Re: [PATCH v2] perf session: add missing evlist__delete when deleting a session

From: Jiri Olsa <jolsa@redhat.com>
To: Riccardo Mancini <rickyman7@gmail.com>
Cc: Ian Rogers <irogers@google.com>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	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>,
	Adrian Hunter <adrian.hunter@intel.com>,
	Kan Liang <kan.liang@linux.intel.com>,
	Leo Yan <leo.yan@linaro.org>,
	linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] perf session: add missing evlist__delete when deleting a session
Date: Sun, 27 Jun 2021 19:29:26 +0200	[thread overview]
Message-ID: <YNi1dspuN/cdNAKr@krava> (raw)
In-Reply-To: <4725ef2249540b6876daea3714a38a6be59223e6.camel@gmail.com>

On Fri, Jun 25, 2021 at 05:45:59PM +0200, Riccardo Mancini wrote:
> Hi,
> 
> thank you both for your comments.
> 
> On Fri, 2021-06-25 at 13:54 +0200, Jiri Olsa wrote:
> > On Thu, Jun 24, 2021 at 10:39:34PM -0700, Ian Rogers wrote:
> > > On Thu, Jun 24, 2021 at 4:20 PM Riccardo Mancini <rickyman7@gmail.com>
> > > wrote:
> > > > 
> > > > ASan reports a memory leak caused by evlist not being deleted on exit in
> > > > perf-report, perf-script and perf-data.
> > > > The problem is caused by evlist->session not being deleted, which is
> > > > allocated in perf_session__read_header, called in perf_session__new if
> > > > perf_data is in read mode.
> > > > In case of write mode, the session->evlist is filled by the caller.
> > > > This patch solves the problem by calling evlist__delete in
> > > > perf_session__delete if perf_data is in read mode.
> > 
> > ugh, I'm surprised we did not free that.. and can't find
> > in git log we ever did ;-) I briefly check commands using
> > sessions and looks like it's correct
> > 
> > Acked-by: Jiri Olsa <jolsa@redhat.com>
> > 
> > > 
> > > Acked-by: Ian Rogers <irogers@google.com>
> > > 
> > > It is messy that in read mode the session owns the evlist, but
> > > otherwise not. Imo, it'd be nice to make the ownership unconditional.
> > 
> > yep, would be nice
> 
> I think the root problem is that perf_session__new has different behaviours
> depending on perf_data and perf_tool and that it probably does too many things
> for a __new function.
> If we split it into multiple functions and then, say, we create two helpers
> perf_session__init_read and perf_session__init_write, with the corresponding
> perf_session__fini_read and perf_session__fini_write, then the conditional
> ownership won't be a big problem due to having these two cases clearly
> separated.
> What do you think?

yes, interesting idea, let's see how the code looks like ;-)

thanks,
jirka