On Thu, 2016-09-29 at 22:55 +1000, Nicholas Piggin wrote:

PG_swapcache - can this be replaced with ane of the private bits, I
wonder?

Perhaps. page->mapping needs to be free to point
at the anon_vma, but from the mapping pointer we
can see that the page is swap backed.

Is there any use for page->private for swap
backed pages that is not the page cache index?

If so, (PageAnon(page) && page->private)
might work as a replacement for PG_swapcache.

That might catch some false positives with
the special swap types used for migration, but
maybe we do not need to care about those (much),
or we can filter them out with a more in-depth
check?

-- 
All rights reversed