On Mon, 2026-11-16 at 21:12 +0100, David Hildenbrand (Arm) wrote:
> On 11/16/26 15:53, Thomas Hellström wrote:
> > On Fri, 2026-11-13 at 22:19 +0100, David Hildenbrand (Arm) wrote:
> > > On 11/13/26 21:56, Thomas Hellström wrote:
> > > > 
> > > > Wouldn't the net effect be that content stays put instead of
> > > > being
> > > > migrated? In both cases a later CPU access will not trigger
> > > > NUMA
> > > > hinting except that the proposed change doesn't send content to
> > > > an
> > > > irrelevant node?
> > > 
> > > My point is that you are essentially disabling NUMA hinting
> > > because
> > > the CPU will
> > > no longer trigger NUMA hinting faults.
> > 
> > Yes, I got that part, but that's is only for a single scan, right?
> 
> Well, one NUMA-hinting protects again, you would get an MMU notifier,
> and fault
> again here?
> 
> It all gets pretty unpredictable for these PTEs, no?

True, it does. But again, my point is that exact same thing happens if
hmm_range_fault() is allowed to take the fault and migrate? Which
happens here.

> 
> 
> > > But then, just disable it entirely. Maybe John knows what to do.
> > 
> > We can't do that globally, really, for these type of apps. But one
> > remedy might be recommending the app to mbind() with a no-
> > migration-on-
> > fault policy.
> 
> Yeah, something like this. I mean, it's hard to see how NUMA
> balancing makes any
> sense here. You're just trying to fight the mechanism because it ends
> up causing
> more harm then good.
> 
> > 
> > Also the hmm O(n²) complexity in this situation isn't really ideal.
> > We
> > might need to look at that as well.
> I didn't quite get that, can you elaborate?

My understanding is that due to the migration invalidations
hmm_range_fault() is restarted from the beginning roughly on each PMD,
leading to O(n_pages²) complexity, although with a small constant.

Thanks,
Thomas


Reply via email to