From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757780AbcDMBHY (ORCPT ); Tue, 12 Apr 2016 21:07:24 -0400 Received: from mail-pf0-f195.google.com ([209.85.192.195]:34074 "EHLO mail-pf0-f195.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754931AbcDMBHW (ORCPT ); Tue, 12 Apr 2016 21:07:22 -0400 Date: Wed, 13 Apr 2016 10:06:57 +0900 From: Namhyung Kim To: Arnaldo Carvalho de Melo Cc: Jiri Olsa , David Ahern , Peter Zijlstra , Milian Wolff , Wang Nan , Linux Kernel Mailing List Subject: Re: zeros in frame pointer callchains Message-ID: <20160413010657.GA10693@danjae.aot.lge.com> References: <20160412133847.GC10972@kernel.org> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <20160412133847.GC10972@kernel.org> User-Agent: Mutt/1.6.0 (2016-04-01) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Arnaldo, On Tue, Apr 12, 2016 at 10:38:47AM -0300, Arnaldo Carvalho de Melo wrote: > Hi, > > Namhyung, while working on supporting callchains in 'perf trace' I > noticed something I think we discussed at some point, zeros in frame pointer > callchains... > > Oh well figured it out, its probably userspace not having frame > pointers, grrr anyway, keeping the rest of the message, for reference, > but now I think we need to have some expressiveness to say that for that > particular event we want to use perf_event_attr.exclude_callchain_user... > > First add a special probe point that 'perf trace' uses, if present: > > # perf probe 'vfs_getname=getname_flags:72 pathname=filename:string' > > Then lets ask to see just when this thing hits, no syscall > printing whatsoever ('trace' == 'perf trace'), sometimes just a few > zeros at the top, sometimes a long series, need to investigate how to > chop those off in the kernel, not to use the ring buffer needlessly, > will start chopping these in userspace (in perf trace): My solution was to look at /proc/sys/vm/mmap_min_addr [1]. For kernel it'd be simply discarding callchains after seeing an address below the mmap_min_addr. For userspace, it requires to carry the mmap_min_addr via feature bit or something. Or Maybe we can just discard addresses under 4096 for simplicity? [1] https://lkml.org/lkml/2015/11/26/40 Thanks, Namhyung > > # trace --no-sys --ev probe:vfs_getname/call-graph=fp/ > > 34033.533 probe:vfs_getname:(ffffffff8124c06c) pathname="/usr/share/pixmaps") > getname_flags+0xfe20009d ([kernel.kallsyms]) > user_path_at_empty+0xfe200023 ([kernel.kallsyms]) > vfs_fstatat+0xfe200066 ([kernel.kallsyms]) > SYSC_newstat+0xfe20002e ([kernel.kallsyms]) > sys_newstat+0xfe20000e ([kernel.kallsyms]) > entry_SYSCALL_64_fastpath+0xfe20001a ([kernel.kallsyms]) > __xstat64+0xffff011e90b98015 (/usr/lib64/libc-2.22.so) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > 36001.681 probe:vfs_getname:(ffffffff8124c06c) pathname="/var/tmp") > getname_flags+0xfe20009d ([kernel.kallsyms]) > user_path_at_empty+0xfe200023 ([kernel.kallsyms]) > vfs_fstatat+0xfe200066 ([kernel.kallsyms]) > SYSC_newstat+0xfe20002e ([kernel.kallsyms]) > sys_newstat+0xfe20000e ([kernel.kallsyms]) > entry_SYSCALL_64_fastpath+0xfe20001a ([kernel.kallsyms]) > __xstat64+0xffff00b816a28015 (/usr/lib64/libc-2.22.so) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > 36001.696 probe:vfs_getname:(ffffffff8124c06c) pathname="/var/tmp") > getname_flags+0xfe20009d ([kernel.kallsyms]) > user_path_at_empty+0xfe200023 ([kernel.kallsyms]) > sys_access+0xfe2000b4 ([kernel.kallsyms]) > entry_SYSCALL_64_fastpath+0xfe20001a ([kernel.kallsyms]) > __GI___access+0xffff00b816a28007 (/usr/lib64/libc-2.22.so) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > [0] ([unknown]) > 8254.597 probe:vfs_getname:(ffffffff8124c06c) pathname="/etc/localtime") > getname_flags+0xfe20009d ([kernel.kallsyms]) > user_path_at_empty+0xfe200023 ([kernel.kallsyms]) > vfs_fstatat+0xfe200066 ([kernel.kallsyms]) > SYSC_newstat+0xfe20002e ([kernel.kallsyms]) > sys_newstat+0xfe20000e ([kernel.kallsyms]) > entry_SYSCALL_64_fastpath+0xfe20001a ([kernel.kallsyms]) > __xstat64+0xffff0104f5696015 (/usr/lib64/libc-2.22.so) > [0] ([unknown]) > [0] ([unknown]) > 9001.567 probe:vfs_getname:(ffffffff8124c06c) pathname="/etc/firewalld/ipsets") > getname_flags+0xfe20009d ([kernel.kallsyms]) > user_path_at_empty+0xfe200023 ([kernel.kallsyms]) > sys_inotify_add_watch+0xfe200098 ([kernel.kallsyms]) > entry_SYSCALL_64_fastpath+0xfe20001a ([kernel.kallsyms]) > inotify_add_watch+0xffff008bb9432007 (/usr/lib64/libc-2.22.so) > 9450.181 probe:vfs_getname:(ffffffff8124c06c) pathname="/root") > getname_flags+0xfe20009d ([kernel.kallsyms]) > user_path_at_empty+0xfe200023 ([kernel.kallsyms]) > sys_inotify_add_watch+0xfe200098 ([kernel.kallsyms]) > entry_SYSCALL_64_fastpath+0xfe20001a ([kernel.kallsyms]) > inotify_add_watch+0xffff01b7aa4f2007 (/usr/lib64/libc-2.22.so)