[PATCH v5 2/2] arm: perf: Add event descriptions

[Will Deacon]

On Thu, Oct 15, 2015 at 08:15:06AM -0700, Drew Richardson wrote:
> On Thu, Oct 15, 2015 at 02:21:12PM +0100, Will Deacon wrote:
> > On Tue, Oct 13, 2015 at 08:36:45AM -0700, Drew Richardson wrote:
> > > Add additional information about the ARM architected hardware events
> > > to make counters self describing. This makes the hardware PMUs easier
> > > to use as perf list contains possible events instead of users having
> > > to refer to documentation like the ARM TRMs.
> > > 
> > > Signed-off-by: Drew Richardson <drew.richardson at arm.com>
> > > ---
> > >  arch/arm/kernel/perf_event_v7.c | 121 ++++++++++++++++++++++++++++++++++++++++
> > >  drivers/perf/arm_pmu.c          |   1 +
> > >  2 files changed, 122 insertions(+)
> > 
> > [...]
> > 
> > > diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
> > > index 2365a32a595e..e933d2dd71c0 100644
> > > --- a/drivers/perf/arm_pmu.c
> > > +++ b/drivers/perf/arm_pmu.c
> > > @@ -548,6 +548,7 @@ static void armpmu_init(struct arm_pmu *armpmu)
> > >  		.stop		= armpmu_stop,
> > >  		.read		= armpmu_read,
> > >  		.filter_match	= armpmu_filter_match,
> > > +		.attr_groups	= armpmu->pmu.attr_groups,
> > 
> > I don't understand this hunk. What's it doing?
> 
> I'm not 100% clear either on what it's doing. But without this line
> the attr_groups don't get passed on and I don't see them on my TC2. I
> debugged the issue down to this but it may not be the proper way to
> solve the problem.

Oh yuck, it's because we call armpmu_init after cpu_pmu_init and the former
uses struct initialisation and ends up zeroing anything set previously.

We should probably tidy all this up:

  * Remove armpmu_register and call perf_pmu_register directly from
    arm_pmu_device_probe instead

  * Call armpmu_init immediately prior to arm_cpu_init

Will