From: Janusz Dziedzic Date: Sun, 10 Mar 2019 17:22:08 +0100 Subject: [PATCH] mac80211: add TX_NEEDS_ALIGNED4_SKBS hw flag The driver should set this flag if the hardware requires tx skb data (starting with the LLC header) to be aligned to 4 bytes. Padding is added after ieee80211_hdr, before IV/LLC. Before this patch, we have to do memmove(hdrlen) twice in the driver: Once before we pass this to HW and once again in tx completion (to fix up the skb for monitor mode). With this patch we can skip this memmove() and thus reduce CPU cycles in the data path. Signed-off-by: Janusz Dziedzic Signed-off-by: Felix Fietkau --- --- a/include/net/mac80211.h +++ b/include/net/mac80211.h @@ -2140,6 +2140,9 @@ struct ieee80211_txq { * @IEEE80211_HW_TX_STATUS_NO_AMPDU_LEN: Driver does not report accurate A-MPDU * length in tx status information * + * @IEEE80211_HW_TX_NEEDS_ALIGNED4_SKBS: Driver need aligned skbs to four-byte. + * Padding will be added after ieee80211_hdr, before IV/LLC. + * * @NUM_IEEE80211_HW_FLAGS: number of hardware flags, used for sizing arrays */ enum ieee80211_hw_flags { @@ -2186,6 +2189,7 @@ enum ieee80211_hw_flags { IEEE80211_HW_DEAUTH_NEED_MGD_TX_PREP, IEEE80211_HW_DOESNT_SUPPORT_QOS_NDP, IEEE80211_HW_TX_STATUS_NO_AMPDU_LEN, + IEEE80211_HW_TX_NEEDS_ALIGNED4_SKBS, /* keep last, obviously */ NUM_IEEE80211_HW_FLAGS @@ -2472,6 +2476,40 @@ ieee80211_get_alt_retry_rate(const struc void ieee80211_free_txskb(struct ieee80211_hw *hw, struct sk_buff *skb); /** + * ieee80211_hdr_padsize - get size of padding between 802.11 header and LLC + * @hw: the hardware + * @hdrlen: 802.11 header length + */ +static inline unsigned int +ieee80211_hdr_padsize(struct ieee80211_hw *hw, unsigned int hdrlen) +{ + /* + * While hdrlen is already aligned to two-byte boundaries, + * simple check with & 2 will return correct padsize. + */ + if (ieee80211_hw_check(hw, TX_NEEDS_ALIGNED4_SKBS)) + return hdrlen & 2; + return 0; +} + +/** + * ieee80211_padded_hdrlen - get padded 802.11 header size + * @hw: the hardware + * @fc: frame control field in little-endian format + */ +static inline unsigned int +ieee80211_padded_hdrlen(struct ieee80211_hw *hw, __le16 fc) +{ + unsigned int hdrlen; + + hdrlen = ieee80211_hdrlen(fc); + hdrlen += ieee80211_hdr_padsize(hw, hdrlen); + + return hdrlen; +} + + +/** * DOC: Hardware crypto acceleration * * mac80211 is capable of taking advantage of many hardware --- a/net/mac80211/iface.c +++ b/net/mac80211/iface.c @@ -1871,6 +1871,10 @@ int ieee80211_if_add(struct ieee80211_lo + 8 /* rfc1042/bridge tunnel */ - ETH_HLEN /* ethernet hard_header_len */ + IEEE80211_ENCRYPT_HEADROOM; + + if (ieee80211_hw_check(&local->hw, TX_NEEDS_ALIGNED4_SKBS)) + ndev->needed_headroom += 2; /* padding */ + ndev->needed_tailroom = IEEE80211_ENCRYPT_TAILROOM; ret = dev_alloc_name(ndev, ndev->name); --- a/net/mac80211/mesh_pathtbl.c +++ b/net/mac80211/mesh_pathtbl.c @@ -105,13 +105,15 @@ void mesh_path_assign_nexthop(struct mes static void prepare_for_gate(struct sk_buff *skb, char *dst_addr, struct mesh_path *gate_mpath) { + struct ieee80211_sub_if_data *sdata = gate_mpath->sdata; + struct ieee80211_hw *hw = &sdata->local->hw; struct ieee80211_hdr *hdr; struct ieee80211s_hdr *mshdr; int mesh_hdrlen, hdrlen; char *next_hop; hdr = (struct ieee80211_hdr *) skb->data; - hdrlen = ieee80211_hdrlen(hdr->frame_control); + hdrlen = ieee80211_padded_hdrlen(hw, hdr->frame_control); mshdr = (struct ieee80211s_hdr *) (skb->data + hdrlen); if (!(mshdr->flags & MESH_FLAGS_AE)) { --- a/net/mac80211/rx.c +++ b/net/mac80211/rx.c @@ -2597,7 +2597,7 @@ ieee80211_rx_h_mesh_fwding(struct ieee80 struct ieee80211_local *local = rx->local; struct ieee80211_sub_if_data *sdata = rx->sdata; struct ieee80211_if_mesh *ifmsh = &sdata->u.mesh; - u16 ac, q, hdrlen; + u16 ac, q, hdrlen, padsize; int tailroom = 0; hdr = (struct ieee80211_hdr *) skb->data; @@ -2690,7 +2690,9 @@ ieee80211_rx_h_mesh_fwding(struct ieee80 if (sdata->crypto_tx_tailroom_needed_cnt) tailroom = IEEE80211_ENCRYPT_TAILROOM; - fwd_skb = skb_copy_expand(skb, local->tx_headroom + + padsize = ieee80211_hdr_padsize(&local->hw, hdrlen); + + fwd_skb = skb_copy_expand(skb, local->tx_headroom + padsize + sdata->encrypt_headroom, tailroom, GFP_ATOMIC); if (!fwd_skb) @@ -2722,6 +2724,12 @@ ieee80211_rx_h_mesh_fwding(struct ieee80 return RX_DROP_MONITOR; } + if (padsize) { + skb_push(fwd_skb, padsize); + memmove(fwd_skb->data, skb->data + padsize, hdrlen); + memset(fwd_skb->data + hdrlen, 0, padsize); + } + IEEE80211_IFSTA_MESH_CTR_INC(ifmsh, fwded_frames); ieee80211_add_pending_skb(local, fwd_skb); out: --- a/net/mac80211/sta_info.h +++ b/net/mac80211/sta_info.h @@ -311,7 +311,7 @@ struct ieee80211_fast_tx { u8 hdr_len; u8 sa_offs, da_offs, pn_offs; u8 band; - u8 hdr[30 + 2 + IEEE80211_FAST_XMIT_MAX_IV + + u8 hdr[30 + 2 + 2 + IEEE80211_FAST_XMIT_MAX_IV + sizeof(rfc1042_header)] __aligned(2); struct rcu_head rcu_head; --- a/net/mac80211/status.c +++ b/net/mac80211/status.c @@ -515,6 +515,7 @@ static void ieee80211_report_used_skb(st { struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); struct ieee80211_hdr *hdr = (void *)skb->data; + struct ieee80211_hw *hw = &local->hw; bool acked = info->flags & IEEE80211_TX_STAT_ACK; if (dropped) @@ -531,7 +532,7 @@ static void ieee80211_report_used_skb(st skb->dev = NULL; } else { unsigned int hdr_size = - ieee80211_hdrlen(hdr->frame_control); + ieee80211_padded_hdrlen(hw, hdr->frame_control); /* Check to see if packet is a TDLS teardown packet */ if (ieee80211_is_data(hdr->frame_control) && @@ -655,9 +656,22 @@ void ieee80211_tx_monitor(struct ieee802 struct sk_buff *skb2; struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); struct ieee80211_sub_if_data *sdata; + struct ieee80211_hdr *hdr = (void *)skb->data; struct net_device *prev_dev = NULL; + unsigned int hdrlen, padsize; int rtap_len; + /* Remove padding if was added */ + if (ieee80211_hw_check(&local->hw, TX_NEEDS_ALIGNED4_SKBS)) { + hdrlen = ieee80211_hdrlen(hdr->frame_control); + padsize = ieee80211_hdr_padsize(&local->hw, hdrlen); + + if (padsize && skb->len > hdrlen + padsize) { + memmove(skb->data + padsize, skb->data, hdrlen); + skb_pull(skb, padsize); + } + } + /* send frame to monitor interfaces now */ rtap_len = ieee80211_tx_radiotap_len(info); if (WARN_ON_ONCE(skb_headroom(skb) < rtap_len)) { --- a/net/mac80211/tkip.c +++ b/net/mac80211/tkip.c @@ -201,10 +201,12 @@ void ieee80211_get_tkip_p2k(struct ieee8 { struct ieee80211_key *key = (struct ieee80211_key *) container_of(keyconf, struct ieee80211_key, conf); + struct ieee80211_hw *hw = &key->local->hw; const u8 *tk = &key->conf.key[NL80211_TKIP_DATA_OFFSET_ENCR_KEY]; struct tkip_ctx *ctx = &key->u.tkip.tx; struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data; - const u8 *data = (u8 *)hdr + ieee80211_hdrlen(hdr->frame_control); + const u8 *data = (u8 *)hdr + ieee80211_padded_hdrlen(hw, + hdr->frame_control); u32 iv32 = get_unaligned_le32(&data[4]); u16 iv16 = data[2] | (data[0] << 8); --- a/net/mac80211/tx.c +++ b/net/mac80211/tx.c @@ -1175,8 +1175,7 @@ ieee80211_tx_prepare(struct ieee80211_su info->flags &= ~IEEE80211_TX_INTFL_NEED_TXPROCESSING; hdr = (struct ieee80211_hdr *) skb->data; - - tx->hdrlen = ieee80211_hdrlen(hdr->frame_control); + tx->hdrlen = ieee80211_padded_hdrlen(&local->hw, hdr->frame_control); if (likely(sta)) { if (!IS_ERR(sta)) @@ -2222,7 +2221,7 @@ netdev_tx_t ieee80211_monitor_start_xmit goto fail; hdr = (struct ieee80211_hdr *)(skb->data + len_rthdr); - hdrlen = ieee80211_hdrlen(hdr->frame_control); + hdrlen = ieee80211_padded_hdrlen(&local->hw, hdr->frame_control); if (skb->len < len_rthdr + hdrlen) goto fail; @@ -2441,7 +2440,7 @@ static struct sk_buff *ieee80211_build_h struct ieee80211_chanctx_conf *chanctx_conf; struct ieee80211_sub_if_data *ap_sdata; enum nl80211_band band; - int ret; + int padsize, ret; if (IS_ERR(sta)) sta = NULL; @@ -2740,7 +2739,9 @@ static struct sk_buff *ieee80211_build_h } skb_pull(skb, skip_header_bytes); + padsize = ieee80211_hdr_padsize(&local->hw, hdrlen); head_need = hdrlen + encaps_len + meshhdrlen - skb_headroom(skb); + head_need += padsize; /* * So we need to modify the skb header and hence need a copy of @@ -2773,6 +2774,9 @@ static struct sk_buff *ieee80211_build_h memcpy(skb_push(skb, meshhdrlen), &mesh_hdr, meshhdrlen); #endif + if (padsize) + memset(skb_push(skb, padsize), 0, padsize); + if (ieee80211_is_data_qos(fc)) { __le16 *qos_control; @@ -2949,6 +2953,8 @@ void ieee80211_check_fast_xmit(struct st fc |= cpu_to_le16(IEEE80211_STYPE_QOS_DATA); } + build.hdr_len += ieee80211_hdr_padsize(&local->hw, build.hdr_len); + /* We store the key here so there's no point in using rcu_dereference() * but that's fine because the code that changes the pointers will call * this function after doing so. For a single CPU that would be enough, @@ -3525,7 +3531,7 @@ begin: tx.local = local; tx.skb = skb; tx.sdata = vif_to_sdata(info->control.vif); - tx.hdrlen = ieee80211_hdrlen(hdr->frame_control); + tx.hdrlen = ieee80211_padded_hdrlen(hw, hdr->frame_control); if (txq->sta) { tx.sta = container_of(txq->sta, struct sta_info, sta); @@ -4029,7 +4035,7 @@ ieee80211_build_data_template(struct iee hdr = (void *)skb->data; tx.sta = sta_info_get(sdata, hdr->addr1); tx.skb = skb; - tx.hdrlen = ieee80211_hdrlen(hdr->frame_control); + tx.hdrlen = ieee80211_padded_hdrlen(&tx.local->hw, hdr->frame_control); if (ieee80211_tx_h_select_key(&tx) != TX_CONTINUE) { rcu_read_unlock(); --- a/net/mac80211/debugfs.c +++ b/net/mac80211/debugfs.c @@ -215,6 +215,7 @@ static const char *hw_flag_names[] = { FLAG(DEAUTH_NEED_MGD_TX_PREP), FLAG(DOESNT_SUPPORT_QOS_NDP), FLAG(TX_STATUS_NO_AMPDU_LEN), + FLAG(TX_NEEDS_ALIGNED4_SKBS), #undef FLAG };