From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andi Kleen Subject: Re: What helps with stacks? Date: Tue, 21 Apr 2015 13:06:34 -0700 Message-ID: <87sibtmek5.fsf@tassilo.jf.intel.com> References: Mime-Version: 1.0 Content-Type: text/plain Return-path: Received: from mga09.intel.com ([134.134.136.24]:48574 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932109AbbDUUGj (ORCPT ); Tue, 21 Apr 2015 16:06:39 -0400 In-Reply-To: (Daniel Speyer's message of "Tue, 21 Apr 2015 19:54:28 +0000 (UTC)") Sender: linux-perf-users-owner@vger.kernel.org List-ID: To: Daniel Speyer Cc: linux-perf-users@vger.kernel.org Daniel Speyer writes: > Hello all, > > I'm trying to get stack traces for a bunch of events using perf record and > having trouble with them getting cut off. I'm using --callgraph dwarf, > which helps a little, and I recompiled libc with --fno-omit-frame-pointer, > which doesn't seem to have helped at all. Very often, the bottom of the > stack is in libc, often __epoll_wait_nocancel. Stacks for > sched:sched_switch events seem to be worse than cycles or sched:sched_wakeup > events, but it isn't 100%. Is there anything else I could try? If you have a Haswell/Broadwell system and uptodate perf (4.0) you can also use --call-graph lbr This uses a new hardware mechanism that can handle many situations where the old ones failed (but also has some situations where it has to give up) -andi -- ak@linux.intel.com -- Speaking for myself only