From: Felix Fietkau Date: Mon, 10 Oct 2022 10:11:10 +0000 (+0200) Subject: mac80211: add patch that gives the driver more control over netdev offloads X-Git-Url: http://git.lede-project.org./?a=commitdiff_plain;h=88803cb0e6e42ea00bd7c3dc8838e660239d3f16;p=openwrt%2Fstaging%2Fwigyori.git mac80211: add patch that gives the driver more control over netdev offloads This can be used to selectively disable checksum, SG or GSO offloads Signed-off-by: Felix Fietkau --- diff --git a/package/kernel/mac80211/patches/subsys/364-mac80211-add-support-for-restricting-netdev-features.patch b/package/kernel/mac80211/patches/subsys/364-mac80211-add-support-for-restricting-netdev-features.patch new file mode 100644 index 0000000000..d133ed91eb --- /dev/null +++ b/package/kernel/mac80211/patches/subsys/364-mac80211-add-support-for-restricting-netdev-features.patch @@ -0,0 +1,513 @@ +From: Felix Fietkau +Date: Sun, 9 Oct 2022 20:15:46 +0200 +Subject: [PATCH] mac80211: add support for restricting netdev features per vif + +This can be used to selectively disable feature flags for checksum offload, +scatter/gather or GSO by changing vif->netdev_features. +Removing features from vif->netdev_features does not affect the netdev +features themselves, but instead fixes up skbs in the tx path so that the +offloads are not needed in the driver. + +Aside from making it easier to deal with vif type based hardware limitations, +this also makes it possible to optimize performance on hardware without native +GSO support by declaring GSO support in hw->netdev_features and removing it +from vif->netdev_features. This allows mac80211 to handle GSO segmentation +after the sta lookup, but before itxq enqueue, thus reducing the number of +unnecessary sta lookups, as well as some other per-packet processing. + +Signed-off-by: Felix Fietkau +--- + +--- a/include/net/fq_impl.h ++++ b/include/net/fq_impl.h +@@ -200,6 +200,7 @@ static void fq_tin_enqueue(struct fq *fq + fq_skb_free_t free_func) + { + struct fq_flow *flow; ++ struct sk_buff *next; + bool oom; + + lockdep_assert_held(&fq->lock); +@@ -214,11 +215,15 @@ static void fq_tin_enqueue(struct fq *fq + } + + flow->tin = tin; +- flow->backlog += skb->len; +- tin->backlog_bytes += skb->len; +- tin->backlog_packets++; +- fq->memory_usage += skb->truesize; +- fq->backlog++; ++ skb_list_walk_safe(skb, skb, next) { ++ skb_mark_not_on_list(skb); ++ flow->backlog += skb->len; ++ tin->backlog_bytes += skb->len; ++ tin->backlog_packets++; ++ fq->memory_usage += skb->truesize; ++ fq->backlog++; ++ __skb_queue_tail(&flow->queue, skb); ++ } + + if (list_empty(&flow->flowchain)) { + flow->deficit = fq->quantum; +@@ -226,7 +231,6 @@ static void fq_tin_enqueue(struct fq *fq + &tin->new_flows); + } + +- __skb_queue_tail(&flow->queue, skb); + oom = (fq->memory_usage > fq->memory_limit); + while (fq->backlog > fq->limit || oom) { + flow = fq_find_fattest_flow(fq); +--- a/include/net/mac80211.h ++++ b/include/net/mac80211.h +@@ -1685,6 +1685,10 @@ enum ieee80211_offload_flags { + * write-protected by sdata_lock and local->mtx so holding either is fine + * for read access. + * @mu_mimo_owner: indicates interface owns MU-MIMO capability ++ * @netdev_features: tx netdev features supported by the hardware for this ++ * vif. mac80211 initializes this to hw->netdev_features, and the driver ++ * can mask out specific tx features. mac80211 will handle software fixup ++ * for masked offloads (GSO, CSUM) + * @driver_flags: flags/capabilities the driver has for this interface, + * these need to be set (or cleared) when the interface is added + * or, if supported by the driver, the interface type is changed +@@ -1736,6 +1740,7 @@ struct ieee80211_vif { + + struct ieee80211_chanctx_conf __rcu *chanctx_conf; + ++ netdev_features_t netdev_features; + u32 driver_flags; + u32 offload_flags; + +--- a/net/mac80211/iface.c ++++ b/net/mac80211/iface.c +@@ -2209,6 +2209,7 @@ int ieee80211_if_add(struct ieee80211_lo + ndev->features |= local->hw.netdev_features; + ndev->hw_features |= ndev->features & + MAC80211_SUPPORTED_FEATURES_TX; ++ sdata->vif.netdev_features = local->hw.netdev_features; + + netdev_set_default_ethtool_ops(ndev, &ieee80211_ethtool_ops); + +--- a/net/mac80211/tx.c ++++ b/net/mac80211/tx.c +@@ -1310,7 +1310,11 @@ static struct txq_info *ieee80211_get_tx + + static void ieee80211_set_skb_enqueue_time(struct sk_buff *skb) + { +- IEEE80211_SKB_CB(skb)->control.enqueue_time = codel_get_time(); ++ struct sk_buff *next; ++ codel_time_t now = codel_get_time(); ++ ++ skb_list_walk_safe(skb, skb, next) ++ IEEE80211_SKB_CB(skb)->control.enqueue_time = now; + } + + static u32 codel_skb_len_func(const struct sk_buff *skb) +@@ -3499,47 +3503,71 @@ ieee80211_xmit_fast_finish(struct ieee80 + return TX_CONTINUE; + } + +-static bool ieee80211_xmit_fast(struct ieee80211_sub_if_data *sdata, +- struct sta_info *sta, +- struct ieee80211_fast_tx *fast_tx, +- struct sk_buff *skb) ++static netdev_features_t ++ieee80211_sdata_netdev_features(struct ieee80211_sub_if_data *sdata) + { +- struct ieee80211_local *local = sdata->local; +- u16 ethertype = (skb->data[12] << 8) | skb->data[13]; +- int extra_head = fast_tx->hdr_len - (ETH_HLEN - 2); +- int hw_headroom = sdata->local->hw.extra_tx_headroom; +- struct ethhdr eth; +- struct ieee80211_tx_info *info; +- struct ieee80211_hdr *hdr = (void *)fast_tx->hdr; +- struct ieee80211_tx_data tx; +- ieee80211_tx_result r; +- struct tid_ampdu_tx *tid_tx = NULL; +- u8 tid = IEEE80211_NUM_TIDS; ++ if (sdata->vif.type != NL80211_IFTYPE_AP_VLAN) ++ return sdata->vif.netdev_features; + +- /* control port protocol needs a lot of special handling */ +- if (cpu_to_be16(ethertype) == sdata->control_port_protocol) +- return false; ++ if (!sdata->bss) ++ return 0; + +- /* only RFC 1042 SNAP */ +- if (ethertype < ETH_P_802_3_MIN) +- return false; ++ sdata = container_of(sdata->bss, struct ieee80211_sub_if_data, u.ap); ++ return sdata->vif.netdev_features; ++} + +- /* don't handle TX status request here either */ +- if (skb->sk && skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS) +- return false; ++static struct sk_buff * ++ieee80211_tx_skb_fixup(struct sk_buff *skb, netdev_features_t features) ++{ ++ if (skb_is_gso(skb)) { ++ struct sk_buff *segs; + +- if (hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) { +- tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK; +- tid_tx = rcu_dereference(sta->ampdu_mlme.tid_tx[tid]); +- if (tid_tx) { +- if (!test_bit(HT_AGG_STATE_OPERATIONAL, &tid_tx->state)) +- return false; +- if (tid_tx->timeout) +- tid_tx->last_tx = jiffies; +- } ++ segs = skb_gso_segment(skb, features); ++ if (!segs) ++ return skb; ++ if (IS_ERR(segs)) ++ goto free; ++ ++ consume_skb(skb); ++ return segs; ++ } ++ ++ if (skb_needs_linearize(skb, features) && __skb_linearize(skb)) ++ goto free; ++ ++ if (skb->ip_summed == CHECKSUM_PARTIAL) { ++ int ofs = skb_checksum_start_offset(skb); ++ ++ if (skb->encapsulation) ++ skb_set_inner_transport_header(skb, ofs); ++ else ++ skb_set_transport_header(skb, ofs); ++ ++ if (skb_csum_hwoffload_help(skb, features)) ++ goto free; + } + +- /* after this point (skb is modified) we cannot return false */ ++ skb_mark_not_on_list(skb); ++ return skb; ++ ++free: ++ kfree_skb(skb); ++ return NULL; ++} ++ ++static void __ieee80211_xmit_fast(struct ieee80211_sub_if_data *sdata, ++ struct sta_info *sta, ++ struct ieee80211_fast_tx *fast_tx, ++ struct sk_buff *skb, u8 tid, bool ampdu) ++{ ++ struct ieee80211_local *local = sdata->local; ++ struct ieee80211_hdr *hdr = (void *)fast_tx->hdr; ++ struct ieee80211_tx_info *info; ++ struct ieee80211_tx_data tx; ++ ieee80211_tx_result r; ++ int hw_headroom = sdata->local->hw.extra_tx_headroom; ++ int extra_head = fast_tx->hdr_len - (ETH_HLEN - 2); ++ struct ethhdr eth; + + if (skb_shared(skb)) { + struct sk_buff *tmp_skb = skb; +@@ -3548,12 +3576,12 @@ static bool ieee80211_xmit_fast(struct i + kfree_skb(tmp_skb); + + if (!skb) +- return true; ++ return; + } + + if ((hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) && + ieee80211_amsdu_aggregate(sdata, sta, fast_tx, skb)) +- return true; ++ return; + + /* will not be crypto-handled beyond what we do here, so use false + * as the may-encrypt argument for the resize to not account for +@@ -3562,10 +3590,8 @@ static bool ieee80211_xmit_fast(struct i + if (unlikely(ieee80211_skb_resize(sdata, skb, + max_t(int, extra_head + hw_headroom - + skb_headroom(skb), 0), +- ENCRYPT_NO))) { +- kfree_skb(skb); +- return true; +- } ++ ENCRYPT_NO))) ++ goto free; + + memcpy(ð, skb->data, ETH_HLEN - 2); + hdr = skb_push(skb, extra_head); +@@ -3579,7 +3605,7 @@ static bool ieee80211_xmit_fast(struct i + info->control.vif = &sdata->vif; + info->flags = IEEE80211_TX_CTL_FIRST_FRAGMENT | + IEEE80211_TX_CTL_DONTFRAG | +- (tid_tx ? IEEE80211_TX_CTL_AMPDU : 0); ++ (ampdu ? IEEE80211_TX_CTL_AMPDU : 0); + info->control.flags = IEEE80211_TX_CTRL_FAST_XMIT; + + #ifdef CPTCFG_MAC80211_DEBUGFS +@@ -3601,16 +3627,14 @@ static bool ieee80211_xmit_fast(struct i + tx.key = fast_tx->key; + + if (ieee80211_queue_skb(local, sdata, sta, skb)) +- return true; ++ return; + + 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 (r == TX_DROP) ++ goto free; + + if (sdata->vif.type == NL80211_IFTYPE_AP_VLAN) + sdata = container_of(sdata->bss, +@@ -3618,6 +3642,55 @@ static bool ieee80211_xmit_fast(struct i + + __skb_queue_tail(&tx.skbs, skb); + ieee80211_tx_frags(local, &sdata->vif, sta, &tx.skbs, false); ++ return; ++ ++free: ++ kfree_skb(skb); ++} ++ ++static bool ieee80211_xmit_fast(struct ieee80211_sub_if_data *sdata, ++ struct sta_info *sta, ++ struct ieee80211_fast_tx *fast_tx, ++ struct sk_buff *skb) ++{ ++ u16 ethertype = (skb->data[12] << 8) | skb->data[13]; ++ struct ieee80211_hdr *hdr = (void *)fast_tx->hdr; ++ struct tid_ampdu_tx *tid_tx = NULL; ++ struct sk_buff *next; ++ u8 tid = IEEE80211_NUM_TIDS; ++ ++ /* control port protocol needs a lot of special handling */ ++ if (cpu_to_be16(ethertype) == sdata->control_port_protocol) ++ return false; ++ ++ /* only RFC 1042 SNAP */ ++ if (ethertype < ETH_P_802_3_MIN) ++ return false; ++ ++ /* don't handle TX status request here either */ ++ if (skb->sk && skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS) ++ return false; ++ ++ if (hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) { ++ tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK; ++ tid_tx = rcu_dereference(sta->ampdu_mlme.tid_tx[tid]); ++ if (tid_tx) { ++ if (!test_bit(HT_AGG_STATE_OPERATIONAL, &tid_tx->state)) ++ return false; ++ if (tid_tx->timeout) ++ tid_tx->last_tx = jiffies; ++ } ++ } ++ ++ skb = ieee80211_tx_skb_fixup(skb, ieee80211_sdata_netdev_features(sdata)); ++ if (!skb) ++ return true; ++ ++ skb_list_walk_safe(skb, skb, next) { ++ skb_mark_not_on_list(skb); ++ __ieee80211_xmit_fast(sdata, sta, fast_tx, skb, tid, tid_tx); ++ } ++ + return true; + } + +@@ -4123,31 +4196,14 @@ void __ieee80211_subif_start_xmit(struct + goto out; + } + +- if (skb_is_gso(skb)) { +- struct sk_buff *segs; +- +- segs = skb_gso_segment(skb, 0); +- if (IS_ERR(segs)) { +- goto out_free; +- } else if (segs) { +- consume_skb(skb); +- skb = segs; +- } +- } else { +- /* we cannot process non-linear frames on this path */ +- if (skb_linearize(skb)) +- goto out_free; +- +- /* the frame could be fragmented, software-encrypted, and other +- * things so we cannot really handle checksum offload with it - +- * fix it up in software before we handle anything else. +- */ +- if (skb->ip_summed == CHECKSUM_PARTIAL) { +- skb_set_transport_header(skb, +- skb_checksum_start_offset(skb)); +- if (skb_checksum_help(skb)) +- goto out_free; +- } ++ /* the frame could be fragmented, software-encrypted, and other ++ * things so we cannot really handle checksum or GSO offload. ++ * fix it up in software before we handle anything else. ++ */ ++ skb = ieee80211_tx_skb_fixup(skb, 0); ++ if (!skb) { ++ len = 0; ++ goto out; + } + + skb_list_walk_safe(skb, skb, next) { +@@ -4310,9 +4366,11 @@ netdev_tx_t ieee80211_subif_start_xmit(s + return NETDEV_TX_OK; + } + +-static bool ieee80211_tx_8023(struct ieee80211_sub_if_data *sdata, +- struct sk_buff *skb, struct sta_info *sta, +- bool txpending) ++ ++ ++static bool __ieee80211_tx_8023(struct ieee80211_sub_if_data *sdata, ++ struct sk_buff *skb, struct sta_info *sta, ++ bool txpending) + { + struct ieee80211_local *local = sdata->local; + struct ieee80211_tx_control control = {}; +@@ -4321,14 +4379,6 @@ static bool ieee80211_tx_8023(struct iee + unsigned long flags; + int q = info->hw_queue; + +- if (sta) +- sk_pacing_shift_update(skb->sk, local->hw.tx_sk_pacing_shift); +- +- ieee80211_tpt_led_trig_tx(local, skb->len); +- +- if (ieee80211_queue_skb(local, sdata, sta, skb)) +- return true; +- + spin_lock_irqsave(&local->queue_stop_reason_lock, flags); + + if (local->queue_stop_reasons[q] || +@@ -4355,27 +4405,50 @@ static bool ieee80211_tx_8023(struct iee + return true; + } + ++static bool ieee80211_tx_8023(struct ieee80211_sub_if_data *sdata, ++ struct sk_buff *skb, struct sta_info *sta, ++ bool txpending) ++{ ++ struct ieee80211_local *local = sdata->local; ++ struct sk_buff *next; ++ bool ret = true; ++ ++ if (ieee80211_queue_skb(local, sdata, sta, skb)) ++ return true; ++ ++ skb_list_walk_safe(skb, skb, next) { ++ skb_mark_not_on_list(skb); ++ if (!__ieee80211_tx_8023(sdata, skb, sta, txpending)) ++ ret = false; ++ } ++ ++ return ret; ++} ++ + static void ieee80211_8023_xmit(struct ieee80211_sub_if_data *sdata, + struct net_device *dev, struct sta_info *sta, + struct ieee80211_key *key, struct sk_buff *skb) + { +- struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); ++ struct ieee80211_tx_info *info; + struct ieee80211_local *local = sdata->local; + struct tid_ampdu_tx *tid_tx; ++ struct sk_buff *seg, *next; ++ unsigned int skbs = 0, len = 0; ++ u16 queue; + u8 tid; + + if (local->ops->wake_tx_queue) { +- u16 queue = __ieee80211_select_queue(sdata, sta, skb); ++ queue = __ieee80211_select_queue(sdata, sta, skb); + skb_set_queue_mapping(skb, queue); + skb_get_hash(skb); ++ } else { ++ queue = skb_get_queue_mapping(skb); + } + + if (unlikely(test_bit(SCAN_SW_SCANNING, &local->scanning)) && + test_bit(SDATA_STATE_OFFCHANNEL, &sdata->state)) + goto out_free; + +- memset(info, 0, sizeof(*info)); +- + ieee80211_aggr_check(sdata, sta, skb); + + tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK; +@@ -4387,22 +4460,20 @@ static void ieee80211_8023_xmit(struct i + return; + } + +- info->flags |= IEEE80211_TX_CTL_AMPDU; + if (tid_tx->timeout) + tid_tx->last_tx = jiffies; + } + +- if (unlikely(skb->sk && +- skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS)) +- info->ack_frame_id = ieee80211_store_ack_skb(local, skb, +- &info->flags, NULL); ++ skb = ieee80211_tx_skb_fixup(skb, ieee80211_sdata_netdev_features(sdata)); ++ if (!skb) ++ return; + +- info->hw_queue = sdata->vif.hw_queue[skb_get_queue_mapping(skb)]; ++ info = IEEE80211_SKB_CB(skb); ++ memset(info, 0, sizeof(*info)); ++ if (tid_tx) ++ info->flags |= IEEE80211_TX_CTL_AMPDU; + +- dev_sw_netstats_tx_add(dev, 1, skb->len); +- +- sta->tx_stats.bytes[skb_get_queue_mapping(skb)] += skb->len; +- sta->tx_stats.packets[skb_get_queue_mapping(skb)]++; ++ info->hw_queue = sdata->vif.hw_queue[queue]; + + if (sdata->vif.type == NL80211_IFTYPE_AP_VLAN) + sdata = container_of(sdata->bss, +@@ -4414,6 +4485,24 @@ static void ieee80211_8023_xmit(struct i + if (key) + info->control.hw_key = &key->conf; + ++ skb_list_walk_safe(skb, seg, next) { ++ skbs++; ++ len += seg->len; ++ if (seg != skb) ++ memcpy(IEEE80211_SKB_CB(seg), info, sizeof(*info)); ++ } ++ ++ if (unlikely(skb->sk && ++ skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS)) ++ info->ack_frame_id = ieee80211_store_ack_skb(local, skb, ++ &info->flags, NULL); ++ ++ dev_sw_netstats_tx_add(dev, skbs, len); ++ sta->tx_stats.packets[queue] += skbs; ++ sta->tx_stats.bytes[queue] += len; ++ ++ ieee80211_tpt_led_trig_tx(local, len); ++ + ieee80211_tx_8023(sdata, skb, sta, false); + + return; +@@ -4455,6 +4544,7 @@ netdev_tx_t ieee80211_subif_start_xmit_8 + key->conf.cipher == WLAN_CIPHER_SUITE_TKIP)) + goto skip_offload; + ++ sk_pacing_shift_update(skb->sk, sdata->local->hw.tx_sk_pacing_shift); + ieee80211_8023_xmit(sdata, dev, sta, key, skb); + goto out; +