Re: reply: [RFC] pin_user_pages_fast failure count increased

[John Hubbard <jhubbard@nvidia.com>]

On 5/22/25 8:07 AM, David Hildenbrand wrote:
...
>>>> Thank you for your comment and patch.
>>>> By the way, what if there are any_unpinnable pages and also pages in the movable_folio_list,
>>>> but migrate_longterm_unpinnable_folios failed to migrate and return other erros instead of -EAGAIN?
>>>> In that case, I think the CMA or other unpinnable pages would be pinned.
>>>
>>> Oh, I think what we have to do is call pofs_unpin(pofs)() in case we
>>> return with -EAGAIN early. That's what
>>> migrate_longterm_unpinnable_folios() would do.
>>
>> I did not understand. Do you mean we need to call pofs_unpin in case of any_unpinnable?
> 
> At least that's what we did before 1aaf8c122 when calling 
> collect_longterm_unpinnable_folios() I think.
> 

Not from what I see in the diffs for that commit, no.

>> I think your following code seems to be good to me.
>>    return any_unpinnable ? -EAGAIN : 0;
> 
> I think we have to call pofs_unpin() here.

That seems correct, although I'd have to look a little deeper to 
understand why 1aaf8c122, and not some other commit, revealed
the problem


thanks,
-- 
John Hubbard

> 
> Assuming we always call migrate_longterm_unpinnable_folios() with a list 
> now, we could drop the list_empty() check in there.
>