Re: [Xen-devel] [PATCH 15/22] vixen: pass through version hypercalls to parent Xen

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 07.01.18 at 16:40, <anthony@xxxxxxxxxxxxx> wrote:
> On Sun, Jan 7, 2018 at 12:31 AM, Roger Pau Monné <roger.pau@xxxxxxxxxx> wrote:
>> On Sat, Jan 06, 2018 at 02:54:30PM -0800, Anthony Liguori wrote:
>>> From: Anthony Liguori <aliguori@xxxxxxxxxx>
>>>
>>> This is necessary to trigger event channel upcalls but it is also
>>
>> I'm lost here, what does version have to do with upcalls?
> 
> In Linux, xen_force_evtchn_callback() does HYPERVISOR_xen_version(0,
> NULL).  This is done when IRQs are re-enabled after being disabled to
> trigger checking pending.
> 
> I'm not 100% confident that it's necessary to pass this all the way
> through to the parent Xen but it seemed like the right thing to do
> since we need the parent to update pending events in order for the
> events in Vixen to get updated.
> 
>>> useful to passthrough the full version information such that the
>>> guest believes it is running on the parent Xen.
>>
>> In any case, I think this is wrong. The interface the guest sees is
>> the interface from vixen, not the interface of the L0. Hence reporting
>> the L0 version is not appropriate.
> 
> I think it depends on what you want.  We were aiming for maximum
> compatibility and many users trigger behavior from Xen version for
> better or worse.

At the example of the plain version number, I think what is being
reported back to the guest may need to be the lower of shim and
actual hypervisor versions. XENVER_get_features may want ANDing
both values (perhaps with some customization in case there are
bits exclusively affected by either party). I didn't think through other
sub-ops yet.

Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel