Return-path: Received: from mail-bk0-f51.google.com ([209.85.214.51]:44432 "EHLO mail-bk0-f51.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750707AbaBYHYv convert rfc822-to-8bit (ORCPT ); Tue, 25 Feb 2014 02:24:51 -0500 Received: by mail-bk0-f51.google.com with SMTP id 6so98294bkj.38 for ; Mon, 24 Feb 2014 23:24:50 -0800 (PST) MIME-Version: 1.0 In-Reply-To: <1393312414-9404-1-git-send-email-janusz.dziedzic@tieto.com> References: <1393312414-9404-1-git-send-email-janusz.dziedzic@tieto.com> Date: Tue, 25 Feb 2014 08:24:49 +0100 Message-ID: (sfid-20140225_082455_070390_B7F76C31) Subject: Re: [PATCH] ath10k: improve rx path when play with attention flags From: Michal Kazior To: Janusz Dziedzic Cc: "ath10k@lists.infradead.org" , linux-wireless Content-Type: text/plain; charset=UTF-8 Sender: linux-wireless-owner@vger.kernel.org List-ID: On 25 February 2014 08:13, Janusz Dziedzic wrote: > Currently when we check attention flags we do __le32_to_cpu() > four times for each packet. This could have performance > impact for BIG endian platforms. This patch improve this > little bit. > > Signed-off-by: Janusz Dziedzic > --- > drivers/net/wireless/ath/ath10k/htt_rx.c | 71 +++++------------------------- > 1 file changed, 11 insertions(+), 60 deletions(-) > > diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c > index 316433f..97e4cd5 100644 > --- a/drivers/net/wireless/ath/ath10k/htt_rx.c > +++ b/drivers/net/wireless/ath/ath10k/htt_rx.c > @@ -810,62 +810,6 @@ static void ath10k_htt_rx_msdu(struct ath10k_htt *htt, struct htt_rx_info *info) > ath10k_process_rx(htt->ar, info); > } > > -static bool ath10k_htt_rx_has_decrypt_err(struct sk_buff *skb) > -{ > - struct htt_rx_desc *rxd; > - u32 flags; > - > - rxd = (void *)skb->data - sizeof(*rxd); > - flags = __le32_to_cpu(rxd->attention.flags); > - > - if (flags & RX_ATTENTION_FLAGS_DECRYPT_ERR) > - return true; > - > - return false; > -} > - > -static bool ath10k_htt_rx_has_fcs_err(struct sk_buff *skb) > -{ > - struct htt_rx_desc *rxd; > - u32 flags; > - > - rxd = (void *)skb->data - sizeof(*rxd); > - flags = __le32_to_cpu(rxd->attention.flags); > - > - if (flags & RX_ATTENTION_FLAGS_FCS_ERR) > - return true; > - > - return false; > -} > - > -static bool ath10k_htt_rx_has_mic_err(struct sk_buff *skb) > -{ > - struct htt_rx_desc *rxd; > - u32 flags; > - > - rxd = (void *)skb->data - sizeof(*rxd); > - flags = __le32_to_cpu(rxd->attention.flags); > - > - if (flags & RX_ATTENTION_FLAGS_TKIP_MIC_ERR) > - return true; > - > - return false; > -} > - > -static bool ath10k_htt_rx_is_mgmt(struct sk_buff *skb) > -{ > - struct htt_rx_desc *rxd; > - u32 flags; > - > - rxd = (void *)skb->data - sizeof(*rxd); > - flags = __le32_to_cpu(rxd->attention.flags); > - > - if (flags & RX_ATTENTION_FLAGS_MGMT_TYPE) > - return true; > - > - return false; > -} > - > static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb) > { > struct htt_rx_desc *rxd; > @@ -929,6 +873,9 @@ static void ath10k_htt_rx_handler(struct ath10k_htt *htt, > struct sk_buff *msdu_head, *msdu_tail; > enum htt_rx_mpdu_status status; > int msdu_chaining; > + struct htt_rx_desc *rxd; > + u32 att_flags; > + No need for an empty line I suppose? I would also prefer `attention` instead of `att_flags`, but no big deal. > msdu_head = NULL; > msdu_tail = NULL; > @@ -950,7 +897,11 @@ static void ath10k_htt_rx_handler(struct ath10k_htt *htt, > continue; > } > > - if (ath10k_htt_rx_has_decrypt_err(msdu_head)) { > + rxd = (void *)msdu_head->data - sizeof(*rxd); I think container_of(msdu_head->data, htt_tx_desc, msdu_payload) is a better fit here :-) > + att_flags = __le32_to_cpu(rxd->attention.flags); Attention flags don't contain bitmasked values so we can skip the conversion altogether here and do __cpu_to_le32() on the constant values. > + > + More empty lines? Michał