Re: [PATCH 1/11] Add __GFP_MOVABLE flag and update callers

[Mel Gorman]

On Fri, 24 Nov 2006, Christoph Lameter wrote:

> On Thu, 23 Nov 2006, Linus Torvalds wrote:
>
> > And the assumption would be that if it's MOVABLE, then it's obviously a
> > USER allocation (it it can fail much more eagerly - that's really what the
> > whole USER bit ends up meaning internally).
>
> We can probably make several types of kernel allocations movable if there
> would be some benefit from it.

Page tables are the major type of allocation that comes to mind. From what 
I've seen, they are the most common long-lived unmovable and unreclaimable 
allocation.

> Mel already has a problem with mlocked user pages in the movable section.
> If this is fixed by using page migration to move the mlocked pages

That is the long-term plan.

> then we
> can likely make addititional classes kernel pages also movable and reduce
> the amount of memory that is unmovable. If we have more movable pages then
> the defrag can work more efficiently.

Indeed, although some sort of placement is still needed to keep these 
movable allocations together.

> Having most pages movable will also
> help to make memory unplug a reality.
>
> So please do not require movable pages to be user allocations.

That is not the intention. It just happens that allocations that are 
directly accessible by userspace are also the ones that are currently 
movable.

--
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/