--- /dev/null
+From: Felix Fietkau <nbd@nbd.name>
+Date: Fri, 9 Feb 2024 19:43:40 +0100
+Subject: [PATCH] mac80211: add AQL support for broadcast packets
+
+Excessive broadcast traffic with little competing unicast traffic can easily
+flood hardware queues, leading to throughput issues. Additionally, filling
+the hardware queues with too many packets breaks FQ for broadcast data.
+Fix this by enabling AQL for broadcast packets.
+
+Signed-off-by: Felix Fietkau <nbd@nbd.name>
+---
+
+--- a/include/net/cfg80211.h
++++ b/include/net/cfg80211.h
+@@ -3324,6 +3324,7 @@ enum wiphy_params_flags {
+ /* The per TXQ device queue limit in airtime */
+ #define IEEE80211_DEFAULT_AQL_TXQ_LIMIT_L 5000
+ #define IEEE80211_DEFAULT_AQL_TXQ_LIMIT_H 12000
++#define IEEE80211_DEFAULT_AQL_TXQ_LIMIT_BC 50000
+
+ /* The per interface airtime threshold to switch to lower queue limit */
+ #define IEEE80211_AQL_THRESHOLD 24000
+--- a/net/mac80211/debugfs.c
++++ b/net/mac80211/debugfs.c
+@@ -215,11 +215,13 @@ static ssize_t aql_pending_read(struct f
+ "VI %u us\n"
+ "BE %u us\n"
+ "BK %u us\n"
++ "BC/MC %u us\n"
+ "total %u us\n",
+ atomic_read(&local->aql_ac_pending_airtime[IEEE80211_AC_VO]),
+ atomic_read(&local->aql_ac_pending_airtime[IEEE80211_AC_VI]),
+ atomic_read(&local->aql_ac_pending_airtime[IEEE80211_AC_BE]),
+ atomic_read(&local->aql_ac_pending_airtime[IEEE80211_AC_BK]),
++ atomic_read(&local->aql_bc_pending_airtime),
+ atomic_read(&local->aql_total_pending_airtime));
+ return simple_read_from_buffer(user_buf, count, ppos,
+ buf, len);
+@@ -245,7 +247,8 @@ static ssize_t aql_txq_limit_read(struct
+ "VO %u %u\n"
+ "VI %u %u\n"
+ "BE %u %u\n"
+- "BK %u %u\n",
++ "BK %u %u\n"
++ "BC/MC %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],
+@@ -253,7 +256,8 @@ static ssize_t aql_txq_limit_read(struct
+ 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->aql_txq_limit_high[IEEE80211_AC_BK],
++ local->aql_txq_limit_bc);
+ return simple_read_from_buffer(user_buf, count, ppos,
+ buf, len);
+ }
+@@ -279,6 +283,11 @@ static ssize_t aql_txq_limit_write(struc
+ else
+ buf[count] = '\0';
+
++ if (sscanf(buf, "mcast %u", &q_limit_low) == 1) {
++ local->aql_txq_limit_bc = q_limit_low;
++ return count;
++ }
++
+ if (sscanf(buf, "%u %u %u", &ac, &q_limit_low, &q_limit_high) != 3)
+ return -EINVAL;
+
+--- a/net/mac80211/ieee80211_i.h
++++ b/net/mac80211/ieee80211_i.h
+@@ -1328,10 +1328,12 @@ struct ieee80211_local {
+ spinlock_t handle_wake_tx_queue_lock;
+
+ u16 airtime_flags;
++ u32 aql_txq_limit_bc;
+ 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;
++ atomic_t aql_bc_pending_airtime;
+ atomic_t aql_ac_pending_airtime[IEEE80211_NUM_ACS];
+
+ const struct ieee80211_ops *ops;
+--- a/net/mac80211/main.c
++++ b/net/mac80211/main.c
+@@ -788,6 +788,7 @@ struct ieee80211_hw *ieee80211_alloc_hw_
+ spin_lock_init(&local->rx_path_lock);
+ spin_lock_init(&local->queue_stop_reason_lock);
+
++ local->aql_txq_limit_bc = IEEE80211_DEFAULT_AQL_TXQ_LIMIT_BC;
+ for (i = 0; i < IEEE80211_NUM_ACS; i++) {
+ INIT_LIST_HEAD(&local->active_txqs[i]);
+ spin_lock_init(&local->active_txq_lock[i]);
+--- a/net/mac80211/sta_info.c
++++ b/net/mac80211/sta_info.c
+@@ -2341,28 +2341,27 @@ void ieee80211_sta_update_pending_airtim
+ struct sta_info *sta, u8 ac,
+ u16 tx_airtime, bool tx_completed)
+ {
++ atomic_t *counter;
+ int tx_pending;
+
+ if (!wiphy_ext_feature_isset(local->hw.wiphy, NL80211_EXT_FEATURE_AQL))
+ return;
+
+- if (!tx_completed) {
+- if (sta)
+- atomic_add(tx_airtime,
+- &sta->airtime[ac].aql_tx_pending);
++ if (sta)
++ counter = &sta->airtime[ac].aql_tx_pending;
++ else
++ counter = &local->aql_bc_pending_airtime;
+
++ if (!tx_completed) {
++ atomic_add(tx_airtime, counter);
+ atomic_add(tx_airtime, &local->aql_total_pending_airtime);
+ atomic_add(tx_airtime, &local->aql_ac_pending_airtime[ac]);
+ return;
+ }
+
+- if (sta) {
+- tx_pending = atomic_sub_return(tx_airtime,
+- &sta->airtime[ac].aql_tx_pending);
+- if (tx_pending < 0)
+- atomic_cmpxchg(&sta->airtime[ac].aql_tx_pending,
+- tx_pending, 0);
+- }
++ tx_pending = atomic_sub_return(tx_airtime, counter);
++ if (tx_pending < 0)
++ atomic_cmpxchg(counter, tx_pending, 0);
+
+ atomic_sub(tx_airtime, &local->aql_total_pending_airtime);
+ tx_pending = atomic_sub_return(tx_airtime,
+--- a/net/mac80211/tx.c
++++ b/net/mac80211/tx.c
+@@ -3958,9 +3958,8 @@ begin:
+ encap_out:
+ IEEE80211_SKB_CB(skb)->control.vif = vif;
+
+- if (tx.sta &&
+- wiphy_ext_feature_isset(local->hw.wiphy, NL80211_EXT_FEATURE_AQL)) {
+- bool ampdu = txq->ac != IEEE80211_AC_VO;
++ if (wiphy_ext_feature_isset(local->hw.wiphy, NL80211_EXT_FEATURE_AQL)) {
++ bool ampdu = txq->sta && txq->ac != IEEE80211_AC_VO;
+ u32 airtime;
+
+ airtime = ieee80211_calc_expected_tx_airtime(hw, vif, txq->sta,
+@@ -4026,6 +4025,7 @@ struct ieee80211_txq *ieee80211_next_txq
+ struct ieee80211_txq *ret = NULL;
+ struct txq_info *txqi = NULL, *head = NULL;
+ bool found_eligible_txq = false;
++ bool aql_check;
+
+ spin_lock_bh(&local->active_txq_lock[ac]);
+
+@@ -4049,26 +4049,26 @@ struct ieee80211_txq *ieee80211_next_txq
+ if (!head)
+ head = txqi;
+
++ aql_check = ieee80211_txq_airtime_check(hw, &txqi->txq);
++ if (aql_check)
++ found_eligible_txq = true;
++
+ if (txqi->txq.sta) {
+ struct sta_info *sta = container_of(txqi->txq.sta,
+ struct sta_info, sta);
+- bool aql_check = ieee80211_txq_airtime_check(hw, &txqi->txq);
+- s32 deficit = ieee80211_sta_deficit(sta, txqi->txq.ac);
+-
+- if (aql_check)
+- found_eligible_txq = true;
+-
+- if (deficit < 0)
++ if (ieee80211_sta_deficit(sta, txqi->txq.ac) < 0) {
+ sta->airtime[txqi->txq.ac].deficit +=
+ sta->airtime_weight << AIRTIME_QUANTUM_SHIFT;
+-
+- if (deficit < 0 || !aql_check) {
+- list_move_tail(&txqi->schedule_order,
+- &local->active_txqs[txqi->txq.ac]);
+- goto begin;
++ aql_check = false;
+ }
+ }
+
++ if (!aql_check) {
++ list_move_tail(&txqi->schedule_order,
++ &local->active_txqs[txqi->txq.ac]);
++ goto begin;
++ }
++
+ if (txqi->schedule_round == local->schedule_round[ac])
+ goto out;
+
+@@ -4133,7 +4133,8 @@ bool ieee80211_txq_airtime_check(struct
+ return true;
+
+ if (!txq->sta)
+- return true;
++ return atomic_read(&local->aql_bc_pending_airtime) <
++ local->aql_txq_limit_bc;
+
+ if (unlikely(txq->tid == IEEE80211_NUM_TIDS))
+ return true;
+@@ -4182,15 +4183,15 @@ bool ieee80211_txq_may_transmit(struct i
+
+ spin_lock_bh(&local->active_txq_lock[ac]);
+
+- if (!txqi->txq.sta)
+- goto out;
+-
+ if (list_empty(&txqi->schedule_order))
+ goto out;
+
+ if (!ieee80211_txq_schedule_airtime_check(local, ac))
+ goto out;
+
++ if (!txqi->txq.sta)
++ goto out;
++
+ list_for_each_entry_safe(iter, tmp, &local->active_txqs[ac],
+ schedule_order) {
+ if (iter == txqi)