From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from mail2.tohojo.dk ([77.235.48.147]:57846 "EHLO mail2.tohojo.dk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752731AbcIBNls (ORCPT ); Fri, 2 Sep 2016 09:41:48 -0400 From: =?UTF-8?q?Toke=20H=C3=B8iland-J=C3=B8rgensen?= To: make-wifi-fast@lists.bufferbloat.net, linux-wireless@vger.kernel.org Cc: =?UTF-8?q?Toke=20H=C3=B8iland-J=C3=B8rgensen?= Subject: [PATCH v6] mac80211: Move reorder-sensitive TX handlers to after TXQ dequeue. Date: Fri, 2 Sep 2016 15:41:04 +0200 Message-Id: <20160902134104.29309-1-toke@toke.dk> (sfid-20160902_154152_974316_885E437A) In-Reply-To: <20160901160312.31540-1-toke@toke.dk> References: <20160901160312.31540-1-toke@toke.dk> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Sender: linux-wireless-owner@vger.kernel.org List-ID: The TXQ intermediate queues can cause packet reordering when more than one flow is active to a single station. Since some of the wifi-specific packet handling (notably sequence number and encryption handling) is sensitive to re-ordering, things break if they are applied before the TXQ. This splits up the TX handlers and fast_xmit logic into two parts: An early part and a late part. The former is applied before TXQ enqueue, and the latter after dequeue. The non-TXQ path just applies both parts at once. Because fragments shouldn't be split up or reordered, the fragmentation handler is run after dequeue. Any fragments are then kept in the TXQ and on subsequent dequeues they take precedence over dequeueing from the FQ structure. This approach avoids having to scatter special cases for when TXQ is enabled, at the cost of making the fast_xmit and TX handler code slightly more complex. Signed-off-by: Toke H=C3=B8iland-J=C3=B8rgensen --- Changes since v5: - Move the fragmentation handler to *after* TXQ dequeue. Fragments are kept in the TXQ for subsequent dequeues. This change also means that the changes to make some of the handlers fragmentation aware are no longer necessary. - One of the TX stats updates in the fast path was done before the enqueue step; move that to xmit_fast_finish(). - Move the rate selection handler to after dequeue, so it's run closer to the time where the packet is actually transmitted. =20 include/net/mac80211.h | 2 + net/mac80211/ieee80211_i.h | 2 + net/mac80211/tx.c | 207 +++++++++++++++++++++++++++++++++++----= ------ 3 files changed, 168 insertions(+), 43 deletions(-) diff --git a/include/net/mac80211.h b/include/net/mac80211.h index cca510a..9a6a3e9 100644 --- a/include/net/mac80211.h +++ b/include/net/mac80211.h @@ -715,6 +715,7 @@ enum mac80211_tx_info_flags { * frame (PS-Poll or uAPSD). * @IEEE80211_TX_CTRL_RATE_INJECT: This frame is injected with rate info= rmation * @IEEE80211_TX_CTRL_AMSDU: This frame is an A-MSDU frame + * @IEEE80211_TX_CTRL_FAST_XMIT: This frame is going through the fast_xm= it path * * These flags are used in tx_info->control.flags. */ @@ -723,6 +724,7 @@ enum mac80211_tx_control_flags { IEEE80211_TX_CTRL_PS_RESPONSE =3D BIT(1), IEEE80211_TX_CTRL_RATE_INJECT =3D BIT(2), IEEE80211_TX_CTRL_AMSDU =3D BIT(3), + IEEE80211_TX_CTRL_FAST_XMIT =3D BIT(4), }; =20 /* diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h index f56d342..de9991d 100644 --- a/net/mac80211/ieee80211_i.h +++ b/net/mac80211/ieee80211_i.h @@ -813,11 +813,13 @@ enum txq_info_flags { * @def_flow: used as a fallback flow when a packet destined to @tin has= hes 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 */ struct txq_info { struct fq_tin tin; struct fq_flow def_flow; struct codel_vars def_cvars; + struct sk_buff_head frags; unsigned long flags; =20 /* keep last! */ diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c index 1d0746d..a3a4593 100644 --- a/net/mac80211/tx.c +++ b/net/mac80211/tx.c @@ -38,6 +38,12 @@ #include "wme.h" #include "rate.h" =20 +static int invoke_tx_handlers_late(struct ieee80211_tx_data *tx); +static bool ieee80211_xmit_fast_finish(struct ieee80211_sub_if_data *sda= ta, + struct sta_info *sta, + struct ieee80211_fast_tx *fast_tx, + struct sk_buff *skb, bool xmit); + /* misc utils */ =20 static inline void ieee80211_tx_stats(struct net_device *dev, u32 len) @@ -1403,6 +1409,7 @@ void ieee80211_txq_init(struct ieee80211_sub_if_dat= a *sdata, fq_tin_init(&txqi->tin); fq_flow_init(&txqi->def_flow); codel_vars_init(&txqi->def_cvars); + __skb_queue_head_init(&txqi->frags); =20 txqi->txq.vif =3D &sdata->vif; =20 @@ -1425,6 +1432,7 @@ void ieee80211_txq_purge(struct ieee80211_local *lo= cal, struct fq_tin *tin =3D &txqi->tin; =20 fq_tin_reset(fq, tin, fq_skb_free_func); + ieee80211_purge_tx_queue(&local->hw, &txqi->frags); } =20 int ieee80211_txq_setup_flows(struct ieee80211_local *local) @@ -1481,33 +1489,62 @@ struct sk_buff *ieee80211_tx_dequeue(struct ieee8= 0211_hw *hw, { struct ieee80211_local *local =3D hw_to_local(hw); struct txq_info *txqi =3D container_of(txq, struct txq_info, txq); - struct ieee80211_hdr *hdr; struct sk_buff *skb =3D NULL; struct fq *fq =3D &local->fq; struct fq_tin *tin =3D &txqi->tin; + struct ieee80211_tx_info *info; =20 spin_lock_bh(&fq->lock); =20 if (test_bit(IEEE80211_TXQ_STOP, &txqi->flags)) goto out; =20 + /* Make sure fragments stay together. */ + skb =3D __skb_dequeue(&txqi->frags); + if (skb) + goto out; + +begin: skb =3D fq_tin_dequeue(fq, tin, fq_tin_dequeue_func); if (!skb) goto out; =20 ieee80211_set_skb_vif(skb, txqi); =20 - hdr =3D (struct ieee80211_hdr *)skb->data; - if (txq->sta && ieee80211_is_data_qos(hdr->frame_control)) { + info =3D IEEE80211_SKB_CB(skb); + if (txq->sta && info->control.flags & IEEE80211_TX_CTRL_FAST_XMIT) { struct sta_info *sta =3D container_of(txq->sta, struct sta_info, sta); - struct ieee80211_tx_info *info =3D IEEE80211_SKB_CB(skb); + struct ieee80211_fast_tx *fast_tx; =20 - hdr->seq_ctrl =3D ieee80211_tx_next_seq(sta, txq->tid); - if (test_bit(IEEE80211_TXQ_AMPDU, &txqi->flags)) - info->flags |=3D IEEE80211_TX_CTL_AMPDU; - else - info->flags &=3D ~IEEE80211_TX_CTL_AMPDU; + fast_tx =3D rcu_dereference(sta->fast_tx); + if (WARN_ON(!fast_tx)) { + /* lost the fast_tx pointer while the packet was queued */ + ieee80211_free_txskb(hw, skb); + goto begin; + } + ieee80211_xmit_fast_finish(sta->sdata, sta, fast_tx, skb, false); + } else { + struct ieee80211_tx_data tx =3D { }; + + __skb_queue_head_init(&tx.skbs); + tx.local =3D local; + tx.skb =3D skb; + if (txq->sta) { + tx.sta =3D container_of(txq->sta, struct sta_info, sta); + tx.sdata =3D tx.sta->sdata; + } else { + tx.sdata =3D vif_to_sdata(info->control.vif); + } + + + if (invoke_tx_handlers_late(&tx)) + goto begin; + + skb =3D __skb_dequeue(&tx.skbs); + + if (!skb_queue_empty(&tx.skbs)) + skb_queue_splice_tail(&tx.skbs, &txqi->frags); } =20 out: @@ -1521,6 +1558,47 @@ out: } EXPORT_SYMBOL(ieee80211_tx_dequeue); =20 +static bool ieee80211_queue_skb(struct ieee80211_local *local, + struct ieee80211_sub_if_data *sdata, + struct sta_info *sta, + struct sk_buff *skb) +{ + struct ieee80211_tx_info *info =3D IEEE80211_SKB_CB(skb); + struct fq *fq =3D &local->fq; + struct ieee80211_vif *vif; + struct txq_info *txqi; + struct ieee80211_sta *pubsta; + + if (!local->ops->wake_tx_queue || + sdata->vif.type =3D=3D NL80211_IFTYPE_MONITOR) + return false; + + if (sta && sta->uploaded) + pubsta =3D &sta->sta; + else + pubsta =3D NULL; + + if (sdata->vif.type =3D=3D NL80211_IFTYPE_AP_VLAN) + sdata =3D container_of(sdata->bss, + struct ieee80211_sub_if_data, u.ap); + + vif =3D &sdata->vif; + txqi =3D ieee80211_get_txq(local, vif, pubsta, skb); + + if (!txqi) + return false; + + info->control.vif =3D vif; + + spin_lock_bh(&fq->lock); + ieee80211_txq_enqueue(local, txqi, skb); + spin_unlock_bh(&fq->lock); + + drv_wake_tx_queue(local, txqi); + + return true; +} + static bool ieee80211_tx_frags(struct ieee80211_local *local, struct ieee80211_vif *vif, struct ieee80211_sta *sta, @@ -1528,9 +1606,7 @@ static bool ieee80211_tx_frags(struct ieee80211_loc= al *local, bool txpending) { struct ieee80211_tx_control control =3D {}; - struct fq *fq =3D &local->fq; struct sk_buff *skb, *tmp; - struct txq_info *txqi; unsigned long flags; =20 skb_queue_walk_safe(skbs, skb, tmp) { @@ -1545,21 +1621,6 @@ static bool ieee80211_tx_frags(struct ieee80211_lo= cal *local, } #endif =20 - txqi =3D ieee80211_get_txq(local, vif, sta, skb); - if (txqi) { - info->control.vif =3D vif; - - __skb_unlink(skb, skbs); - - spin_lock_bh(&fq->lock); - ieee80211_txq_enqueue(local, txqi, skb); - spin_unlock_bh(&fq->lock); - - drv_wake_tx_queue(local, txqi); - - continue; - } - spin_lock_irqsave(&local->queue_stop_reason_lock, flags); if (local->queue_stop_reasons[q] || (!txpending && !skb_queue_empty(&local->pending[q]))) { @@ -1680,10 +1741,13 @@ static bool __ieee80211_tx(struct ieee80211_local= *local, /* * Invoke TX handlers, return 0 on success and non-zero if the * frame was dropped or queued. + * + * The handlers are split into an early and late part. The latter is eve= rything + * that can be sensitive to reordering, and will be deferred to after pa= ckets + * are dequeued from the intermediate queues (when they are enabled). */ -static int invoke_tx_handlers(struct ieee80211_tx_data *tx) +static int invoke_tx_handlers_early(struct ieee80211_tx_data *tx) { - struct ieee80211_tx_info *info =3D IEEE80211_SKB_CB(tx->skb); ieee80211_tx_result res =3D TX_DROP; =20 #define CALL_TXH(txh) \ @@ -1697,7 +1761,28 @@ static int invoke_tx_handlers(struct ieee80211_tx_= data *tx) CALL_TXH(ieee80211_tx_h_check_assoc); CALL_TXH(ieee80211_tx_h_ps_buf); CALL_TXH(ieee80211_tx_h_check_control_port_protocol); - CALL_TXH(ieee80211_tx_h_select_key); + + txh_done: + if (unlikely(res =3D=3D TX_DROP)) { + I802_DEBUG_INC(tx->local->tx_handlers_drop); + if (tx->skb) + ieee80211_free_txskb(&tx->local->hw, tx->skb); + else + ieee80211_purge_tx_queue(&tx->local->hw, &tx->skbs); + return -1; + } else if (unlikely(res =3D=3D TX_QUEUED)) { + I802_DEBUG_INC(tx->local->tx_handlers_queued); + return -1; + } + + return 0; +} + +static int invoke_tx_handlers_late(struct ieee80211_tx_data *tx) +{ + struct ieee80211_tx_info *info =3D IEEE80211_SKB_CB(tx->skb); + ieee80211_tx_result res =3D TX_DROP; + if (!ieee80211_hw_check(&tx->local->hw, HAS_RATE_CONTROL)) CALL_TXH(ieee80211_tx_h_rate_ctrl); =20 @@ -1707,6 +1792,7 @@ static int invoke_tx_handlers(struct ieee80211_tx_d= ata *tx) goto txh_done; } =20 + CALL_TXH(ieee80211_tx_h_select_key); CALL_TXH(ieee80211_tx_h_michael_mic_add); CALL_TXH(ieee80211_tx_h_sequence); CALL_TXH(ieee80211_tx_h_fragment); @@ -1733,6 +1819,15 @@ static int invoke_tx_handlers(struct ieee80211_tx_= data *tx) return 0; } =20 +static int invoke_tx_handlers(struct ieee80211_tx_data *tx) +{ + int r =3D invoke_tx_handlers_early(tx); + if (r) + return r; + + return invoke_tx_handlers_late(tx); +} + bool ieee80211_tx_prepare_skb(struct ieee80211_hw *hw, struct ieee80211_vif *vif, struct sk_buff *skb, int band, struct ieee80211_sta **sta) @@ -1807,7 +1902,13 @@ static bool ieee80211_tx(struct ieee80211_sub_if_d= ata *sdata, info->hw_queue =3D sdata->vif.hw_queue[skb_get_queue_mapping(skb)]; =20 - if (!invoke_tx_handlers(&tx)) + if (invoke_tx_handlers_early(&tx)) + return false; + + if (ieee80211_queue_skb(local, sdata, tx.sta, tx.skb)) + return true; + + if (!invoke_tx_handlers_late(&tx)) result =3D __ieee80211_tx(local, &tx.skbs, led_len, tx.sta, txpending); =20 @@ -3159,7 +3260,7 @@ out: } =20 static bool ieee80211_xmit_fast(struct ieee80211_sub_if_data *sdata, - struct net_device *dev, struct sta_info *sta, + struct sta_info *sta, struct ieee80211_fast_tx *fast_tx, struct sk_buff *skb) { @@ -3170,8 +3271,6 @@ static bool ieee80211_xmit_fast(struct ieee80211_su= b_if_data *sdata, struct ethhdr eth; struct ieee80211_tx_info *info =3D IEEE80211_SKB_CB(skb); struct ieee80211_hdr *hdr =3D (void *)fast_tx->hdr; - struct ieee80211_tx_data tx; - ieee80211_tx_result r; struct tid_ampdu_tx *tid_tx =3D NULL; u8 tid =3D IEEE80211_NUM_TIDS; =20 @@ -3210,8 +3309,6 @@ static bool ieee80211_xmit_fast(struct ieee80211_su= b_if_data *sdata, return true; } =20 - ieee80211_tx_stats(dev, skb->len + extra_head); - if ((hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) && ieee80211_amsdu_aggregate(sdata, sta, fast_tx, skb)) return true; @@ -3240,11 +3337,32 @@ static bool ieee80211_xmit_fast(struct ieee80211_= sub_if_data *sdata, info->flags =3D IEEE80211_TX_CTL_FIRST_FRAGMENT | IEEE80211_TX_CTL_DONTFRAG | (tid_tx ? IEEE80211_TX_CTL_AMPDU : 0); + info->control.flags =3D IEEE80211_TX_CTRL_FAST_XMIT; + + if (ieee80211_queue_skb(local, sdata, sta, skb)) + return true; + + return ieee80211_xmit_fast_finish(sdata, sta, fast_tx, skb, true); +} + +static bool ieee80211_xmit_fast_finish(struct ieee80211_sub_if_data *sda= ta, + struct sta_info *sta, + struct ieee80211_fast_tx *fast_tx, + struct sk_buff *skb, bool xmit) +{ + struct ieee80211_local *local =3D sdata->local; + struct ieee80211_tx_info *info =3D IEEE80211_SKB_CB(skb); + struct ieee80211_hdr *hdr =3D (void *)skb->data; + struct ieee80211_tx_data tx; + ieee80211_tx_result r; + u8 tid =3D IEEE80211_NUM_TIDS; + + ieee80211_tx_stats(skb->dev, skb->len); =20 if (hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) { + tid =3D skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK; *ieee80211_get_qos_ctl(hdr) =3D tid; - if (!sta->sta.txq[0]) - hdr->seq_ctrl =3D ieee80211_tx_next_seq(sta, tid); + hdr->seq_ctrl =3D ieee80211_tx_next_seq(sta, tid); } else { info->flags |=3D IEEE80211_TX_CTL_ASSIGN_SEQ; hdr->seq_ctrl =3D cpu_to_le16(sdata->sequence_number); @@ -3309,12 +3427,15 @@ static bool ieee80211_xmit_fast(struct ieee80211_= sub_if_data *sdata, } } =20 - if (sdata->vif.type =3D=3D NL80211_IFTYPE_AP_VLAN) - sdata =3D container_of(sdata->bss, - struct ieee80211_sub_if_data, u.ap); + if (xmit) { + if (sdata->vif.type =3D=3D NL80211_IFTYPE_AP_VLAN) + sdata =3D container_of(sdata->bss, + struct ieee80211_sub_if_data, u.ap); + + __skb_queue_tail(&tx.skbs, skb); + ieee80211_tx_frags(local, &sdata->vif, &sta->sta, &tx.skbs, false); + } =20 - __skb_queue_tail(&tx.skbs, skb); - ieee80211_tx_frags(local, &sdata->vif, &sta->sta, &tx.skbs, false); return true; } =20 @@ -3342,7 +3463,7 @@ void __ieee80211_subif_start_xmit(struct sk_buff *s= kb, fast_tx =3D rcu_dereference(sta->fast_tx); =20 if (fast_tx && - ieee80211_xmit_fast(sdata, dev, sta, fast_tx, skb)) + ieee80211_xmit_fast(sdata, sta, fast_tx, skb)) goto out; } =20 --=20 2.9.3