Date: Thu, 7 Dec 2006 18:49:13 +0100 From: Paolo Pisati <piso@FreeBSD.org> To: Robert Watson <rwatson@FreeBSD.org> Cc: Perforce Change Reviews <perforce@FreeBSD.org>, Gleb Smirnoff <glebius@FreeBSD.org>, Andre Oppermann <andre@FreeBSD.org>, Paolo Pisati <piso@FreeBSD.org> Subject: Re: PERFORCE change 111230 for review Message-ID: <20061207174913.GB1195@tin.it> In-Reply-To: <20061207161434.O50906@fledge.watson.org> References: <200612062319.kB6NJgsq031755@repoman.freebsd.org> <20061207110225.GU32700@FreeBSD.org> <4578070A.2030609@freebsd.org> <20061207142254.GA1195@tin.it> <20061207161434.O50906@fledge.watson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Dec 07, 2006 at 04:16:41PM +0000, Robert Watson wrote: > On Thu, 7 Dec 2006, Paolo Pisati wrote: > > >Then, after a discussion on irc, pull the plug on any present (and future) > >hackery & half-baked solution, and declare in kernel libalias incompatible > >with tso. > > This seems silly -- why is it not compatible? Because libalias expects a pkt into a contiguous piece of memory, and to achieve this we previously do a pullup of the entire pkt into an mbuf cluster. This worked fine until libalias met a tso capable nic, and i did a 'cvsup' on that box: TSO created pkts bigger than 2k, the pullup function couldn't handle them and pkts were silently discarded. The real solution here is to make libalias use mbuf chain, but i wanted this code to enter the tree without any further delay, and thus libalias was declared incompatible with tso. bye -- Paolo Piso's first law: nothing works as expected!
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20061207174913.GB1195>