Re: [PATCH] mm/mmu_notifier: Remove non_block_start/end() from notifier invocation

[Jason Gunthorpe <jgg@ziepe.ca>]

On Thu, Aug 13, 2026 at 06:59:36AM -0700, Sean Christopherson wrote:

> > On the other hand, they are basically *all* false positives from the
> > point of view of what the check was actually trying to do. And in KVM.
> 
> FWIW, all of the KVM false positives are due to PREEMPT_RT turning mmu_lock into
> a sleepable lock on x86.  So killing off the restriction for PREEMPT_RT would
> "fix" the KVM issues.

Yes..

Still, if you want to get rid of the non-blocking feature then it
should be replaced with something else that still achieves the same
end of preventing recursion into the OOM flow. Maybe some smart
lockdep scheme or something.

The non-blocking thing is pretty nasty, and it would be nice to
upgrade the drivers using sleeping locks without MM entanglements to
be able to work with the OOM killer, but I don't think anyone should
attempt that without some robust scheme to detect the invalid cases.

Jason