Re: pvops dom0 work roadmap

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Thu, Aug 20, 2009 at 05:36:01PM +0100, Gianluca Guida wrote:
> Konrad Rzeszutek Wilk writes:
>  > On Thu, Aug 20, 2009 at 04:49:08PM +0100, Gianluca Guida wrote:
>  > > Konrad Rzeszutek Wilk writes:
>  > >  > I will retry this, but in the meantime I am tracking balloon accessing non-existent pages:
>  > > 
>  > > Hm, what tree of dom0 kernel are you using? Is commit
>  > 
>  > [konrad@phenom linux]$ git checkout origin/rebase/master -b boo
>  > Branch boo set up to track remote branch refs/remotes/origin/rebase/master.
>  > Switched to a new branch 'boo'
>  > [konrad@phenom linux]$ git show  7394b25905929a92e51771b832ee67b70ccebcf0
>  > fatal: bad object 7394b25905929a92e51771b832ee67b70ccebcf0
> 
> Ok, you don't have the revert, do you have
> a472909fe0e3b7bfa34f7c9dd748121b0ede0ee6 in your tree?

Yes.

> 
>  > 
>  > > 7394b25905929a92e51771b832ee67b70ccebcf0 (Revert "Xen: Allow guest to
>  > > boot ballooned.") there?
>  > 
>  > What branch is it in?
> 
> rebase/master, checked in last week.

Hmmm. Let me checkout a new tree and see why I am behind.

> 
>  > 
>  > > 
>  > > The point is that xen_memory_setup should truncate the ram visible in
>  > > the e820 map to xen_start_info->nr_pages, so that the balloon driver
>  > > would never try to add pages on boot.
>  > 
>  > Sure. This problem is for pages above ->nr_pages.
> 
> max_pfn is calculated by the e820 map, so there would be nothing above
> nr_pages.

Duh! Yes

> 
> I was assuming you had an up-to-date tree, where there's no "Allow
> guest to boot ballooned" patch. When that patch is present no
> truncation happens, and you get hit by that bug, which your patch
> fixes.
> 
> Cheers,
> Gianluca