Re: [PATCH v2 5/5] libxl, hvmloader: Don't relocate memory for MMIO hole

[George Dunlap <george.dunlap@eu.citrix.com>]

On 06/25/2013 10:56 AM, Ian Campbell wrote:
> On Thu, 2013-06-20 at 10:22 +0100, George Dunlap wrote:
>> Really, I'm of the opinion that if KVM is using SeaBIOS's pci
>> routines, we should just move do the same.  No sense in duplicating
>> the effort for something like this.
>
> Perhaps I'm misremembering and/or conflating different issues but I
> think I tried this when I initially ported SeaBIOS to Xen and it got
> complicated quickly.
>
> My memory is a bit vague but IIRC there were issues with things like the
> PCI IRQ routing and with the ACPI tables. In both cases these are
> handled in hvmloader because they are counterparts to the implementation
> of the underlying virtualised platform which is tied to Xen. Moving
> things into SeaBIOS would constrain our ability to change stuff going
> forward (e.g. SeaBIOSes decisions about interrupt routing doesn't not
> currently match Xen's implementation) and would lead to a tricky API
> boundary somewhere between hvmloader+xen and SeaBIOS(+-qemu)
>
> Maybe PCI BAR placement isn't inherently linked to all that though, you
> are welcome to try and split it out ;-)

Thanks for the heads up.  Yeah, I guess the interrupt routing stuff is 
all mixed in there together, isn't it... anyway I'll keep that in mind.

  -George