwl_txq *txq; 2118 dma_addr_t tb0_phys, tb1_phys, scratch_phys; 2119 void *tb1_addr; 2120 void *tfd; 2121 u16 len, tb1_len; 2122 bool wait_write_ptr; 2123 __le16 fc; 2124 u8 hdr_len; 2125 u16 wifi_seq; 2126 bool amsdu; 2127 2128 txq = trans_pcie->txqs.txq[txq_id]; 2129 2130 if (WARN_ONCE(!test_bit(txq_id, trans_pcie->txqs.queue_used), 2131 "TX on unused queue %d\n", txq_id)) 2132 return -EINVAL; 2133 2134 if (skb_is_nonlinear(skb) && 2135 skb_shinfo(skb)->nr_frags > IWL_TRANS_PCIE_MAX_FRAGS(trans_pcie) && 2136 __skb_linearize(skb)) 2137 return -ENOMEM; 2138 2139 /* mac80211 always puts the full header into the SKB's head, 2140 * so there's no need to check if it's readable there 2141 */ 2142 hdr = (struct ieee80211_hdr *)skb->data; 2143 fc = hdr->frame_control; 2144 hdr_len = ieee80211_hdrlen(fc); 2145 2146 spin_lock(&txq->lock); 2147 2148 if (iwl_txq_space(trans, txq) < txq->high_mark) { 2149 iwl_txq_stop(trans, txq); 2150 2151 /* don't put the packet on the ring, if there is no room */ 2152 if (unlikely(iwl_txq_space(trans, txq) < 3)) { 2153 struct iwl_device_tx_cmd **dev_cmd_ptr; 2154 2155 dev_cmd_ptr = (void *)((u8 *)skb->cb + 2156 trans->conf.cb_data_offs + 2157 sizeof(void *)); 2158 2159 *dev_cmd_ptr = dev_cmd; 2160 __skb_queue_tail(&txq->overflow_q, skb); 2161 2162 spin_unlock(&txq->lock); 2163 return 0; 2164 } 2165 } 2166 2167 /* In AGG mode, the index in the ring must correspond to the WiFi 2168 * sequence number. This is a HW requirements to help the SCD to parse 2169 * the BA. 2170 * Check here that the packets are in the right place on the ring. 2171 */ 2172 wifi_seq = IEEE80211_SEQ_TO_SN(le16_to_cpu(hdr->seq_ctrl)); 2173 WARN_ONCE(txq->ampdu && 2174 (wifi_seq & 0xff) != txq->write_ptr, 2175 "Q: %d WiFi Seq %d tfdNum %d", 2176 txq_id, wifi_seq, txq->write_ptr); 2177 2178 /* Set up driver data for this TFD */ 2179 txq->entries[txq->write_ptr].skb = skb; 2180 txq->entries[txq->write_ptr].cmd = dev_cmd; 2181 2182 dev_cmd->hdr.sequence = 2183 cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) | 2184 INDEX_TO_SEQ(txq->write_ptr))); 2185 2186 tb0_phys = iwl_txq_get_first_tb_dma(txq, txq->write_ptr); 2187 scratch_phys = tb0_phys + sizeof(struct iwl_cmd_header) + 2188 offsetof(struct iwl_tx_cmd_v6, scratch); 2189 2190 tx_cmd->dram_lsb_ptr = cpu_to_le32(scratch_phys); 2191 tx_cmd->dram_msb_ptr = iwl_get_dma_hi_addr(scratch_phys); 2192 2193 /* Set up first empty entry in queue's array of Tx/cmd buffers */ 2194 out_meta = &txq->entries[txq->write_ptr].meta; 2195 memset(out_meta, 0, sizeof(*out_meta)); 2196 2197 /* 2198 * The second TB (tb1) points to the remainder of the TX command 2199 * and the 802.11 header - dword aligned size 2200 * (This calculation modifies the TX command, so do it before the 2201 * setup of the first TB) 2202 */ 2203 len = sizeof(struct iwl_tx_cmd_v6) + sizeof(struct iwl_cmd_header) + 2204 hdr_len - IWL_FIRST_TB_SIZE; 2205 /* do not align A-MSDU to dword as the subframe header aligns it */ 2206 amsdu = ieee80211_is_data_qos(fc) && 2207 (*ieee80211_get_qos_ctl(hdr) & 2208 IEEE80211_QOS_CTL_A_MSDU_PRESENT); 2209 if (!amsdu) { 2210 tb1_len = ALIGN(len, 4); 2211 /* Tell NIC about any 2-byte padding after MAC header */ 2212 if (tb1_len != len) 2213 tx_cmd->tx_flags |= cpu_to_le32(TX_CMD_FLG_MH_PAD); 2214 } else { 2215 tb1_len = len; 2216 } 2217 2218 /* 2219 * The first TB points to bi-directional DMA data, we'll 2220 * memcpy the data into it later. 2221 */ 2222 iwl_pcie_txq_build_tfd(trans, txq, tb0_phys, 2223 IWL_FIRST_TB_SIZE, true); 2224 2225 /* there must be data left over for TB1 or this code must be changed */ 2226 BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_v6) < IWL_FIRST_TB_SIZE); 2227 BUILD_BUG_ON(sizeof(struct iwl_cmd_header) + 2228 offsetofend(struct iwl_tx_cmd_v6, scratch) > 2229 IWL_FIRST_TB_SIZE); 2230 2231 /* map the data for TB1 */ 2232 tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE; 2233 tb1_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE); 2234 if (unlikely(dma_mapping_error(trans->dev, tb1_phys))) 2235 goto out_err; 2236 iwl_pcie_txq_build_tfd(trans, txq, tb1_phys, tb1_len, false); 2237 2238 trace_iwlwifi_dev_tx(trans->dev, skb, 2239 iwl_txq_get_tfd(trans, txq, txq->write_ptr), 2240 trans_pcie->txqs.tfd.size, 2241 &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 2242 hdr_len); 2243 2244 /* 2245 * If gso_size wasn't set, don't give the frame "amsdu treatment" 2246 * (adding subframes, etc.). 2247 * This can happen in some testing flows when the amsdu was already 2248 * pre-built, and we just need to send the resulting skb. 2249 */ 2250 if (amsdu && skb_shinfo(skb)->gso_size) { 2251 if (unlikely(iwl_fill_data_tbs_amsdu(trans, skb, txq, hdr_len, 2252 out_meta, dev_cmd, 2253 tb1_len))) 2254 goto out_err; 2255 } else { 2256 struct sk_buff *frag; 2257 2258 if (unlikely(iwl_fill_data_tbs(trans, skb, txq, hdr_len, 2259 out_meta))) 2260 goto out_err; 2261 2262 skb_walk_frags(skb, frag) { 2263 if (unlikely(iwl_fill_data_tbs(trans, frag, txq, 0, 2264 out_meta))) 2265 goto out_err; 2266 } 2267 } 2268 2269 /* building the A-MSDU might have changed this data, so memcpy it now */ 2270 memcpy(&txq->first_tb_bufs[txq->write_ptr], dev_cmd, IWL_FIRST_TB_SIZE); 2271 2272 tfd = iwl_txq_get_tfd(trans, txq, txq->write_ptr); 2273 /* Set up entry for this TFD in Tx byte-count array */ 2274 iwl_txq_gen1_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len), 2275 iwl_txq_gen1_tfd_get_num_tbs(tfd)); 2276 2277 wait_write_ptr = ieee80211_has_morefrags(fc); 2278 2279 /* start timer if queue currently empty */ 2280 if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) { 2281 /* 2282 * If the TXQ is active, then set the timer, if not, 2283 * set the timer in remainder so that the timer will 2284 * be armed with the right value when the station will 2285 * wake up. 2286 */ 2287 if (!txq->frozen) 2288 mod_timer(&txq->stuck_timer, 2289 jiffies + txq->wd_timeout); 2290 else 2291 txq->frozen_expiry_remainder = txq->wd_timeout; 2292 } 2293 2294 /* Tell device the write index *just past* this latest filled TFD */ 2295 txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr); 2296 if (!wait_write_ptr) 2297 iwl_pcie_txq_inc_wr_ptr(trans, txq); 2298 2299 /* 2300 * At this point the frame is "transmitted" successfully 2301 * and we will get a TX status notification eventually. 2302 */ 2303 spin_unlock(&txq->lock); 2304 return 0; 2305 out_err: 2306 iwl_txq_gen1_tfd_unmap(trans, out_meta, txq, txq->write_ptr); 2307 spin_unlock(&txq->lock); 2308 return -1; 2309 } 2310 2311 static void iwl_txq_gen1_inval_byte_cnt_tbl(struct iwl_trans *trans, 2312 struct iwl_txq *txq, 2313 int read_ptr) 2314 { 2315 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2316 struct iwl_bc_tbl_entry *scd_bc_tbl = trans_pcie->txqs.scd_bc_tbls.addr; 2317 int txq_id = txq->id; 2318 u8 sta_id = 0; 2319 __le16 bc_ent; 2320 struct iwl_device_tx_cmd *dev_cmd = txq->entries[read_ptr].cmd; 2321 struct iwl_tx_cmd_v6 *tx_cmd = (void *)dev_cmd->payload; 2322 2323 WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX); 2324 2325 if (txq_id != trans->conf.cmd_queue) 2326 sta_id = tx_cmd->sta_id; 2327 2328 bc_ent = cpu_to_le16(1 | (sta_id << 12)); 2329 2330 scd_bc_tbl[txq_id * BC_TABLE_SIZE + read_ptr].tfd_offset = bc_ent; 2331 2332 if (read_ptr < TFD_QUEUE_SIZE_BC_DUP) 2333 scd_bc_tbl[txq_id * BC_TABLE_SIZE + TFD_QUEUE_SIZE_MAX + read_ptr].tfd_offset = 2334 bc_ent; 2335 } 2336 2337 /* Frees buffers until index _not_ inclusive */ 2338 void iwl_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn, 2339 struct sk_buff_head *skbs, bool is_flush) 2340 { 2341 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2342 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 2343 int tfd_num, read_ptr, last_to_free; 2344 int txq_read_ptr, txq_write_ptr; 2345 2346 /* This function is not meant to release cmd queue*/ 2347 if (WARN_ON(txq_id == trans->conf.cmd_queue)) 2348 return; 2349 2350 if (WARN_ON(!txq)) 2351 return; 2352 2353 tfd_num = iwl_txq_get_cmd_index(txq, ssn); 2354 2355 spin_lock_bh(&txq->reclaim_lock); 2356 2357 spin_lock(&txq->lock); 2358 txq_read_ptr = txq->read_ptr; 2359 txq_write_ptr = txq->write_ptr; 2360 spin_unlock(&txq->lock); 2361 2362 /* There is nothing to do if we are flushing an empty queue */ 2363 if (is_flush && txq_write_ptr == txq_read_ptr) 2364 goto out; 2365 2366 read_ptr = iwl_txq_get_cmd_index(txq, txq_read_ptr); 2367 2368 if (!test_bit(txq_id, trans_pcie->txqs.queue_used)) { 2369 IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n", 2370 txq_id, ssn); 2371 goto out; 2372 } 2373 2374 if (read_ptr == tfd_num) 2375 goto out; 2376 2377 IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d (%d) -> %d (%d)\n", 2378 txq_id, read_ptr, txq_read_ptr, tfd_num, ssn); 2379 2380 /* Since we free until index _not_ inclusive, the one before index is 2381 * the last we will free. This one must be used 2382 */ 2383 last_to_free = iwl_txq_dec_wrap(trans, tfd_num); 2384 2385 if (!iwl_txq_used(txq, last_to_free, txq_read_ptr, txq_write_ptr)) { 2386 IWL_ERR(trans, 2387 "%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n", 2388 __func__, txq_id, last_to_free, 2389 trans->mac_cfg->base->max_tfd_queue_size, 2390 txq_write_ptr, txq_read_ptr); 2391 2392 iwl_op_mode_time_point(trans->op_mode, 2393 IWL_FW_INI_TIME_POINT_FAKE_TX, 2394 NULL); 2395 goto out; 2396 } 2397 2398 if (WARN_ON(!skb_queue_empty(skbs))) 2399 goto out; 2400 2401 for (; 2402 read_ptr != tfd_num; 2403 txq_read_ptr = iwl_txq_inc_wrap(trans, txq_read_ptr), 2404 read_ptr = iwl_txq_get_cmd_index(txq, txq_read_ptr)) { 2405 struct iwl_cmd_meta *cmd_meta = &txq->entries[read_ptr].meta; 2406 struct sk_buff *skb = txq->entries[read_ptr].skb; 2407 2408 if (WARN_ONCE(!skb, "no SKB at %d (%d) on queue %d\n", 2409 read_ptr, txq_read_ptr, txq_id)) 2410 continue; 2411 2412 iwl_pcie_free_tso_pages(trans, skb, cmd_meta); 2413 2414 __skb_queue_tail(skbs, skb); 2415 2416 txq->entries[read_ptr].skb = NULL; 2417 2418 if (!trans->mac_cfg->gen2) 2419 iwl_txq_gen1_inval_byte_cnt_tbl(trans, txq, 2420 txq_read_ptr); 2421 2422 iwl_txq_free_tfd(trans, txq, txq_read_ptr); 2423 } 2424 2425 spin_lock(&txq->lock); 2426 txq->read_ptr = txq_read_ptr; 2427 2428 iwl_txq_progress(txq); 2429 2430 if (iwl_txq_space(trans, txq) > txq->low_mark && 2431 test_bit(txq_id, trans_pcie->txqs.queue_stopped)) { 2432 struct sk_buff_head overflow_skbs; 2433 struct sk_buff *skb; 2434 2435 __skb_queue_head_init(&overflow_skbs); 2436 skb_queue_splice_init(&txq->overflow_q, 2437 is_flush ? skbs : &overflow_skbs); 2438 2439 /* 2440 * We are going to transmit from the overflow queue. 2441 * Remember this state so that wait_for_txq_empty will know we 2442 * are adding more packets to the TFD queue. It cannot rely on 2443 * the state of &txq->overflow_q, as we just emptied it, but 2444 * haven't TXed the content yet. 2445 */ 2446 txq->overflow_tx = true; 2447 2448 /* 2449 * This is tricky: we are in reclaim path and are holding 2450 * reclaim_lock, so noone will try to access the txq data 2451 * from that path. We stopped tx, so we can't have tx as well. 2452 * Bottom line, we can unlock and re-lock later. 2453 */ 2454 spin_unlock(&txq->lock); 2455 2456 while ((skb = __skb_dequeue(&overflow_skbs))) { 2457 struct iwl_device_tx_cmd *dev_cmd_ptr; 2458 2459 dev_cmd_ptr = *(void **)((u8 *)skb->cb + 2460 trans->conf.cb_data_offs + 2461 sizeof(void *)); 2462 2463 /* 2464 * Note that we can very well be overflowing again. 2465 * In that case, iwl_txq_space will be small again 2466 * and we won't wake mac80211's queue. 2467 */ 2468 iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id); 2469 } 2470 2471 if (iwl_txq_space(trans, txq) > txq->low_mark) 2472 iwl_trans_pcie_wake_queue(trans, txq); 2473 2474 spin_lock(&txq->lock); 2475 txq->overflow_tx = false; 2476 } 2477 2478 spin_unlock(&txq->lock); 2479 out: 2480 spin_unlock_bh(&txq->reclaim_lock); 2481 } 2482 2483 /* Set wr_ptr of specific device and txq */ 2484 void iwl_pcie_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr) 2485 { 2486 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2487 struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; 2488 2489 spin_lock_bh(&txq->lock); 2490 2491 txq->write_ptr = ptr; 2492 txq->read_ptr = txq->write_ptr; 2493 2494 spin_unlock_bh(&txq->lock); 2495 } 2496 2497 void iwl_pcie_freeze_txq_timer(struct iwl_trans *trans, 2498 unsigned long txqs, bool freeze) 2499 { 2500 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2501 int queue; 2502 2503 for_each_set_bit(queue, &txqs, BITS_PER_LONG) { 2504 struct iwl_txq *txq = trans_pcie->txqs.txq[queue]; 2505 unsigned long now; 2506 2507 spin_lock_bh(&txq->lock); 2508 2509 now = jiffies; 2510 2511 if (txq->frozen == freeze) 2512 goto next_queue; 2513 2514 IWL_DEBUG_TX_QUEUES(trans, "%s TXQ %d\n", 2515 freeze ? "Freezing" : "Waking", queue); 2516 2517 txq->frozen = freeze; 2518 2519 if (txq->read_ptr == txq->write_ptr) 2520 goto next_queue; 2521 2522 if (freeze) { 2523 if (unlikely(time_after(now, 2524 txq->stuck_timer.expires))) { 2525 /* 2526 * The timer should have fired, maybe it is 2527 * spinning right now on the lock. 2528 */ 2529 goto next_queue; 2530 } 2531 /* remember how long until the timer fires */ 2532 txq->frozen_expiry_remainder = 2533 txq->stuck_timer.expires - now; 2534 timer_delete(&txq->stuck_timer); 2535 goto next_queue; 2536 } 2537 2538 /* 2539 * Wake a non-empty queue -> arm timer with the 2540 * remainder before it froze 2541 */ 2542 mod_timer(&txq->stuck_timer, 2543 now + txq->frozen_expiry_remainder); 2544 2545 next_queue: 2546 spin_unlock_bh(&txq->lock); 2547 } 2548 } 2549 2550 #define HOST_COMPLETE_TIMEOUT (2 * HZ) 2551 2552 static int iwl_trans_pcie_send_hcmd_sync(struct iwl_trans *trans, 2553 struct iwl_host_cmd *cmd, 2554 const char *cmd_str) 2555 { 2556 struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); 2557 struct iwl_txq *txq = trans_pcie->txqs.txq[trans->conf.cmd_queue]; 2558 int cmd_idx; 2559 int ret; 2560 2561 IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str); 2562 2563 if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE, 2564 &trans->status), 2565 "Command %s: a command is already active!\n", cmd_str)) 2566 return -EIO; 2567 2568 IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str); 2569 2570 if (trans->mac_cfg->gen2) 2571 cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd); 2572 else 2573 cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd); 2574 2575 if (cmd_idx < 0) { 2576 ret = cmd_idx; 2577 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 2578 IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n", 2579 cmd_str, ret); 2580 return ret; 2581 } 2582 2583 ret = wait_event_timeout(trans_pcie->wait_command_queue, 2584 !test_bit(STATUS_SYNC_HCMD_ACTIVE, 2585 &trans->status), 2586 HOST_COMPLETE_TIMEOUT); 2587 if (!ret) { 2588 IWL_ERR(trans, "Error sending %s: time out after %dms.\n", 2589 cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT)); 2590 2591 IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n", 2592 txq->read_ptr, txq->write_ptr); 2593 2594 clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); 2595 IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n", 2596 cmd_str); 2597 ret = -ETIMEDOUT; 2598 2599 iwl_trans_pcie_sync_nmi(trans); 2600 goto cancel; 2601 } 2602 2603 if (test_bit(STATUS_FW_ERROR, &trans->status)) { 2604 if (!test_and_clear_bit(STATUS_SUPPRESS_CMD_ERROR_ONCE, 2605 &trans->status)) { 2606 IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str); 2607 dump_stack(); 2608 } 2609 ret = -EIO; 2610 goto cancel; 2611 } 2612 2613 if (!(cmd->flags & CMD_SEND_IN_RFKILL) && 2614 test_bit(STATUS_RFKILL_OPMODE, &trans->status)) { 2615 IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n"); 2616 ret = -ERFKILL; 2617 goto cancel; 2618 } 2619 2620 if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) { 2621 IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str); 2622 ret = -EIO; 2623 goto cancel; 2624 } 2625 2626 return 0; 2627 2628 cancel: 2629 if (cmd->flags & CMD_WANT_SKB) { 2630 /* 2631 * Cancel the CMD_WANT_SKB flag for the cmd in the 2632 * TX cmd queue. Otherwise in case the cmd comes 2633 * in later, it will possibly set an invalid 2634 * address (cmd->meta.source). 2635 */ 2636 txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB; 2637 } 2638 2639 if (cmd->resp_pkt) { 2640 iwl_free_resp(cmd); 2641 cmd->resp_pkt = NULL; 2642 } 2643 2644 return ret; 2645 } 2646 2647 int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, 2648 struct iwl_host_cmd *cmd) 2649 { 2650 const char *cmd_str = iwl_get_cmd_string(trans, cmd->id); 2651 2652 /* Make sure the NIC is still alive in the bus */ 2653 if (test_bit(STATUS_TRANS_DEAD, &trans->status)) 2654 return -ENODEV; 2655 2656 if (!(cmd->flags & CMD_SEND_IN_RFKILL) && 2657 test_bit(STATUS_RFKILL_OPMODE, &trans->status)) { 2658 IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n", 2659 cmd->id); 2660 return -ERFKILL; 2661 } 2662 2663 if (cmd->flags & CMD_ASYNC) { 2664 int ret; 2665 2666 IWL_DEBUG_INFO(trans, "Sending async command %s\n", cmd_str); 2667 2668 /* An asynchronous command can not expect an SKB to be set. */ 2669 if (WARN_ON(cmd->flags & CMD_WANT_SKB)) 2670 return -EINVAL; 2671 2672 if (trans->mac_cfg->gen2) 2673 ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd); 2674 else 2675 ret = iwl_pcie_enqueue_hcmd(trans, cmd); 2676 2677 if (ret < 0) { 2678 IWL_ERR(trans, 2679 "Error sending %s: enqueue_hcmd failed: %d\n", 2680 iwl_get_cmd_string(trans, cmd->id), ret); 2681 return ret; 2682 } 2683 return 0; 2684 } 2685 2686 return iwl_trans_pcie_send_hcmd_sync(trans, cmd, cmd_str); 2687 } 2688