Re: [Qemu-devel] [edk2] [PATCH] exec: fix regression by making system-memory region UINT64_MAX size

[Laszlo Ersek]

On 11/07/13 22:21, Paolo Bonzini wrote:
> Il 07/11/2013 22:12, Laszlo Ersek ha scritto:
>> 0000000000000000-7ffffffffffffffe (prio 0, RW): system
>>   [...]
>>   0000000060000000-00000000ffffffff (prio 0, RW): alias pci-hole @pci 
>> 0000000060000000-00000000ffffffff
>>   [...]
>>   00000000ffe00000-00000000ffffffff (prio 0, R-): system.flash
>> [...]
> 
> Priorities are not "transitive" across aliases; once you use an alias to
> map a region, the alias's priority counts, not the target region's
> priority.  So the INT_MIN priority for pci-master-abort counts *within
> the alias*, but the choice between pci-hole and system.flash is only
> affected by the priorities of pci-hole and system.flash.

Right. It's also documented in docs/memory.txt -- Peter's recent
addition I think?

> You could give a smaller priority (-1 or INT_MIN) to pci-hole and just
> let it occupy the whole address space, from 0 to UINT64_MAX.  Or perhaps
> the pci-hole alias is too large and it should end before the system
> flash area.  Both solutions should work.

I did reorder pci-hole and system.flash, but rather than lowering
pci-hole, I raised system.flash. I have no preference.

Thanks
Laszlo