Re: [PATCH splitout] mm: memory-failure: serialize TestSetPageHWPoison with zone->lock

[David Hildenbrand (Arm)]

On 6/11/26 09:36, Miaohe Lin wrote:
> On 2026/6/11 13:43, Michael S. Tsirkin wrote:
>> On Thu, Jun 11, 2026 at 11:35:36AM +0800, Miaohe Lin wrote:
>>>
>>> Do you mean repeating SetPageHWPoison on every branch?
>>
>> Right.
>>
>>> Is it possible
>>> to make __free_pages_prepare changes page->flags atomically or this race
>>> is specified to memory_failure?
>>>
>>> Thanks.
>>> .
>>
>>
>> Adding an atomic op on every fast path page allocation is, I am
>> guessing, going to slow down Linux measureably.
>>
>> Doing it for the benefit of memory_failure, which is the slowest of
>> slow paths, seems unpalatable, to me.
> 
> Agree, it's not worth to do so.
> 
>>
>> Neither am I sure it's the only racy place -
>> grep for __SetPage and __ClearPage - all these have the same issue, I
>> suspect.
>>
>> At the same time, I'm not an mm maintainer. If you disagree, try to
>> upstream a change converting all non atomics in mm to atomics, and see
>> what others say.
> 
> Since memory_failure might be the only place, this change would be unacceptable.
> We should come up with a better solution. Maybe we can try repeating SetPageHWPoison
> and ClearPageHWPoison at a first attempt though it looks somewhat weird to me and makes
> code more complicated.

And I am fairly sure we could still have some remaining races ... it's shaky.

Hm ...

-- 
Cheers,

David