From owner-freebsd-net@FreeBSD.ORG Thu Oct 26 01:05:44 2006 Return-Path: X-Original-To: freebsd-net@freebsd.org Delivered-To: freebsd-net@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 98EFD16A416 for ; Thu, 26 Oct 2006 01:05:44 +0000 (UTC) (envelope-from scottl@samsco.org) Received: from pooker.samsco.org (pooker.samsco.org [168.103.85.57]) by mx1.FreeBSD.org (Postfix) with ESMTP id 9838143D58 for ; Thu, 26 Oct 2006 01:05:43 +0000 (GMT) (envelope-from scottl@samsco.org) Received: from [192.168.254.11] (phobos.samsco.home [192.168.254.11]) (authenticated bits=0) by pooker.samsco.org (8.13.4/8.13.4) with ESMTP id k9Q15ZBC094046; Wed, 25 Oct 2006 19:05:41 -0600 (MDT) (envelope-from scottl@samsco.org) Message-ID: <454009DF.5000704@samsco.org> Date: Wed, 25 Oct 2006 19:05:35 -0600 From: Scott Long User-Agent: Mozilla/5.0 (Macintosh; U; Intel Mac OS X; en-US; rv:1.8.0.7) Gecko/20060910 SeaMonkey/1.0.5 MIME-Version: 1.0 To: Jack Vogel References: <200610251818.k9PIIe7p062530@ambrisko.com> <2a41acea0610251736n16cc4188h489f6d953130f91a@mail.gmail.com> In-Reply-To: <2a41acea0610251736n16cc4188h489f6d953130f91a@mail.gmail.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit X-Spam-Status: No, score=-1.4 required=3.8 tests=ALL_TRUSTED autolearn=failed version=3.1.1 X-Spam-Checker-Version: SpamAssassin 3.1.1 (2006-03-10) on pooker.samsco.org Cc: freebsd-net , John Polstra Subject: Re: em network issues X-BeenThere: freebsd-net@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Networking and TCP/IP with FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 26 Oct 2006 01:05:44 -0000 Jack Vogel wrote: > On 10/25/06, Doug Ambrisko wrote: > >> 3) In em_process_receive_interrupts/em_rxeof always decrement >> the count on every run through the loop. If you notice >> count is an is an int that starts at the passed in value >> of -1. It then count-- until count==0. Doing -1, -2, -3 >> takes awhile until the int rolls over to 0. Passing 100 >> limits it more :-) So this can run 3 * 100 versuses >> infinite * int roll over assuming we don't skip a count--. > > Been chatting with Jesse Brandeburg (one of our senior Linux guys) about > receive side cleaning. Gave me a number of things to think about. First > off, > this change you mention is problematic. The reason it doesnt increment > every time thru the while loop is its meant as a packet counter, NOT a > descriptor counter. If we just fix this number at 100, and have it only > counting descriptors you could get all but the EOP descriptor of a packet > and then exit without finishing it and calling the stack, not a good > tactic. > > Having a limited count is still a good idea, but I think we still want > to base > it on packets and not just descriptors. > > Jesse also talked about their experience with the Linux driver, deciding > where to update the RDT, my current code doesnt do it til after the whole > while loop is completed (havent looked at CURRENT again today yet), > Jesse suggested doing it but not EVERY pass in the loop, maybe making > it mod the number of rx descriptors. Having that AND a fixed limit on the > number of total packets cleaned in a pass might be good. Good idea. Though for 1518 byte frames, I think you'll only have one descriptor per packet. Definitely need to do the right thing for jumbo frames, though. > > I was also thinking, maybe having the taskqueue code be selectable, but > not just a POLL vs TASKQUEUE, rather keep a legacy intr option which > has a POLL option within it. > > My motivation for doing that is I can take the TASKQUEUE code into the > Intel code base, but make it backward compatible, the default would have > it optioned off. > > Jack I had it that way initially, and I think you commented that it was ugly ;-)