From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758250Ab0EGU7A (ORCPT ); Fri, 7 May 2010 16:59:00 -0400 Received: from mail.openrapids.net ([64.15.138.104]:60261 "EHLO blackscsi.openrapids.net" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1755535Ab0EGU66 (ORCPT ); Fri, 7 May 2010 16:58:58 -0400 Date: Fri, 7 May 2010 16:58:56 -0400 From: Mathieu Desnoyers To: Steven Rostedt Cc: Frederic Weisbecker , linux-kernel@vger.kernel.org, Ingo Molnar , Andrew Morton , Thomas Gleixner , Peter Zijlstra , Arnaldo Carvalho de Melo , Lai Jiangshan , Li Zefan , Masami Hiramatsu , Christoph Hellwig , "H. Peter Anvin" , Andi Kleen , "Paul E. McKenney" Subject: Re: [PATCH 3/9 - v2][RFC] tracing: Remove per event trace registering Message-ID: <20100507205855.GB28122@Krystal> References: <20100504034045.085822814@goodmis.org> <20100504034202.282030521@goodmis.org> <20100507042025.GC8069@nowhere> <20100507145438.GB29573@Krystal> <20100507180130.GA5401@nowhere> <1273259305.22438.417.camel@gandalf.stny.rr.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1273259305.22438.417.camel@gandalf.stny.rr.com> X-Editor: vi X-Info: http://www.efficios.com X-Operating-System: Linux/2.6.26-2-686 (i686) X-Uptime: 16:56:28 up 104 days, 23:33, 7 users, load average: 0.20, 0.16, 0.10 User-Agent: Mutt/1.5.18 (2008-05-17) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Steven Rostedt (rostedt@goodmis.org) wrote: > On Fri, 2010-05-07 at 20:01 +0200, Frederic Weisbecker wrote: > > On Fri, May 07, 2010 at 10:54:38AM -0400, Mathieu Desnoyers wrote: > > > The second is this extra parameter passed whether or not it is needed. > > And although we suppose it is safe, I don't feel comfortable with it. > > So if we can find a more proper way to avoid it, I'm all for it. > > Now I'm making the extra parameter mandatory for all tracepoint > probes. ;-) > > But this time, it will be at the start not the end. > > void probe(void *data, proto); > > > Unfortunately we can't avoid it. In order to remove the extra code > (registering and unregistering) and even share the probe among several > events, we need a way to pass the data to the probe to let the probe > know what event it is dealing with (to put in the event id into the > buffer, to let the tracer output code know what event this data is for). > > The current method is that only the proto that the tracepoint uses is > passed to the probe. This gives us no way to add any more information. > > This new method allows data to be assigned at probe register, and the > probe gets this data as the first parameter. > > The register_* functions will still do typechecking of the probes, they > just add the "void *" at the beginning. > > Actually, here is a place that I can see where Mathieu's check does come > in handy. If we add the check test to each probe, and the tracepoint > proto changes, it will flag it. Ah ! finally we are getting on the same page. :-) > > Mathieu, you've been explaining this wrong ;-) Yep, it seems like I've done a terrible job at trying to explain my motivation to you. ;) > > I'm not worried about changes to ftrace.h breaking things. I'm worried > about changes to tracepoint.h breaking ftrace.h. This is where your > check comes in. As I change the void *data from the end to the start, > I'm nervous about catching all the probes that are registered this way. > (ftrace events, syscalls, kprobes, and perf) I'm nervous about catching all tracepoint-related problems that may arise from both tracepoint and tracepoint probes prototype changes, yes. I'm glad we agree then. :-) Thanks, Mathieu > > -- Steve > > -- Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com