Re: possible deadlock in lru_add_drain_all

[Peter Zijlstra]

On Tue, Oct 31, 2017 at 04:45:46PM +0100, Michal Hocko wrote:
> Anyway, this lock dependecy is subtle as hell and I am worried that we
> might have way too many of those. We have so many callers of
> get_online_cpus that dependecies like this are just waiting to blow up.

Yes, the filesystem locks inside hotplug thing is totally annoying. I've
got a few other splats that contain a similar theme and I've no real
clue what to do about.

See for instance this one:

  https://lkml.kernel.org/r/20171027151137.GC3165@xxxxxxxxxxxxxxxxxxxxxx

splice from devtmpfs is another common theme and links it do the
pipe->mutex, which then makes another other splice op invert against
that hotplug crap :/


I'm sure I've suggested simply creating possible_cpus devtmpfs files up
front to get around this... maybe we should just do that.