Re: mm: deadlock between get_online_cpus/pcpu_alloc

[Thomas Gleixner]

On Mon, 6 Feb 2017, Dmitry Vyukov wrote:
> On Mon, Jan 30, 2017 at 4:48 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> Unfortunately it does not seem to help.
> Fuzzer now runs on 510948533b059f4f5033464f9f4a0c32d4ab0c08 of
> mmotm/auto-latest
> (git://git.kernel.org/pub/scm/linux/kernel/git/mhocko/mm.git):
> 
> commit 510948533b059f4f5033464f9f4a0c32d4ab0c08
> Date:   Thu Feb 2 10:08:47 2017 +0100
>     mmotm: userfaultfd-non-cooperative-add-event-for-memory-unmaps-fix
> 
> The commit you referenced is already there:
> 
> commit 806b158031ca0b4714e775898396529a758ebc2c
> Date:   Thu Feb 2 08:53:16 2017 +0100
>     mm, page_alloc: use static global work_struct for draining per-cpu pages

<SNIP>

> Chain exists of:
>  Possible unsafe locking scenario:
> 
>        CPU0                    CPU1
>        ----                    ----
>   lock(pcpu_alloc_mutex);
>                                lock(cpu_hotplug.lock);
>                                lock(pcpu_alloc_mutex);
>   lock(cpu_hotplug.dep_map);

And that's exactly what happens:

				cpu_up()
    alloc_percpu()		 lock(hotplug.lock)
    lock(&pcpu_alloc_mutex)
    ..				   alloc_percpu()
      drain_all_pages()		     lock(&pcpu_alloc_mutex)
        get_online_cpus()
          lock(hotplug.lock)

Classic deadlock, i.e. you _cannot_ call get_online_cpus() while holding
pcpu_alloc_mutex.

Alternatively you can forbid to do per cpu alloc/free while holding
hotplug.lock. I doubt that this will make people happy :)

Thanks,

	tglx