mac80211: Switch to a virtual time-based airtime scheduler

This switches the airtime scheduler in mac80211 to use a virtual
time-based scheduler instead of the round-robin scheduler used before.
This has a couple of advantages:

- No need to sync up the round-robin scheduler in firmware/hardware with
  the round-robin airtime scheduler.

- If several stations are eligible for transmission we can schedule both
  of them; no need to hard-block the scheduling rotation until the head
  of the queue has used up its quantum.

- The check of whether a station is eligible for transmission becomes
  simpler (in ieee80211_txq_may_transmit()).

The drawback is that scheduling becomes slightly more expensive, as we
need to maintain an rbtree of TXQs sorted by virtual time. This means
that ieee80211_register_airtime() becomes O(logN) in the number of
currently scheduled TXQs because it can change the order of the
scheduled stations. We mitigate this overhead by only resorting when a
station changes position in the tree, and hopefully N rarely grows too
big (it's only TXQs currently backlogged, not all associated stations),
so it shouldn't be too big of an issue.

To prevent divisions in the fast path, we maintain both station sums and
pre-computed reciprocals of the sums. This turns the fast-path operation
into a multiplication, with divisions only happening as the number of
active stations change (to re-compute the current sum of all active
station weights). To prevent this re-computation of the reciprocal from
happening too frequently, we use a time-based notion of station
activity, instead of updating the weight every time a station gets
scheduled or de-scheduled. As queues can oscillate between empty and
occupied quite frequently, this can significantly cut down on the number
of re-computations. It also has the added benefit of making the station
airtime calculation independent on whether the queue happened to have
drained at the time an airtime value was accounted.
Co-developed-by: Yibo Zhao <yiboz@codeaurora.org>
Signed-off-by: Yibo Zhao <yiboz@codeaurora.org>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/r/20210623134755.235545-1-toke@redhat.comSigned-off-by: Johannes Berg <johannes.berg@intel.com>

include/net/mac80211.h

@@ -6605,9 +6605,6 @@ static inline void ieee80211_txq_schedule_end(struct ieee80211_hw *hw, u8 ac)
 {
 }
 
-void __ieee80211_schedule_txq(struct ieee80211_hw *hw,
-			      struct ieee80211_txq *txq, bool force);
-
 /**
  * ieee80211_schedule_txq - schedule a TXQ for transmission
  *
@@ -6620,11 +6617,7 @@ void __ieee80211_schedule_txq(struct ieee80211_hw *hw,
  * The driver may call this function if it has buffered packets for
  * this TXQ internally.
  */
-static inline void
-ieee80211_schedule_txq(struct ieee80211_hw *hw, struct ieee80211_txq *txq)
-{
-	__ieee80211_schedule_txq(hw, txq, true);
-}
+void ieee80211_schedule_txq(struct ieee80211_hw *hw, struct ieee80211_txq *txq);
 
 /**
  * ieee80211_return_txq - return a TXQ previously acquired by ieee80211_next_txq()
@@ -6636,12 +6629,8 @@ ieee80211_schedule_txq(struct ieee80211_hw *hw, struct ieee80211_txq *txq)
  * The driver may set force=true if it has buffered packets for this TXQ
  * internally.
  */
-static inline void
-ieee80211_return_txq(struct ieee80211_hw *hw, struct ieee80211_txq *txq,
-		     bool force)
-{
-	__ieee80211_schedule_txq(hw, txq, force);
-}
+void ieee80211_return_txq(struct ieee80211_hw *hw, struct ieee80211_txq *txq,
+			  bool force);
 
 /**
  * ieee80211_txq_may_transmit - check whether TXQ is allowed to transmit

net/mac80211/cfg.c

@@ -1442,6 +1442,38 @@ static void sta_apply_mesh_params(struct ieee80211_local *local,
 #endif
 }
 
+static void sta_apply_airtime_params(struct ieee80211_local *local,
+				     struct sta_info *sta,
+				     struct station_parameters *params)
+{
+	u8 ac;
+
+	for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
+		struct airtime_sched_info *air_sched = &local->airtime[ac];
+		struct airtime_info *air_info = &sta->airtime[ac];
+		struct txq_info *txqi;
+		u8 tid;
+
+		spin_lock_bh(&air_sched->lock);
+		for (tid = 0; tid < IEEE80211_NUM_TIDS + 1; tid++) {
+			if (air_info->weight == params->airtime_weight ||
+			    !sta->sta.txq[tid] ||
+			    ac != ieee80211_ac_from_tid(tid))
+				continue;
+
+			airtime_weight_set(air_info, params->airtime_weight);
+
+			txqi = to_txq_info(sta->sta.txq[tid]);
+			if (RB_EMPTY_NODE(&txqi->schedule_order))
+				continue;
+
+			ieee80211_update_airtime_weight(local, air_sched,
+							0, true);
+		}
+		spin_unlock_bh(&air_sched->lock);
+	}
+}
+
 static int sta_apply_parameters(struct ieee80211_local *local,
 				struct sta_info *sta,
 				struct station_parameters *params)
@@ -1629,7 +1661,8 @@ static int sta_apply_parameters(struct ieee80211_local *local,
 		sta_apply_mesh_params(local, sta, params);
 
 	if (params->airtime_weight)
-		sta->airtime_weight = params->airtime_weight;
+		sta_apply_airtime_params(local, sta, params);
+
 
 	/* set the STA state after all sta info from usermode has been set */
 	if (test_sta_flag(sta, WLAN_STA_TDLS_PEER) ||

net/mac80211/debugfs.c

@@ -216,14 +216,14 @@ static ssize_t aql_txq_limit_read(struct file *file,
 			"VI	%u		%u\n"
 			"BE	%u		%u\n"
 			"BK	%u		%u\n",
-			local->aql_txq_limit_low[IEEE80211_AC_VO],
-			local->aql_txq_limit_high[IEEE80211_AC_VO],
-			local->aql_txq_limit_low[IEEE80211_AC_VI],
-			local->aql_txq_limit_high[IEEE80211_AC_VI],
-			local->aql_txq_limit_low[IEEE80211_AC_BE],
-			local->aql_txq_limit_high[IEEE80211_AC_BE],
-			local->aql_txq_limit_low[IEEE80211_AC_BK],
-			local->aql_txq_limit_high[IEEE80211_AC_BK]);
+			local->airtime[IEEE80211_AC_VO].aql_txq_limit_low,
+			local->airtime[IEEE80211_AC_VO].aql_txq_limit_high,
+			local->airtime[IEEE80211_AC_VI].aql_txq_limit_low,
+			local->airtime[IEEE80211_AC_VI].aql_txq_limit_high,
+			local->airtime[IEEE80211_AC_BE].aql_txq_limit_low,
+			local->airtime[IEEE80211_AC_BE].aql_txq_limit_high,
+			local->airtime[IEEE80211_AC_BK].aql_txq_limit_low,
+			local->airtime[IEEE80211_AC_BK].aql_txq_limit_high);
 	return simple_read_from_buffer(user_buf, count, ppos,
 				       buf, len);
 }
@@ -255,11 +255,11 @@ static ssize_t aql_txq_limit_write(struct file *file,
 	if (ac >= IEEE80211_NUM_ACS)
 		return -EINVAL;
 
-	q_limit_low_old = local->aql_txq_limit_low[ac];
-	q_limit_high_old = local->aql_txq_limit_high[ac];
+	q_limit_low_old = local->airtime[ac].aql_txq_limit_low;
+	q_limit_high_old = local->airtime[ac].aql_txq_limit_high;
 
-	local->aql_txq_limit_low[ac] = q_limit_low;
-	local->aql_txq_limit_high[ac] = q_limit_high;
+	local->airtime[ac].aql_txq_limit_low = q_limit_low;
+	local->airtime[ac].aql_txq_limit_high = q_limit_high;
 
 	mutex_lock(&local->sta_mtx);
 	list_for_each_entry(sta, &local->sta_list, list) {
@@ -382,6 +382,46 @@ static const struct file_operations force_tx_status_ops = {
 	.llseek = default_llseek,
 };
 
+static ssize_t airtime_read(struct file *file,
+			    char __user *user_buf,
+			    size_t count,
+			    loff_t *ppos)
+{
+	struct ieee80211_local *local = file->private_data;
+	char buf[200];
+	u64 v_t[IEEE80211_NUM_ACS];
+	u64 wt[IEEE80211_NUM_ACS];
+	int len = 0, ac;
+
+	for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
+		spin_lock_bh(&local->airtime[ac].lock);
+		v_t[ac] = local->airtime[ac].v_t;
+		wt[ac] = local->airtime[ac].weight_sum;
+		spin_unlock_bh(&local->airtime[ac].lock);
+	}
+	len = scnprintf(buf, sizeof(buf),
+			"\tVO         VI         BE         BK\n"
+			"Virt-t\t%-10llu %-10llu %-10llu %-10llu\n"
+			"Weight\t%-10llu %-10llu %-10llu %-10llu\n",
+			v_t[0],
+			v_t[1],
+			v_t[2],
+			v_t[3],
+			wt[0],
+			wt[1],
+			wt[2],
+			wt[3]);
+
+	return simple_read_from_buffer(user_buf, count, ppos,
+				       buf, len);
+}
+
+static const struct file_operations airtime_ops = {
+	.read = airtime_read,
+	.open = simple_open,
+	.llseek = default_llseek,
+};
+
 #ifdef CONFIG_PM
 static ssize_t reset_write(struct file *file, const char __user *user_buf,
 			   size_t count, loff_t *ppos)
@@ -632,7 +672,11 @@ void debugfs_hw_add(struct ieee80211_local *local)
 	if (local->ops->wake_tx_queue)
 		DEBUGFS_ADD_MODE(aqm, 0600);
 
-	DEBUGFS_ADD_MODE(airtime_flags, 0600);
+	if (wiphy_ext_feature_isset(local->hw.wiphy,
+				    NL80211_EXT_FEATURE_AIRTIME_FAIRNESS)) {
+		DEBUGFS_ADD_MODE(airtime, 0600);
+		DEBUGFS_ADD_MODE(airtime_flags, 0600);
+	}
 
 	DEBUGFS_ADD(aql_txq_limit);
 	debugfs_create_u32("aql_threshold", 0600,

net/mac80211/debugfs_netdev.c

@@ -512,6 +512,34 @@ static ssize_t ieee80211_if_fmt_aqm(
 }
 IEEE80211_IF_FILE_R(aqm);
 
+static ssize_t ieee80211_if_fmt_airtime(
+	const struct ieee80211_sub_if_data *sdata, char *buf, int buflen)
+{
+	struct ieee80211_local *local = sdata->local;
+	struct ieee80211_txq *txq = sdata->vif.txq;
+	struct airtime_info *air_info;
+	int len;
+
+	if (!txq)
+		return 0;
+
+	spin_lock_bh(&local->airtime[txq->ac].lock);
+	air_info = to_airtime_info(txq);
+	len = scnprintf(buf,
+			buflen,
+			"RX: %llu us\nTX: %llu us\nWeight: %u\n"
+			"Virt-T: %lld us\n",
+			air_info->rx_airtime,
+			air_info->tx_airtime,
+			air_info->weight,
+			air_info->v_t);
+	spin_unlock_bh(&local->airtime[txq->ac].lock);
+
+	return len;
+}
+
+IEEE80211_IF_FILE_R(airtime);
+
 IEEE80211_IF_FILE(multicast_to_unicast, u.ap.multicast_to_unicast, HEX);
 
 /* IBSS attributes */
@@ -657,8 +685,10 @@ static void add_common_files(struct ieee80211_sub_if_data *sdata)
 
 	if (sdata->local->ops->wake_tx_queue &&
 	    sdata->vif.type != NL80211_IFTYPE_P2P_DEVICE &&
-	    sdata->vif.type != NL80211_IFTYPE_NAN)
+	    sdata->vif.type != NL80211_IFTYPE_NAN) {
 		DEBUGFS_ADD(aqm);
+		DEBUGFS_ADD(airtime);
+	}
 }
 
 static void add_sta_files(struct ieee80211_sub_if_data *sdata)

net/mac80211/debugfs_sta.c

@@ -202,7 +202,7 @@ static ssize_t sta_airtime_read(struct file *file, char __user *userbuf,
 	size_t bufsz = 400;
 	char *buf = kzalloc(bufsz, GFP_KERNEL), *p = buf;
 	u64 rx_airtime = 0, tx_airtime = 0;
-	s64 deficit[IEEE80211_NUM_ACS];
+	u64 v_t[IEEE80211_NUM_ACS];
 	ssize_t rv;
 	int ac;
 
@@ -210,18 +210,18 @@ static ssize_t sta_airtime_read(struct file *file, char __user *userbuf,
 		return -ENOMEM;
 
 	for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
-		spin_lock_bh(&local->active_txq_lock[ac]);
+		spin_lock_bh(&local->airtime[ac].lock);
 		rx_airtime += sta->airtime[ac].rx_airtime;
 		tx_airtime += sta->airtime[ac].tx_airtime;
-		deficit[ac] = sta->airtime[ac].deficit;
-		spin_unlock_bh(&local->active_txq_lock[ac]);
+		v_t[ac] = sta->airtime[ac].v_t;
+		spin_unlock_bh(&local->airtime[ac].lock);
 	}
 
 	p += scnprintf(p, bufsz + buf - p,
 		"RX: %llu us\nTX: %llu us\nWeight: %u\n"
-		"Deficit: VO: %lld us VI: %lld us BE: %lld us BK: %lld us\n",
-		rx_airtime, tx_airtime, sta->airtime_weight,
-		deficit[0], deficit[1], deficit[2], deficit[3]);
+		"Virt-T: VO: %lld us VI: %lld us BE: %lld us BK: %lld us\n",
+		rx_airtime, tx_airtime, sta->airtime[0].weight,
+		v_t[0], v_t[1], v_t[2], v_t[3]);
 
 	rv = simple_read_from_buffer(userbuf, count, ppos, buf, p - buf);
 	kfree(buf);
@@ -236,11 +236,11 @@ static ssize_t sta_airtime_write(struct file *file, const char __user *userbuf,
 	int ac;
 
 	for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
-		spin_lock_bh(&local->active_txq_lock[ac]);
+		spin_lock_bh(&local->airtime[ac].lock);
 		sta->airtime[ac].rx_airtime = 0;
 		sta->airtime[ac].tx_airtime = 0;
-		sta->airtime[ac].deficit = sta->airtime_weight;
-		spin_unlock_bh(&local->active_txq_lock[ac]);
+		sta->airtime[ac].v_t = 0;
+		spin_unlock_bh(&local->airtime[ac].lock);
 	}
 
 	return count;
@@ -263,10 +263,10 @@ static ssize_t sta_aql_read(struct file *file, char __user *userbuf,
 		return -ENOMEM;
 
 	for (ac = 0; ac < IEEE80211_NUM_ACS; ac++) {
-		spin_lock_bh(&local->active_txq_lock[ac]);
+		spin_lock_bh(&local->airtime[ac].lock);
 		q_limit_l[ac] = sta->airtime[ac].aql_limit_low;
 		q_limit_h[ac] = sta->airtime[ac].aql_limit_high;
-		spin_unlock_bh(&local->active_txq_lock[ac]);
+		spin_unlock_bh(&local->airtime[ac].lock);
 		q_depth[ac] = atomic_read(&sta->airtime[ac].aql_tx_pending);
 	}
 

net/mac80211/ieee80211_i.h

@@ -831,20 +831,16 @@ enum txq_info_flags {
  * @def_flow: used as a fallback flow when a packet destined to @tin hashes to
  *	a fq_flow which is already owned by a different tin
  * @def_cvars: codel vars for @def_flow
- * @frags: used to keep fragments created after dequeue
  * @schedule_order: used with ieee80211_local->active_txqs
- * @schedule_round: counter to prevent infinite loops on TXQ scheduling
+ * @frags: used to keep fragments created after dequeue
  */
 struct txq_info {
 	struct fq_tin tin;
 	struct codel_vars def_cvars;
 	struct codel_stats cstats;
-
-	u16 schedule_round;
-	struct list_head schedule_order;
+	struct rb_node schedule_order;
 
 	struct sk_buff_head frags;
-
 	unsigned long flags;
 
 	/* keep last! */
@@ -921,6 +917,8 @@ struct ieee80211_sub_if_data {
 	struct ieee80211_tx_queue_params tx_conf[IEEE80211_NUM_ACS];
 	struct mac80211_qos_map __rcu *qos_map;
 
+	struct airtime_info airtime[IEEE80211_NUM_ACS];
+
 	struct work_struct csa_finalize_work;
 	bool csa_block_tx; /* write-protected by sdata_lock and local->mtx */
 	struct cfg80211_chan_def csa_chandef;
@@ -1133,6 +1131,44 @@ enum mac80211_scan_state {
 	SCAN_ABORT,
 };
 
+/**
+ * struct airtime_sched_info - state used for airtime scheduling and AQL
+ *
+ * @lock: spinlock that protects all the fields in this struct
+ * @active_txqs: rbtree of currently backlogged queues, sorted by virtual time
+ * @schedule_pos: the current position maintained while a driver walks the tree
+ *                with ieee80211_next_txq()
+ * @active_list: list of struct airtime_info structs that were active within
+ *               the last AIRTIME_ACTIVE_DURATION (100 ms), used to compute
+ *               weight_sum
+ * @last_weight_update: used for rate limiting walking active_list
+ * @last_schedule_time: tracks the last time a transmission was scheduled; used
+ *                      for catching up v_t if no stations are eligible for
+ *                      transmission.
+ * @v_t: global virtual time; queues with v_t < this are eligible for
+ *       transmission
+ * @weight_sum: total sum of all active stations used for dividing airtime
+ * @weight_sum_reciprocal: reciprocal of weight_sum (to avoid divisions in fast
+ *                         path - see comment above
+ *                         IEEE80211_RECIPROCAL_DIVISOR_64)
+ * @aql_txq_limit_low: AQL limit when total outstanding airtime
+ *                     is < IEEE80211_AQL_THRESHOLD
+ * @aql_txq_limit_high: AQL limit when total outstanding airtime
+ *                      is > IEEE80211_AQL_THRESHOLD
+ */
+struct airtime_sched_info {
+	spinlock_t lock;
+	struct rb_root_cached active_txqs;
+	struct rb_node *schedule_pos;
+	struct list_head active_list;
+	u64 last_weight_update;
+	u64 last_schedule_activity;
+	u64 v_t;
+	u64 weight_sum;
+	u64 weight_sum_reciprocal;
+	u32 aql_txq_limit_low;
+	u32 aql_txq_limit_high;
+};
 DECLARE_STATIC_KEY_FALSE(aql_disable);
 
 struct ieee80211_local {
@@ -1146,13 +1182,8 @@ struct ieee80211_local {
 	struct codel_params cparams;
 
 	/* protects active_txqs and txqi->schedule_order */
-	spinlock_t active_txq_lock[IEEE80211_NUM_ACS];
-	struct list_head active_txqs[IEEE80211_NUM_ACS];
-	u16 schedule_round[IEEE80211_NUM_ACS];
-
+	struct airtime_sched_info airtime[IEEE80211_NUM_ACS];
 	u16 airtime_flags;
-	u32 aql_txq_limit_low[IEEE80211_NUM_ACS];
-	u32 aql_txq_limit_high[IEEE80211_NUM_ACS];
 	u32 aql_threshold;
 	atomic_t aql_total_pending_airtime;
 
@@ -1566,6 +1597,125 @@ static inline bool txq_has_queue(struct ieee80211_txq *txq)
 	return !(skb_queue_empty(&txqi->frags) && !txqi->tin.backlog_packets);
 }
 
+static inline struct airtime_info *to_airtime_info(struct ieee80211_txq *txq)
+{
+	struct ieee80211_sub_if_data *sdata;
+	struct sta_info *sta;
+
+	if (txq->sta) {
+		sta = container_of(txq->sta, struct sta_info, sta);
+		return &sta->airtime[txq->ac];
+	}
+
+	sdata = vif_to_sdata(txq->vif);
+	return &sdata->airtime[txq->ac];
+}
+
+/* To avoid divisions in the fast path, we keep pre-computed reciprocals for
+ * airtime weight calculations. There are two different weights to keep track
+ * of: The per-station weight and the sum of weights per phy.
+ *
+ * For the per-station weights (kept in airtime_info below), we use 32-bit
+ * reciprocals with a devisor of 2^19. This lets us keep the multiplications and
+ * divisions for the station weights as 32-bit operations at the cost of a bit
+ * of rounding error for high weights; but the choice of divisor keeps rounding
+ * errors <10% for weights <2^15, assuming no more than 8ms of airtime is
+ * reported at a time.
+ *
+ * For the per-phy sum of weights the values can get higher, so we use 64-bit
+ * operations for those with a 32-bit divisor, which should avoid any
+ * significant rounding errors.
+ */
+#define IEEE80211_RECIPROCAL_DIVISOR_64 0x100000000ULL
+#define IEEE80211_RECIPROCAL_SHIFT_64 32
+#define IEEE80211_RECIPROCAL_DIVISOR_32 0x80000U
+#define IEEE80211_RECIPROCAL_SHIFT_32 19
+
+static inline void airtime_weight_set(struct airtime_info *air_info, u16 weight)
+{
+	if (air_info->weight == weight)
+		return;
+
+	air_info->weight = weight;
+	if (weight) {
+		air_info->weight_reciprocal =
+			IEEE80211_RECIPROCAL_DIVISOR_32 / weight;
+	} else {
+		air_info->weight_reciprocal = 0;
+	}
+}
+
+static inline void airtime_weight_sum_set(struct airtime_sched_info *air_sched,
+					  int weight_sum)
+{
+	if (air_sched->weight_sum == weight_sum)
+		return;
+
+	air_sched->weight_sum = weight_sum;
+	if (air_sched->weight_sum) {
+		air_sched->weight_sum_reciprocal = IEEE80211_RECIPROCAL_DIVISOR_64;
+		do_div(air_sched->weight_sum_reciprocal, air_sched->weight_sum);
+	} else {
+		air_sched->weight_sum_reciprocal = 0;
+	}
+}
+
+/* A problem when trying to enforce airtime fairness is that we want to divide
+ * the airtime between the currently *active* stations. However, basing this on
+ * the instantaneous queue state of stations doesn't work, as queues tend to
+ * oscillate very quickly between empty and occupied, leading to the scheduler
+ * thinking only a single station is active when deciding whether to allow
+ * transmission (and thus not throttling correctly).
+ *
+ * To fix this we use a timer-based notion of activity: a station is considered
+ * active if it has been scheduled within the last 100 ms; we keep a separate
+ * list of all the stations considered active in this manner, and lazily update
+ * the total weight of active stations from this list (filtering the stations in
+ * the list by their 'last active' time).
+ *
+ * We add one additional safeguard to guard against stations that manage to get
+ * scheduled every 100 ms but don't transmit a lot of data, and thus don't use
+ * up any airtime. Such stations would be able to get priority for an extended
+ * period of time if they do start transmitting at full capacity again, and so
+ * we add an explicit maximum for how far behind a station is allowed to fall in
+ * the virtual airtime domain. This limit is set to a relatively high value of
+ * 20 ms because the main mechanism for catching up idle stations is the active
+ * state as described above; i.e., the hard limit should only be hit in
+ * pathological cases.
+ */
+#define AIRTIME_ACTIVE_DURATION (100 * NSEC_PER_MSEC)
+#define AIRTIME_MAX_BEHIND 20000 /* 20 ms */
+
+static inline bool airtime_is_active(struct airtime_info *air_info, u64 now)
+{
+	return air_info->last_scheduled >= now - AIRTIME_ACTIVE_DURATION;
+}
+
+static inline void airtime_set_active(struct airtime_sched_info *air_sched,
+				      struct airtime_info *air_info, u64 now)
+{
+	air_info->last_scheduled = now;
+	air_sched->last_schedule_activity = now;
+	list_move_tail(&air_info->list, &air_sched->active_list);
+}
+
+static inline bool airtime_catchup_v_t(struct airtime_sched_info *air_sched,
+				       u64 v_t, u64 now)
+{
+	air_sched->v_t = v_t;
+	return true;
+}
+
+static inline void init_airtime_info(struct airtime_info *air_info,
+				     struct airtime_sched_info *air_sched)
+{
+	atomic_set(&air_info->aql_tx_pending, 0);
+	air_info->aql_limit_low = air_sched->aql_txq_limit_low;
+	air_info->aql_limit_high = air_sched->aql_txq_limit_high;
+	airtime_weight_set(air_info, IEEE80211_DEFAULT_AIRTIME_WEIGHT);
+	INIT_LIST_HEAD(&air_info->list);
+}
+
 static inline int ieee80211_bssid_match(const u8 *raddr, const u8 *addr)
 {
 	return ether_addr_equal(raddr, addr) ||
@@ -1808,6 +1958,14 @@ int ieee80211_tx_control_port(struct wiphy *wiphy, struct net_device *dev,
 			      u64 *cookie);
 int ieee80211_probe_mesh_link(struct wiphy *wiphy, struct net_device *dev,
 			      const u8 *buf, size_t len);
+void ieee80211_resort_txq(struct ieee80211_hw *hw,
+			  struct ieee80211_txq *txq);
+void ieee80211_unschedule_txq(struct ieee80211_hw *hw,
+			      struct ieee80211_txq *txq,
+			      bool purge);
+void ieee80211_update_airtime_weight(struct ieee80211_local *local,
+				     struct airtime_sched_info *air_sched,
+				     u64 now, bool force);
 
 /* HT */
 void ieee80211_apply_htcap_overrides(struct ieee80211_sub_if_data *sdata,

net/mac80211/iface.c

@@ -1977,6 +1977,9 @@ int ieee80211_if_add(struct ieee80211_local *local, const char *name,
 		}
 	}
 
+	for (i = 0; i < IEEE80211_NUM_ACS; i++)
+		init_airtime_info(&sdata->airtime[i], &local->airtime[i]);
+
 	ieee80211_set_default_queues(sdata);
 
 	sdata->ap_power_level = IEEE80211_UNSET_POWER_LEVEL;

net/mac80211/main.c

@@ -705,10 +705,13 @@ struct ieee80211_hw *ieee80211_alloc_hw_nm(size_t priv_data_len,
 	spin_lock_init(&local->queue_stop_reason_lock);
 
 	for (i = 0; i < IEEE80211_NUM_ACS; i++) {
-		INIT_LIST_HEAD(&local->active_txqs[i]);
-		spin_lock_init(&local->active_txq_lock[i]);
-		local->aql_txq_limit_low[i] = IEEE80211_DEFAULT_AQL_TXQ_LIMIT_L;
-		local->aql_txq_limit_high[i] =
+		struct airtime_sched_info *air_sched = &local->airtime[i];
+
+		air_sched->active_txqs = RB_ROOT_CACHED;
+		INIT_LIST_HEAD(&air_sched->active_list);
+		spin_lock_init(&air_sched->lock);
+		air_sched->aql_txq_limit_low = IEEE80211_DEFAULT_AQL_TXQ_LIMIT_L;
+		air_sched->aql_txq_limit_high =
 			IEEE80211_DEFAULT_AQL_TXQ_LIMIT_H;
 	}
 

net/mac80211/rx.c

@@ -1578,12 +1578,8 @@ static void sta_ps_start(struct sta_info *sta)
 
 	for (tid = 0; tid < IEEE80211_NUM_TIDS; tid++) {
 		struct ieee80211_txq *txq = sta->sta.txq[tid];
-		struct txq_info *txqi = to_txq_info(txq);
 
-		spin_lock(&local->active_txq_lock[txq->ac]);
-		if (!list_empty(&txqi->schedule_order))
-			list_del_init(&txqi->schedule_order);
-		spin_unlock(&local->active_txq_lock[txq->ac]);
+		ieee80211_unschedule_txq(&local->hw, txq, false);
 
 		if (txq_has_queue(txq))
 			set_bit(tid, &sta->txq_buffered_tids);

net/mac80211/sta_info.c

@@ -425,15 +425,11 @@ struct sta_info *sta_info_alloc(struct ieee80211_sub_if_data *sdata,
 	if (sta_prepare_rate_control(local, sta, gfp))
 		goto free_txq;
 
-	sta->airtime_weight = IEEE80211_DEFAULT_AIRTIME_WEIGHT;
 
 	for (i = 0; i < IEEE80211_NUM_ACS; i++) {
 		skb_queue_head_init(&sta->ps_tx_buf[i]);
 		skb_queue_head_init(&sta->tx_filtered[i]);
-		sta->airtime[i].deficit = sta->airtime_weight;
-		atomic_set(&sta->airtime[i].aql_tx_pending, 0);
-		sta->airtime[i].aql_limit_low = local->aql_txq_limit_low[i];
-		sta->airtime[i].aql_limit_high = local->aql_txq_limit_high[i];
+		init_airtime_info(&sta->airtime[i], &local->airtime[i]);
 	}
 
 	for (i = 0; i < IEEE80211_NUM_TIDS; i++)
@@ -1892,24 +1888,59 @@ void ieee80211_sta_set_buffered(struct ieee80211_sta *pubsta,
 }
 EXPORT_SYMBOL(ieee80211_sta_set_buffered);
 
-void ieee80211_sta_register_airtime(struct ieee80211_sta *pubsta, u8 tid,
-				    u32 tx_airtime, u32 rx_airtime)
+void ieee80211_register_airtime(struct ieee80211_txq *txq,
+				u32 tx_airtime, u32 rx_airtime)
 {
-	struct sta_info *sta = container_of(pubsta, struct sta_info, sta);
-	struct ieee80211_local *local = sta->sdata->local;
-	u8 ac = ieee80211_ac_from_tid(tid);
+	struct ieee80211_sub_if_data *sdata = vif_to_sdata(txq->vif);
+	struct ieee80211_local *local = sdata->local;
+	u64 weight_sum, weight_sum_reciprocal;
+	struct airtime_sched_info *air_sched;
+	struct airtime_info *air_info;
 	u32 airtime = 0;
 
-	if (sta->local->airtime_flags & AIRTIME_USE_TX)
+	air_sched = &local->airtime[txq->ac];
+	air_info = to_airtime_info(txq);
+
+	if (local->airtime_flags & AIRTIME_USE_TX)
 		airtime += tx_airtime;
-	if (sta->local->airtime_flags & AIRTIME_USE_RX)
+	if (local->airtime_flags & AIRTIME_USE_RX)
 		airtime += rx_airtime;
 
-	spin_lock_bh(&local->active_txq_lock[ac]);
-	sta->airtime[ac].tx_airtime += tx_airtime;
-	sta->airtime[ac].rx_airtime += rx_airtime;
-	sta->airtime[ac].deficit -= airtime;
-	spin_unlock_bh(&local->active_txq_lock[ac]);
+	/* Weights scale so the unit weight is 256 */
+	airtime <<= 8;
+
+	spin_lock_bh(&air_sched->lock);
+
+	air_info->tx_airtime += tx_airtime;
+	air_info->rx_airtime += rx_airtime;
+
+	if (air_sched->weight_sum) {
+		weight_sum = air_sched->weight_sum;
+		weight_sum_reciprocal = air_sched->weight_sum_reciprocal;
+	} else {
+		weight_sum = air_info->weight;
+		weight_sum_reciprocal = air_info->weight_reciprocal;
+	}
+
+	/* Round the calculation of global vt */
+	air_sched->v_t += (u64)((airtime + (weight_sum >> 1)) *
+				weight_sum_reciprocal) >> IEEE80211_RECIPROCAL_SHIFT_64;
+	air_info->v_t += (u32)((airtime + (air_info->weight >> 1)) *
+			       air_info->weight_reciprocal) >> IEEE80211_RECIPROCAL_SHIFT_32;
+	ieee80211_resort_txq(&local->hw, txq);
+
+	spin_unlock_bh(&air_sched->lock);
+}
+
+void ieee80211_sta_register_airtime(struct ieee80211_sta *pubsta, u8 tid,
+				    u32 tx_airtime, u32 rx_airtime)
+{
+	struct ieee80211_txq *txq = pubsta->txq[tid];
+
+	if (!txq)
+		return;
+
+	ieee80211_register_airtime(txq, tx_airtime, rx_airtime);
 }
 EXPORT_SYMBOL(ieee80211_sta_register_airtime);
 
@@ -2353,7 +2384,7 @@ void sta_set_sinfo(struct sta_info *sta, struct station_info *sinfo,
 	}
 
 	if (!(sinfo->filled & BIT_ULL(NL80211_STA_INFO_AIRTIME_WEIGHT))) {
-		sinfo->airtime_weight = sta->airtime_weight;
+		sinfo->airtime_weight = sta->airtime[0].weight;
 		sinfo->filled |= BIT_ULL(NL80211_STA_INFO_AIRTIME_WEIGHT);
 	}
 

net/mac80211/sta_info.h

@@ -135,18 +135,25 @@ enum ieee80211_agg_stop_reason {
 #define AIRTIME_USE_TX		BIT(0)
 #define AIRTIME_USE_RX		BIT(1)
 
+
 struct airtime_info {
 	u64 rx_airtime;
 	u64 tx_airtime;
-	s64 deficit;
+	u64 v_t;
+	u64 last_scheduled;
+	struct list_head list;
 	atomic_t aql_tx_pending; /* Estimated airtime for frames pending */
 	u32 aql_limit_low;
 	u32 aql_limit_high;
+	u32 weight_reciprocal;
+	u16 weight;
 };
 
 void ieee80211_sta_update_pending_airtime(struct ieee80211_local *local,
 					  struct sta_info *sta, u8 ac,
 					  u16 tx_airtime, bool tx_completed);
+void ieee80211_register_airtime(struct ieee80211_txq *txq,
+				u32 tx_airtime, u32 rx_airtime);
 
 struct sta_info;
 
@@ -515,7 +522,6 @@ struct ieee80211_fragment_cache {
  * @tid_seq: per-TID sequence numbers for sending to this STA
  * @airtime: per-AC struct airtime_info describing airtime statistics for this
  *	station
- * @airtime_weight: station weight for airtime fairness calculation purposes
  * @ampdu_mlme: A-MPDU state machine state
  * @mesh: mesh STA information
  * @debugfs_dir: debug filesystem directory dentry
@@ -646,7 +652,6 @@ struct sta_info {
 	u16 tid_seq[IEEE80211_QOS_CTL_TID_MASK + 1];
 
 	struct airtime_info airtime[IEEE80211_NUM_ACS];
-	u16 airtime_weight;
 
 	/*
 	 * Aggregation information, locked with lock.

net/mac80211/status.c

@@ -970,6 +970,25 @@ static void __ieee80211_tx_status(struct ieee80211_hw *hw,
 		if (!(info->flags & IEEE80211_TX_CTL_INJECTED) && acked)
 			ieee80211_frame_acked(sta, skb);
 
+	} else if (wiphy_ext_feature_isset(local->hw.wiphy,
+					   NL80211_EXT_FEATURE_AIRTIME_FAIRNESS)) {
+		struct ieee80211_sub_if_data *sdata;
+		struct ieee80211_txq *txq;
+		u32 airtime;
+
+		/* Account airtime to multicast queue */
+		sdata = ieee80211_sdata_from_skb(local, skb);
+
+		if (sdata && (txq = sdata->vif.txq)) {
+			airtime = info->status.tx_time ?:
+				ieee80211_calc_expected_tx_airtime(hw,
+								   &sdata->vif,
+								   NULL,
+								   skb->len,
+								   false);
+
+			ieee80211_register_airtime(txq, airtime, 0);
+		}
 	}
 
 	/* SNMP counters