Re: [BUG] perf report: garbage output with event grouping

From: Stephane Eranian
Date: Wed Jun 26 2013 - 08:39:33 EST


On Wed, Jun 26, 2013 at 11:13 AM, Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> On Wed, Jun 26, 2013 at 10:42:49AM +0200, Stephane Eranian wrote:
>> Hi,
>>
>>
>> $ perf record -e 'cpi{cycles,instructions}' noploop 5
>>
>> $ perf report --stdio --group
>> # group: ^P^Q<A0>\{cycles,instructions}
>> # ========
>> #
>> # Samples: 39K of event '^P^Q<A0>\ { cycles, instructions }'
>> # Event count (approx.): 37782563692
>> #
>> # Overhead Command Shared Object Symbol
>> # ................ ....... ................. ...........................
>> #
>> 99.91% 99.98% noploop noploop [.] main
>> 0.06% -nan% noploop [kernel.kallsyms] [k] read_hpet
>> 0.01% 0.01% noploop [kernel.kallsyms] [k] __acct_update_integrals
>> 0.01% -nan% noploop [kernel.kallsyms] [k] raise_softirq
>> 0.01% -nan% noploop [kernel.kallsyms] [k] apic_timer_interrupt
>> 0.01% -nan% noploop [kernel.kallsyms] [k] __do_softirq
>> 0.01% -nan% noploop [kernel.kallsyms] [k] common_interrupt
>> 0.00% 0.00% noploop [kernel.kallsyms] [k] intel_pmu_enable_all
>> 0.00% 0.01% noploop [kernel.kallsyms] [k] cpuacct_account_field
>> 0.00% 0.01% noploop [kernel.kallsyms] [k] __local_bh_enable
>>
>>
>> 1. The group name is garbage if not anonymous group.
>> 2. arithmetic error in the overhead columns leading to NaN.
>> 3. I assume the Overhead column order is the group event order. Is that right?
>
> hi
>
> 1) attached patch fixies that for me
> 2) I cannot reproduce this one on acme's tree
> 3) right, and we do need event names there ;-)
>
> ---
> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> index 738d3b8..673d98c 100644
> --- a/tools/perf/util/header.c
> +++ b/tools/perf/util/header.c
> @@ -2089,7 +2089,7 @@ static int process_group_desc(struct perf_file_section *section __maybe_unused,
> evsel->leader = evsel;
> /* {anon_group} is a dummy name */
> if (strcmp(desc[i].name, "{anon_group}"))
> - evsel->group_name = desc[i].name;
> + evsel->group_name = strdup(desc[i].name);
> evsel->nr_members = desc[i].nr_members;
>
> if (i >= nr_groups || nr > 0) {

Ok, works. Submit the patch please.

Related question: trying to understand the point of naming a group. Is
this used somewhere else in
perf report/annotate?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/