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;
Date:   Wed, 5 Jul 2023 10:38:33 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Namhyung Kim <namhyung@kernel.org>
Cc:     Ingo Molnar <mingo@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Kan Liang <kan.liang@linux.intel.com>,
        Stephane Eranian <eranian@google.com>,
        Greg Thelen <gthelen@google.com>, stable@vger.kernel.org
Subject: Re: [PATCH] perf/x86: Fix lockdep warning in
 for_each_sibling_event() on SPR
Message-ID: <20230705083833.GE462772@hirez.programming.kicks-ass.net>
References: <20230704181516.3293665-1-namhyung@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230704181516.3293665-1-namhyung@kernel.org>
Precedence: bulk

On Tue, Jul 04, 2023 at 11:15:15AM -0700, Namhyung Kim wrote:
> On SPR, the load latency event needs an auxiliary event in the same
> group to work properly.  There's a check in intel_pmu_hw_config()
> for this to iterate sibling events and find a mem-loads-aux event.
> 
> The for_each_sibling_event() has a lockdep assert to make sure if it
> disabled hardirq or hold leader->ctx->mutex.  This works well if the
> given event has a separate leader event since perf_try_init_event()
> grabs the leader->ctx->mutex to protect the sibling list.  But it can
> cause a problem when the event itself is a leader since the event is
> not initialized yet and there's no ctx for the event.
> 
> Actually I got a lockdep warning when I run the below command on SPR,
> but I guess it could be a NULL pointer dereference.
> 
>   $ perf record -d -e cpu/mem-loads/uP true
> 
> The code path to the warning is:
> 
>   sys_perf_event_open()
>     perf_event_alloc()
>       perf_init_event()
>         perf_try_init_event()
>           x86_pmu_event_init()
>             hsw_hw_config()
>               intel_pmu_hw_config()
>                 for_each_sibling_event()
>                   lockdep_assert_event_ctx()
> 
> We don't need for_each_sibling_event() when it's a standalone event.
> Let's return the error code directly.
> 
> Fixes: f3c0eba28704 ("perf: Add a few assertions")
> Reported-by: Greg Thelen <gthelen@google.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> ---
>  arch/x86/events/intel/core.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
> index 0d09245aa8df..933fe4894c32 100644
> --- a/arch/x86/events/intel/core.c
> +++ b/arch/x86/events/intel/core.c
> @@ -3983,6 +3983,14 @@ static int intel_pmu_hw_config(struct perf_event *event)
>  		struct perf_event *leader = event->group_leader;
>  		struct perf_event *sibling = NULL;
>  
> +		/*
> +		 * The event is not fully initialized yet and no ctx is set
> +		 * for the event.  Avoid for_each_sibling_event() since it
> +		 * has a lockdep assert with leader->ctx->mutex.
> +		 */

If I understand things correctly, your patch is indeed correct, however
I don't much like this comment, does something like:

		/*
		 * When this memload event is also the first event (no
		 * group exists yet), then there is no aux event before
		 * it.
		 */

work for you?

> +		if (leader == event)
> +			return -ENODATA;
> +
>  		if (!is_mem_loads_aux_event(leader)) {
>  			for_each_sibling_event(sibling, leader) {
>  				if (is_mem_loads_aux_event(sibling))
> -- 
> 2.41.0.255.g8b1d071c50-goog
>