From mboxrd@z Thu Jan 1 00:00:00 1970 From: Arun Sharma Subject: Re: [RFC] Sort inclusive Date: Tue, 6 Mar 2012 11:41:38 -0800 Message-ID: <4F566872.9010408@fb.com> References: <20120305221253.GA30250@dev3310.snc6.facebook.com> <20120306065753.GB5100@elte.hu> Mime-Version: 1.0 Content-Type: text/plain; charset="ISO-8859-1"; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from mx0b-00082601.pphosted.com ([67.231.153.30]:56827 "EHLO mx0a-00082601.pphosted.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S965176Ab2CFTmR (ORCPT ); Tue, 6 Mar 2012 14:42:17 -0500 In-Reply-To: <20120306065753.GB5100@elte.hu> Sender: linux-perf-users-owner@vger.kernel.org List-ID: To: Ingo Molnar Cc: Arnaldo Carvalho de Melo , Frederic Weisbecker , Mike Galbraith , Paul Mackerras , Peter Zijlstra , Stephane Eranian , Tom Zanussi , linux-kernel@vger.kernel.org, linux-perf-users@vger.kernel.org On 3/5/12 10:57 PM, Ingo Molnar wrote: > Btw., assuming that you are profiling user-space applications > via call-chains, how do you deal with the lack of dwarf decoding > done by perf, in particular on 64-bit x86 systems where most > distros don't compile call-chains into libraries and > application? In what practical situations does that limitation > hinder you and what do you do about it - rebuild your apps with > frame pointers included? Yes - we're rebuilding everything with frame pointers so we could get callchains. While this works for C/C++, I'm not convinced that we have a good solution for JITs where there may be multiple stacks (eg: one each for interpreted and JIT code), and the kernel may not be able to unwind without dwarf decoding. For now, we're going with the /tmp/perf-$pid.txt based solution for JITs. I'll post a new patch with the code refactored into smaller functions. -Arun