Return-path: Received: from sabertooth02.qualcomm.com ([65.197.215.38]:35022 "EHLO sabertooth02.qualcomm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751542Ab3IPRFN (ORCPT ); Mon, 16 Sep 2013 13:05:13 -0400 From: Kalle Valo To: Michal Kazior CC: , Subject: Re: [PATCHv2 0/7] ath10k: improve TX path References: <1378821003-22925-1-git-send-email-michal.kazior@tieto.com> <1379074618-30534-1-git-send-email-michal.kazior@tieto.com> Date: Mon, 16 Sep 2013 20:05:11 +0300 In-Reply-To: <1379074618-30534-1-git-send-email-michal.kazior@tieto.com> (Michal Kazior's message of "Fri, 13 Sep 2013 14:16:51 +0200") Message-ID: <871u4on1dk.fsf@qca.qualcomm.com> (sfid-20130916_190524_933964_5ED61CAD) MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Sender: linux-wireless-owner@vger.kernel.org List-ID: Michal Kazior writes: > Hi, > > This patchset addresses two issues: > > * system/userspace starvation on heavy briding > UDP TX > * unstable/inconsistent UDP TX throughput > > In short the patchset simplifies TX path by > removing HTC TX workers, makes WMI commands block > and makes ath10k more responsive to queues > becoming full. This contributes to both improved > throughput and makes the system more responsive > under heavy UDP TX load. > > Max stable briding TX (ath10k as TX): > UDP w/o patchset: 520mbps > UDP w/ patchset: 570mbps > > TCP w/o patchset: 350mbps > TCP w/ patchset: 400mbps > > Measured with two 2x2 cards, one acting as an AP > on AP135 board, the other as a STA on a core i5 > laptop. The AP135 was passing traffic from a > station on LAN/eth1 to the STA/wlan0. > > v2: > * use DIV_ROUND_UP > * comment usage of ath10k_wmi_tx_beacons_nowait > * additional check for tx_credits_flow_enabled > > > Michal Kazior (7): > ath10k: simplify HTC credits calculation > ath10k: add HTC TX credits replenishing notification > ath10k: make WMI commands block by design > ath10k: simplify HTC command submitting > ath10k: improve beacon submission latency > ath10k: remove wmi pending count limit > ath10k: remove wmi event worker thread Thanks, applied. -- Kalle Valo