Re: [RFC] Remove unswappable anonymous pages off the LRU

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 16 Feb 2007 13:57:14 +0900 KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Thu, 15 Feb 2007 20:15:46 -0800 (PST)
> Christoph Lameter <clameter@sgi.com> wrote:
> 
> > On Thu, 15 Feb 2007, Rik van Riel wrote:
> > 
> > > Christoph Lameter wrote:
> > > 
> > > > I tinkered with some similar radical ideas lately. Maybe a bit vector
> > > > could be used instead? For 1G of memory we would need 
> > > > 2^(30 - PAGE_SHIFT / 8 = 2^(30-12-3) = 2^15 = 32k bytes of a bitmap.
> > > > 
> > > > Seems to be reasonable?
> > > 
> > > At that point, wouldn't it be easier to simply increase
> > > the size of struct page?  I don't think they're power of
> > > two sized anyway, at least on 64 bit architectures.
> > 
> > On 64 bit platforms we can add one unsigned long to get from 56 to 64 
> > bytes.
> > 
> 
> I sometimes dreams 
> ==
> struct page {
> 	...
> 	struct zone	*zone;
> 	...
> };
> #define page_zone(page)		(page)->zone
> ==
> but never tried ;)

hm.  We can calculate page_zone(page) from the pfn.  And I suspect we can
do that locklessly too.  I bet a nice tight implementation of that would be
efficient enough and it'll reclaim heaps of flags.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>