[libvirt] [PATCH 12/12] food for thought

[Eric Blake]

On 09/29/2010 06:09 PM, Eric Blake wrote:
> Here's where I ran out of time for the day.  I'm much less familiar
> with xen than with qemu, so I have no idea how to tell if xen's
> documented domain/vcpu_avail (which is what we want for current vcpus)
> is usable in contrast to domain/vcpus (the maximum amount).  For that
> matter, I'm not even sure if modifying the Sxpr parsing/generating
> code is enough to make xen use the new attribute, or what else might
> be involved.  Hints on what I need to do from here are greatly
> appreciated.
>
> ---
>   src/xen/xend_internal.c   |    6 ++++++
>   src/xen/xm_internal.c     |    6 ++++++
>   src/xenapi/xenapi_utils.c |    2 +-
>   3 files changed, 13 insertions(+), 1 deletions(-)

Xen is more complex than I first thought.

I'm debating whether to support xendConfigVersion <= 2 (xm_internal.h) 
or just xendConfigVersion >= 3 (xend_internal.h).

Additional things I've discovered:

http://lists.xensource.com/archives/html/xen-devel/2009-11/msg01061.html

documents that the config file (xm_internal.h) has two parameters, vcpus 
is the max, and vcpu_avail is a bitmask of which vcpus to enable at 
boot.  I'm assuming that xen cannot support more than 32 vcpus?

http://www.cl.cam.ac.uk/research/srg/netos/xen/readmes/interface/interface.html

documents that each /domain sexpr contains vcpus (maximum), online_vcpus 
(current), and vcpu_avail (bitmask of online vcpus).

The xm vcpu-set command can only change the current vcpus of a running 
domain, so I'm not sure how that interacts with persistent storage.  I 
can't figure out whether I have to modify xm_internal.h to set the 
vcpu_avail bitmask to get this to happen, or whether changing the 
/domain sexpr also affects the persistent configuration.

-- 
Eric Blake   eblake at redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org