mac80211: update to wireless-testing 2015-03-09
[openwrt/staging/dedeckeh.git] / package / kernel / mac80211 / patches / 300-mac80211-add-an-intermediate-software-queue-implemen.patch
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
4
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.
8
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.
13
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.
16
17 Signed-off-by: Felix Fietkau <nbd@openwrt.org>
18 ---
19
20 --- a/include/net/mac80211.h
21 +++ b/include/net/mac80211.h
22 @@ -1257,6 +1257,8 @@ struct ieee80211_vif {
23 u8 cab_queue;
24 u8 hw_queue[IEEE80211_NUM_ACS];
25
26 + struct ieee80211_txq *txq;
27 +
28 struct ieee80211_chanctx_conf __rcu *chanctx_conf;
29
30 u32 driver_flags;
31 @@ -1519,6 +1521,8 @@ struct ieee80211_sta {
32 bool tdls_initiator;
33 bool mfp;
34
35 + struct ieee80211_txq *txq[IEEE80211_NUM_TIDS];
36 +
37 /* must be last */
38 u8 drv_priv[0] __aligned(sizeof(void *));
39 };
40 @@ -1547,6 +1551,27 @@ struct ieee80211_tx_control {
41 };
42
43 /**
44 + * struct ieee80211_txq - Software intermediate tx queue
45 + *
46 + * @vif: &struct ieee80211_vif pointer from the add_interface callback.
47 + * @sta: station table entry, may be NULL for per-vif queue
48 + * @tid: the TID for this queue (unset for per-vif queue)
49 + * @ac: the AC for this queue
50 + *
51 + * The driver can obtain packets from this queue by calling
52 + * ieee80211_tx_dequeue().
53 + */
54 +struct ieee80211_txq {
55 + struct ieee80211_vif *vif;
56 + struct ieee80211_sta *sta;
57 + u8 tid;
58 + u8 ac;
59 +
60 + /* must be last */
61 + u8 drv_priv[0] __aligned(sizeof(void *));
62 +};
63 +
64 +/**
65 * enum ieee80211_hw_flags - hardware flags
66 *
67 * These flags are used to indicate hardware capabilities to
68 @@ -1770,6 +1795,8 @@ enum ieee80211_hw_flags {
69 * within &struct ieee80211_sta.
70 * @chanctx_data_size: size (in bytes) of the drv_priv data area
71 * within &struct ieee80211_chanctx_conf.
72 + * @txq_data_size: size (in bytes) of the drv_priv data area
73 + * within @struct ieee80211_txq.
74 *
75 * @max_rates: maximum number of alternate rate retry stages the hw
76 * can handle.
77 @@ -1818,6 +1845,9 @@ enum ieee80211_hw_flags {
78 * @n_cipher_schemes: a size of an array of cipher schemes definitions.
79 * @cipher_schemes: a pointer to an array of cipher scheme definitions
80 * supported by HW.
81 + *
82 + * @txq_ac_max_pending: maximum number of frames per AC pending in all txq
83 + * entries for a vif.
84 */
85 struct ieee80211_hw {
86 struct ieee80211_conf conf;
87 @@ -1830,6 +1860,7 @@ struct ieee80211_hw {
88 int vif_data_size;
89 int sta_data_size;
90 int chanctx_data_size;
91 + int txq_data_size;
92 u16 queues;
93 u16 max_listen_interval;
94 s8 max_signal;
95 @@ -1846,6 +1877,7 @@ struct ieee80211_hw {
96 u8 uapsd_max_sp_len;
97 u8 n_cipher_schemes;
98 const struct ieee80211_cipher_scheme *cipher_schemes;
99 + int txq_ac_max_pending;
100 };
101
102 /**
103 @@ -3007,6 +3039,8 @@ enum ieee80211_reconfig_type {
104 * response template is provided, together with the location of the
105 * switch-timing IE within the template. The skb can only be used within
106 * the function call.
107 + *
108 + * @wake_tx_queue: Called when new packets have been added to the queue.
109 */
110 struct ieee80211_ops {
111 void (*tx)(struct ieee80211_hw *hw,
112 @@ -3238,6 +3272,9 @@ struct ieee80211_ops {
113 void (*tdls_recv_channel_switch)(struct ieee80211_hw *hw,
114 struct ieee80211_vif *vif,
115 struct ieee80211_tdls_ch_sw_params *params);
116 +
117 + void (*wake_tx_queue)(struct ieee80211_hw *hw,
118 + struct ieee80211_txq *txq);
119 };
120
121 /**
122 @@ -5249,4 +5286,17 @@ void ieee80211_unreserve_tid(struct ieee
123 */
124 size_t ieee80211_ie_split(const u8 *ies, size_t ielen,
125 const u8 *ids, int n_ids, size_t offset);
126 +
127 +/**
128 + * ieee80211_tx_dequeue - dequeue a packet from a software tx queue
129 + *
130 + * @hw: pointer as obtained from ieee80211_alloc_hw()
131 + * @txq: pointer obtained from .add_tx_queue() call
132 + *
133 + * Returns the sjb if successful, ERR_PTR(-EAGAIN) if no frame was available.
134 + */
135 +struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,
136 + struct ieee80211_txq *txq);
137 +
138 +
139 #endif /* MAC80211_H */
140 --- a/net/mac80211/driver-ops.h
141 +++ b/net/mac80211/driver-ops.h
142 @@ -1367,4 +1367,21 @@ drv_tdls_recv_channel_switch(struct ieee
143 trace_drv_return_void(local);
144 }
145
146 +static inline void drv_wake_tx_queue(struct ieee80211_local *local,
147 + struct txq_info *txq)
148 +{
149 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(txq->txq.vif);
150 +
151 + if (!check_sdata_in_driver(sdata))
152 + return;
153 +
154 + if (txq->txq.sta)
155 + trace_drv_wake_sta_tx_queue(local, sdata, txq->txq.sta,
156 + txq->txq.tid);
157 + else
158 + trace_drv_wake_vif_tx_queue(local, sdata);
159 +
160 + local->ops->wake_tx_queue(&local->hw, &txq->txq);
161 +}
162 +
163 #endif /* __MAC80211_DRIVER_OPS */
164 --- a/net/mac80211/ieee80211_i.h
165 +++ b/net/mac80211/ieee80211_i.h
166 @@ -809,6 +809,13 @@ struct mac80211_qos_map {
167 struct rcu_head rcu_head;
168 };
169
170 +struct txq_info {
171 + struct sk_buff_head queue;
172 +
173 + /* keep last! */
174 + struct ieee80211_txq txq;
175 +};
176 +
177 struct ieee80211_sub_if_data {
178 struct list_head list;
179
180 @@ -853,6 +860,7 @@ struct ieee80211_sub_if_data {
181 bool control_port_no_encrypt;
182 int encrypt_headroom;
183
184 + atomic_t txqs_len[IEEE80211_NUM_ACS];
185 struct ieee80211_tx_queue_params tx_conf[IEEE80211_NUM_ACS];
186 struct mac80211_qos_map __rcu *qos_map;
187
188 @@ -1905,6 +1913,12 @@ static inline bool ieee80211_can_run_wor
189 return true;
190 }
191
192 +void ieee80211_init_tx_queue(struct ieee80211_sub_if_data *sdata,
193 + struct sta_info *sta,
194 + struct txq_info *txq, int tid);
195 +void ieee80211_flush_tx_queue(struct ieee80211_local *local,
196 + struct ieee80211_txq *txq);
197 +
198 void ieee80211_send_auth(struct ieee80211_sub_if_data *sdata,
199 u16 transaction, u16 auth_alg, u16 status,
200 const u8 *extra, size_t extra_len, const u8 *bssid,
201 --- a/net/mac80211/iface.c
202 +++ b/net/mac80211/iface.c
203 @@ -969,6 +969,9 @@ static void ieee80211_do_stop(struct iee
204 }
205 spin_unlock_irqrestore(&local->queue_stop_reason_lock, flags);
206
207 + if (sdata->vif.txq)
208 + ieee80211_flush_tx_queue(local, sdata->vif.txq);
209 +
210 if (local->open_count == 0)
211 ieee80211_clear_tx_pending(local);
212
213 @@ -1773,6 +1776,15 @@ int ieee80211_if_add(struct ieee80211_lo
214 ieee80211_setup_sdata(sdata, type);
215
216 if (ndev) {
217 + struct txq_info *txqi = NULL;
218 +
219 + if (local->ops->wake_tx_queue) {
220 + txqi = kzalloc(sizeof(*txqi) +
221 + local->hw.txq_data_size, GFP_KERNEL);
222 + if (txqi)
223 + ieee80211_init_tx_queue(sdata, NULL, txqi, 0);
224 + }
225 +
226 if (params) {
227 ndev->ieee80211_ptr->use_4addr = params->use_4addr;
228 if (type == NL80211_IFTYPE_STATION)
229 @@ -1785,6 +1797,7 @@ int ieee80211_if_add(struct ieee80211_lo
230
231 ret = register_netdevice(ndev);
232 if (ret) {
233 + kfree(txqi);
234 free_netdev(ndev);
235 return ret;
236 }
237 @@ -1802,6 +1815,7 @@ int ieee80211_if_add(struct ieee80211_lo
238
239 void ieee80211_if_remove(struct ieee80211_sub_if_data *sdata)
240 {
241 + struct txq_info *txqi;
242 ASSERT_RTNL();
243
244 mutex_lock(&sdata->local->iflist_mtx);
245 @@ -1810,6 +1824,11 @@ void ieee80211_if_remove(struct ieee8021
246
247 synchronize_rcu();
248
249 + if (sdata->vif.txq) {
250 + txqi = container_of(sdata->vif.txq, struct txq_info, txq);
251 + kfree(txqi);
252 + }
253 +
254 if (sdata->dev) {
255 unregister_netdevice(sdata->dev);
256 } else {
257 @@ -1833,6 +1852,7 @@ void ieee80211_sdata_stop(struct ieee802
258 void ieee80211_remove_interfaces(struct ieee80211_local *local)
259 {
260 struct ieee80211_sub_if_data *sdata, *tmp;
261 + struct txq_info *txqi;
262 LIST_HEAD(unreg_list);
263 LIST_HEAD(wdev_list);
264
265 @@ -1851,6 +1871,12 @@ void ieee80211_remove_interfaces(struct
266 list_for_each_entry_safe(sdata, tmp, &local->interfaces, list) {
267 list_del(&sdata->list);
268
269 + if (sdata->vif.txq) {
270 + txqi = container_of(sdata->vif.txq, struct txq_info,
271 + txq);
272 + kfree(txqi);
273 + }
274 +
275 if (sdata->dev)
276 unregister_netdevice_queue(sdata->dev, &unreg_list);
277 else
278 --- a/net/mac80211/main.c
279 +++ b/net/mac80211/main.c
280 @@ -1019,6 +1019,9 @@ int ieee80211_register_hw(struct ieee802
281
282 local->dynamic_ps_forced_timeout = -1;
283
284 + if (!local->hw.txq_ac_max_pending)
285 + local->hw.txq_ac_max_pending = 64;
286 +
287 result = ieee80211_wep_init(local);
288 if (result < 0)
289 wiphy_debug(local->hw.wiphy, "Failed to initialize wep: %d\n",
290 --- a/net/mac80211/sta_info.c
291 +++ b/net/mac80211/sta_info.c
292 @@ -118,6 +118,11 @@ static void __cleanup_single_sta(struct
293 atomic_dec(&ps->num_sta_ps);
294 }
295
296 + if (sta->txq) {
297 + for (i = 0; i < IEEE80211_NUM_TIDS; i++)
298 + ieee80211_flush_tx_queue(local, sta->sta.txq[i]);
299 + }
300 +
301 for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
302 local->total_ps_buffered -= skb_queue_len(&sta->ps_tx_buf[ac]);
303 ieee80211_purge_tx_queue(&local->hw, &sta->ps_tx_buf[ac]);
304 @@ -234,6 +239,7 @@ void sta_info_free(struct ieee80211_loca
305
306 sta_dbg(sta->sdata, "Destroyed STA %pM\n", sta->sta.addr);
307
308 + kfree(sta->txq);
309 kfree(rcu_dereference_raw(sta->sta.rates));
310 kfree(sta);
311 }
312 @@ -285,11 +291,12 @@ struct sta_info *sta_info_alloc(struct i
313 const u8 *addr, gfp_t gfp)
314 {
315 struct ieee80211_local *local = sdata->local;
316 + struct ieee80211_hw *hw = &local->hw;
317 struct sta_info *sta;
318 struct timespec uptime;
319 int i;
320
321 - sta = kzalloc(sizeof(*sta) + local->hw.sta_data_size, gfp);
322 + sta = kzalloc(sizeof(*sta) + hw->sta_data_size, gfp);
323 if (!sta)
324 return NULL;
325
326 @@ -321,11 +328,23 @@ struct sta_info *sta_info_alloc(struct i
327 for (i = 0; i < ARRAY_SIZE(sta->chain_signal_avg); i++)
328 ewma_init(&sta->chain_signal_avg[i], 1024, 8);
329
330 - if (sta_prepare_rate_control(local, sta, gfp)) {
331 - kfree(sta);
332 - return NULL;
333 + if (local->ops->wake_tx_queue) {
334 + int size = sizeof(struct txq_info) +
335 + ALIGN(hw->txq_data_size, sizeof(void *));
336 +
337 + sta->txq = kcalloc(IEEE80211_NUM_TIDS, size, gfp);
338 + if (!sta->txq)
339 + goto free;
340 +
341 + for (i = 0; i < IEEE80211_NUM_TIDS; i++) {
342 + struct txq_info *txq = sta->txq + i * size;
343 + ieee80211_init_tx_queue(sdata, sta, txq, i);
344 + }
345 }
346
347 + if (sta_prepare_rate_control(local, sta, gfp))
348 + goto free_txq;
349 +
350 for (i = 0; i < IEEE80211_NUM_TIDS; i++) {
351 /*
352 * timer_to_tid must be initialized with identity mapping
353 @@ -346,7 +365,7 @@ struct sta_info *sta_info_alloc(struct i
354 if (sdata->vif.type == NL80211_IFTYPE_AP ||
355 sdata->vif.type == NL80211_IFTYPE_AP_VLAN) {
356 struct ieee80211_supported_band *sband =
357 - local->hw.wiphy->bands[ieee80211_get_sdata_band(sdata)];
358 + hw->wiphy->bands[ieee80211_get_sdata_band(sdata)];
359 u8 smps = (sband->ht_cap.cap & IEEE80211_HT_CAP_SM_PS) >>
360 IEEE80211_HT_CAP_SM_PS_SHIFT;
361 /*
362 @@ -371,6 +390,12 @@ struct sta_info *sta_info_alloc(struct i
363 sta_dbg(sdata, "Allocated STA %pM\n", sta->sta.addr);
364
365 return sta;
366 +
367 +free_txq:
368 + kfree(sta->txq);
369 +free:
370 + kfree(sta);
371 + return NULL;
372 }
373
374 static int sta_info_insert_check(struct sta_info *sta)
375 --- a/net/mac80211/sta_info.h
376 +++ b/net/mac80211/sta_info.h
377 @@ -368,6 +368,7 @@ struct sta_info {
378 struct sk_buff_head ps_tx_buf[IEEE80211_NUM_ACS];
379 struct sk_buff_head tx_filtered[IEEE80211_NUM_ACS];
380 unsigned long driver_buffered_tids;
381 + struct txq_info *txq;
382
383 /* Updated from RX path only, no locking requirements */
384 unsigned long rx_packets;
385 --- a/net/mac80211/trace.h
386 +++ b/net/mac80211/trace.h
387 @@ -2312,6 +2312,40 @@ TRACE_EVENT(drv_tdls_recv_channel_switch
388 )
389 );
390
391 +DEFINE_EVENT(local_sdata_evt, drv_wake_vif_tx_queue,
392 + TP_PROTO(struct ieee80211_local *local,
393 + struct ieee80211_sub_if_data *sdata),
394 + TP_ARGS(local, sdata)
395 +);
396 +
397 +TRACE_EVENT(drv_wake_sta_tx_queue,
398 + TP_PROTO(struct ieee80211_local *local,
399 + struct ieee80211_sub_if_data *sdata,
400 + struct ieee80211_sta *sta,
401 + u8 tid),
402 +
403 + TP_ARGS(local, sdata, sta, tid),
404 +
405 + TP_STRUCT__entry(
406 + LOCAL_ENTRY
407 + VIF_ENTRY
408 + STA_ENTRY
409 + __field(u8, tid)
410 + ),
411 +
412 + TP_fast_assign(
413 + LOCAL_ASSIGN;
414 + VIF_ASSIGN;
415 + STA_ASSIGN;
416 + __entry->tid = tid;
417 + ),
418 +
419 + TP_printk(
420 + LOCAL_PR_FMT VIF_PR_FMT STA_PR_FMT " tid: 0x%x",
421 + LOCAL_PR_ARG, VIF_PR_ARG, STA_PR_ARG, __entry->tid
422 + )
423 +);
424 +
425 #ifdef CPTCFG_MAC80211_MESSAGE_TRACING
426 #undef TRACE_SYSTEM
427 #define TRACE_SYSTEM mac80211_msg
428 --- a/net/mac80211/tx.c
429 +++ b/net/mac80211/tx.c
430 @@ -1201,13 +1201,76 @@ ieee80211_tx_prepare(struct ieee80211_su
431 return TX_CONTINUE;
432 }
433
434 +static void ieee80211_drv_tx(struct ieee80211_local *local,
435 + struct ieee80211_vif *vif,
436 + struct ieee80211_sta *pubsta,
437 + struct sk_buff *skb)
438 +{
439 + struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
440 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
441 + struct ieee80211_tx_control control = {
442 + .sta = pubsta
443 + };
444 + struct ieee80211_txq *txq = NULL;
445 + struct txq_info *txqi;
446 + u8 ac;
447 +
448 + if (ieee80211_is_mgmt(hdr->frame_control) ||
449 + ieee80211_is_ctl(hdr->frame_control))
450 + goto tx_normal;
451 +
452 + if (pubsta) {
453 + u8 tid = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
454 + txq = pubsta->txq[tid];
455 + } else if (vif) {
456 + txq = vif->txq;
457 + }
458 +
459 + if (!txq)
460 + goto tx_normal;
461 +
462 + ac = txq->ac;
463 + txqi = container_of(txq, struct txq_info, txq);
464 + atomic_inc(&sdata->txqs_len[ac]);
465 + if (atomic_read(&sdata->txqs_len[ac]) >= local->hw.txq_ac_max_pending)
466 + netif_stop_subqueue(sdata->dev, ac);
467 +
468 + skb_queue_tail(&txqi->queue, skb);
469 + drv_wake_tx_queue(local, txqi);
470 +
471 + return;
472 +
473 +tx_normal:
474 + drv_tx(local, &control, skb);
475 +}
476 +
477 +struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,
478 + struct ieee80211_txq *txq)
479 +{
480 + struct ieee80211_local *local = hw_to_local(hw);
481 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(txq->vif);
482 + struct txq_info *txqi = container_of(txq, struct txq_info, txq);
483 + struct sk_buff *skb;
484 + u8 ac = txq->ac;
485 +
486 + skb = skb_dequeue(&txqi->queue);
487 + if (!skb)
488 + return ERR_PTR(-EAGAIN);
489 +
490 + atomic_dec(&sdata->txqs_len[ac]);
491 + if (__netif_subqueue_stopped(sdata->dev, ac))
492 + ieee80211_propagate_queue_wake(local, sdata->vif.hw_queue[ac]);
493 +
494 + return skb;
495 +}
496 +EXPORT_SYMBOL(ieee80211_tx_dequeue);
497 +
498 static bool ieee80211_tx_frags(struct ieee80211_local *local,
499 struct ieee80211_vif *vif,
500 struct ieee80211_sta *sta,
501 struct sk_buff_head *skbs,
502 bool txpending)
503 {
504 - struct ieee80211_tx_control control;
505 struct sk_buff *skb, *tmp;
506 unsigned long flags;
507
508 @@ -1265,10 +1328,9 @@ static bool ieee80211_tx_frags(struct ie
509 spin_unlock_irqrestore(&local->queue_stop_reason_lock, flags);
510
511 info->control.vif = vif;
512 - control.sta = sta;
513
514 __skb_unlink(skb, skbs);
515 - drv_tx(local, &control, skb);
516 + ieee80211_drv_tx(local, vif, sta, skb);
517 }
518
519 return true;
520 --- a/net/mac80211/util.c
521 +++ b/net/mac80211/util.c
522 @@ -308,6 +308,11 @@ void ieee80211_propagate_queue_wake(stru
523 for (ac = 0; ac < n_acs; ac++) {
524 int ac_queue = sdata->vif.hw_queue[ac];
525
526 + if (local->ops->wake_tx_queue &&
527 + (atomic_read(&sdata->txqs_len[ac]) >
528 + local->hw.txq_ac_max_pending))
529 + continue;
530 +
531 if (ac_queue == queue ||
532 (sdata->vif.cab_queue == queue &&
533 local->queue_stop_reasons[ac_queue] == 0 &&
534 @@ -3307,3 +3312,36 @@ u8 *ieee80211_add_wmm_info_ie(u8 *buf, u
535
536 return buf;
537 }
538 +
539 +void ieee80211_init_tx_queue(struct ieee80211_sub_if_data *sdata,
540 + struct sta_info *sta,
541 + struct txq_info *txqi, int tid)
542 +{
543 + skb_queue_head_init(&txqi->queue);
544 + txqi->txq.vif = &sdata->vif;
545 +
546 + if (sta) {
547 + txqi->txq.sta = &sta->sta;
548 + sta->sta.txq[tid] = &txqi->txq;
549 + txqi->txq.ac = ieee802_1d_to_ac[tid & 7];
550 + } else {
551 + sdata->vif.txq = &txqi->txq;
552 + txqi->txq.ac = IEEE80211_AC_BE;
553 + }
554 +}
555 +
556 +void ieee80211_flush_tx_queue(struct ieee80211_local *local,
557 + struct ieee80211_txq *txq)
558 +{
559 + struct txq_info *txqi = container_of(txq, struct txq_info, txq);
560 + struct ieee80211_sub_if_data *sdata = vif_to_sdata(txq->vif);
561 + struct sk_buff *skb;
562 + int n = 0;
563 +
564 + while ((skb = skb_dequeue(&txqi->queue)) != NULL) {
565 + n++;
566 + ieee80211_free_txskb(&local->hw, skb);
567 + }
568 +
569 + atomic_sub(n, &sdata->txqs_len[txq->ac]);
570 +}