Re: [PATCH v6 6/7] tools: add example application to initialize dom0less PV drivers

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Fri, 13 May 2022, Julien Grall wrote:
> On 13/05/2022 02:09, Stefano Stabellini wrote:
> > On Wed, 11 May 2022, Julien Grall wrote:
> > > > +    rc = xs_introduce_domain(xsh, info->domid,
> > > > +            (GUEST_MAGIC_BASE >> XC_PAGE_SHIFT) + XENSTORE_PFN_OFFSET,
> > > > +            xenstore_evtchn);
> > > 
> > > I might be missing something here. In an ealier version, I pointed out
> > > that
> > > xs_introduce_domain() would fail in the case the dom0less domain doesn't
> > > have
> > > "xen,enhanced".
> > > 
> > > AFAICT, you agreed that the (part?) of initialization should be skipped.
> > > But I
> > > don't see the change in the code. Regarding the placement, we could either
> > > fully skip init_domain() or just xs_introduce_domain(). The latter might
> > > be
> > > better so all the domains are listed using xenstore-ls.
> > > 
> > > So something like below should work:
> > > 
> > > if ( xenstore_evtchn )
> > > {
> > >      rc = xs_introduce...();
> > >      ...
> > > }
> > 
> > Yes, good point. xenstore_evtchn could be zero validly (first evtchn is
> > zero),
> 
> Event channel 0 is always reserved when initialization the event channel
> subsystem (evtchn_init()):
> 
> evtchn_from_port(d, 0)->state = ECS_RESERVED;
> 
> > so instead I'll check on xenstore_pfn to be zero (xenstore_pfn is
> > 0 for non-enhanced domUs.)
> 
> I spotted that difference but decided to not comment on it as Linux is already
> considering the values 0 and ~0 as invalid. However, I am not in favor on any
> code to rely on 0 means Xenstore will never be available while ~0 means that
> it might be available.
> 
> Anyway, as I wrote above, the event channel 0 is always reserved. So you can
> safely use this value to detect whether we allocated the event channel for
> Xenstore.

Thanks I didn't know that. In that case, using xenstore_evtchn is
better, I'll do that.