From owner-freebsd-net@FreeBSD.ORG Tue Jul 15 07:24:59 2014 Return-Path: Delivered-To: freebsd-net@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [8.8.178.115]) (using TLSv1 with cipher ADH-AES256-SHA (256/256 bits)) (No client certificate requested) by hub.freebsd.org (Postfix) with ESMTPS id 54B67F83; Tue, 15 Jul 2014 07:24:59 +0000 (UTC) Received: from mail-pa0-x232.google.com (mail-pa0-x232.google.com [IPv6:2607:f8b0:400e:c03::232]) (using TLSv1 with cipher ECDHE-RSA-RC4-SHA (128/128 bits)) (Client CN "smtp.gmail.com", Issuer "Google Internet Authority G2" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 220EF2588; Tue, 15 Jul 2014 07:24:59 +0000 (UTC) Received: by mail-pa0-f50.google.com with SMTP id et14so1339518pad.23 for ; Tue, 15 Jul 2014 00:24:58 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=from:date:to:cc:subject:message-id:reply-to:references:mime-version :content-type:content-disposition:in-reply-to:user-agent; bh=PcczKJlQXd2Nid+bZBFSQ4jg9PHTO4QQC7uyR+d+dzs=; b=n0G/sFcfLwGXg+iHiKx62RdtTf0C9UVz6GgFRgkWNHVxAG1GGzvZCVqTVch/+XRm8D 3qe9pJav28Fpk/lYZSoBGixb1k7S6SMCstfGiG3G87XU+CzSE8o733EqF0V6OTNUWJt2 sQKYtNw2bSlwg/SliO4XRDMyFxHUXamciXnCkT323CJwKXkP/L/0Cwn95w8ZDiTxANvy ELhO3gSKdxWVIb5TnXmJdFbiasm+zI7O+PJwl6tbYuxSv2aosSZKCO4RgyQWyOI898Uh 4tODqNEKu0X37JepmcBl44HjBWLXdaT5eglFKee/9BP2UwTUM2Y2da/ZxmXtlvuqig83 waJg== X-Received: by 10.66.65.204 with SMTP id z12mr21094161pas.60.1405409098609; Tue, 15 Jul 2014 00:24:58 -0700 (PDT) Received: from pyunyh@gmail.com ([106.247.248.2]) by mx.google.com with ESMTPSA id ei4sm12967689pbb.42.2014.07.15.00.24.55 for (version=TLSv1 cipher=RC4-SHA bits=128/128); Tue, 15 Jul 2014 00:24:57 -0700 (PDT) From: Yonghyeon PYUN X-Google-Original-From: "Yonghyeon PYUN" Received: by pyunyh@gmail.com (sSMTP sendmail emulation); Tue, 15 Jul 2014 16:24:49 +0900 Date: Tue, 15 Jul 2014 16:24:49 +0900 To: Rick Macklem Subject: Re: NFS client READ performance on -current Message-ID: <20140715072449.GA1488@michelle.fasterthan.com> Reply-To: pyunyh@gmail.com References: <20140712060538.GA3649@michelle.fasterthan.com> <2136988575.13956627.1405199640153.JavaMail.root@uoguelph.ca> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <2136988575.13956627.1405199640153.JavaMail.root@uoguelph.ca> User-Agent: Mutt/1.4.2.3i Cc: "Russell L. Carter" , freebsd-net@freebsd.org, John Baldwin X-BeenThere: freebsd-net@freebsd.org X-Mailman-Version: 2.1.18 Precedence: list List-Id: Networking and TCP/IP with FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 15 Jul 2014 07:24:59 -0000 On Sat, Jul 12, 2014 at 05:14:00PM -0400, Rick Macklem wrote: > Yonghyeon Pyun wrote: > > On Fri, Jul 11, 2014 at 09:54:23AM -0400, John Baldwin wrote: > > > On Thursday, July 10, 2014 6:31:43 pm Rick Macklem wrote: > > > > John Baldwin wrote: > > > > > On Thursday, July 03, 2014 8:51:01 pm Rick Macklem wrote: > > > > > > Russell L. Carter wrote: > > > > > > > > > > > > > > > > > > > > > On 07/02/14 19:09, Rick Macklem wrote: > > > > > > > > > > > > > > > Could you please post the dmesg stuff for the network > > > > > > > > interface, > > > > > > > > so I can tell what driver is being used? I'll take a look > > > > > > > > at > > > > > > > > it, > > > > > > > > in case it needs to be changed to use m_defrag(). > > > > > > > > > > > > > > em0: port > > > > > > > 0xd020-0xd03f > > > > > > > mem > > > > > > > 0xfe4a0000-0xfe4bffff,0xfe480000-0xfe49ffff irq 44 at > > > > > > > device 0.0 > > > > > > > on > > > > > > > pci2 > > > > > > > em0: Using an MSI interrupt > > > > > > > em0: Ethernet address: 00:15:17:bc:29:ba > > > > > > > 001.000007 [2323] netmap_attach success for em0 > > > > > > > tx > > > > > > > 1/1024 > > > > > > > rx > > > > > > > 1/1024 queues/slots > > > > > > > > > > > > > > This is one of those dual nic cards, so there is em1 as > > > > > > > well... > > > > > > > > > > > > > Well, I took a quick look at the driver and it does use > > > > > > m_defrag(), > > > > > > but > > > > > > I think that the "retry:" label it does a goto after doing so > > > > > > might > > > > > > be in > > > > > > the wrong place. > > > > > > > > > > > > The attached untested patch might fix this. > > > > > > > > > > > > Is it convenient to build a kernel with this patch applied > > > > > > and then > > > > > > try > > > > > > it with TSO enabled? > > > > > > > > > > > > rick > > > > > > ps: It does have the transmit segment limit set to 32. I have > > > > > > no > > > > > > idea if > > > > > > this is a hardware limitation. > > > > > > > > > > I think the retry is not in the wrong place, but the overhead > > > > > of all > > > > > those > > > > > pullups is apparently quite severe. > > > > The m_defrag() call after the first failure will just barely > > > > squeeze > > > > the just under 64K TSO segment into 32 mbuf clusters. Then I > > > > think any > > > > m_pullup() done during the retry will allocate an mbuf > > > > (at a glance it seems to always do this when the old mbuf is a > > > > cluster) > > > > and prepend that to the list. > > > > --> Now the list is > 32 mbufs again and the > > > > bus_dmammap_load_mbuf_sg() > > > > will fail again on the retry, this time fatally, I think? > > > > > > > > I can't see any reason to re-do all the stuff using m_pullup() > > > > and Russell > > > > reported that moving the "retry:" fixed his problem, from what I > > > > understood. > > > > > > Ah, I had assumed (incorrectly) that the m_pullup()s would all be > > > nops in this > > > case. It seems the NIC would really like to have all those things > > > in a single > > > segment, but it is not required, so I agree that your patch is > > > fine. > > > > > > > I recall em(4) controllers have various limitation in TSO. Driver > > has to update IP header to make TSO work so driver has to get a > > writable mbufs. bpf(4) consumers will see IP packet length is 0 > > after this change. I think tcpdump has a compile time option to > > guess correct IP packet length. The firmware of controller also > > should be able to access complete IP/TCP header in a single buffer. > > I don't remember more details in TSO limitation but I guess you may > > be able to get more details TSO limitation from publicly available > > Intel data sheet. > I think that the patch should handle this ok. All of the m_pullup() > stuff gets done the first time. Then, if the result is more than 32 > mbufs in the list, m_defrag() is called to copy the chain. This should > result in all the header stuff in the first mbuf cluster and the map > call is done again with this list of clusters. (Without the patch, > m_pullup() would allocate another prepended mbuf and make the chain > more than 32mbufs again.) > Yes, your patch looks right. > Russell seemed to confirm that the patch fixed the problem for him, > but since I don't have em(4) hardware, it would be nice to have someone > with commit privilege and access to em(4) hardware test and commit it. > Due to breakage of power supply on a box with em(4) controller, I can't test the patch. But I guess it's ok to commit it and Russel already tested it. Thanks for your patch.