Re: [PATCH v2 21/21] perf metric: Allow modifiers on metrics.

[Arnaldo Carvalho de Melo]

On October 19, 2021 5:00:23 PM GMT-03:00, Ian Rogers <irogers@xxxxxxxxxx> wrote:
>On Tue, Oct 19, 2021 at 8:18 AM Arnaldo Carvalho de Melo
><acme@xxxxxxxxxx> wrote:
>>
>> Em Tue, Oct 19, 2021 at 12:17:10PM -0300, Arnaldo Carvalho de Melo escreveu:
>> > Em Tue, Oct 19, 2021 at 12:13:52PM -0300, Arnaldo Carvalho de Melo escreveu:
>> > > Em Tue, Oct 19, 2021 at 12:06:17PM -0300, Arnaldo Carvalho de Melo escreveu:
>> > > > Em Fri, Oct 15, 2021 at 10:21:32AM -0700, Ian Rogers escreveu:
>> > > > > By allowing modifiers on metrics we can, for example, gather the
>> > > > > same metric for kernel and user mode. On a SkylakeX with
>> > > > > TopDownL1 this gives:
>> > > > >
>> > > > > $ perf stat -M TopDownL1:u,TopDownL1:k -a sleep 2
>> > > > >
>> > > > >  Performance counter stats for 'system wide':
>> > > >
>> > > > Hi Ian, can you please take a look on this? this is on my perf/core
>> > > > branch.
>> > >
>> > > I processed the first version of this series, reviewed by Andi, can you
>> > > please submit the diff from one to the other?
>> >
>> > The interdiff from the 21st patch on the first batch versus on the v2
>> > batch is below, but it doesn't apply to my current perf/core branch,
>> > lemme push it to tmp.perf/core...
>>
>> It was there already, what I have locally is what is in tmp.perf/core.
>
>Hi Arnaldo,
>
>The last change I see in tmp.perf/core is:
>https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/commit/?h=tmp.perf/core&id=5f43061b84d815c0f6477c1a8836bf0c6fec15a2
>
>The difference between v2 and v1 which fixes this bug is:
>
>+@@ -1500,7 +1568,10 @@ int metricgroup__copy_metric_events(struct
>evlist *evlist, struct cgroup *cgrp,
>+                               return -ENOMEM;
>+
>+                       new_expr->metric_expr = old_expr->metric_expr;
>+-                      new_expr->metric_name = old_expr->metric_name;
>++                      new_expr->metric_name = strdup(old_expr->metric_name);
>++                      if (!new_expr->metric_name)
>++                              return -ENOMEM;
>++
>+                       new_expr->metric_unit = old_expr->metric_unit;
>+                       new_expr->runtime = old_expr->runtime;
>+
>
>I also cleaned up some checkpatch line length warnings in v2, which I
>think is the reason for the other changes. Ideally I'd prefer the v2
>patch set over the v1, but they are largely identical. Both were based
>on acme/perf/core. Let me know how I can help.
>

I'm trying not to rewrite the perf/core branch after it's made public, so the best at this point is to add a new patch fixing the problem, if you can please send it, that would help.

If it was just on tmp.perf/core, then it would be ok to fix it.

- Arnaldo