Re: [PATCH v9 08/15] hugetlb: Use the has_hwpoisoned flag

[Matthew Wilcox <willy@infradead.org>]

On Thu, Sep 24, 2026 at 12:32:13PM +0200, David Hildenbrand (Arm) wrote:
> > Oh, no, it's awful and complicated.  We could strip out the two bugfixes
> > which are at the start of the series, but why would we want to do that?
> 
> Andrew usually wants us to separate bugfixes (especially stable ones) from
> other work. Unless we all agree that they are not urgent and can go in in
> the next cycle. I assume that's the case here.

I don't understand that preference, to be honest.  As long as they're
first in the patch series, they're easy to pick out and don't depend
on the other patches in the series.

> I am not convinced about the huge_poison thing though, that originally caught my
> attention.
> 
> So trying to understand the race we are trying to solve (that is independent of
> the original idea of the patch set), I realize that these races are only on some 
> slow paths we don't particularly care about for performance.

They are not hot paths, I agree, but they are user-triggerable paths.
So I'm not happy about the ability to spin up thousands of threads and
delay other accesses to hugetlb_lock indefinitely.