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;
From:   Valentin Schneider <vschneid@redhat.com>
To:     Peter Zijlstra <peterz@infradead.org>,
        Wander Lairson Costa <wander@redhat.com>
Cc:     Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Ian Rogers <irogers@google.com>,
        Adrian Hunter <adrian.hunter@intel.com>,
        Will Deacon <will@kernel.org>,
        Waiman Long <longman@redhat.com>,
        Boqun Feng <boqun.feng@gmail.com>,
        Juri Lelli <juri.lelli@redhat.com>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        "Eric W. Biederman" <ebiederm@xmission.com>,
        Oleg Nesterov <oleg@redhat.com>,
        Brian Cain <bcain@quicinc.com>,
        Kefeng Wang <wangkefeng.wang@huawei.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        "Liam R. Howlett" <Liam.Howlett@oracle.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        Christian Brauner <brauner@kernel.org>,
        Andrei Vagin <avagin@gmail.com>,
        Shakeel Butt <shakeelb@google.com>,
        open list <linux-kernel@vger.kernel.org>,
        "open list:PERFORMANCE EVENTS SUBSYSTEM" 
        <linux-perf-users@vger.kernel.org>, Hu Chunyu <chuhu@redhat.com>,
        Paul McKenney <paulmck@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>
Subject: Re: [PATCH v7 2/3] sched/task: Add the
 put_task_struct_atomic_safe() function
In-Reply-To: <20230504084229.GI1734100@hirez.programming.kicks-ass.net>
References: <20230425114307.36889-1-wander@redhat.com>
 <20230425114307.36889-3-wander@redhat.com>
 <20230504084229.GI1734100@hirez.programming.kicks-ass.net>
Date:   Thu, 04 May 2023 10:32:31 +0100
Message-ID: <xhsmha5ykjvbk.mognet@vschneid.remote.csb>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

On 04/05/23 10:42, Peter Zijlstra wrote:
> On Tue, Apr 25, 2023 at 08:43:02AM -0300, Wander Lairson Costa wrote:
>> diff --git a/include/linux/sched/task.h b/include/linux/sched/task.h
>> index b597b97b1f8f..cf774b83b2ec 100644
>> --- a/include/linux/sched/task.h
>> +++ b/include/linux/sched/task.h
>> @@ -141,6 +141,41 @@ static inline void put_task_struct_many(struct task_struct *t, int nr)
>>
>>  void put_task_struct_rcu_user(struct task_struct *task);
>>
>> +extern void __delayed_put_task_struct(struct rcu_head *rhp);
>> +
>> +static inline void put_task_struct_atomic_safe(struct task_struct *task)
>> +{
>> +	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
>> +		/*
>> +		 * Decrement the refcount explicitly to avoid unnecessarily
>> +		 * calling call_rcu.
>> +		 */
>> +		if (refcount_dec_and_test(&task->usage))
>> +			/*
>> +			 * under PREEMPT_RT, we can't call put_task_struct
>> +			 * in atomic context because it will indirectly
>> +			 * acquire sleeping locks.
>> +			 * call_rcu() will schedule __delayed_put_task_struct()
>> +			 * to be called in process context.
>> +			 *
>> +			 * __put_task_struct() is called when
>> +			 * refcount_dec_and_test(&t->usage) succeeds.
>> +			 *
>> +			 * This means that it can't conflict with
>> +			 * put_task_struct_rcu_user() which abuses ->rcu the same
>> +			 * way; rcu_users has a reference so task->usage can't be
>> +			 * zero after rcu_users 1 -> 0 transition.
>> +			 *
>> +			 * delayed_free_task() also uses ->rcu, but it is only called
>> +			 * when it fails to fork a process. Therefore, there is no
>> +			 * way it can conflict with put_task_struct().
>> +			 */
>> +			call_rcu(&task->rcu, __delayed_put_task_struct);
>> +	} else {
>> +		put_task_struct(task);
>> +	}
>> +}
>
> Urgh.. that's plenty horrible. And I'm sure everybody plus kitchen sink
> has already asked why can't we just rcu free the thing unconditionally.
>
> Google only found me an earlier version of this same patch set, but I'm
> sure we've had that discussion many times over the past several years.
> The above and your follow up patch is just horrible.
>

So on v3/v4 we got to doing that unconditionally for PREEMPT_RT, but per
[1] Wander went back to hand-fixing the problematic callsites.

Now that I'm looking at it again, I couldn't find a concrete argument from
Oleg against doing this unconditionally - as Wander is pointing out in the
changelog and comments, reusing task_struct.rcu for that purpose is safe
(although not necessarily obviously so).

Is this just miscommunication, or is there a genuine issue with doing this
unconditionally? As argued before, I'd also much rather have this be an
unconditional call_rcu() (regardless of context or PREEMPT_RT).