mac80211: improve rate control performance

Call rate control handler after intermediate queueuing
Includes follow-up fixes

Signed-off-by: Felix Fietkau <nbd@nbd.name>

cherry-picked from commits:
- 7dd8829ef915f1c5fc728be8f8360c61ddaadf1b
- a603e82dd342680d584c4eb5f1b222e056379890
- 8bb4437c01ca35a5ac67e391630a1b24cb52dbb7

4 files changed, 353 insertions, 1 deletions

diff --git a/package/kernel/mac80211/patches/subsys/374-mac80211-move-A-MPDU-session-check-from-minstrel_ht-.patch b/package/kernel/mac80211/patches/subsys/374-mac80211-move-A-MPDU-session-check-from-minstrel_ht-.patch
new file mode 100644
index 0000000000..4e3242e6a9
--- /dev/null
+++ b/package/kernel/mac80211/patches/subsys/374-mac80211-move-A-MPDU-session-check-from-minstrel_ht-.patch
@@ -0,0 +1,126 @@
+From: Felix Fietkau <nbd@nbd.name>
+Date: Thu, 17 Jun 2021 17:56:54 +0200
+Subject: [PATCH] mac80211: move A-MPDU session check from minstrel_ht to
+ mac80211
+
+This avoids calling back into tx handlers from within the rate control module.
+Preparation for deferring rate control until tx dequeue
+
+Signed-off-by: Felix Fietkau <nbd@nbd.name>
+---
+
+--- a/include/net/mac80211.h
++++ b/include/net/mac80211.h
+@@ -6160,6 +6160,11 @@ enum rate_control_capabilities {
+ 	 * otherwise the NSS difference doesn't bother us.
+ 	 */
+ 	RATE_CTRL_CAPA_VHT_EXT_NSS_BW = BIT(0),
++	/**
++	 * @RATE_CTRL_CAPA_AMPDU_TRIGGER:
++	 * mac80211 should start A-MPDU sessions on tx
++	 */
++	RATE_CTRL_CAPA_AMPDU_TRIGGER = BIT(1),
+ };
+ 
+ struct rate_control_ops {
+--- a/net/mac80211/rc80211_minstrel_ht.c
++++ b/net/mac80211/rc80211_minstrel_ht.c
+@@ -1153,29 +1153,6 @@ minstrel_downgrade_prob_rate(struct mins
+ }
+ 
+ static void
+-minstrel_aggr_check(struct ieee80211_sta *pubsta, struct sk_buff *skb)
+-{
+-	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
+-	struct sta_info *sta = container_of(pubsta, struct sta_info, sta);
+-	u16 tid;
+-
+-	if (skb_get_queue_mapping(skb) == IEEE80211_AC_VO)
+-		return;
+-
+-	if (unlikely(!ieee80211_is_data_qos(hdr->frame_control)))
+-		return;
+-
+-	if (unlikely(skb->protocol == cpu_to_be16(ETH_P_PAE)))
+-		return;
+-
+-	tid = ieee80211_get_tid(hdr);
+-	if (likely(sta->ampdu_mlme.tid_tx[tid]))
+-		return;
+-
+-	ieee80211_start_tx_ba_session(pubsta, tid, 0);
+-}
+-
+-static void
+ minstrel_ht_tx_status(void *priv, struct ieee80211_supported_band *sband,
+                       void *priv_sta, struct ieee80211_tx_status *st)
+ {
+@@ -1477,10 +1454,6 @@ minstrel_ht_get_rate(void *priv, struct
+ 	struct minstrel_priv *mp = priv;
+ 	u16 sample_idx;
+ 
+-	if (!(info->flags & IEEE80211_TX_CTL_AMPDU) &&
+-	    !minstrel_ht_is_legacy_group(MI_RATE_GROUP(mi->max_prob_rate)))
+-		minstrel_aggr_check(sta, txrc->skb);
+-
+ 	info->flags |= mi->tx_flags;
+ 
+ #ifdef CPTCFG_MAC80211_DEBUGFS
+@@ -1894,6 +1867,7 @@ static u32 minstrel_ht_get_expected_thro
+ 
+ static const struct rate_control_ops mac80211_minstrel_ht = {
+ 	.name = "minstrel_ht",
++	.capa = RATE_CTRL_CAPA_AMPDU_TRIGGER,
+ 	.tx_status_ext = minstrel_ht_tx_status,
+ 	.get_rate = minstrel_ht_get_rate,
+ 	.rate_init = minstrel_ht_rate_init,
+--- a/net/mac80211/tx.c
++++ b/net/mac80211/tx.c
+@@ -3933,6 +3933,29 @@ void ieee80211_txq_schedule_start(struct
+ }
+ EXPORT_SYMBOL(ieee80211_txq_schedule_start);
+ 
++static void
++ieee80211_aggr_check(struct ieee80211_sub_if_data *sdata,
++		     struct sta_info *sta,
++		     struct sk_buff *skb)
++{
++	struct rate_control_ref *ref = sdata->local->rate_ctrl;
++	u16 tid;
++
++	if (!ref || !(ref->ops->capa & RATE_CTRL_CAPA_AMPDU_TRIGGER))
++		return;
++
++	if (!sta || !sta->sta.ht_cap.ht_supported ||
++	    !sta->sta.wme || skb_get_queue_mapping(skb) == IEEE80211_AC_VO ||
++	    skb->protocol == sdata->control_port_protocol)
++		return;
++
++	tid = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
++	if (likely(sta->ampdu_mlme.tid_tx[tid]))
++		return;
++
++	ieee80211_start_tx_ba_session(&sta->sta, tid, 0);
++}
++
+ void __ieee80211_subif_start_xmit(struct sk_buff *skb,
+ 				  struct net_device *dev,
+ 				  u32 info_flags,
+@@ -3963,6 +3986,8 @@ void __ieee80211_subif_start_xmit(struct
+ 		skb_get_hash(skb);
+ 	}
+ 
++	ieee80211_aggr_check(sdata, sta, skb);
++
+ 	if (sta) {
+ 		struct ieee80211_fast_tx *fast_tx;
+ 
+@@ -4226,6 +4251,8 @@ static void ieee80211_8023_xmit(struct i
+ 
+ 	memset(info, 0, sizeof(*info));
+ 
++	ieee80211_aggr_check(sdata, sta, skb);
++
+ 	tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK;
+ 	tid_tx = rcu_dereference(sta->ampdu_mlme.tid_tx[tid]);
+ 	if (tid_tx) {
diff --git a/package/kernel/mac80211/patches/subsys/375-mac80211-call-ieee80211_tx_h_rate_ctrl-when-dequeue.patch b/package/kernel/mac80211/patches/subsys/375-mac80211-call-ieee80211_tx_h_rate_ctrl-when-dequeue.patch
new file mode 100644
index 0000000000..49b1212213
--- /dev/null
+++ b/package/kernel/mac80211/patches/subsys/375-mac80211-call-ieee80211_tx_h_rate_ctrl-when-dequeue.patch
@@ -0,0 +1,114 @@
+From: Ryder Lee <ryder.lee@mediatek.com>
+Date: Fri, 28 May 2021 14:05:41 +0800
+Subject: [PATCH] mac80211: call ieee80211_tx_h_rate_ctrl() when dequeue
+
+Make ieee80211_tx_h_rate_ctrl() get called on dequeue to improve
+performance since it reduces the turnaround time for rate control.
+
+Signed-off-by: Ryder Lee <ryder.lee@mediatek.com>
+---
+
+--- a/net/mac80211/tx.c
++++ b/net/mac80211/tx.c
+@@ -1780,8 +1780,6 @@ static int invoke_tx_handlers_early(stru
+ 	CALL_TXH(ieee80211_tx_h_ps_buf);
+ 	CALL_TXH(ieee80211_tx_h_check_control_port_protocol);
+ 	CALL_TXH(ieee80211_tx_h_select_key);
+-	if (!ieee80211_hw_check(&tx->local->hw, HAS_RATE_CONTROL))
+-		CALL_TXH(ieee80211_tx_h_rate_ctrl);
+ 
+  txh_done:
+ 	if (unlikely(res == TX_DROP)) {
+@@ -1814,6 +1812,9 @@ static int invoke_tx_handlers_late(struc
+ 		goto txh_done;
+ 	}
+ 
++	if (!ieee80211_hw_check(&tx->local->hw, HAS_RATE_CONTROL))
++		CALL_TXH(ieee80211_tx_h_rate_ctrl);
++
+ 	CALL_TXH(ieee80211_tx_h_michael_mic_add);
+ 	CALL_TXH(ieee80211_tx_h_sequence);
+ 	CALL_TXH(ieee80211_tx_h_fragment);
+@@ -3384,15 +3385,21 @@ out:
+  * Can be called while the sta lock is held. Anything that can cause packets to
+  * be generated will cause deadlock!
+  */
+-static void ieee80211_xmit_fast_finish(struct ieee80211_sub_if_data *sdata,
+-				       struct sta_info *sta, u8 pn_offs,
+-				       struct ieee80211_key *key,
+-				       struct sk_buff *skb)
++static ieee80211_tx_result
++ieee80211_xmit_fast_finish(struct ieee80211_sub_if_data *sdata,
++			   struct sta_info *sta, u8 pn_offs,
++			   struct ieee80211_key *key,
++			   struct ieee80211_tx_data *tx)
+ {
++	struct sk_buff *skb = tx->skb;
+ 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+ 	struct ieee80211_hdr *hdr = (void *)skb->data;
+ 	u8 tid = IEEE80211_NUM_TIDS;
+ 
++	if (!ieee80211_hw_check(&tx->local->hw, HAS_RATE_CONTROL) &&
++	    ieee80211_tx_h_rate_ctrl(tx) != TX_CONTINUE)
++		return TX_DROP;
++
+ 	if (key)
+ 		info->control.hw_key = &key->conf;
+ 
+@@ -3441,6 +3448,8 @@ static void ieee80211_xmit_fast_finish(s
+ 			break;
+ 		}
+ 	}
++
++	return TX_CONTINUE;
+ }
+ 
+ static bool ieee80211_xmit_fast(struct ieee80211_sub_if_data *sdata,
+@@ -3544,24 +3553,17 @@ static bool ieee80211_xmit_fast(struct i
+ 	tx.sta = sta;
+ 	tx.key = fast_tx->key;
+ 
+-	if (!ieee80211_hw_check(&local->hw, HAS_RATE_CONTROL)) {
+-		tx.skb = skb;
+-		r = ieee80211_tx_h_rate_ctrl(&tx);
+-		skb = tx.skb;
+-		tx.skb = NULL;
+-
+-		if (r != TX_CONTINUE) {
+-			if (r != TX_QUEUED)
+-				kfree_skb(skb);
+-			return true;
+-		}
+-	}
+-
+ 	if (ieee80211_queue_skb(local, sdata, sta, skb))
+ 		return true;
+ 
+-	ieee80211_xmit_fast_finish(sdata, sta, fast_tx->pn_offs,
+-				   fast_tx->key, skb);
++	tx.skb = skb;
++	r = ieee80211_xmit_fast_finish(sdata, sta, fast_tx->pn_offs,
++				       fast_tx->key, &tx);
++	tx.skb = NULL;
++	if (r == TX_DROP) {
++		kfree_skb(skb);
++		return true;
++	}
+ 
+ 	if (sdata->vif.type == NL80211_IFTYPE_AP_VLAN)
+ 		sdata = container_of(sdata->bss,
+@@ -3672,8 +3674,12 @@ begin:
+ 		    (tx.key->conf.flags & IEEE80211_KEY_FLAG_GENERATE_IV))
+ 			pn_offs = ieee80211_hdrlen(hdr->frame_control);
+ 
+-		ieee80211_xmit_fast_finish(sta->sdata, sta, pn_offs,
+-					   tx.key, skb);
++		r = ieee80211_xmit_fast_finish(sta->sdata, sta, pn_offs,
++					       tx.key, &tx);
++		if (r != TX_CONTINUE) {
++			ieee80211_free_txskb(&local->hw, skb);
++			goto begin;
++		}
+ 	} else {
+ 		if (invoke_tx_handlers_late(&tx))
+ 			goto begin;
diff --git a/package/kernel/mac80211/patches/subsys/377-mac80211-minstrel_ht-fix-sample-time-check.patch b/package/kernel/mac80211/patches/subsys/377-mac80211-minstrel_ht-fix-sample-time-check.patch
index 787254b4df..d4b327c69c 100644
--- a/package/kernel/mac80211/patches/subsys/377-mac80211-minstrel_ht-fix-sample-time-check.patch
+++ b/package/kernel/mac80211/patches/subsys/377-mac80211-minstrel_ht-fix-sample-time-check.patch
@@ -12,7 +12,7 @@ Signed-off-by: Felix Fietkau <nbd@nbd.name>
 
 --- a/net/mac80211/rc80211_minstrel_ht.c
 +++ b/net/mac80211/rc80211_minstrel_ht.c
-@@ -1498,7 +1498,7 @@ minstrel_ht_get_rate(void *priv, struct
+@@ -1466,7 +1466,7 @@ minstrel_ht_get_rate(void *priv, struct
  	    (info->control.flags & IEEE80211_TX_CTRL_PORT_CTRL_PROTO))
  		return;
  
diff --git a/package/kernel/mac80211/patches/subsys/379-mac80211-fix-starting-aggregation-sessions-on-mesh-i.patch b/package/kernel/mac80211/patches/subsys/379-mac80211-fix-starting-aggregation-sessions-on-mesh-i.patch
new file mode 100644
index 0000000000..56eafaf847
--- /dev/null
+++ b/package/kernel/mac80211/patches/subsys/379-mac80211-fix-starting-aggregation-sessions-on-mesh-i.patch
@@ -0,0 +1,112 @@
+From: Felix Fietkau <nbd@nbd.name>
+Date: Tue, 29 Jun 2021 13:25:09 +0200
+Subject: [PATCH] mac80211: fix starting aggregation sessions on mesh
+ interfaces
+
+The logic for starting aggregation sessions was recently moved from minstrel_ht
+to mac80211, into the subif tx handler just after the sta lookup.
+Unfortunately this didn't work for mesh interfaces, since the sta lookup is
+deferred until a much later point in time on those.
+Fix this by also calling the aggregation check right after the deferred sta
+lookup.
+
+Fixes: 08a46c642001 ("mac80211: move A-MPDU session check from minstrel_ht to mac80211")
+Signed-off-by: Felix Fietkau <nbd@nbd.name>
+---
+
+--- a/net/mac80211/tx.c
++++ b/net/mac80211/tx.c
+@@ -1159,6 +1159,29 @@ static bool ieee80211_tx_prep_agg(struct
+ 	return queued;
+ }
+ 
++static void
++ieee80211_aggr_check(struct ieee80211_sub_if_data *sdata,
++		     struct sta_info *sta,
++		     struct sk_buff *skb)
++{
++	struct rate_control_ref *ref = sdata->local->rate_ctrl;
++	u16 tid;
++
++	if (!ref || !(ref->ops->capa & RATE_CTRL_CAPA_AMPDU_TRIGGER))
++		return;
++
++	if (!sta || !sta->sta.ht_cap.ht_supported ||
++	    !sta->sta.wme || skb_get_queue_mapping(skb) == IEEE80211_AC_VO ||
++	    skb->protocol == sdata->control_port_protocol)
++		return;
++
++	tid = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
++	if (likely(sta->ampdu_mlme.tid_tx[tid]))
++		return;
++
++	ieee80211_start_tx_ba_session(&sta->sta, tid, 0);
++}
++
+ /*
+  * initialises @tx
+  * pass %NULL for the station if unknown, a valid pointer if known
+@@ -1172,6 +1195,7 @@ ieee80211_tx_prepare(struct ieee80211_su
+ 	struct ieee80211_local *local = sdata->local;
+ 	struct ieee80211_hdr *hdr;
+ 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
++	bool aggr_check = false;
+ 	int tid;
+ 
+ 	memset(tx, 0, sizeof(*tx));
+@@ -1202,8 +1226,10 @@ ieee80211_tx_prepare(struct ieee80211_su
+ 			   tx->sdata->control_port_protocol == tx->skb->protocol) {
+ 			tx->sta = sta_info_get_bss(sdata, hdr->addr1);
+ 		}
+-		if (!tx->sta && !is_multicast_ether_addr(hdr->addr1))
++		if (!tx->sta && !is_multicast_ether_addr(hdr->addr1)) {
+ 			tx->sta = sta_info_get(sdata, hdr->addr1);
++			aggr_check = true;
++		}
+ 	}
+ 
+ 	if (tx->sta && ieee80211_is_data_qos(hdr->frame_control) &&
+@@ -1213,8 +1239,12 @@ ieee80211_tx_prepare(struct ieee80211_su
+ 		struct tid_ampdu_tx *tid_tx;
+ 
+ 		tid = ieee80211_get_tid(hdr);
+-
+ 		tid_tx = rcu_dereference(tx->sta->ampdu_mlme.tid_tx[tid]);
++		if (!tid_tx && aggr_check) {
++			ieee80211_aggr_check(sdata, tx->sta, skb);
++			tid_tx = rcu_dereference(tx->sta->ampdu_mlme.tid_tx[tid]);
++		}
++
+ 		if (tid_tx) {
+ 			bool queued;
+ 
+@@ -3949,29 +3979,6 @@ void ieee80211_txq_schedule_start(struct
+ }
+ EXPORT_SYMBOL(ieee80211_txq_schedule_start);
+ 
+-static void
+-ieee80211_aggr_check(struct ieee80211_sub_if_data *sdata,
+-		     struct sta_info *sta,
+-		     struct sk_buff *skb)
+-{
+-	struct rate_control_ref *ref = sdata->local->rate_ctrl;
+-	u16 tid;
+-
+-	if (!ref || !(ref->ops->capa & RATE_CTRL_CAPA_AMPDU_TRIGGER))
+-		return;
+-
+-	if (!sta || !sta->sta.ht_cap.ht_supported ||
+-	    !sta->sta.wme || skb_get_queue_mapping(skb) == IEEE80211_AC_VO ||
+-	    skb->protocol == sdata->control_port_protocol)
+-		return;
+-
+-	tid = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
+-	if (likely(sta->ampdu_mlme.tid_tx[tid]))
+-		return;
+-
+-	ieee80211_start_tx_ba_session(&sta->sta, tid, 0);
+-}
+-
+ void __ieee80211_subif_start_xmit(struct sk_buff *skb,
+ 				  struct net_device *dev,
+ 				  u32 info_flags,