From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from s3.sipsolutions.net ([5.9.151.49]:56944 "EHLO sipsolutions.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755362AbbDNPNt (ORCPT ); Tue, 14 Apr 2015 11:13:49 -0400 Received: by sipsolutions.net with esmtpsa (TLS1.2:ECDHE_RSA_AES_128_GCM_SHA256:128) (Exim 4.84) (envelope-from ) id 1Yi2X6-0000G0-C3 for linux-wireless@vger.kernel.org; Tue, 14 Apr 2015 17:13:48 +0200 Message-ID: <1429024427.3019.18.camel@sipsolutions.net> (sfid-20150414_171353_232842_7E8DDE08) Subject: Re: [RFC/RFT 00/11] mac80211: enable fast-xmit and some offloads From: Johannes Berg To: linux-wireless@vger.kernel.org Date: Tue, 14 Apr 2015 17:13:47 +0200 In-Reply-To: <1429023829-3991-1-git-send-email-johannes@sipsolutions.net> (sfid-20150414_170402_357343_42C8C498) References: <1429023829-3991-1-git-send-email-johannes@sipsolutions.net> (sfid-20150414_170402_357343_42C8C498) Content-Type: text/plain; charset="UTF-8" Mime-Version: 1.0 Sender: linux-wireless-owner@vger.kernel.org List-ID: On Tue, 2015-04-14 at 17:03 +0200, Johannes Berg wrote: > First, add the TX fastpath ("fast-xmit") that I've been kicking around for > a while. I'm pretty happy with the abstraction since it allows me to not > have to worry about a lot of details in the regular TX path... > > Secondly, I want to enable more offloads. So the first thing to do is to > actually fix checksum offload - the TI driver has a bug in that if it fails > to upload keys or so, checksums will be completely bogus. > > So the first step to fix that bug is to enable checksum offload only on > the fast-xmit path, while having a software fallback on the regular TX path > to make software crypto etc. do the right thing. To not cause regressions > for the TI/ath10k drivers, this requires extending fast-xmit to cover more > cases, but that's easy. > > Secondly then, we can add code to do in software what the driver might do, > which is actually what the network stack does for us anyway today, but now > we can do it only on the non-fast-xmit path so that the fast-xmit path can > pass packets with required checksumming or segmentation to the device. > > So far I've only tested the software fallbacks, not the actual hardware > offloads. Forgot to say - for those who do want to play with this, this is also available in my mac80211-next tree in the fast-xmit branch. johannes