Re: [Qemu-devel] [Xen-devel] [PATCH 3/4] xen: Import Xen public headers used by xen-hvm.c

[Paul Durrant]

> -----Original Message-----
> From: Anthony PERARD [mailto:address@hidden]
> Sent: 17 June 2019 18:19
> To: Paul Durrant <address@hidden>
> Cc: address@hidden; Stefano Stabellini <address@hidden>; address@hidden
> Subject: Re: [Xen-devel] [PATCH 3/4] xen: Import Xen public headers used by 
> xen-hvm.c
> 
> On Mon, Jun 17, 2019 at 05:45:44PM +0100, Anthony PERARD wrote:
> > On Mon, Jun 17, 2019 at 05:15:51PM +0100, Paul Durrant wrote:
> > > AFAICT the only place (apart from legacy code in xen_common.h) that
> > > params.h is necessary is in xen_suspend_notifier(). I wonder whether
> > > that would be better moved into xen_common.h too (since it's just a
> > > wrapper round xc_set_hvm_param() and then the inclusion of params.h
> > > can be moved there as well.
> >
> > :(, I didn't realised that xen_common.h needed params.h too. That patch
> > series might not build on older version of Xen. I'll move the inclusion
> > of params.h to xen_common.h, and fix params.h to have all the needed
> > #defines.
> 
> Actually, it works fine without including hvm/params.h in xen_common.h
> because xen_common.h includes xenctrl.h which includes hvm/params.h.
> 
> So I think instead I'll remove all includes of params.h, because
> HVM_PARAM_* are parameters for xc_hvm_param_{get,set}, which is in a
> library and not an interface with a guest. Including xenctrl.h should be
> enough.

The guest/tools separation is messy. ioreq.h is also an interface with Xen and 
not with a guest, so really I think it should be treated the same way as 
params.h. e820.h is referring to guest memory layout, which IMO shouldn't 
really be hardcoded at all.

  Paul

> 
> --
> Anthony PERARD