Received-SPF: pass (google.com: domain of linux-kernel+bounces-182093-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Date: Fri, 17 May 2024 14:17:59 +0200
From: Mark Rutland <mark.rutland@arm.com>
To: Will Deacon <will@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>, Ingo Molnar <mingo@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	LKML <linux-kernel@vger.kernel.org>,
	linux-arm-kernel@lists.infradead.org,
	Greg Thelen <gthelen@google.com>,
	Robin Murphy <robin.murphy@arm.com>,
	Tuan Phan <tuanphan@os.amperecomputing.com>
Subject: Re: [PATCH] perf/arm-dmc620: Fix lockdep assert in ->event_init()
Message-ID: <ZkdK9xe3jsc3OxOF@J2N7QTR9R3>
References: <20240514180050.182454-1-namhyung@kernel.org>
 <20240517120234.GA32598@willie-the-truck>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240517120234.GA32598@willie-the-truck>

On Fri, May 17, 2024 at 01:02:34PM +0100, Will Deacon wrote:
> On Tue, May 14, 2024 at 11:00:50AM -0700, Namhyung Kim wrote:
> > for_each_sibling_event() checks leader's ctx but it doesn't have the ctx
> > yet if it's the leader.  Like in perf_event_validate_size(), we should
> > skip checking siblings in that case.
> > 
> > Fixes: f3c0eba287049 ("perf: Add a few assertions")
> > Reported-by: Greg Thelen <gthelen@google.com>
> > Cc: Robin Murphy <robin.murphy@arm.com>
> > Cc: Tuan Phan <tuanphan@os.amperecomputing.com>
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> > ---
> >  drivers/perf/arm_dmc620_pmu.c | 9 ++++++---
> >  1 file changed, 6 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/perf/arm_dmc620_pmu.c b/drivers/perf/arm_dmc620_pmu.c
> > index 8a81be2dd5ec..88c17c1d6d49 100644
> > --- a/drivers/perf/arm_dmc620_pmu.c
> > +++ b/drivers/perf/arm_dmc620_pmu.c
> > @@ -542,12 +542,16 @@ static int dmc620_pmu_event_init(struct perf_event *event)
> >  	if (event->cpu < 0)
> >  		return -EINVAL;
> >  
> > +	hwc->idx = -1;
> > +
> > +	if (event->group_leader == event)
> > +		return 0;
> > +
> >  	/*
> >  	 * We can't atomically disable all HW counters so only one event allowed,
> >  	 * although software events are acceptable.
> >  	 */
> > -	if (event->group_leader != event &&
> > -			!is_software_event(event->group_leader))
> > +	if (!is_software_event(event->group_leader))
> >  		return -EINVAL;
> >  
> >  	for_each_sibling_event(sibling, event->group_leader) {
> > @@ -556,7 +560,6 @@ static int dmc620_pmu_event_init(struct perf_event *event)
> >  			return -EINVAL;
> >  	}
> >  
> > -	hwc->idx = -1;
> >  	return 0;
> >  }
> 
> Thanks, I'll pick this up, although Mark reckoned he'd found some other
> issues over at:
> 
> https://lore.kernel.org/r/Zg0l642PgQ7T3a8Z@FVFF77S0Q05N
> 
> but didn't elaborate on what exactly he'd found :/

Sorry; what I was referring to was that some drivers (including this
one) also forgot to validate that the group size could actually fit on
the PMU, which we're *supposed* to check, but if we fail to do so the
only fallout is that events won't count and we waste a bit of time
trying to schedule events unnecessarily.

For this patch as-is:

Acked-by: Mark Rutland <mark.rutland@arm.com>

.. and I'll try to take a look at the rest when I'm back in the UK.

Mark.