Return-path: Received: from mail-pg0-f44.google.com ([74.125.83.44]:34818 "EHLO mail-pg0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751228AbdESQmg (ORCPT ); Fri, 19 May 2017 12:42:36 -0400 Received: by mail-pg0-f44.google.com with SMTP id q125so40564508pgq.2 for ; Fri, 19 May 2017 09:42:36 -0700 (PDT) Date: Fri, 19 May 2017 09:42:33 -0700 From: Brian Norris To: Xinming Hu Cc: Linux Wireless , Kalle Valo , Dmitry Torokhov , rajatja@google.com, Zhiyuan Yang , Cathy Luo , Xinming Hu , Ganapathi Bhat Subject: Re: [PATCH v3 5/6] mwifiex: do not aggregate tcp ack in usb tx aggregation queue Message-ID: <20170519164232.GB125761@google.com> (sfid-20170519_184240_506439_70BF691F) References: <1495184805-6631-1-git-send-email-huxinming820@gmail.com> <1495184805-6631-5-git-send-email-huxinming820@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1495184805-6631-5-git-send-email-huxinming820@gmail.com> Sender: linux-wireless-owner@vger.kernel.org List-ID: On Fri, May 19, 2017 at 09:06:44AM +0000, Xinming Hu wrote: > From: Xinming Hu > > Tcp ack should be send as soon to avoid throuput drop during receive tcp > traffic. > > Signed-off-by: Xinming Hu > Signed-off-by: Cathy Luo > Signed-off-by: Ganapathi Bhat > --- > v2: same as v1 > v3: avoid to use magic tcp flag (Kalle) > --- > drivers/net/wireless/marvell/mwifiex/11n_aggr.c | 4 ++++ > drivers/net/wireless/marvell/mwifiex/decl.h | 1 + > drivers/net/wireless/marvell/mwifiex/main.c | 26 +++++++++++++++++++++++++ > drivers/net/wireless/marvell/mwifiex/main.h | 1 + > drivers/net/wireless/marvell/mwifiex/usb.c | 6 ++++-- > 5 files changed, 36 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/wireless/marvell/mwifiex/11n_aggr.c b/drivers/net/wireless/marvell/mwifiex/11n_aggr.c > index e8ffb26..cbf3bc2 100644 > --- a/drivers/net/wireless/marvell/mwifiex/11n_aggr.c > +++ b/drivers/net/wireless/marvell/mwifiex/11n_aggr.c > @@ -194,6 +194,10 @@ > > if (tx_info_src->flags & MWIFIEX_BUF_FLAG_TDLS_PKT) > tx_info_aggr->flags |= MWIFIEX_BUF_FLAG_TDLS_PKT; > + > + if (tx_info_src->flags & MWIFIEX_BUF_FLAG_TCP_ACK) > + tx_info_aggr->flags |= MWIFIEX_BUF_FLAG_TCP_ACK; > + > tx_info_aggr->flags |= MWIFIEX_BUF_FLAG_AGGR_PKT; > skb_aggr->priority = skb_src->priority; > skb_aggr->tstamp = skb_src->tstamp; > diff --git a/drivers/net/wireless/marvell/mwifiex/decl.h b/drivers/net/wireless/marvell/mwifiex/decl.h > index 188e4c3..0487420 100644 > --- a/drivers/net/wireless/marvell/mwifiex/decl.h > +++ b/drivers/net/wireless/marvell/mwifiex/decl.h > @@ -89,6 +89,7 @@ > #define MWIFIEX_BUF_FLAG_EAPOL_TX_STATUS BIT(3) > #define MWIFIEX_BUF_FLAG_ACTION_TX_STATUS BIT(4) > #define MWIFIEX_BUF_FLAG_AGGR_PKT BIT(5) > +#define MWIFIEX_BUF_FLAG_TCP_ACK BIT(6) > > #define MWIFIEX_BRIDGED_PKTS_THR_HIGH 1024 > #define MWIFIEX_BRIDGED_PKTS_THR_LOW 128 > diff --git a/drivers/net/wireless/marvell/mwifiex/main.c b/drivers/net/wireless/marvell/mwifiex/main.c > index 2c42191..b7d835c 100644 > --- a/drivers/net/wireless/marvell/mwifiex/main.c > +++ b/drivers/net/wireless/marvell/mwifiex/main.c > @@ -851,6 +851,30 @@ struct sk_buff * > return skb; > } > > +static bool is_piggyback_tcp_ack(struct sk_buff *skb) > +{ > + struct ethhdr *ethh = NULL; > + struct iphdr *iph = NULL; > + struct tcphdr *tcph = NULL; > + > + ethh = (struct ethhdr *)skb->data; > + if (ntohs(ethh->h_proto) != ETH_P_IP) > + return false; What about IPv6? Also, what about all the protocols that you're *not* hacking this for? Is the latency induced by aggregation really that poor? I'm not really a networking expert here, but this whole patch smells bad to me. > + > + iph = (struct iphdr *)((u8 *)ethh + sizeof(struct ethhdr)); > + if (iph->protocol != IPPROTO_TCP) > + return false; > + > + tcph = (struct tcphdr *)((u8 *)iph + iph->ihl * 4); > + /* Piggyback ack without payload*/ > + if (tcp_flag_word(tcph) == TCP_FLAG_ACK && > + ntohs(iph->tot_len) <= (iph->ihl + tcph->doff) * 4) { > + return true; > + } > + > + return false; > +} > + > /* > * CFG802.11 network device handler for data transmission. > */ > @@ -904,6 +928,8 @@ struct sk_buff * > tx_info->bss_num = priv->bss_num; > tx_info->bss_type = priv->bss_type; > tx_info->pkt_len = skb->len; > + if (is_piggyback_tcp_ack(skb)) > + tx_info->flags |= MWIFIEX_BUF_FLAG_TCP_ACK; > > multicast = is_multicast_ether_addr(skb->data); > > diff --git a/drivers/net/wireless/marvell/mwifiex/main.h b/drivers/net/wireless/marvell/mwifiex/main.h > index 1f7e6c3..b0010a9 100644 > --- a/drivers/net/wireless/marvell/mwifiex/main.h > +++ b/drivers/net/wireless/marvell/mwifiex/main.h > @@ -49,6 +49,7 @@ > #include > #include > #include > +#include I dislike polluting the (driver-global) main.h with #include's that only are used in one file. Especially when you talk about seemingly layer-violating TCP hacks. Can you move this to main.c instead? Brian > > #include "decl.h" > #include "ioctl.h" > diff --git a/drivers/net/wireless/marvell/mwifiex/usb.c b/drivers/net/wireless/marvell/mwifiex/usb.c > index f230017..e9c3d43 100644 > --- a/drivers/net/wireless/marvell/mwifiex/usb.c > +++ b/drivers/net/wireless/marvell/mwifiex/usb.c > @@ -934,6 +934,7 @@ static int mwifiex_usb_aggr_tx_data(struct mwifiex_adapter *adapter, u8 ep, > struct urb_context *context = NULL; > struct txpd *local_tx_pd = > (struct txpd *)((u8 *)skb->data + adapter->intf_hdr_len); > + struct mwifiex_txinfo *tx_info = MWIFIEX_SKB_TXCB(skb); > u8 f_send_aggr_buf = 0; > u8 f_send_cur_buf = 0; > u8 f_precopy_cur_buf = 0; > @@ -989,8 +990,9 @@ static int mwifiex_usb_aggr_tx_data(struct mwifiex_adapter *adapter, u8 ep, > } > } > > - if (local_tx_pd->flags & MWIFIEX_TxPD_POWER_MGMT_NULL_PACKET) { > - /* Send NULL packet immediately*/ > + if (local_tx_pd->flags & MWIFIEX_TxPD_POWER_MGMT_NULL_PACKET || > + tx_info->flags & MWIFIEX_BUF_FLAG_TCP_ACK) { > + /* Send NULL data/TCP ACK packet immediately*/ > if (f_precopy_cur_buf) { > if (skb_queue_empty(&port->tx_aggr.aggr_list)) { > f_precopy_cur_buf = 0; > -- > 1.9.1 >