perf: buildid not managed properly when cmd path is relative
in [Kernel]
|
Prev: [GIT PULL] more Ceph bits for 2.6.35-rc1
Next: [PATCH v2] act_nat: fix the wrong checksum when addr isn't in old_addr/mask
From: Arnaldo Carvalho de Melo on 29 May 2010 19:00 Em Thu, May 27, 2010 at 03:46:16PM +0200, Stephane Eranian escreveu: > I ran into another problem while running more tests with > perf record, perf buildid-list. > > I do the following: > > $ perf record foo/noploop 5 > $ perf buildid-list > 54b1e7cc3cf52e0db255aab44ce7538eb62655b8 [kernel.kallsyms] > 875ae61623e89f408b425ca0486a9ec99e3ac73e > /home/eranian/perfmon/official/tip/build/tools/perf/foo/noploop > > I know I have samples in noploop: > $ perf report -D > ... > 0x10a0 [0x20]: PERF_RECORD_SAMPLE(IP, 2): 14721/14721: 0x4006d6 period: 2351576 > ... thread: noploop:14721 > ...... dso: ./foo/noploop > > But if I ask with buildid-list (like per-archive is doing) then I get: > > $ perf buildid-list --with-hits > 54b1e7cc3cf52e0db255aab44ce7538eb62655b8 [kernel.kallsyms] > 0000000000000000000000000000000000000000 ./foo/noploop > > The builid is bogus for noploop and it is relative path not full anymore. > > I instrumented __dsos__fprintf_buildid() and I get: > hit=0 name=/home/eranian/perfmon/official/tip/build/tools/perf/foo/noploop > hit=1 name=./foo/noploop > 0000000000000000000000000000000000000000 ./foo/noploop > > So it looks like when cmd is relative there are two entries but the one > that counts the hits is the one with relative path. But is does not have > the buildid, the full path entry does. > > This is an issue because perf-archive only packages the > content of .debug with hits. > > The problem does not exists when cmd are found from PATH. I reproduced the problem, and yeah, annoying, will work on it, thanks for the report. - Arnaldo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Arnaldo Carvalho de Melo on 1 Jun 2010 14:40 Em Thu, May 27, 2010 at 03:46:16PM +0200, Stephane Eranian escreveu: > I ran into another problem while running more tests with perf record, > perf buildid-list. > I do the following: > $ perf record foo/noploop 5 ; perf buildid-list > 54b1e7cc3cf52e0db255aab44ce7538eb62655b8 [kernel.kallsyms] > 875ae61623e89f408b425ca0486a9ec99e3ac73e > /home/eranian/perfmon/official/tip/build/tools/perf/foo/noploop > > I know I have samples in noploop: > $ perf report -D > 0x10a0 [0x20]: PERF_RECORD_SAMPLE(IP, 2): 14721/14721: 0x4006d6 period: 2351576 > ... thread: noploop:14721 > ...... dso: ./foo/noploop > But if I ask with buildid-list (like per-archive is doing) then I get: > $ perf buildid-list --with-hits > 54b1e7cc3cf52e0db255aab44ce7538eb62655b8 [kernel.kallsyms] > 0000000000000000000000000000000000000000 ./foo/noploop > The builid is bogus for noploop and it is relative path not full anymore. Hi Stephane, Can you please try the following patch? Thanks in advance, - Arnaldo From 9d146ff3c994634c529396442a4a96b0e58809e2 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo <acme(a)redhat.com> Date: Tue, 1 Jun 2010 12:37:05 -0300 Subject: [PATCH 1/1] perf buildid-list: Fix --with-hits event processing MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When we use plain 'perf buildid-list' we use only what is in the buildid table in the perf.data header. And those have absolute pathnames because at 'perf record' time we used __perf_session__process_events and that doesn't sets up the path shortening code in map__new() that happens if symbol_conf.full_paths is false, the default. On the other hand, when we use 'perf buildid-list --with-hits' we process all the events using perf_session__process_events, adding entries to the global DSO list _after_ removing the current directory from the DSO name, for presentation purposes. Because of that we end up having two entries in the DSO list when recording events for binaries using relative pathnames. Fix it minimally by setting symbol_conf.full_paths to true when marking the DSOs with hits in 'perf buildid-list --with-hits', as used by 'perf archive' Right fix longer term is to shorten the path only at presentation time. Will be done for 2.6.36. Reported-by: Stephane Eranian <eranian(a)google.com> Cc: David S. Miller <davem(a)davemloft.net> Cc: Fr�d�ric Weisbecker <fweisbec(a)gmail.com> Cc: Ingo Molnar <mingo(a)elte.hu> Cc: Mike Galbraith <efault(a)gmx.de> Cc: Paul Mackerras <paulus(a)samba.org> Cc: Peter Zijlstra <peterz(a)infradead.org> Cc: Stephane Eranian <eranian(a)google.com> Cc: Tom Zanussi <tzanussi(a)gmail.com> LKML-Reference: <new-submission> Signed-off-by: Arnaldo Carvalho de Melo <acme(a)redhat.com> --- tools/perf/builtin-buildid-list.c | 4 +++- 1 files changed, 3 insertions(+), 1 deletions(-) diff --git a/tools/perf/builtin-buildid-list.c b/tools/perf/builtin-buildid-list.c index 44a47e1..9989072 100644 --- a/tools/perf/builtin-buildid-list.c +++ b/tools/perf/builtin-buildid-list.c @@ -43,8 +43,10 @@ static int __cmd_buildid_list(void) if (session == NULL) return -1; - if (with_hits) + if (with_hits) { + symbol_conf.full_paths = true; perf_session__process_events(session, &build_id__mark_dso_hit_ops); + } perf_session__fprintf_dsos_buildid(session, stdout, with_hits); -- 1.6.5.2 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Arnaldo Carvalho de Melo on 1 Jun 2010 16:30
Em Tue, Jun 01, 2010 at 10:05:24PM +0200, Stephane Eranian escreveu: > Hi, > > You patch does seem to fix the problem. thanks for confirming. > However, I think there may be another issue, maybe > caused by the patch. > > If you monitor a program WITHOUT buildids, and you run > > $ perf record tmp/foo > [ perf record: Woken up 1 times to write data ] > [ perf record: Captured and wrote 0.077 MB perf.data (~3370 samples) ] > > $ ./perf buildid-list > 86200ab2bca285344a55156a079debbbe172bcc5 [kernel.kallsyms] > > foo does not appear, yet we know it got samples. > > $ ./perf buildid-list --with-hits > 86200ab2bca285344a55156a079debbbe172bcc5 [kernel.kallsyms] > 0000000000000000000000000000000000000000 /home/eranian/tmp/foo > > With the --with-hits option, I get more output than without.Supposedly > no option means 'prints all entries', including those with hits or > without buildids. Right, I noticed that too, but concentrated on fixing the problem that was clearly a bug, will add this to my TODO list as I agree this is inconsistent and/or incomplete behaviour. Will add a Tested-by: you and push it to Ingo via perf/urgent. Thanks for testing, - Arnaldo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |