Re: [Xen-devel] [PATCH for-next RFC 0/8] Port Xen to Hyper-V

[Wei Liu <wl@xxxxxxx>]

On Mon, Sep 23, 2019 at 12:11:26PM +0000, Paul Durrant wrote:
> > -----Original Message-----
> > From: Wei Liu <wl@xxxxxxx>
> > Sent: 23 September 2019 12:27
> > To: Paul Durrant <Paul.Durrant@xxxxxxxxxx>
> > Cc: 'Wei Liu' <wl@xxxxxxx>; Xen Development List 
> > <xen-devel@xxxxxxxxxxxxxxxxxxxx>; Wei Liu
> > <liuwe@xxxxxxxxxxxxx>; Andrew Cooper <Andrew.Cooper3@xxxxxxxxxx>; Michael 
> > Kelley
> > <mikelley@xxxxxxxxxxxxx>; Jan Beulich <jbeulich@xxxxxxxx>; Roger Pau Monne 
> > <roger.pau@xxxxxxxxxx>
> > Subject: Re: [Xen-devel] [PATCH for-next RFC 0/8] Port Xen to Hyper-V
> > 
> > On Mon, Sep 23, 2019 at 10:48:45AM +0000, Paul Durrant wrote:
> > > > -----Original Message-----
> > > > From: Xen-devel <xen-devel-bounces@xxxxxxxxxxxxxxxxxxxx> On Behalf Of 
> > > > Wei Liu
> > > > Sent: 23 September 2019 11:09
> > > > To: Xen Development List <xen-devel@xxxxxxxxxxxxxxxxxxxx>
> > > > Cc: Wei Liu <liuwe@xxxxxxxxxxxxx>; Wei Liu <wl@xxxxxxx>; Andrew Cooper
> > <Andrew.Cooper3@xxxxxxxxxx>;
> > > > Michael Kelley <mikelley@xxxxxxxxxxxxx>; Jan Beulich 
> > > > <jbeulich@xxxxxxxx>; Roger Pau Monne
> > > > <roger.pau@xxxxxxxxxx>
> > > > Subject: [Xen-devel] [PATCH for-next RFC 0/8] Port Xen to Hyper-V
> > > >
> > > > Hi all
> > > >
> > > > In case you're wondering, I can already run a fully fledged Xen system 
> > > > on
> > > > Hyper-V with emulated disk and network.
> > > >
> > > > This is the very first stage for porting Xen to run on Hyper-V with all 
> > > > the
> > > > goodies Hyper-V has to offer.  With this series, Xen can successfully 
> > > > detect
> > > > Hyper-V and prints out a message.  I would like to first get the 
> > > > directory
> > > > structure and kconfig options agreed upon.
> > > >
> > > > There are two major areas to be worked on:
> > > >   * Make Dom0 able to use Hyper-V's synthetic devices.
> > > >   * Make Xen use of the synthetic timer, reference TSC and 
> > > > enlightenment VMCS
> > > >     and other interfaces.
> > > >
> > > > They aren't trivial, and time can be scarce on my side, so I intend to 
> > > > post
> > > > patches piece meal when they are ready.
> > > >
> > > > Questions and comments are welcome.
> > >
> > > Yay! Glad you found the time to do this :-) Making use of the synic,
> > > stimer, will be useful. To get the benefit for nested guests though, I
> > > assume we'll need to exposed enlightened VMCS, EPT flush, etc.
> > > through?
> > 
> > Per TLFS, eVMCS should be used by L1 Xen.
> 
> Yes, I guess it only needs to be used by L1, but Windows is using an
> increasing number of VMs for various purposes so I think making it
> stack well would be useful.
> 
> > 
> > Not sure what you meant by EPT flush. If you mean things like
> > HvFlushGuestPhysicalAddressSpace, I think they should be used by Xen
> > only as well. I'm not sure whether they should be expose to L2 guests.
> > 
> 
> Yes, that's what I meant and, again, I was thinking about the nested
> Windows guest using VMs itself.

So this is basically nested on nested. I'm not sure how to make that
work yet.  I'm also not sure how well these interfaces stack. TLFS only
describes one level of nesting. I will need to ask if we can make nested
on nested work.

> 
> > There are hypercalls for virtual TLB management. We may choose to expose
> > them to L2 guests for better performance.
> > 
> 
> Do you mean 'Direct Virtual Flush'? The hypercalls in question are already 
> handled in viridian for the purposes of remote TLB flush, so I guess it would 
> be a question of wiring them through.

Yes. I meant that.

Wei.

> 
>   Paul
> 
> > Wei.

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