1 From: Felix Fietkau <nbd@openwrt.org>
2 Date: Tue, 18 Nov 2014 23:58:51 +0100
3 Subject: [PATCH] mac80211: add an intermediate software queue implementation
5 This allows drivers to request per-vif and per-sta-tid queues from which
6 they can pull frames. This makes it easier to keep the hardware queues
7 short, and to improve fairness between clients and vifs.
9 The task of scheduling packet transmission is left up to the driver -
10 queueing is controlled by mac80211. Drivers can only dequeue packets by
11 calling ieee80211_tx_dequeue. This makes it possible to add active queue
12 management later without changing drivers using this code.
14 This can also be used as a starting point to implement A-MSDU
15 aggregation in a way that does not add artificially induced latency.
17 Signed-off-by: Felix Fietkau <nbd@openwrt.org>
20 --- a/include/net/mac80211.h
21 +++ b/include/net/mac80211.h
27 + * DOC: mac80211 software tx queueing
29 + * mac80211 provides an optional intermediate queueing implementation designed
30 + * to allow the driver to keep hardware queues short and provide some fairness
31 + * between different stations/interfaces.
32 + * In this model, the driver pulls data frames from the mac80211 queue instead
33 + * of letting mac80211 push them via drv_tx().
34 + * Other frames (e.g. control or management) are still pushed using drv_tx().
36 + * Intermediate queues (struct ieee80211_txq) are kept per-sta per-tid, with a
37 + * single per-vif queue for multicast data frames.
39 + * The driver is expected to initialize its private per-queue data for stations
40 + * and interfaces in the .add_interface and .sta_add ops.
42 + * The driver can not access the queue directly. To dequeue a frame, it calls
43 + * ieee80211_tx_dequeue(). Whenever mac80211 adds a new frame to a queue, it
44 + * calls the .wake_tx_queue driver op.
46 + * For AP powersave TIM handling, the driver only needs to indicate if it has
47 + * buffered packets in the driver specific data structures by calling
48 + * ieee80211_sta_set_buffered(). For frames buffered in the ieee80211_txq
49 + * struct, mac80211 sets the appropriate TIM PVB bits and calls
50 + * .release_buffered_frames().
51 + * That callback is expected to release its own buffered frames and afterwards
52 + * also frames from the ieee80211_txq (obtained via ieee80211_tx_dequeue).
58 @@ -1257,6 +1286,8 @@ struct ieee80211_vif {
60 u8 hw_queue[IEEE80211_NUM_ACS];
62 + struct ieee80211_txq *txq;
64 struct ieee80211_chanctx_conf __rcu *chanctx_conf;
67 @@ -1519,6 +1550,8 @@ struct ieee80211_sta {
71 + struct ieee80211_txq *txq[IEEE80211_NUM_TIDS];
74 u8 drv_priv[0] __aligned(sizeof(void *));
76 @@ -1547,6 +1580,27 @@ struct ieee80211_tx_control {
80 + * struct ieee80211_txq - Software intermediate tx queue
82 + * @vif: &struct ieee80211_vif pointer from the add_interface callback.
83 + * @sta: station table entry, %NULL for per-vif queue
84 + * @tid: the TID for this queue (unused for per-vif queue)
85 + * @ac: the AC for this queue
87 + * The driver can obtain packets from this queue by calling
88 + * ieee80211_tx_dequeue().
90 +struct ieee80211_txq {
91 + struct ieee80211_vif *vif;
92 + struct ieee80211_sta *sta;
97 + u8 drv_priv[0] __aligned(sizeof(void *));
101 * enum ieee80211_hw_flags - hardware flags
103 * These flags are used to indicate hardware capabilities to
104 @@ -1770,6 +1824,8 @@ enum ieee80211_hw_flags {
105 * within &struct ieee80211_sta.
106 * @chanctx_data_size: size (in bytes) of the drv_priv data area
107 * within &struct ieee80211_chanctx_conf.
108 + * @txq_data_size: size (in bytes) of the drv_priv data area
109 + * within @struct ieee80211_txq.
111 * @max_rates: maximum number of alternate rate retry stages the hw
113 @@ -1818,6 +1874,9 @@ enum ieee80211_hw_flags {
114 * @n_cipher_schemes: a size of an array of cipher schemes definitions.
115 * @cipher_schemes: a pointer to an array of cipher scheme definitions
118 + * @txq_ac_max_pending: maximum number of frames per AC pending in all txq
119 + * entries for a vif.
121 struct ieee80211_hw {
122 struct ieee80211_conf conf;
123 @@ -1830,6 +1889,7 @@ struct ieee80211_hw {
126 int chanctx_data_size;
129 u16 max_listen_interval;
131 @@ -1846,6 +1906,7 @@ struct ieee80211_hw {
134 const struct ieee80211_cipher_scheme *cipher_schemes;
135 + int txq_ac_max_pending;
139 @@ -3007,6 +3068,8 @@ enum ieee80211_reconfig_type {
140 * response template is provided, together with the location of the
141 * switch-timing IE within the template. The skb can only be used within
144 + * @wake_tx_queue: Called when new packets have been added to the queue.
146 struct ieee80211_ops {
147 void (*tx)(struct ieee80211_hw *hw,
148 @@ -3238,6 +3301,9 @@ struct ieee80211_ops {
149 void (*tdls_recv_channel_switch)(struct ieee80211_hw *hw,
150 struct ieee80211_vif *vif,
151 struct ieee80211_tdls_ch_sw_params *params);
153 + void (*wake_tx_queue)(struct ieee80211_hw *hw,
154 + struct ieee80211_txq *txq);
158 @@ -5249,4 +5315,17 @@ void ieee80211_unreserve_tid(struct ieee
160 size_t ieee80211_ie_split(const u8 *ies, size_t ielen,
161 const u8 *ids, int n_ids, size_t offset);
164 + * ieee80211_tx_dequeue - dequeue a packet from a software tx queue
166 + * @hw: pointer as obtained from ieee80211_alloc_hw()
167 + * @txq: pointer obtained from .add_tx_queue() call
169 + * Returns the skb if successful, %NULL if no frame was available.
171 +struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,
172 + struct ieee80211_txq *txq);
175 #endif /* MAC80211_H */
176 --- a/net/mac80211/driver-ops.h
177 +++ b/net/mac80211/driver-ops.h
178 @@ -1367,4 +1367,16 @@ drv_tdls_recv_channel_switch(struct ieee
179 trace_drv_return_void(local);
182 +static inline void drv_wake_tx_queue(struct ieee80211_local *local,
183 + struct txq_info *txq)
185 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(txq->txq.vif);
187 + if (!check_sdata_in_driver(sdata))
190 + trace_drv_wake_tx_queue(local, sdata, txq->txq.sta, txq->txq.tid);
191 + local->ops->wake_tx_queue(&local->hw, &txq->txq);
194 #endif /* __MAC80211_DRIVER_OPS */
195 --- a/net/mac80211/ieee80211_i.h
196 +++ b/net/mac80211/ieee80211_i.h
197 @@ -809,6 +809,19 @@ struct mac80211_qos_map {
198 struct rcu_head rcu_head;
201 +enum txq_info_flags {
202 + IEEE80211_TXQ_STOP,
203 + IEEE80211_TXQ_AMPDU,
207 + struct sk_buff_head queue;
208 + unsigned long flags;
211 + struct ieee80211_txq txq;
214 struct ieee80211_sub_if_data {
215 struct list_head list;
217 @@ -853,6 +866,7 @@ struct ieee80211_sub_if_data {
218 bool control_port_no_encrypt;
219 int encrypt_headroom;
221 + atomic_t txqs_len[IEEE80211_NUM_ACS];
222 struct ieee80211_tx_queue_params tx_conf[IEEE80211_NUM_ACS];
223 struct mac80211_qos_map __rcu *qos_map;
225 @@ -1453,6 +1467,10 @@ static inline struct ieee80211_local *hw
226 return container_of(hw, struct ieee80211_local, hw);
229 +static inline struct txq_info *to_txq_info(struct ieee80211_txq *txq)
231 + return container_of(txq, struct txq_info, txq);
234 static inline int ieee80211_bssid_match(const u8 *raddr, const u8 *addr)
236 @@ -1905,6 +1923,9 @@ static inline bool ieee80211_can_run_wor
240 +void ieee80211_init_tx_queue(struct ieee80211_sub_if_data *sdata,
241 + struct sta_info *sta,
242 + struct txq_info *txq, int tid);
243 void ieee80211_send_auth(struct ieee80211_sub_if_data *sdata,
244 u16 transaction, u16 auth_alg, u16 status,
245 const u8 *extra, size_t extra_len, const u8 *bssid,
246 --- a/net/mac80211/iface.c
247 +++ b/net/mac80211/iface.c
248 @@ -969,6 +969,13 @@ static void ieee80211_do_stop(struct iee
250 spin_unlock_irqrestore(&local->queue_stop_reason_lock, flags);
252 + if (sdata->vif.txq) {
253 + struct txq_info *txqi = to_txq_info(sdata->vif.txq);
255 + ieee80211_purge_tx_queue(&local->hw, &txqi->queue);
256 + atomic_set(&sdata->txqs_len[txqi->txq.ac], 0);
259 if (local->open_count == 0)
260 ieee80211_clear_tx_pending(local);
262 @@ -1773,6 +1780,15 @@ int ieee80211_if_add(struct ieee80211_lo
263 ieee80211_setup_sdata(sdata, type);
266 + struct txq_info *txqi = NULL;
268 + if (local->ops->wake_tx_queue) {
269 + txqi = kzalloc(sizeof(*txqi) +
270 + local->hw.txq_data_size, GFP_KERNEL);
272 + ieee80211_init_tx_queue(sdata, NULL, txqi, 0);
276 ndev->ieee80211_ptr->use_4addr = params->use_4addr;
277 if (type == NL80211_IFTYPE_STATION)
278 @@ -1785,6 +1801,7 @@ int ieee80211_if_add(struct ieee80211_lo
280 ret = register_netdevice(ndev);
286 @@ -1810,6 +1827,9 @@ void ieee80211_if_remove(struct ieee8021
290 + if (sdata->vif.txq)
291 + kfree(to_txq_info(sdata->vif.txq));
294 unregister_netdevice(sdata->dev);
296 @@ -1851,6 +1871,9 @@ void ieee80211_remove_interfaces(struct
297 list_for_each_entry_safe(sdata, tmp, &local->interfaces, list) {
298 list_del(&sdata->list);
300 + if (sdata->vif.txq)
301 + kfree(to_txq_info(sdata->vif.txq));
304 unregister_netdevice_queue(sdata->dev, &unreg_list);
306 --- a/net/mac80211/main.c
307 +++ b/net/mac80211/main.c
308 @@ -1019,6 +1019,9 @@ int ieee80211_register_hw(struct ieee802
310 local->dynamic_ps_forced_timeout = -1;
312 + if (!local->hw.txq_ac_max_pending)
313 + local->hw.txq_ac_max_pending = 64;
315 result = ieee80211_wep_init(local);
317 wiphy_debug(local->hw.wiphy, "Failed to initialize wep: %d\n",
318 --- a/net/mac80211/sta_info.c
319 +++ b/net/mac80211/sta_info.c
320 @@ -118,6 +118,16 @@ static void __cleanup_single_sta(struct
321 atomic_dec(&ps->num_sta_ps);
324 + if (sta->sta.txq[0]) {
325 + for (i = 0; i < ARRAY_SIZE(sta->sta.txq); i++) {
326 + struct txq_info *txqi = to_txq_info(sta->sta.txq[i]);
327 + int n = skb_queue_len(&txqi->queue);
329 + ieee80211_purge_tx_queue(&local->hw, &txqi->queue);
330 + atomic_sub(n, &sdata->txqs_len[txqi->txq.ac]);
334 for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
335 local->total_ps_buffered -= skb_queue_len(&sta->ps_tx_buf[ac]);
336 ieee80211_purge_tx_queue(&local->hw, &sta->ps_tx_buf[ac]);
337 @@ -234,6 +244,8 @@ void sta_info_free(struct ieee80211_loca
339 sta_dbg(sta->sdata, "Destroyed STA %pM\n", sta->sta.addr);
341 + if (sta->sta.txq[0])
342 + kfree(to_txq_info(sta->sta.txq[0]));
343 kfree(rcu_dereference_raw(sta->sta.rates));
346 @@ -285,11 +297,12 @@ struct sta_info *sta_info_alloc(struct i
347 const u8 *addr, gfp_t gfp)
349 struct ieee80211_local *local = sdata->local;
350 + struct ieee80211_hw *hw = &local->hw;
351 struct sta_info *sta;
352 struct timespec uptime;
355 - sta = kzalloc(sizeof(*sta) + local->hw.sta_data_size, gfp);
356 + sta = kzalloc(sizeof(*sta) + hw->sta_data_size, gfp);
360 @@ -321,11 +334,25 @@ struct sta_info *sta_info_alloc(struct i
361 for (i = 0; i < ARRAY_SIZE(sta->chain_signal_avg); i++)
362 ewma_init(&sta->chain_signal_avg[i], 1024, 8);
364 - if (sta_prepare_rate_control(local, sta, gfp)) {
367 + if (local->ops->wake_tx_queue) {
369 + int size = sizeof(struct txq_info) +
370 + ALIGN(hw->txq_data_size, sizeof(void *));
372 + txq_data = kcalloc(ARRAY_SIZE(sta->sta.txq), size, gfp);
376 + for (i = 0; i < ARRAY_SIZE(sta->sta.txq); i++) {
377 + struct txq_info *txq = txq_data + i * size;
379 + ieee80211_init_tx_queue(sdata, sta, txq, i);
383 + if (sta_prepare_rate_control(local, sta, gfp))
386 for (i = 0; i < IEEE80211_NUM_TIDS; i++) {
388 * timer_to_tid must be initialized with identity mapping
389 @@ -346,7 +373,7 @@ struct sta_info *sta_info_alloc(struct i
390 if (sdata->vif.type == NL80211_IFTYPE_AP ||
391 sdata->vif.type == NL80211_IFTYPE_AP_VLAN) {
392 struct ieee80211_supported_band *sband =
393 - local->hw.wiphy->bands[ieee80211_get_sdata_band(sdata)];
394 + hw->wiphy->bands[ieee80211_get_sdata_band(sdata)];
395 u8 smps = (sband->ht_cap.cap & IEEE80211_HT_CAP_SM_PS) >>
396 IEEE80211_HT_CAP_SM_PS_SHIFT;
398 @@ -371,6 +398,13 @@ struct sta_info *sta_info_alloc(struct i
399 sta_dbg(sdata, "Allocated STA %pM\n", sta->sta.addr);
404 + if (sta->sta.txq[0])
405 + kfree(to_txq_info(sta->sta.txq[0]));
411 static int sta_info_insert_check(struct sta_info *sta)
412 @@ -640,6 +674,8 @@ static void __sta_info_recalc_tim(struct
415 sta->driver_buffered_tids & tids;
417 + sta->txq_buffered_tids & tids;
421 @@ -1071,7 +1107,7 @@ void ieee80211_sta_ps_deliver_wakeup(str
422 struct ieee80211_sub_if_data *sdata = sta->sdata;
423 struct ieee80211_local *local = sdata->local;
424 struct sk_buff_head pending;
425 - int filtered = 0, buffered = 0, ac;
426 + int filtered = 0, buffered = 0, ac, i;
430 @@ -1090,10 +1126,22 @@ void ieee80211_sta_ps_deliver_wakeup(str
432 BUILD_BUG_ON(BITS_TO_LONGS(IEEE80211_NUM_TIDS) > 1);
433 sta->driver_buffered_tids = 0;
434 + sta->txq_buffered_tids = 0;
436 if (!(local->hw.flags & IEEE80211_HW_AP_LINK_PS))
437 drv_sta_notify(local, sdata, STA_NOTIFY_AWAKE, &sta->sta);
439 + if (sta->sta.txq[0]) {
440 + for (i = 0; i < ARRAY_SIZE(sta->sta.txq); i++) {
441 + struct txq_info *txqi = to_txq_info(sta->sta.txq[i]);
443 + if (!skb_queue_len(&txqi->queue))
446 + drv_wake_tx_queue(local, txqi);
450 skb_queue_head_init(&pending);
452 /* sync with ieee80211_tx_h_unicast_ps_buf */
453 @@ -1254,7 +1302,7 @@ ieee80211_sta_ps_deliver_response(struct
454 struct ieee80211_sub_if_data *sdata = sta->sdata;
455 struct ieee80211_local *local = sdata->local;
456 bool more_data = false;
459 unsigned long driver_release_tids = 0;
460 struct sk_buff_head frames;
462 @@ -1275,8 +1323,10 @@ ieee80211_sta_ps_deliver_response(struct
463 /* if we already have frames from software, then we can't also
464 * release from hardware queues
466 - if (skb_queue_empty(&frames))
467 + if (skb_queue_empty(&frames)) {
468 driver_release_tids |= sta->driver_buffered_tids & tids;
469 + driver_release_tids |= sta->txq_buffered_tids & tids;
472 if (driver_release_tids) {
473 /* If the driver has data on more than one TID then
474 @@ -1447,6 +1497,8 @@ ieee80211_sta_ps_deliver_response(struct
476 sta_info_recalc_tim(sta);
478 + unsigned long tids = sta->txq_buffered_tids & driver_release_tids;
481 * We need to release a frame that is buffered somewhere in the
482 * driver ... it'll have to handle that.
483 @@ -1466,8 +1518,22 @@ ieee80211_sta_ps_deliver_response(struct
484 * that the TID(s) became empty before returning here from the
486 * Either way, however, when the driver tells us that the TID(s)
487 - * became empty we'll do the TIM recalculation.
488 + * became empty or we find that a txq became empty, we'll do the
489 + * TIM recalculation.
492 + if (!sta->sta.txq[0])
495 + for (tid = 0; tid < ARRAY_SIZE(sta->sta.txq); tid++) {
496 + struct txq_info *txqi = to_txq_info(sta->sta.txq[tid]);
498 + if (!(tids & BIT(tid)) || skb_queue_len(&txqi->queue))
501 + sta_info_recalc_tim(sta);
507 --- a/net/mac80211/sta_info.h
508 +++ b/net/mac80211/sta_info.h
509 @@ -274,6 +274,7 @@ struct sta_ampdu_mlme {
510 * entered power saving state, these are also delivered to
511 * the station when it leaves powersave or polls for frames
512 * @driver_buffered_tids: bitmap of TIDs the driver has data buffered on
513 + * @txq_buffered_tids: bitmap of TIDs that mac80211 has txq data buffered on
514 * @rx_packets: Number of MSDUs received from this STA
515 * @rx_bytes: Number of bytes received from this STA
516 * @last_rx: time (in jiffies) when last frame was received from this STA
517 @@ -368,6 +369,7 @@ struct sta_info {
518 struct sk_buff_head ps_tx_buf[IEEE80211_NUM_ACS];
519 struct sk_buff_head tx_filtered[IEEE80211_NUM_ACS];
520 unsigned long driver_buffered_tids;
521 + unsigned long txq_buffered_tids;
523 /* Updated from RX path only, no locking requirements */
524 unsigned long rx_packets;
525 --- a/net/mac80211/trace.h
526 +++ b/net/mac80211/trace.h
527 @@ -2312,6 +2312,34 @@ TRACE_EVENT(drv_tdls_recv_channel_switch
531 +TRACE_EVENT(drv_wake_tx_queue,
532 + TP_PROTO(struct ieee80211_local *local,
533 + struct ieee80211_sub_if_data *sdata,
534 + struct ieee80211_sta *sta,
537 + TP_ARGS(local, sdata, sta, tid),
550 + __entry->tid = tid;
554 + LOCAL_PR_FMT VIF_PR_FMT STA_PR_FMT " tid: 0x%x",
555 + LOCAL_PR_ARG, VIF_PR_ARG, STA_PR_ARG, __entry->tid
559 #ifdef CPTCFG_MAC80211_MESSAGE_TRACING
561 #define TRACE_SYSTEM mac80211_msg
562 --- a/net/mac80211/tx.c
563 +++ b/net/mac80211/tx.c
564 @@ -776,12 +776,23 @@ ieee80211_tx_h_rate_ctrl(struct ieee8021
569 +ieee80211_tx_next_seq(struct sta_info *sta, int tid)
571 + u16 *seq = &sta->tid_seq[tid];
572 + u16 ret = cpu_to_le16(*seq);
574 + /* Increase the sequence number. */
575 + *seq = (*seq + 0x10) & IEEE80211_SCTL_SEQ;
580 static ieee80211_tx_result debug_noinline
581 ieee80211_tx_h_sequence(struct ieee80211_tx_data *tx)
583 struct ieee80211_tx_info *info = IEEE80211_SKB_CB(tx->skb);
584 struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)tx->skb->data;
589 @@ -832,13 +843,10 @@ ieee80211_tx_h_sequence(struct ieee80211
591 qc = ieee80211_get_qos_ctl(hdr);
592 tid = *qc & IEEE80211_QOS_CTL_TID_MASK;
593 - seq = &tx->sta->tid_seq[tid];
594 tx->sta->tx_msdu[tid]++;
596 - hdr->seq_ctrl = cpu_to_le16(*seq);
598 - /* Increase the sequence number. */
599 - *seq = (*seq + 0x10) & IEEE80211_SCTL_SEQ;
600 + if (!tx->sta->sta.txq[0])
601 + hdr->seq_ctrl = ieee80211_tx_next_seq(tx->sta, tid);
605 @@ -1067,7 +1075,7 @@ static bool ieee80211_tx_prep_agg(struct
606 * nothing -- this aggregation session is being started
607 * but that might still fail with the driver
610 + } else if (!tx->sta->sta.txq[tid]) {
611 spin_lock(&tx->sta->lock);
613 * Need to re-check now, because we may get here
614 @@ -1201,13 +1209,102 @@ ieee80211_tx_prepare(struct ieee80211_su
618 +static void ieee80211_drv_tx(struct ieee80211_local *local,
619 + struct ieee80211_vif *vif,
620 + struct ieee80211_sta *pubsta,
621 + struct sk_buff *skb)
623 + struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
624 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
625 + struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
626 + struct ieee80211_tx_control control = {
629 + struct ieee80211_txq *txq = NULL;
630 + struct txq_info *txqi;
633 + if (info->control.flags & IEEE80211_TX_CTRL_PS_RESPONSE)
636 + if (!ieee80211_is_data(hdr->frame_control))
640 + u8 tid = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
642 + txq = pubsta->txq[tid];
651 + txqi = to_txq_info(txq);
652 + atomic_inc(&sdata->txqs_len[ac]);
653 + if (atomic_read(&sdata->txqs_len[ac]) >= local->hw.txq_ac_max_pending)
654 + netif_stop_subqueue(sdata->dev, ac);
656 + skb_queue_tail(&txqi->queue, skb);
657 + drv_wake_tx_queue(local, txqi);
662 + drv_tx(local, &control, skb);
665 +struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,
666 + struct ieee80211_txq *txq)
668 + struct ieee80211_local *local = hw_to_local(hw);
669 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(txq->vif);
670 + struct txq_info *txqi = container_of(txq, struct txq_info, txq);
671 + struct ieee80211_hdr *hdr;
672 + struct sk_buff *skb = NULL;
675 + spin_lock_bh(&txqi->queue.lock);
677 + if (test_bit(IEEE80211_TXQ_STOP, &txqi->flags))
680 + skb = __skb_dequeue(&txqi->queue);
684 + atomic_dec(&sdata->txqs_len[ac]);
685 + if (__netif_subqueue_stopped(sdata->dev, ac))
686 + ieee80211_propagate_queue_wake(local, sdata->vif.hw_queue[ac]);
688 + hdr = (struct ieee80211_hdr *)skb->data;
689 + if (txq->sta && ieee80211_is_data_qos(hdr->frame_control)) {
690 + struct sta_info *sta = container_of(txq->sta, struct sta_info,
692 + struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
694 + hdr->seq_ctrl = ieee80211_tx_next_seq(sta, txq->tid);
695 + if (test_bit(IEEE80211_TXQ_AMPDU, &txqi->flags))
696 + info->flags |= IEEE80211_TX_CTL_AMPDU;
698 + info->flags &= ~IEEE80211_TX_CTL_AMPDU;
702 + spin_unlock_bh(&txqi->queue.lock);
706 +EXPORT_SYMBOL(ieee80211_tx_dequeue);
708 static bool ieee80211_tx_frags(struct ieee80211_local *local,
709 struct ieee80211_vif *vif,
710 struct ieee80211_sta *sta,
711 struct sk_buff_head *skbs,
714 - struct ieee80211_tx_control control;
715 struct sk_buff *skb, *tmp;
718 @@ -1265,10 +1362,9 @@ static bool ieee80211_tx_frags(struct ie
719 spin_unlock_irqrestore(&local->queue_stop_reason_lock, flags);
721 info->control.vif = vif;
724 __skb_unlink(skb, skbs);
725 - drv_tx(local, &control, skb);
726 + ieee80211_drv_tx(local, vif, sta, skb);
730 --- a/net/mac80211/util.c
731 +++ b/net/mac80211/util.c
732 @@ -308,6 +308,11 @@ void ieee80211_propagate_queue_wake(stru
733 for (ac = 0; ac < n_acs; ac++) {
734 int ac_queue = sdata->vif.hw_queue[ac];
736 + if (local->ops->wake_tx_queue &&
737 + (atomic_read(&sdata->txqs_len[ac]) >
738 + local->hw.txq_ac_max_pending))
741 if (ac_queue == queue ||
742 (sdata->vif.cab_queue == queue &&
743 local->queue_stop_reasons[ac_queue] == 0 &&
744 @@ -3307,3 +3312,20 @@ u8 *ieee80211_add_wmm_info_ie(u8 *buf, u
749 +void ieee80211_init_tx_queue(struct ieee80211_sub_if_data *sdata,
750 + struct sta_info *sta,
751 + struct txq_info *txqi, int tid)
753 + skb_queue_head_init(&txqi->queue);
754 + txqi->txq.vif = &sdata->vif;
757 + txqi->txq.sta = &sta->sta;
758 + sta->sta.txq[tid] = &txqi->txq;
759 + txqi->txq.ac = ieee802_1d_to_ac[tid & 7];
761 + sdata->vif.txq = &txqi->txq;
762 + txqi->txq.ac = IEEE80211_AC_BE;
765 --- a/net/mac80211/rx.c
766 +++ b/net/mac80211/rx.c
767 @@ -1176,6 +1176,7 @@ static void sta_ps_start(struct sta_info
768 struct ieee80211_sub_if_data *sdata = sta->sdata;
769 struct ieee80211_local *local = sdata->local;
773 if (sta->sdata->vif.type == NL80211_IFTYPE_AP ||
774 sta->sdata->vif.type == NL80211_IFTYPE_AP_VLAN)
775 @@ -1189,6 +1190,18 @@ static void sta_ps_start(struct sta_info
776 drv_sta_notify(local, sdata, STA_NOTIFY_SLEEP, &sta->sta);
777 ps_dbg(sdata, "STA %pM aid %d enters power save mode\n",
778 sta->sta.addr, sta->sta.aid);
780 + if (!sta->sta.txq[0])
783 + for (tid = 0; tid < ARRAY_SIZE(sta->sta.txq); tid++) {
784 + struct txq_info *txqi = to_txq_info(sta->sta.txq[tid]);
786 + if (!skb_queue_len(&txqi->queue))
787 + set_bit(tid, &sta->txq_buffered_tids);
789 + clear_bit(tid, &sta->txq_buffered_tids);
793 static void sta_ps_end(struct sta_info *sta)
794 --- a/net/mac80211/agg-tx.c
795 +++ b/net/mac80211/agg-tx.c
796 @@ -188,6 +188,43 @@ ieee80211_wake_queue_agg(struct ieee8021
797 __release(agg_queue);
801 +ieee80211_agg_stop_txq(struct sta_info *sta, int tid)
803 + struct ieee80211_txq *txq = sta->sta.txq[tid];
804 + struct txq_info *txqi;
809 + txqi = to_txq_info(txq);
811 + /* Lock here to protect against further seqno updates on dequeue */
812 + spin_lock_bh(&txqi->queue.lock);
813 + set_bit(IEEE80211_TXQ_STOP, &txqi->flags);
814 + spin_unlock_bh(&txqi->queue.lock);
818 +ieee80211_agg_start_txq(struct sta_info *sta, int tid, bool enable)
820 + struct ieee80211_txq *txq = sta->sta.txq[tid];
821 + struct txq_info *txqi;
826 + txqi = to_txq_info(txq);
829 + set_bit(IEEE80211_TXQ_AMPDU, &txqi->flags);
831 + clear_bit(IEEE80211_TXQ_AMPDU, &txqi->flags);
833 + clear_bit(IEEE80211_TXQ_STOP, &txqi->flags);
834 + drv_wake_tx_queue(sta->sdata->local, txqi);
838 * splice packets from the STA's pending to the local pending,
839 * requires a call to ieee80211_agg_splice_finish later
840 @@ -247,6 +284,7 @@ static void ieee80211_remove_tid_tx(stru
841 ieee80211_assign_tid_tx(sta, tid, NULL);
843 ieee80211_agg_splice_finish(sta->sdata, tid);
844 + ieee80211_agg_start_txq(sta, tid, false);
846 kfree_rcu(tid_tx, rcu_head);
848 @@ -418,6 +456,8 @@ void ieee80211_tx_ba_session_handle_star
850 clear_bit(HT_AGG_STATE_WANT_START, &tid_tx->state);
852 + ieee80211_agg_stop_txq(sta, tid);
855 * Make sure no packets are being processed. This ensures that
856 * we have a valid starting sequence number and that in-flight
857 @@ -440,6 +480,8 @@ void ieee80211_tx_ba_session_handle_star
858 ieee80211_agg_splice_finish(sdata, tid);
859 spin_unlock_bh(&sta->lock);
861 + ieee80211_agg_start_txq(sta, tid, false);
863 kfree_rcu(tid_tx, rcu_head);
866 @@ -666,6 +708,8 @@ static void ieee80211_agg_tx_operational
867 ieee80211_agg_splice_finish(sta->sdata, tid);
869 spin_unlock_bh(&sta->lock);
871 + ieee80211_agg_start_txq(sta, tid, true);
874 void ieee80211_start_tx_ba_cb(struct ieee80211_vif *vif, u8 *ra, u16 tid)