Re: [PATCH v5 10/11] arm/libxl: Emulated PCI device tree node in libxl

[Ian Jackson <iwj@xxxxxxxxxxxxxx>]

Julien Grall writes ("Re: [PATCH v5 10/11] arm/libxl: Emulated PCI device tree 
node in libxl"):
> On 06/10/2021 19:40, Rahul Singh wrote:
> > diff --git a/tools/libs/light/libxl_types.idl 
> > b/tools/libs/light/libxl_types.idl
> > index 3f9fff653a..78b1ddf0b8 100644
> > --- a/tools/libs/light/libxl_types.idl
> > +++ b/tools/libs/light/libxl_types.idl
> > @@ -644,6 +644,7 @@ libxl_domain_build_info = Struct("domain_build_info",[
> >   
> >       ("arch_arm", Struct(None, [("gic_version", libxl_gic_version),
> >                                  ("vuart", libxl_vuart_type),
> > +                               ("vpci", libxl_defbool),
> 
> I have posted some comments regarding the field in v4. To summarize, 
> AFAICT, this option is meant to be only set by libxl but you still let 
> the toolstack (e.g. xl, libvirt) to set it.
> 
> If you still want to expose to the toolstack, then I think the option 
> should be outside of arch_arm. Otherwise, this should be moved in an 
> internal structure (Ian, do you have any suggestion?).

If it should be in an internal structure, probably the libxl create
context.

But I'm not convinced yet.  In particular, if enabling VPCI is
necessary on ARM for hotplugged PCI devices[1], then there has to be
a way for the admin to say "while this domain may not have any PCI
devices right now, I may wish to hotplug some".  That's what the
"passthrough=" option is for.

See my other mail.

[1] I think this is all true even if PCI hotplug for ARM is not
currently implemented.

Ian.