Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200626194720.2915044-1-jolsa@kernel.org> <20200626212522.GF818054@tassilo.jf.intel.com>
 <CAP-5=fVMs4Ok3=gYmzheNTzbBUGGHbCr0cpJSm9TV45aeZb4Ng@mail.gmail.com>
 <20200626215759.GG818054@tassilo.jf.intel.com> <20200627124821.GF29008@kernel.org>
In-Reply-To: <20200627124821.GF29008@kernel.org>
From:   Ian Rogers <irogers@google.com>
Date:   Sat, 27 Jun 2020 16:25:54 -0700
Message-ID: <CAP-5=fXxrNVfiCd6pVmM9BPErXcVU4PKgX8Fvbzq63H=xh=2Hw@mail.gmail.com>
Subject: Re: [RFC 00/10] perf tools: Add support to reuse metric
To:     Arnaldo Carvalho de Melo <acme@kernel.org>
Cc:     Andi Kleen <ak@linux.intel.com>, Jiri Olsa <jolsa@kernel.org>,
        lkml <linux-kernel@vger.kernel.org>,
        Ingo Molnar <mingo@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Peter Zijlstra <a.p.zijlstra@chello.nl>,
        Michael Petlan <mpetlan@redhat.com>,
        Kajol Jain <kjain@linux.ibm.com>,
        John Garry <john.garry@huawei.com>,
        "Paul A. Clarke" <pc@us.ibm.com>,
        Stephane Eranian <eranian@google.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Jun 27, 2020 at 5:48 AM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> Em Fri, Jun 26, 2020 at 02:57:59PM -0700, Andi Kleen escreveu:
> > > The name could be a metric or an event, the logic for each is quite
> >
> > I would say collisions are unlikely. Event names follow quite structured
> > patterns.
>
> And when introducing a new metric the build process can detect that
> clash and fail.
>
> > > different. You could look up an event and when it fails assume it was
> > > a metric, but I like the simplicity of this approach.
>
> > I don't think it's simpler for the user.
>
> Agreed.
>
> > > Maybe this
> > > change could be adopted more widely with something like "perf stat -e
> > > metric:IPC -a -I 1000" rather than the current "perf stat -M IPC -a -I
> > > 1000".
> >
> > I thought about just adding metrics to -e, without metric: of course.
>
> Ditto.

Thanks, while we're thinking about this I'd like there to be support
for flags on metrics. Such as 'perf stat -M IPC:u ...' where the ':u'
is specifying user only as with events.

Fwiw, another point of pain is lining up events with cgroups. Being
able to have the cgroup be a flag on an event or metric would be nice.

Ian

> - Arnaldo