Re: [PATCH v4 3/8] x86/domain: allocate d->{iomem,irq}_caps before arch-specific initialization

[dmkhn@xxxxxxxxx]

On Mon, Aug 04, 2025 at 12:20:11PM +0200, Jan Beulich wrote:
> On 31.07.2025 21:21, dmkhn@xxxxxxxxx wrote:
> > From: Denis Mukhin <dmukhin@xxxxxxxx>
> >
> > Move IRQ/IOMEM rangesets allocation before arch_domain_create().
> >
> > That guarantees that arch-specific code could access those rangesets to
> > register traps for emulation.
> >
> > It is necessary for those emulators registering trap handlers and ensuring
> > that emulated IRQs are not shared with the physical IRQs.
> >
> > Move dom0_setup_permissions() call right after I/O rangesets are allocated.
> >
> > Move pvh_setup_mmcfg() inside dom0_setup_permissions() close to the place
> > where MMCFG ranges are initialized.
> 
> And this is just because you like doing so? I ask because first and foremost
> I can't make the connection between this and the purpose of this patch.

No, that is a bad leftover from the MMIO-based emulator.

> 
> > --- a/xen/arch/x86/dom0_build.c
> > +++ b/xen/arch/x86/dom0_build.c
> > @@ -471,6 +471,24 @@ static void __init process_dom0_ioports_disable(struct 
> > domain *dom0)
> >      }
> >  }
> >
> > +static void __hwdom_init setup_mmcfg(struct domain *d)
> 
> Why __hwdom_init when ...
> 
> > @@ -480,6 +498,14 @@ int __init dom0_setup_permissions(struct domain *d)
> >      if ( pv_shim )
> >          return 0;
> >
> > +    /*
> > +     * MMCFG initialization must be performed before setting domain
> > +     * permissions, as the MCFG areas must not be part of the domain IOMEM
> > +     * accessible regions.
> > +     */
> > +    if ( is_hvm_domain(d) )
> > +        setup_mmcfg(d);
> 
> ... the sole caller is __init?
> 
> Jan