]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/commit
cgroup: css_set_lock should nest inside tasklist_lock
authorTejun Heo <tj@kernel.org>
Wed, 23 May 2018 18:04:54 +0000 (11:04 -0700)
committerTejun Heo <tj@kernel.org>
Wed, 23 May 2018 18:04:54 +0000 (11:04 -0700)
commitd8742e22902186e30c346b1ba881cb52942ae3e4
treea4dd3d558a5eb43dd5ce547ec10a00e9c9bc993a
parentcc659e76f375ab158fb682c1e39b2a22bf9f7657
cgroup: css_set_lock should nest inside tasklist_lock

cgroup_enable_task_cg_lists() incorrectly nests non-irq-safe
tasklist_lock inside irq-safe css_set_lock triggering the following
lockdep warning.

  WARNING: possible irq lock inversion dependency detected
  4.17.0-rc1-00027-gb37d049 #6 Not tainted
  --------------------------------------------------------
  systemd/1 just changed the state of lock:
  00000000fe57773b (css_set_lock){..-.}, at: cgroup_free+0xf2/0x12a
  but this lock took another, SOFTIRQ-unsafe lock in the past:
   (tasklist_lock){.+.+}

  and interrupts could create inverse lock ordering between them.

  other info that might help us debug this:
   Possible interrupt unsafe locking scenario:

 CPU0                    CPU1
 ----                    ----
    lock(tasklist_lock);
 local_irq_disable();
 lock(css_set_lock);
 lock(tasklist_lock);
    <Interrupt>
      lock(css_set_lock);

   *** DEADLOCK ***

The condition is highly unlikely to actually happen especially given
that the path is executed only once per boot.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Boqun Feng <boqun.feng@gmail.com>
kernel/cgroup/cgroup.c