Re: [Qemu-devel] [PATCH] exec: fix regression by making system-memory region UINT64_MAX size

[Marcel Apfelbaum <marcel.a@redhat.com>]

On Thu, 2013-11-07 at 21:51 +0000, Peter Maydell wrote:
> On 7 November 2013 21:38, Marcel Apfelbaum <marcel.a@redhat.com> wrote:
> > Thanks Paolo,
> > Let me just point out what I know (or I think I know):
> > 1. Not all architectures have the behavior: "Address space that is not RAM(and friends) is for sure PCI".
> >    Only x86 behaves like this (I think).
> 
> More specifically, the x86 pc behaves like this. Other
> x86 based systems could in theory behave differently
> (not that we actually model any, I think).
> 
> > That means that you cannot have a 64bit wide pci-hole
> >    with lower priority that catches all accesses that are not for RAM(and firends).
> 
> ...but this conclusion is wrong, because the pci-hole
> region is created by the pc model. So we should create
Yes... It think you are right. I was thinking for some
reason of master-abort region belonging to the the pci bus
that is not specific to x86 pc...

> it at the correct priority and size to give the behaviour
> relative to other devices in the pc model that is required
> (ie that the hardware has). This doesn't affect any other
> target architecture or board.
> 
> That said, I don't know enough about the PC to know what
> the exact details of the pci-hole are, so I'm not making a
> statement about what the correct model is. I'm just saying
> that what you do with the pci-hole and the container it lives
> in and the other devices in that container is not going to
> change the behaviour of any other target board.
> 
> > 2. If the above is right, and making pci-hole 64 bit wide is not an option,
> >    playing with pci-holes/other-region priorities it would be just wrong,
> >    it would be only to "fight" with the locality of the memory region's priority.
> 
> I have no idea what you mean by "fighting" here. MR priorities
By "fighting" I was referring exactly to the fact that because 
priorities are container specific we need to use them twice to
get the wanted behavior (master abort being with the lowest priority)
1. master-abort region priority for pci-address-space
2. pci-hole priority for system-address-space
But this is as designed...

Thanks,
Marcel

> apply only within a specific container region[*], to set which of
> that container's children appears 'above' another. They're
> totally local to the container (which in this case is part of the
> PC model, not the generic PCI code) and so the PC model
> can freely set them to whatever makes most sense.
> 
> [*] if you didn't already know this, see the recently committed
> updates to doc/memory.txt for a more detailed explanation.
> 
> -- PMM
>