On Thu, Aug 28, 2014 at 10:33:02PM +0100, Peter Maydell wrote:
> On 28 August 2014 22:21, Michael Roth <mdr...@linux.vnet.ibm.com> wrote:
> > But I guess there's still a separate issue, where there's a high liklihood
> > that
> > a 0 address would conflict with some hard-wired IO address? Wouldn't this
> > be a
> > guest bug though?
Real hardware behaves in a specific way.
the problem is we don't always emulate it properly, and PCI
has some work arounds for that.
> Even if it's a guest bug, we should act like the hardware does
> if the guest does this. If that differs between PCI controllers
> then we need a flag so the host controller model can select
> the required behaviour. (The versatile PB PCI controller we
> model does have "address 0 is valid", and we'd need to have
> this working if we implemented DMA accesses properly.)
>
> -- PMM
Exactly.
Actually, I forgot that I might have already fixed it for PC:
83d08f2673504a299194dcac1657a13754b5932a
pc: map PCI address space as catchall region for not mapped addresses
But need to go back and re-check other systems.
--
MST
