From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752047AbZFYJ4e (ORCPT ); Thu, 25 Jun 2009 05:56:34 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751411AbZFYJ41 (ORCPT ); Thu, 25 Jun 2009 05:56:27 -0400 Received: from senator.holtmann.net ([87.106.208.187]:38640 "EHLO mail.holtmann.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751225AbZFYJ40 (ORCPT ); Thu, 25 Jun 2009 05:56:26 -0400 Subject: Re: [PATCH 1/6] staging: android: binder: Remove some funny && usage From: Marcel Holtmann To: Alan Cox Cc: Daniel Walker , Linus Walleij , Brian Swetland , Arve =?ISO-8859-1?Q?Hj=F8nnev=E5g?= , Jeremy Fitzhardinge , Greg Kroah-Hartman , linux-kernel@vger.kernel.org, hackbod@android.com In-Reply-To: <20090625091519.23bc42ff@lxorguk.ukuu.org.uk> References: <1244832678-30329-1-git-send-email-dwalker@fifo99.com> <1245254936.5982.261.camel@desktop> <4A391A54.7000109@goop.org> <1245274308.5982.268.camel@desktop> <1245451983.32124.25.camel@desktop> <1245849223.32124.112.camel@desktop> <63386a3d0906241701o720fa667t662b9e3d4f080397@mail.gmail.com> <1245889219.32124.280.camel@desktop> <20090625091519.23bc42ff@lxorguk.ukuu.org.uk> Content-Type: text/plain Date: Thu, 25 Jun 2009 11:56:30 +0200 Message-Id: <1245923790.12994.9.camel@localhost.localdomain> Mime-Version: 1.0 X-Mailer: Evolution 2.26.2 (2.26.2-1.fc11) Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Alan, > > > What I really want to know, is how this relates to the vmsplice() and > > > other zero-copy buffer passing schemes already in the kernel. I was > > > sort of dreaming that D-Bus and other IPC could be accelerated on > > > top of that. > > > > Marcel had mentioned earlier in this thread that D-Bus could be > > accelerated with shared memory or moving the dbus-daemon into the > > kernel. splice() and vmplice() seem like fairly robust system calls. I > > would think they could be used also .. > > Except for very large amounts of data what makes you think zero copy > buffer passing will be fast ? TLB shootdowns are expensive and they scale > horribly badly with threaded apps on multiprocessor systems ? there is always the problem if we have really stupidly written apps that just copy megabyte of data from one app to the other. These just need fixing and Lennart posted patches to integrate file descriptor passing into the D-Bus protocol which will make it more versatile. And for most cases it will be just good enough to move file descriptors around. Then having the possibility to pass bigger data blobs without too many penalties would be an extra bonus. Regards Marcel