Re: Paging/sharing in 4.2 (Was: Re: 4.2 TODO update)

[Olaf Hering <olaf@aepfle.de>]

On Mon, Mar 12, Ian Campbell wrote:

> On Mon, 2012-03-12 at 13:04 +0000, Olaf Hering wrote:
> > On Mon, Mar 12, Ian Campbell wrote:
> > 
> > > On Mon, 2012-03-12 at 12:11 +0000, Ian Campbell wrote:
> > > > hypervisor, nice to have:
> > > >       * solid implementation of sharing/paging/mem-events (using work
> > > >         queues) (Tim Deegan, Olaf Herring et al -- is this happening? is
> > > >         it DONE even?)
> > 
> > The last patch to use a waitqueue in __get_gfn_type_access() from Tim
> > works.  However, there are a few users who call __get_gfn_type_access
> > with the domain_lock held.
> 
> Thanks. 
> 
> >  This part needs to be addressed in some way.
> 
> This is happening?

To me its not clear what the domain_lock() protects:

Just a guest: XENMEM_remove_from_physmap should be ok with the gfn_lock
taken in guest_physmap_remove_page(), so the domain_lock() could be
removed there.

I'm not sure about the domain_lock in xenmem_add_to_physmap_once() and
emulate_privileged_op().


> > > > tools, nice to have:
> > > > [...]
> > > >       * Configure/control paging via xl/libxl (Olaf Herring, lots of
> > > >         discussion around interface)
> > > 
> > > Is this a reasonable summary of the state of the paging/sharing &
> > > friends world for 4.2? Is there anything missing?
> > 
> > There are indeed alot of ideas regarding the interface, and its
> > currently not clear to me if there is an agreement yet. Was there any
> > outcome where to proceed?
> 
> I've mostly swapped it out while I was away so I need to go back to the
> thread and figure out where we got to. I don't think we had discussed
> the libxl interface, only xl. I think the proposal Andres made sense in
> this respect when he described it to me (although I've not read the mail
> yet).

So is this something we should work out for the 4.2 release?

If I understand it right, there should be a .cfg option
mem_policy=<string>, where string could be "paging" or "balloon".
In case of balloon the target value would be whats in use now
(memory+maxmem). In case of paging these values could be reused in the
same way, just that maxmem+memory would not create a PoD guest.
If mem_policy= is not given it will default to balloon to retain current
behaviour.


Olaf