Re: [RFC 09/11] coresight: etm-perf: Disable the path before capturing the trace data

[Suzuki K Poulose]

On 11/23/20 6:08 AM, Anshuman Khandual wrote:
> On 11/12/20 2:57 PM, Suzuki K Poulose wrote:
> > On 11/10/20 12:45 PM, Anshuman Khandual wrote:
> > > perf handle structure needs to be shared with the TRBE IRQ handler for
> > > capturing trace data and restarting the handle. There is a probability
> > > of an undefined reference based crash when etm event is being stopped
> > > while a TRBE IRQ also getting processed. This happens due the release
> > > of perf handle via perf_aux_output_end(). This stops the sinks via the
> > > link before releasing the handle, which will ensure that a simultaneous
> > > TRBE IRQ could not happen.
> > >
> > > Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
> > > ---
> > > This might cause problem with traditional sink devices which can be
> > > operated in both sysfs and perf mode. This needs to be addressed
> > > correctly. One option would be to move the update_buffer callback
> > > into the respective sink devices. e.g, disable().
> > >
> > >    drivers/hwtracing/coresight/coresight-etm-perf.c | 2 ++
> > >    1 file changed, 2 insertions(+)
> > >
> > > diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c
> > > index 534e205..1a37991 100644
> > > --- a/drivers/hwtracing/coresight/coresight-etm-perf.c
> > > +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c
> > > @@ -429,7 +429,9 @@ static void etm_event_stop(struct perf_event *event, int mode)
> > >              size = sink_ops(sink)->update_buffer(sink, handle,
> > >                              event_data->snk_config);
> > > +        coresight_disable_path(path);
> > >            perf_aux_output_end(handle, size);
> > > +        return;
> > >        }
> >
> > As you mentioned, this is not ideal where another session could be triggered on
> > the sink from a different ETM (not for per-CPU sink) in a different mode before
> > you collect the buffer. I believe the best option is to leave the
> > update_buffer() to disable_hw. This would need to pass on the "handle" to the
> > disable_path.
>
> Passing 'handle' into coresight_ops_sink->disable() would enable pushing
> updated trace data into perf aux buffer. But do you propose to drop the
> update_buffer() call back completely or just move it into disable() call
> back (along with PERF_EF_UPDATE mode check) for all individual sinks for
> now. May be, later it can be dropped off completely.

Yes, once we update the buffer from within the sink_ops->disable(), we don't
need the update buffer anymore. It is pointless to have a function that
is provided to the external user.

> > That way the races can be handled inside the sinks. Also, this aligns the
> > perf mode of the sinks with that of the sysfs mode.
>
> Did not get that, could you please elaborate ?

In sysfs mode, we already do an action similar to "update buffer" for all
the sinks. (e.g, see tmc_etr_sync_sysfs_buf() ). i.e, update the buffer
before the sink is disabled. That is the same we propose above.

Suzuki