Re: [PATCH] sched/schedutil: Fix deadlock between cpuset and cpu hotplug when using schedutil

[Waiman Long]

On 7/11/22 17:11, Steven Rostedt wrote:
> On Mon, 11 Jul 2022 10:58:28 -1000
> Tejun Heo <tj@xxxxxxxxxx> wrote:
>
> > I don't think lockdep would be able to track CPU1 -> CPU2 dependency here
> > unfortunately.
> >
> > > AFAIU:
> > >
> > >
> > > CPU0                                     CPU1                                   CPU2
> > >
> > > // attach task to a different
> > > // cpuset cgroup via sysfs
> > > __acquire(cgroup_threadgroup_rwsem)
> > >
> > >                                           // pring up CPU2 online
> > >                                           __acquire(cpu_hotplug_lock)
> > >                                           // wait for CPU2 to come online
> Should there be some annotation here that tells lockdep that CPU1 is now
> blocked on CPU2?
>
> Then this case would be caught by lockdep.
>
> -- Steve
>
>
> > >                                                                                  // bringup cpu online
> > >                                                                                  // call cpufreq_online() which tries to create sugov kthread
> > > __acquire(cpu_hotplug_lock)                                                     copy_process()
> > >                                                                                     cgroup_can_fork()
> > >                                                                                        cgroup_css_set_fork()
> > >                                                                                        __acquire(cgroup_threadgroup_rwsem)
> > > // blocks forever                        // blocks forever                            // blocks forever

Actually, the dependency can probably be coded by calling 
lockdep_acquire_cpus_lock() in cpu2 (task 2) before acquiring 
cgroup_threadgroup_rwsem to indicate the dependency between CPU1 and 
CPU2 (task 1 and task 2). lockdep_release_cpus_lock() can then called 
after release the rwsem.

Cheers,
Longman