From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeremy Jackson Subject: solutions for forcedeth jumbo rx and order:2 alloc? Date: Fri, 18 Dec 2009 12:52:22 -0500 Message-ID: <1261158742.4932.190.camel@ragnarok> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit To: netdev@vger.kernel.org Return-path: Received: from titan.coplanar.net ([70.47.139.2]:37431 "EHLO titan.coplanar.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932237AbZLRSFi (ORCPT ); Fri, 18 Dec 2009 13:05:38 -0500 Received: from [70.47.139.174] (ragnarok.coplanar.net [70.47.139.174] (may be forged)) (authenticated bits=0) by titan.coplanar.net (8.14.2/8.14.2/Debian-2build1) with ESMTP id nBIHqZ4b006817 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT) for ; Fri, 18 Dec 2009 12:52:36 -0500 Sender: netdev-owner@vger.kernel.org List-ID: Currently, forcedeth seems to want an order:2 alloc for each frame rx. I find this behaviour makes using 9100 byte jumbos somewhat suicidal O.o Frequent kernel messages, failure of bridge STP (the goggles, they do nothing!) due to missed packets, etc. I realize that it's a reverse engineered driver, and that changing the way hardware descriptors works to use paged SKBs might be very difficult, so I'm wondering if the following would work: Allocate linear static buffers for use in rx ring descriptors once at device init time (ifconfig up), and copy into paged skbs when handing > page sized SKBs up to higher layer network stack? (hmm did i just ask to copy every rx packet?) -- Jeremy Jackson Coplanar Networks (519)489-4903 http://www.coplanar.net jerj@coplanar.net