Messages in this thread | | | Date | Tue, 21 Jul 2009 21:17:20 +0300 | From | Izik Eidus <> | Subject | Re: [PATCH 06/10] ksm: identify PageKsm pages |
| |
Andrea Arcangeli wrote: > On Tue, Jul 21, 2009 at 01:55:13PM -0400, Rik van Riel wrote: > >> I guess that if they are to remain unswappable, they >> should go onto the unevictable list. >> > > They should indeed. Not urgent but it will optimize the vm (as in > virtual memory) cpu load a bit. > > >> Then again, I'm guessing this is all about to change >> in not too much time :) >> > > That's my point, current implementation of PageKsm don't seem to last > long, and if we keep logic the same it'll likely happen soon that > PageKsm != PageAnon on a Ksm page. So I'd rather keep it different > even now, given I doubt it's moving the needle anywhere in ksm code. > Hugh mentioned that he specially moved the ksm pages to be anonymous beacuse he felt it is more right... About PageExternal(): If I understand right, you both want to see the ksm pages swapped in the same way: try_to_unmap() will call the stable tree as rmap walker to know what ptes it should unpresent, but while probably Hugh wanted to allow such thing only for ksm (using PageKsm()) you probably talking about PageExternal() that will work with modified function pointers and will let any driver register for such usage...,
So if it is PageKsm() or PageExtrnal() probably related to whatever you allow such rmap walking just for ksm or for every driver.
| |