Re: [Xen-devel] [RFC] Extending numbers of event channels

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Tue, 2012-12-04 at 08:05 +0000, Jan Beulich wrote:
> >>> On 03.12.12 at 19:09, Wei Liu <Wei.Liu2@xxxxxxxxxx> wrote:
> > On Mon, 2012-12-03 at 18:00 +0000, Jan Beulich wrote:
> >> >>> On 03.12.12 at 18:52, Wei Liu <Wei.Liu2@xxxxxxxxxx> wrote:
> >> > On Mon, 2012-12-03 at 17:35 +0000, Jan Beulich wrote:
> >> >> Doesn't the guest also need to set up space for the 2nd level?
> >> >> 
> >> > 
> >> > Yes. That will be embedded in percpu struct vcpu_info, which will be
> >> > also register via the same hypercall op.
> >> 
> >> "struct vcpu_info"? Same hypercall? Or are you mixing up types?
> >> 
> > 
> > What I meant was the second level will be embedded in struct vcpu_info,
> > and the 2nd level will be registered via some hypercall (not the struct
> > vcpu_info).
> 
> I would strongly recommend against embedding in
> struct vcpu_info, particularly in the context of intending to
> allow for having further levels in the future.
> 
> Plus I don't think you really can embed this - there's just not
> enough space left (I'm sure you're aware that you can't
> extend the structure in size), in fact there's no space left at
> all in the architecture independent part of the structure.
> 
> The only option you have is to declare the array you need to
> add to immediately follow the structure when having used
> the placement hypercall. That would probably be acceptable
> for the second from the top level, but you'd again run into
> (space) issues when wanting more than 3 levels (as you
> validly said, all but the leaf level ought to be per-vCPU).

The amount of space needed for levels 1..N-1 ought to be pretty
formulaic for any N, so we could choose to require that enough space be
available after vcpu_info for levels 1..N-1?

With larger N that might become multiple pages, is that a problem? It
might be preferable to have an explicit call which takes potentially
non-contiguous addresses.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel