Re: [PATCH] xen/tmem: cleanup

["Jan Beulich" <JBeulich@suse.com>]

>>> On 06.02.12 at 18:02, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> On Mon, Feb 06, 2012 at 04:49:42PM +0000, Jan Beulich wrote:
>> >>> On 06.02.12 at 17:37, Dan Magenheimer <dan.magenheimer@oracle.com> wrote:
>> >>  From: Jan Beulich [mailto:JBeulich@suse.com]
>> >> > tmem_enabled is used in xen/drivers/xen-selfballoon.c
>> >> 
>> >> Which can't be built as a module, and hence the symbol doesn't need
>> >> exporting. This patch (of course, I'm tempted to say) survived build
>> >> testing.
>> > 
>> > Yes, correct.  BUT... I think only the reason xen-selfballoon.c
>> > can't be built as a module is because the MM variable vm_committed_as
>> > (or an access function) is not exported.  Ideally xen-selfballoon.c
>> > probably should be a module but putting a core MM change in
>> > the critical path of a Xen-only-related enhancement seemed
>> > a recipe for sure failure.
>> 
>> No, this isn't the main reason (as you say further down, this could
>> easily be adjusted for) afaict. The main reason is that
>> register_xen_selfballooning() is being called from non-modular
>> code (xen-balloon.c), which could be made a module too, but in
>> turn has at least one reference that wouldn't be nice to become
>> an export (balloon_set_new_target()).
> 
> It would be nice if all of it could become modules. That way HVM
> device driver domains could load the whole thing without having much
> built-in code in the kernel.
> 
> Is it possible to do that?

As outlined above, it is possible, but I'm not certain it is a good idea
to have balloon_set_new_target() exported. If you think that's
acceptable, I can certainly put together a patch doing that (on top
of the one here, and probably not immediately).

Jan