linux-kernel - [RFC] Deadlock via recursive wakeup via RCU with threadirqs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190626135447.y24mvfuid5fifwjc@linutronix.de>
Date:   Wed, 26 Jun 2019 15:54:47 +0200
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     rcu@...r.kernel.org, linux-kernel@...r.kernel.org
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>,
        "Paul E. McKenney" <paulmck@...ux.ibm.com>,
        Josh Triplett <josh@...htriplett.org>,
        Steven Rostedt <rostedt@...dmis.org>,
        Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
        Lai Jiangshan <jiangshanlai@...il.com>,
        Joel Fernandes <joel@...lfernandes.org>
Subject: [RFC] Deadlock via recursive wakeup via RCU with threadirqs

one of my boxes boots with "threadirqs" and since commit 05f415715ce45
("rcu: Speed up expedited GPs when interrupting RCU reader") I run
reliably into the following deadlock:

| ============================================
| WARNING: possible recursive locking detected
| 5.2.0-rc6 #279 Not tainted
| --------------------------------------------
| (cron)/2109 is trying to acquire lock:
| 0000000088464daa (&p->pi_lock){-.-.}, at: try_to_wake_up+0x37/0x700
|
| but task is already holding lock:
| 0000000088464daa (&p->pi_lock){-.-.}, at: try_to_wake_up+0x37/0x700
|
| other info that might help us debug this:
|  Possible unsafe locking scenario:
|
|        CPU0
|        ----
|   lock(&p->pi_lock);  
|   lock(&p->pi_lock);  
|
|  *** DEADLOCK ***
|
|  May be due to missing lock nesting notation
|
| 4 locks held by (cron)/2109:
|  #0: 00000000c0ae63d9 (&sb->s_type->i_mutex_key){++++}, at: iterate_dir+0x3d/0x170
|  #1: 0000000088464daa (&p->pi_lock){-.-.}, at: try_to_wake_up+0x37/0x700
|  #2: 00000000f62f14cf (&rq->lock){-.-.}, at: try_to_wake_up+0x209/0x700
|  #3: 000000000d32568e (rcu_read_lock){....}, at: cpuacct_charge+0x37/0x1e0
|
| stack backtrace:
| CPU: 3 PID: 2109 Comm: (cron) Not tainted 5.2.0-rc6 #279
| Call Trace:
|  <IRQ>
|  dump_stack+0x67/0x90 
|  __lock_acquire.cold.63+0x142/0x23a
|  lock_acquire+0x9b/0x1a0
|  ? try_to_wake_up+0x37/0x700
|  _raw_spin_lock_irqsave+0x33/0x50
|  ? try_to_wake_up+0x37/0x700
|  try_to_wake_up+0x37/0x700
wake up ksoftirqd

|  rcu_read_unlock_special+0x61/0xa0
|  __rcu_read_unlock+0x58/0x60
|  cpuacct_charge+0xeb/0x1e0
|  update_curr+0x15d/0x350
|  enqueue_entity+0x115/0x7e0
|  enqueue_task_fair+0x78/0x450
|  activate_task+0x41/0x90
|  ttwu_do_activate+0x49/0x80
|  try_to_wake_up+0x23f/0x700

wake up ksoftirqd

|  irq_exit+0xba/0xc0   
|  smp_apic_timer_interrupt+0xb2/0x2a0
|  apic_timer_interrupt+0xf/0x20
|  </IRQ>

based one the commit it seems the problem was always there but now the
mix of raise_softirq_irqoff() and set_tsk_need_resched() seems to hit
the window quite reliably. Replacing it with 

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index 1102765f91fd1..baab36f4d0f45 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -627,14 +627,7 @@ static void rcu_read_unlock_special(struct task_struct *t)
        if (preempt_bh_were_disabled || irqs_were_disabled) {
                WRITE_ONCE(t->rcu_read_unlock_special.b.exp_hint, false);
                /* Need to defer quiescent state until everything is enabled. */
-               if (irqs_were_disabled) {
-                       /* Enabling irqs does not reschedule, so... */
-                       raise_softirq_irqoff(RCU_SOFTIRQ);
-               } else {
-                       /* Enabling BH or preempt does reschedule, so... */
-                       set_tsk_need_resched(current);
-                       set_preempt_need_resched();
-               }
+               raise_softirq_irqoff(RCU_SOFTIRQ);
                local_irq_restore(flags);
                return;
        }

will make it go away.
Any suggestions?

Sebastian