Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230328235543.1082207-1-irogers@google.com> <20230328235543.1082207-2-irogers@google.com>
 <ZCQ4nERbit/f6yhh@kernel.org> <ZCQ6kqhODsxSCXdP@kernel.org>
 <CAM9d7cjSEx_=UTMpDHMwGb=5H6Yf8UdHTMt1xO=4CVToh60oSA@mail.gmail.com> <ZCVxXTc0kyRAQ+XA@kernel.org>
In-Reply-To: <ZCVxXTc0kyRAQ+XA@kernel.org>
From:   Ian Rogers <irogers@google.com>
Date:   Thu, 30 Mar 2023 09:59:08 -0700
Message-ID: <CAP-5=fWZpfo62iN3YWWKm_VZZqMFCyddyg_JQTGe7j6nV+DbmQ@mail.gmail.com>
Subject: Re: [PATCH v1 1/6] perf annotate: Delete session for debug builds
To:     Arnaldo Carvalho de Melo <acme@kernel.org>
Cc:     Namhyung Kim <namhyung@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Adrian Hunter <adrian.hunter@intel.com>,
        Nathan Chancellor <nathan@kernel.org>,
        Nick Desaulniers <ndesaulniers@google.com>,
        Tom Rix <trix@redhat.com>, James Clark <james.clark@arm.com>,
        Andi Kleen <ak@linux.intel.com>,
        Kan Liang <kan.liang@linux.intel.com>,
        German Gomez <german.gomez@arm.com>,
        Sandipan Das <sandipan.das@amd.com>,
        Andres Freund <andres@anarazel.de>,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org,
        llvm@lists.linux.dev
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Mar 30, 2023 at 4:24=E2=80=AFAM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> Em Wed, Mar 29, 2023 at 05:13:17PM -0700, Namhyung Kim escreveu:
> > On Wed, Mar 29, 2023 at 6:18=E2=80=AFAM Arnaldo Carvalho de Melo
> > <acme@kernel.org> wrote:
> > >
> > > Em Wed, Mar 29, 2023 at 10:09:48AM -0300, Arnaldo Carvalho de Melo es=
creveu:
> > > > Em Tue, Mar 28, 2023 at 04:55:38PM -0700, Ian Rogers escreveu:
> > > > > Use the debug build indicator as the guide to free the session. T=
his
> > > > > implements a behavior described in a comment, which is consequent=
ially
> > > > > removed.
> > > > >
> > > > > Signed-off-by: Ian Rogers <irogers@google.com>
> > > > > ---
> > > > >  tools/perf/builtin-annotate.c | 16 ++++++----------
> > > > >  1 file changed, 6 insertions(+), 10 deletions(-)
> > > > >
> > > > > diff --git a/tools/perf/builtin-annotate.c b/tools/perf/builtin-a=
nnotate.c
> > > > > index 4750fac7bf93..98d1b6379230 100644
> > > > > --- a/tools/perf/builtin-annotate.c
> > > > > +++ b/tools/perf/builtin-annotate.c
> > > > > @@ -692,16 +692,12 @@ int cmd_annotate(int argc, const char **arg=
v)
> > > > >
> > > > >  out_delete:
> > > > >     /*
> > > > > -    * Speed up the exit process, for large files this can
> > > > > -    * take quite a while.
> > > > > -    *
> > > > > -    * XXX Enable this when using valgrind or if we ever
> > > > > -    * librarize this command.
> > > > > -    *
> > > > > -    * Also experiment with obstacks to see how much speed
> > > > > -    * up we'll get here.
> > > > > -    *
> > > > > -    * perf_session__delete(session);
> > > > > +    * Speed up the exit process by only deleting for debug build=
s. For
> > > > > +    * large files this can save time.
> > > > >      */
> > > > > +#ifndef NDEBUG
> > > > > +   perf_session__delete(annotate.session);
> > > > > +#endif
> > > >
> > > > So now, but default, we will call this, as we don't have this defin=
ed
> > > > only if DEBUG=3D1 is set, right?
> > > >
> > > > =E2=AC=A2[acme@toolbox perf-tools-next]$ find tools/perf/ -type f |=
 xargs grep NDEBUG
> > > > tools/perf/util/mutex.c:#ifndef NDEBUG
> > > > =E2=AC=A2[acme@toolbox perf-tools-next]$
> > >
> > > We can discuss this later, applied the series with just that zfree()
> > > change to annotation_options__exit().
> >
> > I don't think it's just an issue in the perf annotate.  Maybe we can
> > do the same for perf report and so on.
>
> Yes, I thought at some point of setting some flag, perf_exiting, and
> then we would stop releasing memory, zfree comes to mind, but then we
> would still be traversing data structures, taking locks, etc.
>
> And we can't just exit() as we may need to flush caches, etc.
>
> IIRC this specific case appeared in profiles, so was commented out.
>
> > Anyway we could define NDEBUG=3D1 for release builds from now on.
>
> Yes, we should do it.

I'll send out a patch. Just a heads up that this will also disable
asserts for non-debug builds.

Thanks,
Ian

> - Arnaldo