From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757733Ab2EDMdF (ORCPT ); Fri, 4 May 2012 08:33:05 -0400 Received: from mga03.intel.com ([143.182.124.21]:28421 "EHLO mga03.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757116Ab2EDMdD (ORCPT ); Fri, 4 May 2012 08:33:03 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.71,315,1320652800"; d="scan'208";a="96383921" Message-ID: <4FA3CC71.50404@intel.com> Date: Fri, 04 May 2012 20:32:49 +0800 From: "Yan, Zheng" User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:12.0) Gecko/20120424 Thunderbird/12.0 MIME-Version: 1.0 To: Jiri Olsa CC: Peter Zijlstra , mingo@elte.hu, andi@firstfloor.org, eranian@google.com, ming.m.lin@intel.com, linux-kernel@vger.kernel.org Subject: Re: [PATCH 9/9] perf tool: Add pmu event alias support References: <1335924440-11242-1-git-send-email-zheng.z.yan@intel.com> <1335924440-11242-10-git-send-email-zheng.z.yan@intel.com> <20120503105655.GB1671@m.brq.redhat.com> <1336044261.22523.11.camel@twins> <20120503200518.GC1671@m.brq.redhat.com> In-Reply-To: <20120503200518.GC1671@m.brq.redhat.com> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 05/04/2012 04:05 AM, Jiri Olsa wrote: > On Thu, May 03, 2012 at 01:24:21PM +0200, Peter Zijlstra wrote: >> On Thu, 2012-05-03 at 12:56 +0200, Jiri Olsa wrote: >>> - in sysfs you would have directory with aliases (now called 'events') >>> - each alias is sysfs dir, with file attrs: >>> file name = term name, file value = term value >>> eg.: >>> events/ >>> CAS_COUNT_RD/ >>> # files: >>> config - value 1 >>> config1 - value 2 >>> mask - value ... >> >> I'd prefer the thing Yan proposed (if the sysfs folks let us), >> >> $foo/events/QHL_REQUEST_REMOTE_READS >> >> with contents: "event=0x20,umask=0x4" >> >>> this way it's also possible to add extra terms to existing alias >>> in command line if needed... might be handy >>> >> That should always be possible, if you modify the parser to take things >> like: >> >> event=0x20,umask=0x4,event=0x21 >> >> and have latter values override earlier values, so it collapses into: >> >> umask=0x4,event=0x21 >> >> you can simply take whatever comes out of the event file and stick extra >> bits at the end. > > I discussed this with Peter on irc, so I'll try to sum it up > > we have following options so far: > > with event alias 'al' with definition 'config=1,config1=1,config2=2' > > 1) inside parse_events_add_pmu function > once alias term is detected as part of event definition 'pmu/al/mod' we > construct new event 'pmu/config=1,config1=1,config2=2/mod' and rerun the > event parser on that > > 2) inside parse_events_add_pmu function > once alias term is detected as part of event definition 'pmu/al/mod' we > replace that term with list of terms for that alias definition and run > perf_pmu__config with this new term list > > 3) during bison/flex processing > have option 2) embeded inside flex/bison rules. Once alias term > is detected, insert the aliased terms directly to the list of terms, > not replacing expos as in option 2. > > > - option 1 is currently implemented > - options 2 and 3 requires the aliased config is loaded/parsed from pmu > sysfs tree in form of terms list > - option 3 is a little fuzzy for me now on how to integrate this with > flex/bison > > My interest here is to go with option 2 or 3 if possible - preferrably 2 ;), > because I think it's better/cleaner to deal with terms in one place once they > are parsed - in parse_events_add_pmu function. > > I think there's no need to re run the whole parser (option 1) when we > have the whole thing ready by adding just extra terms. > > thoughts? > I agree with you, option 2 is cleaner than option 1. I will try implementing it Thank you Yan, Zheng