Return-path: Received: from mail-wi0-f179.google.com ([209.85.212.179]:36361 "EHLO mail-wi0-f179.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932241AbaKXOgE (ORCPT ); Mon, 24 Nov 2014 09:36:04 -0500 Received: by mail-wi0-f179.google.com with SMTP id ex7so5996521wid.12 for ; Mon, 24 Nov 2014 06:36:03 -0800 (PST) From: Emmanuel Grumbach To: linux-wireless@vger.kernel.org Cc: Johannes Berg , Emmanuel Grumbach Subject: [PATCH 40/75] iwlwifi: mvm: pull SNAP header into skb->head Date: Mon, 24 Nov 2014 16:34:16 +0200 Message-Id: <1416839691-28533-40-git-send-email-egrumbach@gmail.com> (sfid-20141124_153623_207964_4EE4DC81) In-Reply-To: References: Sender: linux-wireless-owner@vger.kernel.org List-ID: From: Johannes Berg When we pre-populate the skb->head for the stack, we only pull in the 802.11 header including crypto (assuming the packet isn't short enough to be in there completely.) This is fine, but in ieee80211_data_to_8023() we later unconditionally pull 8 more bytes for the SNAP header and ethertype field (except for mesh or 4-addr, where it's even more, but we don't care as much about them). Avoid the additional later pull by pulling in those 8 bytes here. Signed-off-by: Johannes Berg Reviewed-by: IdoX Yariv Signed-off-by: Emmanuel Grumbach --- drivers/net/wireless/iwlwifi/mvm/rx.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/drivers/net/wireless/iwlwifi/mvm/rx.c b/drivers/net/wireless/iwlwifi/mvm/rx.c index 69fa765..94b6e72 100644 --- a/drivers/net/wireless/iwlwifi/mvm/rx.c +++ b/drivers/net/wireless/iwlwifi/mvm/rx.c @@ -104,10 +104,19 @@ static void iwl_mvm_pass_packet_to_mac80211(struct iwl_mvm *mvm, unsigned int hdrlen, fraglen; /* If frame is small enough to fit in skb->head, pull it completely. - * If not, only pull ieee80211_hdr (including crypto if present) so - * that splice() or TCP coalesce are more efficient. + * If not, only pull ieee80211_hdr (including crypto if present, and + * an additional 8 bytes for SNAP/ethertype, see below) so that + * splice() or TCP coalesce are more efficient. + * + * Since, in addition, ieee80211_data_to_8023() always pull in at + * least 8 bytes (possibly more for mesh) we can do the same here + * to save the cost of doing it later. That still doesn't pull in + * the actual IP header since the typical case has a SNAP header. + * If the latter changes (there are efforts in the standards group + * to do so) we should revisit this and ieee80211_data_to_8023(). */ - hdrlen = (len <= skb_tailroom(skb)) ? len : sizeof(*hdr) + crypt_len; + hdrlen = (len <= skb_tailroom(skb)) ? len : + sizeof(*hdr) + crypt_len + 8; memcpy(skb_put(skb, hdrlen), hdr, hdrlen); fraglen = len - hdrlen; -- 1.9.1