| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <1b0464bf-0258-44d3-adca-4346a9fdfd31@linux.intel.com>
Date: Fri, 16 Aug 2024 15:34:41 -0400
From: "Liang, Kan" <kan.liang@...ux.intel.com>
To: Namhyung Kim <namhyung@...nel.org>
Cc: acme@...nel.org, irogers@...gle.com, peterz@...radead.org,
mingo@...nel.org, linux-kernel@...r.kernel.org,
linux-perf-users@...r.kernel.org
Subject: Re: [PATCH] perf report: Support --total-cycles --group in the tui
mode
On 2024-08-15 6:44 p.m., Namhyung Kim wrote:
> Hi Kan,
>
> On Wed, Aug 14, 2024 at 7:19 AM <kan.liang@...ux.intel.com> wrote:
>>
>> From: Kan Liang <kan.liang@...ux.intel.com>
>>
>> With the --total-cycles, the --group is only supported in the stdio
>> mode, but not supported in the tui mode. The output is inconsistent
>> with different modes.
>>
>> if symbol_conf.event_group is applied, always output the event group
>> information together in tui mode as well.
>>
>> $perf record -e "{cycles,instructions}",cache-misses -b sleep 1
>> $perf report --total-cycles --group --tui
>>
>> Before the patch,
>>
>> Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Pro
>> 6.45% 2.0K 0.57% 667 [dl-cacheinfo.h:164 -> dl
>> 5.62% 1.7K 0.74% 871 [dl-cache.c:230 -
>> 5.21% 1.6K 1.37% 1.6K [setup-vdso.h:37 ->
>> 4.92% 1.5K 0.09% 109 [dl-cache.c:367 -
>>
>> Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Pro
>> 5.62% 1.7K 2.76% 871 [dl-cache.c:230 -
>> 4.92% 1.5K 0.35% 109 [dl-cache.c:367 -
>> 1.12% 346 0.55% 173
>> 0.87% 270 0.43% 135 [rtld.c:5
>> 0.64% 197 0.03% 8 [dl-tunables.c:305 -> d
>> 0.60% 185 0.01% 3 [dl-tunables.c:305 -> d
>>
>> Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Pro
>> 5.90% 1.8K 1.28% 1.8K [strtod.c:8
>> 5.70% 1.8K 1.24% 1.8K [strtod_l.c:681 -
>> 5.68% 1.8K 1.24% 1.8K [strtod_l.c:508 -
>> 5.48% 1.7K 1.19% 1.7K [strtod_l.c:1175 ->
>> 5.21% 1.6K 1.13% 1.6K [setup-vdso.h:37 ->
>>
>> With the patch,
>>
>> Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Pro
>> 6.45% 2.0K 0.57% 667 [dl-cacheinfo.h:164 -> dl
>> 5.62% 1.7K 0.74% 871 [dl-cache.c:230 -
>> 5.21% 1.6K 1.37% 1.6K [setup-vdso.h:37 ->
>> 4.92% 1.5K 0.09% 109 [dl-cache.c:367 -
>
> Hmm.. it seems the output just removed the second one.
> I guess it should combine the first and second output somehow.
>
The patch is just to make the behavior/output the same between --stdio
and the tui mode for --group. I didn't change the algorithm of
calculating the cycles.
Yes, it looks suspicious. I will take a deep look and see how the group
information are processed.
Thanks,
Kan
> Thanks,
> Namhyung
>
>>
>> Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Pro
>> 5.90% 1.8K 1.28% 1.8K [strtod.c:8
>> 5.70% 1.8K 1.24% 1.8K [strtod_l.c:681 -
>> 5.68% 1.8K 1.24% 1.8K [strtod_l.c:508 -
>> 5.48% 1.7K 1.19% 1.7K [strtod_l.c:1175 ->
>> 5.21% 1.6K 1.13% 1.6K [setup-vdso.h:37 ->
>>
>> Fixes: 7fa46cbf20d3 ("perf report: Sort by sampled cycles percent per block for tui")
>> Closes: https://lore.kernel.org/lkml/ZrupfUSZwem-hCZm@x1/
>> Signed-off-by: Kan Liang <kan.liang@...ux.intel.com>
>> ---
>> tools/perf/builtin-report.c | 6 +++++-
>> tools/perf/ui/browsers/hists.c | 12 ++++++++++--
>> 2 files changed, 15 insertions(+), 3 deletions(-)
>>
>> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
>> index 1643113616f4..574342fb7269 100644
>> --- a/tools/perf/builtin-report.c
>> +++ b/tools/perf/builtin-report.c
>> @@ -541,7 +541,11 @@ static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *
>> int i = 0, ret;
>>
>> evlist__for_each_entry(evlist, pos) {
>> - ret = report__browse_block_hists(&rep->block_reports[i++].hist,
>> + i++;
>> + if (symbol_conf.event_group && !evsel__is_group_leader(pos))
>> + continue;
>> +
>> + ret = report__browse_block_hists(&rep->block_reports[i - 1].hist,
>> rep->min_percent, pos,
>> &rep->session->header.env);
>> if (ret != 0)
>> diff --git a/tools/perf/ui/browsers/hists.c b/tools/perf/ui/browsers/hists.c
>> index 49ba82bf3391..22af1a6e29d6 100644
>> --- a/tools/perf/ui/browsers/hists.c
>> +++ b/tools/perf/ui/browsers/hists.c
>> @@ -3665,11 +3665,19 @@ single_entry: {
>> static int block_hists_browser__title(struct hist_browser *browser, char *bf,
>> size_t size)
>> {
>> - struct hists *hists = evsel__hists(browser->block_evsel);
>> - const char *evname = evsel__name(browser->block_evsel);
>> + struct evsel *evsel = browser->block_evsel;
>> + struct hists *hists = evsel__hists(evsel);
>> unsigned long nr_samples = hists->stats.nr_samples;
>> + const char *evname;
>> + char buf[512];
>> int ret;
>>
>> + if (evsel__is_group_event(evsel)) {
>> + evsel__group_desc(evsel, buf, sizeof(buf));
>> + evname = buf;
>> + } else
>> + evname = evsel__name(evsel);
>> +
>> ret = scnprintf(bf, size, "# Samples: %lu", nr_samples);
>> if (evname)
>> scnprintf(bf + ret, size - ret, " of event '%s'", evname);
>> --
>> 2.38.1
>>
>
Powered by blists - more mailing lists