From owner-svn-src-all@FreeBSD.ORG Mon Jun 6 11:14:38 2011 Return-Path: Delivered-To: svn-src-all@FreeBSD.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 5B7C61065675 for ; Mon, 6 Jun 2011 11:14:38 +0000 (UTC) (envelope-from vadim@nuclight.avtf.net) Received: from nuclight.avtf.net (nuclight.avtf.net [82.117.70.99]) by mx1.freebsd.org (Postfix) with ESMTP id 7BD808FC18 for ; Mon, 6 Jun 2011 11:14:36 +0000 (UTC) Received: from kernblitz.nuclight.avtf.net (vadim@localhost [127.0.0.1]) by nuclight.avtf.net (8.14.4/8.14.4) with ESMTP id p56BEY6s037687; Mon, 6 Jun 2011 18:14:34 +0700 (NOVST) (envelope-from vadim@kernblitz.nuclight.avtf.net) Received: (from vadim@localhost) by kernblitz.nuclight.avtf.net (8.14.4/8.14.4/Submit) id p56BEYws037684; Mon, 6 Jun 2011 18:14:34 +0700 (NOVST) (envelope-from vadim) Message-Id: <201106061114.p56BEYws037684@kernblitz.nuclight.avtf.net> To: Robert Watson From: Vadim Goncharov In-Reply-To: References: <201106042331.p54NVfC0030751@svn.freebsd.org> <4DEB3C32.9070709@freebsd.org> X-Comment-To: Robert Watson Date: Mon, 6 Jun 2011 18:14:33 +0700 User-Agent: slrn/0.9.9p1 (FreeBSD) Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: svn-src-all@FreeBSD.org Subject: Re: svn commit: r222702 - head/sys/sys X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list Reply-To: vadim_nuclight@mail.ru List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 06 Jun 2011 11:14:38 -0000 Hi Robert Watson! On Sun, 5 Jun 2011 10:20:34 +0100 (BST); Robert Watson wrote: >> hmm note that my stealing of 4 bits for FIB was supposed to go away for 8 >> and beyond. We need to do an mbuf redesign.. at this stage, I guess it's a >> project for 10. > Indeed -- it would be nice not just to reclaim the FIB bits, but also to > support more simultaneous FIBs. We still need to do a KBI analysis of mbuf > for 9.0 across various architectures, but it would be pretty late to fix the > FIB thing. On a redesign, clearly too late for that as well. I still hope we > can get the variable size mbuf cluster in for 10 though. An observation for redesign: we also need a somewhat garbage collection in the networking subsytem, for example, a generation count field in an mbuf (pkthdr?) header. The problem is when mbuf or some of it's mbuf_tags(9) holds pointer to an entity which is already gone while mbuf was in-the-fly (e.g. delayed in dummynet). There are at lease two PRs with this problem for the case of destroyed network interfaces: http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/153255 http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/153671 They are closed, but only because patches were used for specific production use with specific value of net.isr.direct. Originator confirmed yesterday that system still panics in such cases with INVARIANTS. Also, there were voices lining idea to have more than 256 FIBs, e.g. 4096. May be a new uint32_t with bitfields for FIBs, hash and generation count?.. -- WBR, Vadim Goncharov. ICQ#166852181 mailto:vadim_nuclight@mail.ru [Moderator of RU.ANTI-ECOLOGY][FreeBSD][http://antigreen.org][LJ:/nuclight]