From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: An inconsistency/bug in ingress netem timestamps Date: Wed, 15 Apr 2009 16:41:42 -0700 (PDT) Message-ID: <20090415.164142.193730680.davem@davemloft.net> References: <20090415202620.GB3322@ami.dom.local> <20090415132918.7579ffa6@nehalam> <200904151700.59906.alexandre.sidorenko@hp.com> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: shemminger@vyatta.com, jarkao2@gmail.com, netdev@vger.kernel.org To: alexandre.sidorenko@hp.com Return-path: Received: from 74-93-104-97-Washington.hfc.comcastbusiness.net ([74.93.104.97]:32874 "EHLO sunset.davemloft.net" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1751761AbZDOXlv (ORCPT ); Wed, 15 Apr 2009 19:41:51 -0400 In-Reply-To: <200904151700.59906.alexandre.sidorenko@hp.com> Sender: netdev-owner@vger.kernel.org List-ID: From: Alex Sidorenko Date: Wed, 15 Apr 2009 17:00:59 -0400 > the timestamps change depending on whether there are any ptype_all handlers > registered. Just starting tcpdump changes the behaviour, this probably > means 'inconsistent' ? It changes whether there is a "timestamp user" and packet sniffers are currently defined as such as user. The argument is whether the overhead of making this type of use a "timestamp user" is warranted or not. Turning on timestamps is heavily optimized like this because taking the timestamp on every packet is extremely expensive, especially on large classes of x86 systems. Therefore if we make changes here, they have to have a very specific and limited scope in order to avoid turning this expensive operation on when it's not really necessary.