On 24.11.21 15:37, Vlastimil Babka wrote:

On 11/24/21 09:57, David Hildenbrand wrote:

quoted

On 24.11.21 00:59, Jason Gunthorpe wrote:

quoted

quoted

Similarly for io-uring we could be migrating pages to be pinned so that
the end up consolidated close together, and prevent pathologic
situations like in David's reproducer. 

It is an interesting idea to have GUP do some kind of THP preserving
migration.

Unfortunately it will only be a band aid AFAIU. I can rewrite my
reproducer fairly easily to pin the whole 2M range first, pin a second
time only a single page, and then unpin the 2M range, resulting in the
very same way to block THP. (I can block some THP less because I always
need the possibility to memlock 2M first, though).

Hm I see, then we could also condsider making it possible to migrate the
pinned pages - of course io-uring would have to be cooperative here,
similarly to anything that supports PageMovable.

I might be wrong but that would then essentially be an actual mlock+mmu
notifier mechanism, and no longer FOLL_LONGTERM. And the mlock could
actually be done by user space and would be optional.

I'd be very happy to see something like that instead ... but so far
people don't even agree that it's an issue worth fixing.

So ...

-- 
Thanks,

David / dhildenb