On Sun, Dec 19, 2021 at 01:53:36PM -0800, Linus Torvalds wrote:
On Sun, Dec 19, 2021 at 1:48 PM Matthew Wilcox willy@infradead.org wrote:
Yes, agreed, I was thinking that we could use "not mapped at all" as an optimisation to avoid doing rmap walks. eg __unmap_and_move().
So the thing is, it's a very dodgy optimization for a rather simple reason: what if somebody pages the page in?
So even "not mapped at all" is questionable.
You have to check that it's also not a swapcache page, and hold the page lock for that check, at the very least.
And by then, you're really in a very unusual situation - and my gut feel says not one worth optimizing for (because anon pages are _usually_ mapped at least once).
I'd like to get rid of ->mapcount for file pages too. And those are definitely never mapped in the majority of cases.