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:   Tue, 18 Apr 2023 12:25:45 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Yang Jihong <yangjihong1@huawei.com>
Cc:     mingo@redhat.com, acme@kernel.org, mark.rutland@arm.com,
        alexander.shishkin@linux.intel.com, jolsa@kernel.org,
        namhyung@kernel.org, irogers@google.com, adrian.hunter@intel.com,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] perf/core: Fix perf_sample_data not properly initialized
 for different swevents in perf_tp_event()
Message-ID: <20230418102545.GX4253@hirez.programming.kicks-ass.net>
References: <20230412095240.181516-1-yangjihong1@huawei.com>
 <20230417114512.GK83892@hirez.programming.kicks-ass.net>
 <15805714-27c0-b8ff-143a-8f768704a673@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <15805714-27c0-b8ff-143a-8f768704a673@huawei.com>
Precedence: bulk

On Tue, Apr 18, 2023 at 09:35:23AM +0800, Yang Jihong wrote:

> > I'm thinking perhaps those flags that update ->dyn_size are the problem?
> > At the same time, Should you not also then clear dyn_size?

> Yes, according to the code, dyn_size should also be cleared.
> Maybe we need to change it to the following, which would be more
> appropriate?
> 
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -10144,14 +10144,14 @@ void perf_tp_event(u16 event_type, u64 count, void
> *record, int entry_size,
>                 },
>         };
> 
> -       perf_sample_data_init(&data, 0, 0);
> -       perf_sample_save_raw_data(&data, &raw);
> -
>         perf_trace_buf_update(record, event_type);
> 
>         hlist_for_each_entry_rcu(event, head, hlist_entry) {
> -               if (perf_tp_event_match(event, &data, regs))
> +               if (perf_tp_event_match(event, &data, regs)) {
> +                       perf_sample_data_init(&data, 0, 0);
> +                       perf_sample_save_raw_data(&data, &raw);
>                         perf_swevent_event(event, count, &data, regs);
> +               }
>         }

That is certainly the safe option. I just went through the list and
while there's certainly a number of options we'll recompute for naught,
most of them are indeed either dyn_size or event specific :/

So yeah, please send the above as v2.