Subject: Re: metadata cache and memory fragmentation
To: matthew green <>
From: Manuel Bouyer <>
List: tech-kern
Date: 05/14/2004 11:21:06
On Fri, May 14, 2004 at 12:26:18PM +1000, matthew green wrote:
> while this may be more prevelent now, this has always been a problem.
> i think we need some way to steal pages for contig allocations.  find
> the largest segment that has owners you can steal from.  i'm not sure
> about the details.  possibly a simple version that could still fail
> would find r/o non-wired pages.  a more complete version that was
> able to clean dirty pages in the largest otherwise clean segment would
> be useful but it depends on the user being able to sleep...

Not sure. If there is some free memory, you can just copy the pages to
some other place and update the mappin(s), isn't it ?
It this case it shouldn't be needed to sleep ?

Manuel Bouyer <>
     NetBSD: 26 ans d'experience feront toujours la difference