Re: [Qemu-devel] [PATCH] hw/arm/virt: Allow zero address for PCI IO space

[Michael S. Tsirkin]

On Tue, Oct 13, 2015 at 02:47:33PM +0200, Laurent Vivier wrote:
> 
> 
> On 13/10/2015 14:33, Peter Maydell wrote:
> > On 13 October 2015 at 13:48, Alexander Gordeev <address@hidden> wrote:
> >> On Tue, Oct 13, 2015 at 09:16:34AM +0100, Peter Maydell wrote:
> >>> In any case, setting pci_allow_0_address is the right thing,
> >>> so we can just change the commit message in this patch.
> >>
> >> I will post v2 with an updated changelog then.
> >>
> >>> Incidentally, why is this a property on the machine
> >>> and not on the PCI controller device?
> >>
> >> I am CC-ing Laurent Vivier who introduced the flag.
> >>
> >> But IMO it *is* a machine property, not PCI controller's
> >> one, unless I am missing something.
> > 
> > I think "does this PCI controller handle BARs with
> > zero addresses, or does it treat them as if the BAR
> > was unmapped" is definitely a controller property...
> > you might have in theory a machine with two PCI
> > controllers, one of which could deal with zero-addresses
> > and one of which could not.
> 
> MST asked for a global flag:
> 
> https://lists.gnu.org/archive/html/qemu-ppc/2015-07/msg00364.html
> 
> But perhaps the machine is not the good place for this global flag ?
> 
> CC: Michael for that.

Whether controllers treat zero specially should be expressed
using priorities in memory APIs.
This is just about buggy machine types that do
not specify priority for overlapping regions correctly.

As a temporary hacky work-around, a global seems sufficient.

> > thanks
> > -- PMM
> >