Gabriel Paubert writes:

> But I'm not sure that OOO cannot play tricks on you, what guarantees
> that the lwz is done after lwarx?

Nothing, but it doesn't matter, because we have the
mm->page_table_lock, and anything that is changing the top 32 bits, or
anything in the bottom 32 bits other than the _PAGE_HASHPTE bit,
must also take the mm->page_table_lock.  The low-level hash_page
routine can change the _PAGE_HASHPTE bit without having that lock,
which is why we need the atomic sequence.

Paul.

Reply via email to