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:   Mon, 25 Apr 2022 20:33:43 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Oleg Nesterov <oleg@redhat.com>
Cc:     rjw@rjwysocki.net, mingo@kernel.org, vincent.guittot@linaro.org,
        dietmar.eggemann@arm.com, rostedt@goodmis.org, mgorman@suse.de,
        ebiederm@xmission.com, bigeasy@linutronix.de,
        Will Deacon <will@kernel.org>, linux-kernel@vger.kernel.org,
        tj@kernel.org, linux-pm@vger.kernel.org
Subject: Re: [PATCH v2 2/5] sched,ptrace: Fix ptrace_check_attach() vs
 PREEMPT_RT
Message-ID: <20220425183343.GM2731@worktop.programming.kicks-ass.net>
References: <20220421150248.667412396@infradead.org>
 <20220421150654.817117821@infradead.org>
 <20220425143537.GA12412@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220425143537.GA12412@redhat.com>
Precedence: bulk

On Mon, Apr 25, 2022 at 04:35:37PM +0200, Oleg Nesterov wrote:
> On 04/21, Peter Zijlstra wrote:
> >
> > +static void clear_traced_quiesce(void)
> > +{
> > +	spin_lock_irq(&current->sighand->siglock);
> > +	WARN_ON_ONCE(!(current->jobctl & JOBCTL_TRACED_QUIESCE));
> 
> This WARN_ON_ONCE() doesn't look right, the task can be killed right
> after ptrace_stop() sets JOBCTL_TRACED | JOBCTL_TRACED_QUIESCE and
> drops siglock.

OK, will look at that.

> > @@ -2290,14 +2303,26 @@ static int ptrace_stop(int exit_code, in
> >  		/*
> >  		 * Don't want to allow preemption here, because
> >  		 * sys_ptrace() needs this task to be inactive.
> > -		 *
> > -		 * XXX: implement read_unlock_no_resched().
> >  		 */
> >  		preempt_disable();
> >  		read_unlock(&tasklist_lock);
> > -		cgroup_enter_frozen();
> > +		cgroup_enter_frozen(); // XXX broken on PREEMPT_RT !!!
> > +
> > +		/*
> > +		 * JOBCTL_TRACE_QUIESCE bridges the gap between
> > +		 * set_current_state(TASK_TRACED) above and schedule() below.
> > +		 * There must not be any blocking (specifically anything that
> > +		 * touched ->saved_state on PREEMPT_RT) between here and
> > +		 * schedule().
> > +		 *
> > +		 * ptrace_check_attach() relies on this with its
> > +		 * wait_task_inactive() usage.
> > +		 */
> > +		clear_traced_quiesce();
> 
> Well, I think it should be called earlier under tasklist_lock,
> before preempt_disable() above.
> 
> We need tasklist_lock to protect ->parent, debugger can be killed
> and go away right after read_unlock(&tasklist_lock).
> 
> Still trying to convince myself everything is right with
> JOBCTL_STOPPED/TRACED ...

Can't do it earlier, since cgroup_enter_frozen() can do spinlock (eg.
use ->saved_state).