On Wed, May 16, 2018 at 11:38:16AM -0400, Tyler Baicar wrote:

I haven't seen a deadlock from that, but it looks possible. What if
the ghes_proc() call in ghes_probe() is moved before the second switch
statement? That way it is before the NMI/IRQ/poll is setup. At quick
glance I think that should avoid the deadlock and still provide the
functionality that call was added for. I can test that out if you all
agree.

Makes sense but please audit it properly before doing the change. That
code is full of landmines and could use a proper scrubbing first.

Thx.

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.