1 /* 2 * Atheros CARL9170 driver 3 * 4 * 802.11 xmit & status routines 5 * 6 * Copyright 2008, Johannes Berg <johannes@sipsolutions.net> 7 * Copyright 2009, 2010, Christian Lamparter <chunkeey@googlemail.com> 8 * 9 * This program is free software; you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License as published by 11 * the Free Software Foundation; either version 2 of the License, or 12 * (at your option) any later version. 13 * 14 * This program is distributed in the hope that it will be useful, 15 * but WITHOUT ANY WARRANTY; without even the implied warranty of 16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 17 * GNU General Public License for more details. 18 * 19 * You should have received a copy of the GNU General Public License 20 * along with this program; see the file COPYING. If not, see 21 * http://www.gnu.org/licenses/. 22 * 23 * This file incorporates work covered by the following copyright and 24 * permission notice: 25 * Copyright (c) 2007-2008 Atheros Communications, Inc. 26 * 27 * Permission to use, copy, modify, and/or distribute this software for any 28 * purpose with or without fee is hereby granted, provided that the above 29 * copyright notice and this permission notice appear in all copies. 30 * 31 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 32 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 33 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 34 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 35 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 36 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 37 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 38 */ 39 40 #include <linux/init.h> 41 #include <linux/slab.h> 42 #include <linux/module.h> 43 #include <linux/etherdevice.h> 44 #include <net/mac80211.h> 45 #include "carl9170.h" 46 #include "hw.h" 47 #include "cmd.h" 48 49 static inline unsigned int __carl9170_get_queue(struct ar9170 *ar, 50 unsigned int queue) 51 { 52 if (unlikely(modparam_noht)) { 53 return queue; 54 } else { 55 /* 56 * This is just another workaround, until 57 * someone figures out how to get QoS and 58 * AMPDU to play nicely together. 59 */ 60 61 return 2; /* AC_BE */ 62 } 63 } 64 65 static inline unsigned int carl9170_get_queue(struct ar9170 *ar, 66 struct sk_buff *skb) 67 { 68 return __carl9170_get_queue(ar, skb_get_queue_mapping(skb)); 69 } 70 71 static bool is_mem_full(struct ar9170 *ar) 72 { 73 return (DIV_ROUND_UP(IEEE80211_MAX_FRAME_LEN, ar->fw.mem_block_size) > 74 atomic_read(&ar->mem_free_blocks)); 75 } 76 77 static void carl9170_tx_accounting(struct ar9170 *ar, struct sk_buff *skb) 78 { 79 int queue, i; 80 bool mem_full; 81 82 atomic_inc(&ar->tx_total_queued); 83 84 queue = skb_get_queue_mapping(skb); 85 spin_lock_bh(&ar->tx_stats_lock); 86 87 /* 88 * The driver has to accept the frame, regardless if the queue is 89 * full to the brim, or not. We have to do the queuing internally, 90 * since mac80211 assumes that a driver which can operate with 91 * aggregated frames does not reject frames for this reason. 92 */ 93 ar->tx_stats[queue].len++; 94 ar->tx_stats[queue].count++; 95 96 mem_full = is_mem_full(ar); 97 for (i = 0; i < ar->hw->queues; i++) { 98 if (mem_full || ar->tx_stats[i].len >= ar->tx_stats[i].limit) { 99 ieee80211_stop_queue(ar->hw, i); 100 ar->queue_stop_timeout[i] = jiffies; 101 } 102 } 103 104 spin_unlock_bh(&ar->tx_stats_lock); 105 } 106 107 /* needs rcu_read_lock */ 108 static struct ieee80211_sta *__carl9170_get_tx_sta(struct ar9170 *ar, 109 struct sk_buff *skb) 110 { 111 struct _carl9170_tx_superframe *super = (void *) skb->data; 112 struct ieee80211_hdr *hdr = (void *) super->frame_data; 113 struct ieee80211_vif *vif; 114 unsigned int vif_id; 115 116 vif_id = (super->s.misc & CARL9170_TX_SUPER_MISC_VIF_ID) >> 117 CARL9170_TX_SUPER_MISC_VIF_ID_S; 118 119 if (WARN_ON_ONCE(vif_id >= AR9170_MAX_VIRTUAL_MAC)) 120 return NULL; 121 122 vif = rcu_dereference(ar->vif_priv[vif_id].vif); 123 if (unlikely(!vif)) 124 return NULL; 125 126 /* 127 * Normally we should use wrappers like ieee80211_get_DA to get 128 * the correct peer ieee80211_sta. 129 * 130 * But there is a problem with indirect traffic (broadcasts, or 131 * data which is designated for other stations) in station mode. 132 * The frame will be directed to the AP for distribution and not 133 * to the actual destination. 134 */ 135 136 return ieee80211_find_sta(vif, hdr->addr1); 137 } 138 139 static void carl9170_tx_ps_unblock(struct ar9170 *ar, struct sk_buff *skb) 140 { 141 struct ieee80211_sta *sta; 142 struct carl9170_sta_info *sta_info; 143 144 rcu_read_lock(); 145 sta = __carl9170_get_tx_sta(ar, skb); 146 if (unlikely(!sta)) 147 goto out_rcu; 148 149 sta_info = (struct carl9170_sta_info *) sta->drv_priv; 150 if (atomic_dec_return(&sta_info->pending_frames) == 0) 151 ieee80211_sta_block_awake(ar->hw, sta, false); 152 153 out_rcu: 154 rcu_read_unlock(); 155 } 156 157 static void carl9170_tx_accounting_free(struct ar9170 *ar, struct sk_buff *skb) 158 { 159 int queue; 160 161 queue = skb_get_queue_mapping(skb); 162 163 spin_lock_bh(&ar->tx_stats_lock); 164 165 ar->tx_stats[queue].len--; 166 167 if (!is_mem_full(ar)) { 168 unsigned int i; 169 for (i = 0; i < ar->hw->queues; i++) { 170 if (ar->tx_stats[i].len >= CARL9170_NUM_TX_LIMIT_SOFT) 171 continue; 172 173 if (ieee80211_queue_stopped(ar->hw, i)) { 174 unsigned long tmp; 175 176 tmp = jiffies - ar->queue_stop_timeout[i]; 177 if (tmp > ar->max_queue_stop_timeout[i]) 178 ar->max_queue_stop_timeout[i] = tmp; 179 } 180 181 ieee80211_wake_queue(ar->hw, i); 182 } 183 } 184 185 spin_unlock_bh(&ar->tx_stats_lock); 186 187 if (atomic_dec_and_test(&ar->tx_total_queued)) 188 complete(&ar->tx_flush); 189 } 190 191 static int carl9170_alloc_dev_space(struct ar9170 *ar, struct sk_buff *skb) 192 { 193 struct _carl9170_tx_superframe *super = (void *) skb->data; 194 unsigned int chunks; 195 int cookie = -1; 196 197 atomic_inc(&ar->mem_allocs); 198 199 chunks = DIV_ROUND_UP(skb->len, ar->fw.mem_block_size); 200 if (unlikely(atomic_sub_return(chunks, &ar->mem_free_blocks) < 0)) { 201 atomic_add(chunks, &ar->mem_free_blocks); 202 return -ENOSPC; 203 } 204 205 spin_lock_bh(&ar->mem_lock); 206 cookie = bitmap_find_free_region(ar->mem_bitmap, ar->fw.mem_blocks, 0); 207 spin_unlock_bh(&ar->mem_lock); 208 209 if (unlikely(cookie < 0)) { 210 atomic_add(chunks, &ar->mem_free_blocks); 211 return -ENOSPC; 212 } 213 214 super = (void *) skb->data; 215 216 /* 217 * Cookie #0 serves two special purposes: 218 * 1. The firmware might use it generate BlockACK frames 219 * in responds of an incoming BlockAckReqs. 220 * 221 * 2. Prevent double-free bugs. 222 */ 223 super->s.cookie = (u8) cookie + 1; 224 return 0; 225 } 226 227 static void carl9170_release_dev_space(struct ar9170 *ar, struct sk_buff *skb) 228 { 229 struct _carl9170_tx_superframe *super = (void *) skb->data; 230 int cookie; 231 232 /* make a local copy of the cookie */ 233 cookie = super->s.cookie; 234 /* invalidate cookie */ 235 super->s.cookie = 0; 236 237 /* 238 * Do a out-of-bounds check on the cookie: 239 * 240 * * cookie "0" is reserved and won't be assigned to any 241 * out-going frame. Internally however, it is used to 242 * mark no longer/un-accounted frames and serves as a 243 * cheap way of preventing frames from being freed 244 * twice by _accident_. NB: There is a tiny race... 245 * 246 * * obviously, cookie number is limited by the amount 247 * of available memory blocks, so the number can 248 * never execeed the mem_blocks count. 249 */ 250 if (unlikely(WARN_ON_ONCE(cookie == 0) || 251 WARN_ON_ONCE(cookie > ar->fw.mem_blocks))) 252 return; 253 254 atomic_add(DIV_ROUND_UP(skb->len, ar->fw.mem_block_size), 255 &ar->mem_free_blocks); 256 257 spin_lock_bh(&ar->mem_lock); 258 bitmap_release_region(ar->mem_bitmap, cookie - 1, 0); 259 spin_unlock_bh(&ar->mem_lock); 260 } 261 262 /* Called from any context */ 263 static void carl9170_tx_release(struct kref *ref) 264 { 265 struct ar9170 *ar; 266 struct carl9170_tx_info *arinfo; 267 struct ieee80211_tx_info *txinfo; 268 struct sk_buff *skb; 269 270 arinfo = container_of(ref, struct carl9170_tx_info, ref); 271 txinfo = container_of((void *) arinfo, struct ieee80211_tx_info, 272 rate_driver_data); 273 skb = container_of((void *) txinfo, struct sk_buff, cb); 274 275 ar = arinfo->ar; 276 if (WARN_ON_ONCE(!ar)) 277 return; 278 279 BUILD_BUG_ON( 280 offsetof(struct ieee80211_tx_info, status.ack_signal) != 20); 281 282 memset(&txinfo->status.ack_signal, 0, 283 sizeof(struct ieee80211_tx_info) - 284 offsetof(struct ieee80211_tx_info, status.ack_signal)); 285 286 if (atomic_read(&ar->tx_total_queued)) 287 ar->tx_schedule = true; 288 289 if (txinfo->flags & IEEE80211_TX_CTL_AMPDU) { 290 if (!atomic_read(&ar->tx_ampdu_upload)) 291 ar->tx_ampdu_schedule = true; 292 293 if (txinfo->flags & IEEE80211_TX_STAT_AMPDU) { 294 struct _carl9170_tx_superframe *super; 295 296 super = (void *)skb->data; 297 txinfo->status.ampdu_len = super->s.rix; 298 txinfo->status.ampdu_ack_len = super->s.cnt; 299 } else if ((txinfo->flags & IEEE80211_TX_STAT_ACK) && 300 !(txinfo->flags & IEEE80211_TX_CTL_REQ_TX_STATUS)) { 301 /* 302 * drop redundant tx_status reports: 303 * 304 * 1. ampdu_ack_len of the final tx_status does 305 * include the feedback of this particular frame. 306 * 307 * 2. tx_status_irqsafe only queues up to 128 308 * tx feedback reports and discards the rest. 309 * 310 * 3. minstrel_ht is picky, it only accepts 311 * reports of frames with the TX_STATUS_AMPDU flag. 312 * 313 * 4. mac80211 is not particularly interested in 314 * feedback either [CTL_REQ_TX_STATUS not set] 315 */ 316 317 ieee80211_free_txskb(ar->hw, skb); 318 return; 319 } else { 320 /* 321 * Either the frame transmission has failed or 322 * mac80211 requested tx status. 323 */ 324 } 325 } 326 327 skb_pull(skb, sizeof(struct _carl9170_tx_superframe)); 328 ieee80211_tx_status_irqsafe(ar->hw, skb); 329 } 330 331 void carl9170_tx_get_skb(struct sk_buff *skb) 332 { 333 struct carl9170_tx_info *arinfo = (void *) 334 (IEEE80211_SKB_CB(skb))->rate_driver_data; 335 kref_get(&arinfo->ref); 336 } 337 338 int carl9170_tx_put_skb(struct sk_buff *skb) 339 { 340 struct carl9170_tx_info *arinfo = (void *) 341 (IEEE80211_SKB_CB(skb))->rate_driver_data; 342 343 return kref_put(&arinfo->ref, carl9170_tx_release); 344 } 345 346 /* Caller must hold the tid_info->lock & rcu_read_lock */ 347 static void carl9170_tx_shift_bm(struct ar9170 *ar, 348 struct carl9170_sta_tid *tid_info, u16 seq) 349 { 350 u16 off; 351 352 off = SEQ_DIFF(seq, tid_info->bsn); 353 354 if (WARN_ON_ONCE(off >= CARL9170_BAW_BITS)) 355 return; 356 357 /* 358 * Sanity check. For each MPDU we set the bit in bitmap and 359 * clear it once we received the tx_status. 360 * But if the bit is already cleared then we've been bitten 361 * by a bug. 362 */ 363 WARN_ON_ONCE(!test_and_clear_bit(off, tid_info->bitmap)); 364 365 off = SEQ_DIFF(tid_info->snx, tid_info->bsn); 366 if (WARN_ON_ONCE(off >= CARL9170_BAW_BITS)) 367 return; 368 369 if (!bitmap_empty(tid_info->bitmap, off)) 370 off = find_first_bit(tid_info->bitmap, off); 371 372 tid_info->bsn += off; 373 tid_info->bsn &= 0x0fff; 374 375 bitmap_shift_right(tid_info->bitmap, tid_info->bitmap, 376 off, CARL9170_BAW_BITS); 377 } 378 379 static void carl9170_tx_status_process_ampdu(struct ar9170 *ar, 380 struct sk_buff *skb, struct ieee80211_tx_info *txinfo) 381 { 382 struct _carl9170_tx_superframe *super = (void *) skb->data; 383 struct ieee80211_hdr *hdr = (void *) super->frame_data; 384 struct ieee80211_sta *sta; 385 struct carl9170_sta_info *sta_info; 386 struct carl9170_sta_tid *tid_info; 387 u8 tid; 388 389 if (!(txinfo->flags & IEEE80211_TX_CTL_AMPDU) || 390 txinfo->flags & IEEE80211_TX_CTL_INJECTED || 391 (!(super->f.mac_control & cpu_to_le16(AR9170_TX_MAC_AGGR)))) 392 return; 393 394 rcu_read_lock(); 395 sta = __carl9170_get_tx_sta(ar, skb); 396 if (unlikely(!sta)) 397 goto out_rcu; 398 399 tid = get_tid_h(hdr); 400 401 sta_info = (void *) sta->drv_priv; 402 tid_info = rcu_dereference(sta_info->agg[tid]); 403 if (!tid_info) 404 goto out_rcu; 405 406 spin_lock_bh(&tid_info->lock); 407 if (likely(tid_info->state >= CARL9170_TID_STATE_IDLE)) 408 carl9170_tx_shift_bm(ar, tid_info, get_seq_h(hdr)); 409 410 if (sta_info->stats[tid].clear) { 411 sta_info->stats[tid].clear = false; 412 sta_info->stats[tid].req = false; 413 sta_info->stats[tid].ampdu_len = 0; 414 sta_info->stats[tid].ampdu_ack_len = 0; 415 } 416 417 sta_info->stats[tid].ampdu_len++; 418 if (txinfo->status.rates[0].count == 1) 419 sta_info->stats[tid].ampdu_ack_len++; 420 421 if (!(txinfo->flags & IEEE80211_TX_STAT_ACK)) 422 sta_info->stats[tid].req = true; 423 424 if (super->f.mac_control & cpu_to_le16(AR9170_TX_MAC_IMM_BA)) { 425 super->s.rix = sta_info->stats[tid].ampdu_len; 426 super->s.cnt = sta_info->stats[tid].ampdu_ack_len; 427 txinfo->flags |= IEEE80211_TX_STAT_AMPDU; 428 if (sta_info->stats[tid].req) 429 txinfo->flags |= IEEE80211_TX_STAT_AMPDU_NO_BACK; 430 431 sta_info->stats[tid].clear = true; 432 } 433 spin_unlock_bh(&tid_info->lock); 434 435 out_rcu: 436 rcu_read_unlock(); 437 } 438 439 static void carl9170_tx_bar_status(struct ar9170 *ar, struct sk_buff *skb, 440 struct ieee80211_tx_info *tx_info) 441 { 442 struct _carl9170_tx_superframe *super = (void *) skb->data; 443 struct ieee80211_bar *bar = (void *) super->frame_data; 444 445 /* 446 * Unlike all other frames, the status report for BARs does 447 * not directly come from the hardware as it is incapable of 448 * matching a BA to a previously send BAR. 449 * Instead the RX-path will scan for incoming BAs and set the 450 * IEEE80211_TX_STAT_ACK if it sees one that was likely 451 * caused by a BAR from us. 452 */ 453 454 if (unlikely(ieee80211_is_back_req(bar->frame_control)) && 455 !(tx_info->flags & IEEE80211_TX_STAT_ACK)) { 456 struct carl9170_bar_list_entry *entry; 457 int queue = skb_get_queue_mapping(skb); 458 459 rcu_read_lock(); 460 list_for_each_entry_rcu(entry, &ar->bar_list[queue], list) { 461 if (entry->skb == skb) { 462 spin_lock_bh(&ar->bar_list_lock[queue]); 463 list_del_rcu(&entry->list); 464 spin_unlock_bh(&ar->bar_list_lock[queue]); 465 kfree_rcu(entry, head); 466 goto out; 467 } 468 } 469 470 WARN(1, "bar not found in %d - ra:%pM ta:%pM c:%x ssn:%x\n", 471 queue, bar->ra, bar->ta, bar->control, 472 bar->start_seq_num); 473 out: 474 rcu_read_unlock(); 475 } 476 } 477 478 void carl9170_tx_status(struct ar9170 *ar, struct sk_buff *skb, 479 const bool success) 480 { 481 struct ieee80211_tx_info *txinfo; 482 483 carl9170_tx_accounting_free(ar, skb); 484 485 txinfo = IEEE80211_SKB_CB(skb); 486 487 carl9170_tx_bar_status(ar, skb, txinfo); 488 489 if (success) 490 txinfo->flags |= IEEE80211_TX_STAT_ACK; 491 else 492 ar->tx_ack_failures++; 493 494 if (txinfo->flags & IEEE80211_TX_CTL_AMPDU) 495 carl9170_tx_status_process_ampdu(ar, skb, txinfo); 496 497 carl9170_tx_ps_unblock(ar, skb); 498 carl9170_tx_put_skb(skb); 499 } 500 501 /* This function may be called form any context */ 502 void carl9170_tx_callback(struct ar9170 *ar, struct sk_buff *skb) 503 { 504 struct ieee80211_tx_info *txinfo = IEEE80211_SKB_CB(skb); 505 506 atomic_dec(&ar->tx_total_pending); 507 508 if (txinfo->flags & IEEE80211_TX_CTL_AMPDU) 509 atomic_dec(&ar->tx_ampdu_upload); 510 511 if (carl9170_tx_put_skb(skb)) 512 tasklet_hi_schedule(&ar->usb_tasklet); 513 } 514 515 static struct sk_buff *carl9170_get_queued_skb(struct ar9170 *ar, u8 cookie, 516 struct sk_buff_head *queue) 517 { 518 struct sk_buff *skb; 519 520 spin_lock_bh(&queue->lock); 521 skb_queue_walk(queue, skb) { 522 struct _carl9170_tx_superframe *txc = (void *) skb->data; 523 524 if (txc->s.cookie != cookie) 525 continue; 526 527 __skb_unlink(skb, queue); 528 spin_unlock_bh(&queue->lock); 529 530 carl9170_release_dev_space(ar, skb); 531 return skb; 532 } 533 spin_unlock_bh(&queue->lock); 534 535 return NULL; 536 } 537 538 static void carl9170_tx_fill_rateinfo(struct ar9170 *ar, unsigned int rix, 539 unsigned int tries, struct ieee80211_tx_info *txinfo) 540 { 541 unsigned int i; 542 543 for (i = 0; i < IEEE80211_TX_MAX_RATES; i++) { 544 if (txinfo->status.rates[i].idx < 0) 545 break; 546 547 if (i == rix) { 548 txinfo->status.rates[i].count = tries; 549 i++; 550 break; 551 } 552 } 553 554 for (; i < IEEE80211_TX_MAX_RATES; i++) { 555 txinfo->status.rates[i].idx = -1; 556 txinfo->status.rates[i].count = 0; 557 } 558 } 559 560 static void carl9170_check_queue_stop_timeout(struct ar9170 *ar) 561 { 562 int i; 563 struct sk_buff *skb; 564 struct ieee80211_tx_info *txinfo; 565 struct carl9170_tx_info *arinfo; 566 bool restart = false; 567 568 for (i = 0; i < ar->hw->queues; i++) { 569 spin_lock_bh(&ar->tx_status[i].lock); 570 571 skb = skb_peek(&ar->tx_status[i]); 572 573 if (!skb) 574 goto next; 575 576 txinfo = IEEE80211_SKB_CB(skb); 577 arinfo = (void *) txinfo->rate_driver_data; 578 579 if (time_is_before_jiffies(arinfo->timeout + 580 msecs_to_jiffies(CARL9170_QUEUE_STUCK_TIMEOUT)) == true) 581 restart = true; 582 583 next: 584 spin_unlock_bh(&ar->tx_status[i].lock); 585 } 586 587 if (restart) { 588 /* 589 * At least one queue has been stuck for long enough. 590 * Give the device a kick and hope it gets back to 591 * work. 592 * 593 * possible reasons may include: 594 * - frames got lost/corrupted (bad connection to the device) 595 * - stalled rx processing/usb controller hiccups 596 * - firmware errors/bugs 597 * - every bug you can think of. 598 * - all bugs you can't... 599 * - ... 600 */ 601 carl9170_restart(ar, CARL9170_RR_STUCK_TX); 602 } 603 } 604 605 static void carl9170_tx_ampdu_timeout(struct ar9170 *ar) 606 { 607 struct carl9170_sta_tid *iter; 608 struct sk_buff *skb; 609 struct ieee80211_tx_info *txinfo; 610 struct carl9170_tx_info *arinfo; 611 struct ieee80211_sta *sta; 612 613 rcu_read_lock(); 614 list_for_each_entry_rcu(iter, &ar->tx_ampdu_list, list) { 615 if (iter->state < CARL9170_TID_STATE_IDLE) 616 continue; 617 618 spin_lock_bh(&iter->lock); 619 skb = skb_peek(&iter->queue); 620 if (!skb) 621 goto unlock; 622 623 txinfo = IEEE80211_SKB_CB(skb); 624 arinfo = (void *)txinfo->rate_driver_data; 625 if (time_is_after_jiffies(arinfo->timeout + 626 msecs_to_jiffies(CARL9170_QUEUE_TIMEOUT))) 627 goto unlock; 628 629 sta = __carl9170_get_tx_sta(ar, skb); 630 if (WARN_ON(!sta)) 631 goto unlock; 632 633 ieee80211_stop_tx_ba_session(sta, iter->tid); 634 unlock: 635 spin_unlock_bh(&iter->lock); 636 637 } 638 rcu_read_unlock(); 639 } 640 641 void carl9170_tx_janitor(struct work_struct *work) 642 { 643 struct ar9170 *ar = container_of(work, struct ar9170, 644 tx_janitor.work); 645 if (!IS_STARTED(ar)) 646 return; 647 648 ar->tx_janitor_last_run = jiffies; 649 650 carl9170_check_queue_stop_timeout(ar); 651 carl9170_tx_ampdu_timeout(ar); 652 653 if (!atomic_read(&ar->tx_total_queued)) 654 return; 655 656 ieee80211_queue_delayed_work(ar->hw, &ar->tx_janitor, 657 msecs_to_jiffies(CARL9170_TX_TIMEOUT)); 658 } 659 660 static void __carl9170_tx_process_status(struct ar9170 *ar, 661 const uint8_t cookie, const uint8_t info) 662 { 663 struct sk_buff *skb; 664 struct ieee80211_tx_info *txinfo; 665 unsigned int r, t, q; 666 bool success = true; 667 668 q = ar9170_qmap[info & CARL9170_TX_STATUS_QUEUE]; 669 670 skb = carl9170_get_queued_skb(ar, cookie, &ar->tx_status[q]); 671 if (!skb) { 672 /* 673 * We have lost the race to another thread. 674 */ 675 676 return ; 677 } 678 679 txinfo = IEEE80211_SKB_CB(skb); 680 681 if (!(info & CARL9170_TX_STATUS_SUCCESS)) 682 success = false; 683 684 r = (info & CARL9170_TX_STATUS_RIX) >> CARL9170_TX_STATUS_RIX_S; 685 t = (info & CARL9170_TX_STATUS_TRIES) >> CARL9170_TX_STATUS_TRIES_S; 686 687 carl9170_tx_fill_rateinfo(ar, r, t, txinfo); 688 carl9170_tx_status(ar, skb, success); 689 } 690 691 void carl9170_tx_process_status(struct ar9170 *ar, 692 const struct carl9170_rsp *cmd) 693 { 694 unsigned int i; 695 696 for (i = 0; i < cmd->hdr.ext; i++) { 697 if (WARN_ON(i > ((cmd->hdr.len / 2) + 1))) { 698 print_hex_dump_bytes("UU:", DUMP_PREFIX_NONE, 699 (void *) cmd, cmd->hdr.len + 4); 700 break; 701 } 702 703 __carl9170_tx_process_status(ar, cmd->_tx_status[i].cookie, 704 cmd->_tx_status[i].info); 705 } 706 } 707 708 static void carl9170_tx_rate_tpc_chains(struct ar9170 *ar, 709 struct ieee80211_tx_info *info, struct ieee80211_tx_rate *txrate, 710 unsigned int *phyrate, unsigned int *tpc, unsigned int *chains) 711 { 712 struct ieee80211_rate *rate = NULL; 713 u8 *txpower; 714 unsigned int idx; 715 716 idx = txrate->idx; 717 *tpc = 0; 718 *phyrate = 0; 719 720 if (txrate->flags & IEEE80211_TX_RC_MCS) { 721 if (txrate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH) { 722 /* +1 dBm for HT40 */ 723 *tpc += 2; 724 725 if (info->band == IEEE80211_BAND_2GHZ) 726 txpower = ar->power_2G_ht40; 727 else 728 txpower = ar->power_5G_ht40; 729 } else { 730 if (info->band == IEEE80211_BAND_2GHZ) 731 txpower = ar->power_2G_ht20; 732 else 733 txpower = ar->power_5G_ht20; 734 } 735 736 *phyrate = txrate->idx; 737 *tpc += txpower[idx & 7]; 738 } else { 739 if (info->band == IEEE80211_BAND_2GHZ) { 740 if (idx < 4) 741 txpower = ar->power_2G_cck; 742 else 743 txpower = ar->power_2G_ofdm; 744 } else { 745 txpower = ar->power_5G_leg; 746 idx += 4; 747 } 748 749 rate = &__carl9170_ratetable[idx]; 750 *tpc += txpower[(rate->hw_value & 0x30) >> 4]; 751 *phyrate = rate->hw_value & 0xf; 752 } 753 754 if (ar->eeprom.tx_mask == 1) { 755 *chains = AR9170_TX_PHY_TXCHAIN_1; 756 } else { 757 if (!(txrate->flags & IEEE80211_TX_RC_MCS) && 758 rate && rate->bitrate >= 360) 759 *chains = AR9170_TX_PHY_TXCHAIN_1; 760 else 761 *chains = AR9170_TX_PHY_TXCHAIN_2; 762 } 763 764 *tpc = min_t(unsigned int, *tpc, ar->hw->conf.power_level * 2); 765 } 766 767 static __le32 carl9170_tx_physet(struct ar9170 *ar, 768 struct ieee80211_tx_info *info, struct ieee80211_tx_rate *txrate) 769 { 770 unsigned int power = 0, chains = 0, phyrate = 0; 771 __le32 tmp; 772 773 tmp = cpu_to_le32(0); 774 775 if (txrate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH) 776 tmp |= cpu_to_le32(AR9170_TX_PHY_BW_40MHZ << 777 AR9170_TX_PHY_BW_S); 778 /* this works because 40 MHz is 2 and dup is 3 */ 779 if (txrate->flags & IEEE80211_TX_RC_DUP_DATA) 780 tmp |= cpu_to_le32(AR9170_TX_PHY_BW_40MHZ_DUP << 781 AR9170_TX_PHY_BW_S); 782 783 if (txrate->flags & IEEE80211_TX_RC_SHORT_GI) 784 tmp |= cpu_to_le32(AR9170_TX_PHY_SHORT_GI); 785 786 if (txrate->flags & IEEE80211_TX_RC_MCS) { 787 SET_VAL(AR9170_TX_PHY_MCS, phyrate, txrate->idx); 788 789 /* heavy clip control */ 790 tmp |= cpu_to_le32((txrate->idx & 0x7) << 791 AR9170_TX_PHY_TX_HEAVY_CLIP_S); 792 793 tmp |= cpu_to_le32(AR9170_TX_PHY_MOD_HT); 794 795 /* 796 * green field preamble does not work. 797 * 798 * if (txrate->flags & IEEE80211_TX_RC_GREEN_FIELD) 799 * tmp |= cpu_to_le32(AR9170_TX_PHY_GREENFIELD); 800 */ 801 } else { 802 if (info->band == IEEE80211_BAND_2GHZ) { 803 if (txrate->idx <= AR9170_TX_PHY_RATE_CCK_11M) 804 tmp |= cpu_to_le32(AR9170_TX_PHY_MOD_CCK); 805 else 806 tmp |= cpu_to_le32(AR9170_TX_PHY_MOD_OFDM); 807 } else { 808 tmp |= cpu_to_le32(AR9170_TX_PHY_MOD_OFDM); 809 } 810 811 /* 812 * short preamble seems to be broken too. 813 * 814 * if (txrate->flags & IEEE80211_TX_RC_USE_SHORT_PREAMBLE) 815 * tmp |= cpu_to_le32(AR9170_TX_PHY_SHORT_PREAMBLE); 816 */ 817 } 818 carl9170_tx_rate_tpc_chains(ar, info, txrate, 819 &phyrate, &power, &chains); 820 821 tmp |= cpu_to_le32(SET_CONSTVAL(AR9170_TX_PHY_MCS, phyrate)); 822 tmp |= cpu_to_le32(SET_CONSTVAL(AR9170_TX_PHY_TX_PWR, power)); 823 tmp |= cpu_to_le32(SET_CONSTVAL(AR9170_TX_PHY_TXCHAIN, chains)); 824 return tmp; 825 } 826 827 static bool carl9170_tx_rts_check(struct ar9170 *ar, 828 struct ieee80211_tx_rate *rate, 829 bool ampdu, bool multi) 830 { 831 switch (ar->erp_mode) { 832 case CARL9170_ERP_AUTO: 833 if (ampdu) 834 break; 835 836 case CARL9170_ERP_MAC80211: 837 if (!(rate->flags & IEEE80211_TX_RC_USE_RTS_CTS)) 838 break; 839 840 case CARL9170_ERP_RTS: 841 if (likely(!multi)) 842 return true; 843 844 default: 845 break; 846 } 847 848 return false; 849 } 850 851 static bool carl9170_tx_cts_check(struct ar9170 *ar, 852 struct ieee80211_tx_rate *rate) 853 { 854 switch (ar->erp_mode) { 855 case CARL9170_ERP_AUTO: 856 case CARL9170_ERP_MAC80211: 857 if (!(rate->flags & IEEE80211_TX_RC_USE_CTS_PROTECT)) 858 break; 859 860 case CARL9170_ERP_CTS: 861 return true; 862 863 default: 864 break; 865 } 866 867 return false; 868 } 869 870 static int carl9170_tx_prepare(struct ar9170 *ar, 871 struct ieee80211_sta *sta, 872 struct sk_buff *skb) 873 { 874 struct ieee80211_hdr *hdr; 875 struct _carl9170_tx_superframe *txc; 876 struct carl9170_vif_info *cvif; 877 struct ieee80211_tx_info *info; 878 struct ieee80211_tx_rate *txrate; 879 struct carl9170_tx_info *arinfo; 880 unsigned int hw_queue; 881 int i; 882 __le16 mac_tmp; 883 u16 len; 884 bool ampdu, no_ack; 885 886 BUILD_BUG_ON(sizeof(*arinfo) > sizeof(info->rate_driver_data)); 887 BUILD_BUG_ON(sizeof(struct _carl9170_tx_superdesc) != 888 CARL9170_TX_SUPERDESC_LEN); 889 890 BUILD_BUG_ON(sizeof(struct _ar9170_tx_hwdesc) != 891 AR9170_TX_HWDESC_LEN); 892 893 BUILD_BUG_ON(IEEE80211_TX_MAX_RATES < CARL9170_TX_MAX_RATES); 894 895 BUILD_BUG_ON(AR9170_MAX_VIRTUAL_MAC > 896 ((CARL9170_TX_SUPER_MISC_VIF_ID >> 897 CARL9170_TX_SUPER_MISC_VIF_ID_S) + 1)); 898 899 hw_queue = ar9170_qmap[carl9170_get_queue(ar, skb)]; 900 901 hdr = (void *)skb->data; 902 info = IEEE80211_SKB_CB(skb); 903 len = skb->len; 904 905 /* 906 * Note: If the frame was sent through a monitor interface, 907 * the ieee80211_vif pointer can be NULL. 908 */ 909 if (likely(info->control.vif)) 910 cvif = (void *) info->control.vif->drv_priv; 911 else 912 cvif = NULL; 913 914 txc = (void *)skb_push(skb, sizeof(*txc)); 915 memset(txc, 0, sizeof(*txc)); 916 917 SET_VAL(CARL9170_TX_SUPER_MISC_QUEUE, txc->s.misc, hw_queue); 918 919 if (likely(cvif)) 920 SET_VAL(CARL9170_TX_SUPER_MISC_VIF_ID, txc->s.misc, cvif->id); 921 922 if (unlikely(info->flags & IEEE80211_TX_CTL_SEND_AFTER_DTIM)) 923 txc->s.misc |= CARL9170_TX_SUPER_MISC_CAB; 924 925 if (unlikely(info->flags & IEEE80211_TX_CTL_ASSIGN_SEQ)) 926 txc->s.misc |= CARL9170_TX_SUPER_MISC_ASSIGN_SEQ; 927 928 if (unlikely(ieee80211_is_probe_resp(hdr->frame_control))) 929 txc->s.misc |= CARL9170_TX_SUPER_MISC_FILL_IN_TSF; 930 931 mac_tmp = cpu_to_le16(AR9170_TX_MAC_HW_DURATION | 932 AR9170_TX_MAC_BACKOFF); 933 mac_tmp |= cpu_to_le16((hw_queue << AR9170_TX_MAC_QOS_S) & 934 AR9170_TX_MAC_QOS); 935 936 no_ack = !!(info->flags & IEEE80211_TX_CTL_NO_ACK); 937 if (unlikely(no_ack)) 938 mac_tmp |= cpu_to_le16(AR9170_TX_MAC_NO_ACK); 939 940 if (info->control.hw_key) { 941 len += info->control.hw_key->icv_len; 942 943 switch (info->control.hw_key->cipher) { 944 case WLAN_CIPHER_SUITE_WEP40: 945 case WLAN_CIPHER_SUITE_WEP104: 946 case WLAN_CIPHER_SUITE_TKIP: 947 mac_tmp |= cpu_to_le16(AR9170_TX_MAC_ENCR_RC4); 948 break; 949 case WLAN_CIPHER_SUITE_CCMP: 950 mac_tmp |= cpu_to_le16(AR9170_TX_MAC_ENCR_AES); 951 break; 952 default: 953 WARN_ON(1); 954 goto err_out; 955 } 956 } 957 958 ampdu = !!(info->flags & IEEE80211_TX_CTL_AMPDU); 959 if (ampdu) { 960 unsigned int density, factor; 961 962 if (unlikely(!sta || !cvif)) 963 goto err_out; 964 965 factor = min_t(unsigned int, 1u, sta->ht_cap.ampdu_factor); 966 density = sta->ht_cap.ampdu_density; 967 968 if (density) { 969 /* 970 * Watch out! 971 * 972 * Otus uses slightly different density values than 973 * those from the 802.11n spec. 974 */ 975 976 density = max_t(unsigned int, density + 1, 7u); 977 } 978 979 SET_VAL(CARL9170_TX_SUPER_AMPDU_DENSITY, 980 txc->s.ampdu_settings, density); 981 982 SET_VAL(CARL9170_TX_SUPER_AMPDU_FACTOR, 983 txc->s.ampdu_settings, factor); 984 985 for (i = 0; i < CARL9170_TX_MAX_RATES; i++) { 986 txrate = &info->control.rates[i]; 987 if (txrate->idx >= 0) { 988 txc->s.ri[i] = 989 CARL9170_TX_SUPER_RI_AMPDU; 990 991 if (WARN_ON(!(txrate->flags & 992 IEEE80211_TX_RC_MCS))) { 993 /* 994 * Not sure if it's even possible 995 * to aggregate non-ht rates with 996 * this HW. 997 */ 998 goto err_out; 999 } 1000 continue; 1001 } 1002 1003 txrate->idx = 0; 1004 txrate->count = ar->hw->max_rate_tries; 1005 } 1006 1007 mac_tmp |= cpu_to_le16(AR9170_TX_MAC_AGGR); 1008 } 1009 1010 /* 1011 * NOTE: For the first rate, the ERP & AMPDU flags are directly 1012 * taken from mac_control. For all fallback rate, the firmware 1013 * updates the mac_control flags from the rate info field. 1014 */ 1015 for (i = 1; i < CARL9170_TX_MAX_RATES; i++) { 1016 txrate = &info->control.rates[i]; 1017 if (txrate->idx < 0) 1018 break; 1019 1020 SET_VAL(CARL9170_TX_SUPER_RI_TRIES, txc->s.ri[i], 1021 txrate->count); 1022 1023 if (carl9170_tx_rts_check(ar, txrate, ampdu, no_ack)) 1024 txc->s.ri[i] |= (AR9170_TX_MAC_PROT_RTS << 1025 CARL9170_TX_SUPER_RI_ERP_PROT_S); 1026 else if (carl9170_tx_cts_check(ar, txrate)) 1027 txc->s.ri[i] |= (AR9170_TX_MAC_PROT_CTS << 1028 CARL9170_TX_SUPER_RI_ERP_PROT_S); 1029 1030 txc->s.rr[i - 1] = carl9170_tx_physet(ar, info, txrate); 1031 } 1032 1033 txrate = &info->control.rates[0]; 1034 SET_VAL(CARL9170_TX_SUPER_RI_TRIES, txc->s.ri[0], txrate->count); 1035 1036 if (carl9170_tx_rts_check(ar, txrate, ampdu, no_ack)) 1037 mac_tmp |= cpu_to_le16(AR9170_TX_MAC_PROT_RTS); 1038 else if (carl9170_tx_cts_check(ar, txrate)) 1039 mac_tmp |= cpu_to_le16(AR9170_TX_MAC_PROT_CTS); 1040 1041 txc->s.len = cpu_to_le16(skb->len); 1042 txc->f.length = cpu_to_le16(len + FCS_LEN); 1043 txc->f.mac_control = mac_tmp; 1044 txc->f.phy_control = carl9170_tx_physet(ar, info, txrate); 1045 1046 arinfo = (void *)info->rate_driver_data; 1047 arinfo->timeout = jiffies; 1048 arinfo->ar = ar; 1049 kref_init(&arinfo->ref); 1050 return 0; 1051 1052 err_out: 1053 skb_pull(skb, sizeof(*txc)); 1054 return -EINVAL; 1055 } 1056 1057 static void carl9170_set_immba(struct ar9170 *ar, struct sk_buff *skb) 1058 { 1059 struct _carl9170_tx_superframe *super; 1060 1061 super = (void *) skb->data; 1062 super->f.mac_control |= cpu_to_le16(AR9170_TX_MAC_IMM_BA); 1063 } 1064 1065 static void carl9170_set_ampdu_params(struct ar9170 *ar, struct sk_buff *skb) 1066 { 1067 struct _carl9170_tx_superframe *super; 1068 int tmp; 1069 1070 super = (void *) skb->data; 1071 1072 tmp = (super->s.ampdu_settings & CARL9170_TX_SUPER_AMPDU_DENSITY) << 1073 CARL9170_TX_SUPER_AMPDU_DENSITY_S; 1074 1075 /* 1076 * If you haven't noticed carl9170_tx_prepare has already filled 1077 * in all ampdu spacing & factor parameters. 1078 * Now it's the time to check whenever the settings have to be 1079 * updated by the firmware, or if everything is still the same. 1080 * 1081 * There's no sane way to handle different density values with 1082 * this hardware, so we may as well just do the compare in the 1083 * driver. 1084 */ 1085 1086 if (tmp != ar->current_density) { 1087 ar->current_density = tmp; 1088 super->s.ampdu_settings |= 1089 CARL9170_TX_SUPER_AMPDU_COMMIT_DENSITY; 1090 } 1091 1092 tmp = (super->s.ampdu_settings & CARL9170_TX_SUPER_AMPDU_FACTOR) << 1093 CARL9170_TX_SUPER_AMPDU_FACTOR_S; 1094 1095 if (tmp != ar->current_factor) { 1096 ar->current_factor = tmp; 1097 super->s.ampdu_settings |= 1098 CARL9170_TX_SUPER_AMPDU_COMMIT_FACTOR; 1099 } 1100 } 1101 1102 static bool carl9170_tx_rate_check(struct ar9170 *ar, struct sk_buff *_dest, 1103 struct sk_buff *_src) 1104 { 1105 struct _carl9170_tx_superframe *dest, *src; 1106 1107 dest = (void *) _dest->data; 1108 src = (void *) _src->data; 1109 1110 /* 1111 * The mac80211 rate control algorithm expects that all MPDUs in 1112 * an AMPDU share the same tx vectors. 1113 * This is not really obvious right now, because the hardware 1114 * does the AMPDU setup according to its own rulebook. 1115 * Our nicely assembled, strictly monotonic increasing mpdu 1116 * chains will be broken up, mashed back together... 1117 */ 1118 1119 return (dest->f.phy_control == src->f.phy_control); 1120 } 1121 1122 static void carl9170_tx_ampdu(struct ar9170 *ar) 1123 { 1124 struct sk_buff_head agg; 1125 struct carl9170_sta_tid *tid_info; 1126 struct sk_buff *skb, *first; 1127 unsigned int i = 0, done_ampdus = 0; 1128 u16 seq, queue, tmpssn; 1129 1130 atomic_inc(&ar->tx_ampdu_scheduler); 1131 ar->tx_ampdu_schedule = false; 1132 1133 if (atomic_read(&ar->tx_ampdu_upload)) 1134 return; 1135 1136 if (!ar->tx_ampdu_list_len) 1137 return; 1138 1139 __skb_queue_head_init(&agg); 1140 1141 rcu_read_lock(); 1142 tid_info = rcu_dereference(ar->tx_ampdu_iter); 1143 if (WARN_ON_ONCE(!tid_info)) { 1144 rcu_read_unlock(); 1145 return; 1146 } 1147 1148 retry: 1149 list_for_each_entry_continue_rcu(tid_info, &ar->tx_ampdu_list, list) { 1150 i++; 1151 1152 if (tid_info->state < CARL9170_TID_STATE_PROGRESS) 1153 continue; 1154 1155 queue = TID_TO_WME_AC(tid_info->tid); 1156 1157 spin_lock_bh(&tid_info->lock); 1158 if (tid_info->state != CARL9170_TID_STATE_XMIT) 1159 goto processed; 1160 1161 tid_info->counter++; 1162 first = skb_peek(&tid_info->queue); 1163 tmpssn = carl9170_get_seq(first); 1164 seq = tid_info->snx; 1165 1166 if (unlikely(tmpssn != seq)) { 1167 tid_info->state = CARL9170_TID_STATE_IDLE; 1168 1169 goto processed; 1170 } 1171 1172 while ((skb = skb_peek(&tid_info->queue))) { 1173 /* strict 0, 1, ..., n - 1, n frame sequence order */ 1174 if (unlikely(carl9170_get_seq(skb) != seq)) 1175 break; 1176 1177 /* don't upload more than AMPDU FACTOR allows. */ 1178 if (unlikely(SEQ_DIFF(tid_info->snx, tid_info->bsn) >= 1179 (tid_info->max - 1))) 1180 break; 1181 1182 if (!carl9170_tx_rate_check(ar, skb, first)) 1183 break; 1184 1185 atomic_inc(&ar->tx_ampdu_upload); 1186 tid_info->snx = seq = SEQ_NEXT(seq); 1187 __skb_unlink(skb, &tid_info->queue); 1188 1189 __skb_queue_tail(&agg, skb); 1190 1191 if (skb_queue_len(&agg) >= CARL9170_NUM_TX_AGG_MAX) 1192 break; 1193 } 1194 1195 if (skb_queue_empty(&tid_info->queue) || 1196 carl9170_get_seq(skb_peek(&tid_info->queue)) != 1197 tid_info->snx) { 1198 /* 1199 * stop TID, if A-MPDU frames are still missing, 1200 * or whenever the queue is empty. 1201 */ 1202 1203 tid_info->state = CARL9170_TID_STATE_IDLE; 1204 } 1205 done_ampdus++; 1206 1207 processed: 1208 spin_unlock_bh(&tid_info->lock); 1209 1210 if (skb_queue_empty(&agg)) 1211 continue; 1212 1213 /* apply ampdu spacing & factor settings */ 1214 carl9170_set_ampdu_params(ar, skb_peek(&agg)); 1215 1216 /* set aggregation push bit */ 1217 carl9170_set_immba(ar, skb_peek_tail(&agg)); 1218 1219 spin_lock_bh(&ar->tx_pending[queue].lock); 1220 skb_queue_splice_tail_init(&agg, &ar->tx_pending[queue]); 1221 spin_unlock_bh(&ar->tx_pending[queue].lock); 1222 ar->tx_schedule = true; 1223 } 1224 if ((done_ampdus++ == 0) && (i++ == 0)) 1225 goto retry; 1226 1227 rcu_assign_pointer(ar->tx_ampdu_iter, tid_info); 1228 rcu_read_unlock(); 1229 } 1230 1231 static struct sk_buff *carl9170_tx_pick_skb(struct ar9170 *ar, 1232 struct sk_buff_head *queue) 1233 { 1234 struct sk_buff *skb; 1235 struct ieee80211_tx_info *info; 1236 struct carl9170_tx_info *arinfo; 1237 1238 BUILD_BUG_ON(sizeof(*arinfo) > sizeof(info->rate_driver_data)); 1239 1240 spin_lock_bh(&queue->lock); 1241 skb = skb_peek(queue); 1242 if (unlikely(!skb)) 1243 goto err_unlock; 1244 1245 if (carl9170_alloc_dev_space(ar, skb)) 1246 goto err_unlock; 1247 1248 __skb_unlink(skb, queue); 1249 spin_unlock_bh(&queue->lock); 1250 1251 info = IEEE80211_SKB_CB(skb); 1252 arinfo = (void *) info->rate_driver_data; 1253 1254 arinfo->timeout = jiffies; 1255 return skb; 1256 1257 err_unlock: 1258 spin_unlock_bh(&queue->lock); 1259 return NULL; 1260 } 1261 1262 void carl9170_tx_drop(struct ar9170 *ar, struct sk_buff *skb) 1263 { 1264 struct _carl9170_tx_superframe *super; 1265 uint8_t q = 0; 1266 1267 ar->tx_dropped++; 1268 1269 super = (void *)skb->data; 1270 SET_VAL(CARL9170_TX_SUPER_MISC_QUEUE, q, 1271 ar9170_qmap[carl9170_get_queue(ar, skb)]); 1272 __carl9170_tx_process_status(ar, super->s.cookie, q); 1273 } 1274 1275 static bool carl9170_tx_ps_drop(struct ar9170 *ar, struct sk_buff *skb) 1276 { 1277 struct ieee80211_sta *sta; 1278 struct carl9170_sta_info *sta_info; 1279 struct ieee80211_tx_info *tx_info; 1280 1281 rcu_read_lock(); 1282 sta = __carl9170_get_tx_sta(ar, skb); 1283 if (!sta) 1284 goto out_rcu; 1285 1286 sta_info = (void *) sta->drv_priv; 1287 tx_info = IEEE80211_SKB_CB(skb); 1288 1289 if (unlikely(sta_info->sleeping) && 1290 !(tx_info->flags & (IEEE80211_TX_CTL_NO_PS_BUFFER | 1291 IEEE80211_TX_CTL_CLEAR_PS_FILT))) { 1292 rcu_read_unlock(); 1293 1294 if (tx_info->flags & IEEE80211_TX_CTL_AMPDU) 1295 atomic_dec(&ar->tx_ampdu_upload); 1296 1297 tx_info->flags |= IEEE80211_TX_STAT_TX_FILTERED; 1298 carl9170_release_dev_space(ar, skb); 1299 carl9170_tx_status(ar, skb, false); 1300 return true; 1301 } 1302 1303 out_rcu: 1304 rcu_read_unlock(); 1305 return false; 1306 } 1307 1308 static void carl9170_bar_check(struct ar9170 *ar, struct sk_buff *skb) 1309 { 1310 struct _carl9170_tx_superframe *super = (void *) skb->data; 1311 struct ieee80211_bar *bar = (void *) super->frame_data; 1312 1313 if (unlikely(ieee80211_is_back_req(bar->frame_control)) && 1314 skb->len >= sizeof(struct ieee80211_bar)) { 1315 struct carl9170_bar_list_entry *entry; 1316 unsigned int queue = skb_get_queue_mapping(skb); 1317 1318 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1319 if (!WARN_ON_ONCE(!entry)) { 1320 entry->skb = skb; 1321 spin_lock_bh(&ar->bar_list_lock[queue]); 1322 list_add_tail_rcu(&entry->list, &ar->bar_list[queue]); 1323 spin_unlock_bh(&ar->bar_list_lock[queue]); 1324 } 1325 } 1326 } 1327 1328 static void carl9170_tx(struct ar9170 *ar) 1329 { 1330 struct sk_buff *skb; 1331 unsigned int i, q; 1332 bool schedule_garbagecollector = false; 1333 1334 ar->tx_schedule = false; 1335 1336 if (unlikely(!IS_STARTED(ar))) 1337 return; 1338 1339 carl9170_usb_handle_tx_err(ar); 1340 1341 for (i = 0; i < ar->hw->queues; i++) { 1342 while (!skb_queue_empty(&ar->tx_pending[i])) { 1343 skb = carl9170_tx_pick_skb(ar, &ar->tx_pending[i]); 1344 if (unlikely(!skb)) 1345 break; 1346 1347 if (unlikely(carl9170_tx_ps_drop(ar, skb))) 1348 continue; 1349 1350 carl9170_bar_check(ar, skb); 1351 1352 atomic_inc(&ar->tx_total_pending); 1353 1354 q = __carl9170_get_queue(ar, i); 1355 /* 1356 * NB: tx_status[i] vs. tx_status[q], 1357 * TODO: Move into pick_skb or alloc_dev_space. 1358 */ 1359 skb_queue_tail(&ar->tx_status[q], skb); 1360 1361 /* 1362 * increase ref count to "2". 1363 * Ref counting is the easiest way to solve the 1364 * race between the urb's completion routine: 1365 * carl9170_tx_callback 1366 * and wlan tx status functions: 1367 * carl9170_tx_status/janitor. 1368 */ 1369 carl9170_tx_get_skb(skb); 1370 1371 carl9170_usb_tx(ar, skb); 1372 schedule_garbagecollector = true; 1373 } 1374 } 1375 1376 if (!schedule_garbagecollector) 1377 return; 1378 1379 ieee80211_queue_delayed_work(ar->hw, &ar->tx_janitor, 1380 msecs_to_jiffies(CARL9170_TX_TIMEOUT)); 1381 } 1382 1383 static bool carl9170_tx_ampdu_queue(struct ar9170 *ar, 1384 struct ieee80211_sta *sta, struct sk_buff *skb) 1385 { 1386 struct _carl9170_tx_superframe *super = (void *) skb->data; 1387 struct carl9170_sta_info *sta_info; 1388 struct carl9170_sta_tid *agg; 1389 struct sk_buff *iter; 1390 u16 tid, seq, qseq, off; 1391 bool run = false; 1392 1393 tid = carl9170_get_tid(skb); 1394 seq = carl9170_get_seq(skb); 1395 sta_info = (void *) sta->drv_priv; 1396 1397 rcu_read_lock(); 1398 agg = rcu_dereference(sta_info->agg[tid]); 1399 1400 if (!agg) 1401 goto err_unlock_rcu; 1402 1403 spin_lock_bh(&agg->lock); 1404 if (unlikely(agg->state < CARL9170_TID_STATE_IDLE)) 1405 goto err_unlock; 1406 1407 /* check if sequence is within the BA window */ 1408 if (unlikely(!BAW_WITHIN(agg->bsn, CARL9170_BAW_BITS, seq))) 1409 goto err_unlock; 1410 1411 if (WARN_ON_ONCE(!BAW_WITHIN(agg->snx, CARL9170_BAW_BITS, seq))) 1412 goto err_unlock; 1413 1414 off = SEQ_DIFF(seq, agg->bsn); 1415 if (WARN_ON_ONCE(test_and_set_bit(off, agg->bitmap))) 1416 goto err_unlock; 1417 1418 if (likely(BAW_WITHIN(agg->hsn, CARL9170_BAW_BITS, seq))) { 1419 __skb_queue_tail(&agg->queue, skb); 1420 agg->hsn = seq; 1421 goto queued; 1422 } 1423 1424 skb_queue_reverse_walk(&agg->queue, iter) { 1425 qseq = carl9170_get_seq(iter); 1426 1427 if (BAW_WITHIN(qseq, CARL9170_BAW_BITS, seq)) { 1428 __skb_queue_after(&agg->queue, iter, skb); 1429 goto queued; 1430 } 1431 } 1432 1433 __skb_queue_head(&agg->queue, skb); 1434 queued: 1435 1436 if (unlikely(agg->state != CARL9170_TID_STATE_XMIT)) { 1437 if (agg->snx == carl9170_get_seq(skb_peek(&agg->queue))) { 1438 agg->state = CARL9170_TID_STATE_XMIT; 1439 run = true; 1440 } 1441 } 1442 1443 spin_unlock_bh(&agg->lock); 1444 rcu_read_unlock(); 1445 1446 return run; 1447 1448 err_unlock: 1449 spin_unlock_bh(&agg->lock); 1450 1451 err_unlock_rcu: 1452 rcu_read_unlock(); 1453 super->f.mac_control &= ~cpu_to_le16(AR9170_TX_MAC_AGGR); 1454 carl9170_tx_status(ar, skb, false); 1455 ar->tx_dropped++; 1456 return false; 1457 } 1458 1459 void carl9170_op_tx(struct ieee80211_hw *hw, 1460 struct ieee80211_tx_control *control, 1461 struct sk_buff *skb) 1462 { 1463 struct ar9170 *ar = hw->priv; 1464 struct ieee80211_tx_info *info; 1465 struct ieee80211_sta *sta = control->sta; 1466 bool run; 1467 1468 if (unlikely(!IS_STARTED(ar))) 1469 goto err_free; 1470 1471 info = IEEE80211_SKB_CB(skb); 1472 1473 if (unlikely(carl9170_tx_prepare(ar, sta, skb))) 1474 goto err_free; 1475 1476 carl9170_tx_accounting(ar, skb); 1477 /* 1478 * from now on, one has to use carl9170_tx_status to free 1479 * all ressouces which are associated with the frame. 1480 */ 1481 1482 if (sta) { 1483 struct carl9170_sta_info *stai = (void *) sta->drv_priv; 1484 atomic_inc(&stai->pending_frames); 1485 } 1486 1487 if (info->flags & IEEE80211_TX_CTL_AMPDU) { 1488 /* to static code analyzers and reviewers: 1489 * mac80211 guarantees that a valid "sta" 1490 * reference is present, if a frame is to 1491 * be part of an ampdu. Hence any extra 1492 * sta == NULL checks are redundant in this 1493 * special case. 1494 */ 1495 run = carl9170_tx_ampdu_queue(ar, sta, skb); 1496 if (run) 1497 carl9170_tx_ampdu(ar); 1498 1499 } else { 1500 unsigned int queue = skb_get_queue_mapping(skb); 1501 1502 skb_queue_tail(&ar->tx_pending[queue], skb); 1503 } 1504 1505 carl9170_tx(ar); 1506 return; 1507 1508 err_free: 1509 ar->tx_dropped++; 1510 ieee80211_free_txskb(ar->hw, skb); 1511 } 1512 1513 void carl9170_tx_scheduler(struct ar9170 *ar) 1514 { 1515 1516 if (ar->tx_ampdu_schedule) 1517 carl9170_tx_ampdu(ar); 1518 1519 if (ar->tx_schedule) 1520 carl9170_tx(ar); 1521 } 1522 1523 int carl9170_update_beacon(struct ar9170 *ar, const bool submit) 1524 { 1525 struct sk_buff *skb = NULL; 1526 struct carl9170_vif_info *cvif; 1527 struct ieee80211_tx_info *txinfo; 1528 struct ieee80211_tx_rate *rate; 1529 __le32 *data, *old = NULL; 1530 unsigned int plcp, power, chains; 1531 u32 word, ht1, off, addr, len; 1532 int i = 0, err = 0; 1533 1534 rcu_read_lock(); 1535 cvif = rcu_dereference(ar->beacon_iter); 1536 retry: 1537 if (ar->vifs == 0 || !cvif) 1538 goto out_unlock; 1539 1540 list_for_each_entry_continue_rcu(cvif, &ar->vif_list, list) { 1541 if (cvif->active && cvif->enable_beacon) 1542 goto found; 1543 } 1544 1545 if (!ar->beacon_enabled || i++) 1546 goto out_unlock; 1547 1548 goto retry; 1549 1550 found: 1551 rcu_assign_pointer(ar->beacon_iter, cvif); 1552 1553 skb = ieee80211_beacon_get_tim(ar->hw, carl9170_get_vif(cvif), 1554 NULL, NULL); 1555 1556 if (!skb) { 1557 err = -ENOMEM; 1558 goto err_free; 1559 } 1560 1561 txinfo = IEEE80211_SKB_CB(skb); 1562 spin_lock_bh(&ar->beacon_lock); 1563 data = (__le32 *)skb->data; 1564 if (cvif->beacon) 1565 old = (__le32 *)cvif->beacon->data; 1566 1567 off = cvif->id * AR9170_MAC_BCN_LENGTH_MAX; 1568 addr = ar->fw.beacon_addr + off; 1569 len = roundup(skb->len + FCS_LEN, 4); 1570 1571 if ((off + len) > ar->fw.beacon_max_len) { 1572 if (net_ratelimit()) { 1573 wiphy_err(ar->hw->wiphy, "beacon does not " 1574 "fit into device memory!\n"); 1575 } 1576 err = -EINVAL; 1577 goto err_unlock; 1578 } 1579 1580 if (len > AR9170_MAC_BCN_LENGTH_MAX) { 1581 if (net_ratelimit()) { 1582 wiphy_err(ar->hw->wiphy, "no support for beacons " 1583 "bigger than %d (yours:%d).\n", 1584 AR9170_MAC_BCN_LENGTH_MAX, len); 1585 } 1586 1587 err = -EMSGSIZE; 1588 goto err_unlock; 1589 } 1590 1591 ht1 = AR9170_MAC_BCN_HT1_TX_ANT0; 1592 rate = &txinfo->control.rates[0]; 1593 carl9170_tx_rate_tpc_chains(ar, txinfo, rate, &plcp, &power, &chains); 1594 if (!(txinfo->control.rates[0].flags & IEEE80211_TX_RC_MCS)) { 1595 if (plcp <= AR9170_TX_PHY_RATE_CCK_11M) 1596 plcp |= ((skb->len + FCS_LEN) << (3 + 16)) + 0x0400; 1597 else 1598 plcp |= ((skb->len + FCS_LEN) << 16) + 0x0010; 1599 } else { 1600 ht1 |= AR9170_MAC_BCN_HT1_HT_EN; 1601 if (rate->flags & IEEE80211_TX_RC_SHORT_GI) 1602 plcp |= AR9170_MAC_BCN_HT2_SGI; 1603 1604 if (rate->flags & IEEE80211_TX_RC_40_MHZ_WIDTH) { 1605 ht1 |= AR9170_MAC_BCN_HT1_BWC_40M_SHARED; 1606 plcp |= AR9170_MAC_BCN_HT2_BW40; 1607 } 1608 if (rate->flags & IEEE80211_TX_RC_DUP_DATA) { 1609 ht1 |= AR9170_MAC_BCN_HT1_BWC_40M_DUP; 1610 plcp |= AR9170_MAC_BCN_HT2_BW40; 1611 } 1612 1613 SET_VAL(AR9170_MAC_BCN_HT2_LEN, plcp, skb->len + FCS_LEN); 1614 } 1615 1616 SET_VAL(AR9170_MAC_BCN_HT1_PWR_CTRL, ht1, 7); 1617 SET_VAL(AR9170_MAC_BCN_HT1_TPC, ht1, power); 1618 SET_VAL(AR9170_MAC_BCN_HT1_CHAIN_MASK, ht1, chains); 1619 if (chains == AR9170_TX_PHY_TXCHAIN_2) 1620 ht1 |= AR9170_MAC_BCN_HT1_TX_ANT1; 1621 1622 carl9170_async_regwrite_begin(ar); 1623 carl9170_async_regwrite(AR9170_MAC_REG_BCN_HT1, ht1); 1624 if (!(txinfo->control.rates[0].flags & IEEE80211_TX_RC_MCS)) 1625 carl9170_async_regwrite(AR9170_MAC_REG_BCN_PLCP, plcp); 1626 else 1627 carl9170_async_regwrite(AR9170_MAC_REG_BCN_HT2, plcp); 1628 1629 for (i = 0; i < DIV_ROUND_UP(skb->len, 4); i++) { 1630 /* 1631 * XXX: This accesses beyond skb data for up 1632 * to the last 3 bytes!! 1633 */ 1634 1635 if (old && (data[i] == old[i])) 1636 continue; 1637 1638 word = le32_to_cpu(data[i]); 1639 carl9170_async_regwrite(addr + 4 * i, word); 1640 } 1641 carl9170_async_regwrite_finish(); 1642 1643 dev_kfree_skb_any(cvif->beacon); 1644 cvif->beacon = NULL; 1645 1646 err = carl9170_async_regwrite_result(); 1647 if (!err) 1648 cvif->beacon = skb; 1649 spin_unlock_bh(&ar->beacon_lock); 1650 if (err) 1651 goto err_free; 1652 1653 if (submit) { 1654 err = carl9170_bcn_ctrl(ar, cvif->id, 1655 CARL9170_BCN_CTRL_CAB_TRIGGER, 1656 addr, skb->len + FCS_LEN); 1657 1658 if (err) 1659 goto err_free; 1660 } 1661 out_unlock: 1662 rcu_read_unlock(); 1663 return 0; 1664 1665 err_unlock: 1666 spin_unlock_bh(&ar->beacon_lock); 1667 1668 err_free: 1669 rcu_read_unlock(); 1670 dev_kfree_skb_any(skb); 1671 return err; 1672 } 1673