[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH RFC v2 1/7] xen/vNUMA: vNUMA support for PV guests.

Hi Elena,
Thank you for your great work first  :)

In your patch, guest can use a hypercall name: "XENMEM_get_vnuma_info" to get the structure below:
+struct vnuma_topology_info {
+    domid_t domid;
+    uint16_t nr_vnodes;
+    uint32_t _pad;
+    XEN_GUEST_HANDLE_64(uint) vdistance;
+    XEN_GUEST_HANDLE_64(uint) vcpu_to_vnode;
+    XEN_GUEST_HANDLE_64(vnuma_
memblk_t) vnuma_memblks;

I see that Xen have save a vnode_to_pnode[] array for mapping
However, this hypercall won't give this mapping to Guest VM.
Is that right?

On Tue, Sep 17, 2013 at 5:04 PM, Dario Faggioli <dario.faggioli@xxxxxxxxxx> wrote:
On mar, 2013-09-17 at 03:19 -0400, Elena Ufimtseva wrote:
> On Tue, Sep 17, 2013 at 3:11 AM, Dario Faggioli
> <dario.faggioli@xxxxxxxxxx> wrote:
> > On mar, 2013-09-17 at 08:05 +0100, Jan Beulich wrote:
> >> But - all this is only for the internal representations. Anything in
> >> the public interface should be wide enough to allow future
> >> extension.
> >>
> > And, in fact, 'node_to_node_distance' in xen/include/public/sysctl.h
> > (http://xenbits.xen.org/gitweb/?p=xen.git;a=blob;f=xen/include/public/sysctl.h)
> > is uint32.
> >
> Linux has u8 for distance. Ok, thank you for pointing that out.
EhEh... So, very hard to be consistent with  every actor in the
play! :-)

Anyway, I think the point here is, as Jan was saying, to distinguish
internal representation from interface. Within Xen, we should store
everything in the smallest and nicest possible way (which, BTW, is what
Linux does by using u8 for distances).

OTOH, when it comes to exported interfaces, we should be much more
cautious, since changing the way Xen stores distances internally is
trivial, changing the interface (either API or ABI) could be a

In this case, what we (Xen) tell Linux is the interface, it is up to him
(Linux) to convert that in a way that suits its own internals. In fact,
despite Linux being the only one OS using this interface for now, it's
not wise to design the interface itself specifically for Linux, since
many OSes may want to support it in the future.


<<This happens because I choose it to happen!>> (Raistlin Majere)
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

Yechen Li

Team of System Virtualization and Cloud Computing 
School of Electronic Engineering  and Computer Science
Peking University, China

Nothing is impossible because impossible itself  says: " I'm possible "
lccycc From PKU
Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.