Re: [patch] 'sticky pages' support in the VM, futex-2.5.38-C5

[Linus Torvalds]

On Thu, 26 Sep 2002, Linus Torvalds wrote:
> 
> and then when we pin a page, we do
> 
> 	/* This is part of the 
> 	struct page_change_struct pinned_data;

That "This is part of the " comment should continue with "struct futex_q", 
but I went off and was supposed to check what the futex data structure was 
called, and forgot about updating it.

Anyway, the point being that this needs no new allocations in _any_ path,
and only extends a structure that we already need for the slow case for
futexes. It does imply a new lock, though, and the COW path would have to
check that hash (which should scale pretty well, since we only have
entries here when somebody is blocked on a futex).

Oh, and we need a new hash table, since the native futex hash can't just
be re-used due to different indexing - the futex hash is based on physical
page and offset, while the page_change_struct hash is based on virtual
address and the mm.

(I initially thought we could make the page_change_struct hash be based on 
physical page, but there can be multiple instances of the same physical 
page being mapped into the same VM, so that wouldn't be a good thing - 
we'd get callbacks for the wrong virtual address being COW'ed).

		Linus