Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
From:   "Eric W. Biederman" <ebiederm@xmission.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     rjw@rjwysocki.net, oleg@redhat.com, mingo@kernel.org,
        vincent.guittot@linaro.org, dietmar.eggemann@arm.com,
        rostedt@goodmis.org, mgorman@suse.de, bigeasy@linutronix.de,
        Will Deacon <will@kernel.org>, linux-kernel@vger.kernel.org,
        tj@kernel.org, linux-pm@vger.kernel.org
References: <20220421150248.667412396@infradead.org>
        <20220421150654.817117821@infradead.org>
Date:   Thu, 21 Apr 2022 13:40:57 -0500
In-Reply-To: <20220421150654.817117821@infradead.org> (Peter Zijlstra's
        message of "Thu, 21 Apr 2022 17:02:50 +0200")
Message-ID: <87czhap9dy.fsf@email.froward.int.ebiederm.org>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.1 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Subject: Re: [PATCH v2 2/5] sched,ptrace: Fix ptrace_check_attach() vs
 PREEMPT_RT
Precedence: bulk

Peter Zijlstra <peterz@infradead.org> writes:

> Rework ptrace_check_attach() / ptrace_unfreeze_traced() to not rely on
> task->__state as much.
>
> Due to how PREEMPT_RT is changing the rules vs task->__state with the
> introduction of task->saved_state while TASK_RTLOCK_WAIT (the whole
> blocking spinlock thing), the way ptrace freeze tries to do things no
> longer works.


The problem with ptrace_stop and do_signal_stop that requires dropping
siglock and grabbing tasklist_lock is that do_notify_parent_cldstop
needs tasklist_lock to keep parent and real_parent stable.

With just some very modest code changes it looks like we can use
a processes own siglock to keep parent and real_parent stable.  The
siglock is already acquired in all of those places it is just not held
over the changing parent and real_parent.

Then make a rule that a child's siglock must be grabbed before a parents
siglock and do_notify_parent_cldstop can be always be called under the
childs siglock.

This means ptrace_stop can be significantly simplified, and the
notifications can be moved far enough up that set_special_state
can be called after do_notify_parent_cldstop.  With the result
that there is simply no PREEMPT_RT issue to worry about and
wait_task_inactive can be used as is.

I remember Oleg suggesting a change something like this a long
time ago.


I need to handle the case where the parent and the child share
the same sighand but that is just remembering to handle it in
do_notify_parent_cldstop, as the handling is simply not taking
the lock twice.

I am going to play with that and see if I there are any gotcha's
I missed when looking through the code.

Eric