Re: [PATCH] mm/oom_kill: Ensure MMU notifier range_end() is paired with range_start()

[Michal Hocko]

On Wed 10-03-21 20:28:07, Jason Gunthorpe wrote:
> On Wed, Mar 10, 2021 at 01:31:17PM -0800, Sean Christopherson wrote:
> > Invoke the MMU notifier's .invalidate_range_end() callbacks even if one
> > of the .invalidate_range_start() callbacks failed.  If there are multiple
> > notifiers, the notifier that did not fail may have performed actions in
> > its ...start() that it expects to unwind via ...end().  Per the
> > mmu_notifier_ops documentation, ...start() and ...end() must be paired.
> 
> No this is not OK, if invalidate_start returns EBUSY invalidate_end
> should *not* be called.

Yes, this is what I remember when introducing nonblock interface. So I
agree with Jason this patch is not correct. The interface is subtle but
I remember we couldn't come up with something more robust and still
memory with notifiers to be reapable.
-- 
Michal Hocko
SUSE Labs