[Cbe-oss-dev] [Lhms-devel] Re: [RFC] spufs: memory-add fix for CONFIG_NUMA

[Dave Hansen]

On Wed, 2006-04-12 at 00:25 +0200, Arnd Bergmann wrote:
> Am Tuesday 11 April 2006 23:58 schrieb Dave Hansen:
> > How about we poison the unused mem_map[] pages in  the areas of the
> > section which aren't full?
> >
> > That way, we can keep track in case any goofballs try to add the same
> > memory a couple of times, or try to remove memory that wasn't there.
> 
> Yes, that would make sense. I was under the impression that memory-remove
> is far from the state where it gets merged to the mainline kernel so we
> would not have to worry about that in the first place. What is the state
> of that now?

Not far enough away that we should make it harder now. :)

I'm just using the remove case because it is obvious.  The same issues
exist for trying to add the same memory twice.  If you think you've
added a particular physical address range, but it was really previously
present, you may accidentally do some nasty things.

I think the important part is that one and only one caller to
add_memory/pages() should succeed for any given address range.

-- Dave