Return-path: Received: from mail.deathmatch.net ([72.66.92.28]:4123 "EHLO mail.deathmatch.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1759128AbZDOMgF (ORCPT ); Wed, 15 Apr 2009 08:36:05 -0400 Date: Wed, 15 Apr 2009 08:35:06 -0400 From: Bob Copeland To: Daniel J Blueman Cc: linville@tuxdriver.com, jirislaby@gmail.com, mickflemm@gmail.com, lrodriguez@atheros.com, ath5k-devel@lists.ath5k.org, linux-wireless@vger.kernel.org Subject: Re: [ath5k-devel] [PATCH 4/5] ath5k: use rx hw descriptor pointer for self-linked check Message-ID: <20090415123506.GA18774@hash.localnet> (sfid-20090415_143614_634633_A761584C) References: <1239796656-20646-1-git-send-email-me@bobcopeland.com> <1239796656-20646-5-git-send-email-me@bobcopeland.com> <6278d2220904150520x2e385e66s146a3dc95db26e3c@mail.gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <6278d2220904150520x2e385e66s146a3dc95db26e3c@mail.gmail.com> Sender: linux-wireless-owner@vger.kernel.org List-ID: On Wed, Apr 15, 2009 at 01:20:09PM +0100, Daniel J Blueman wrote: > Just an observation: Each synchronous read over PCI will take at > minimum 0.9us to return. Accessing tables in the host's cache > hierarchy/memory will likely be much faster - possibly why it is done > like this already. > > Is this a worthwhile tradeoff? This is a good point -- but I think in this case, it is worth it from a correctness rather than performance standpoint. See http://lkml.org/lkml/2009/3/8/99 for why I think the current code doesn't handle every case. FWIW, madwifi does this too. We could eliminate it if we dropped the self-linked buffer thing (but then we'd need to handle EOL interrupts). -- Bob Copeland %% www.bobcopeland.com