Re: [Qemu-devel] [RFC 3/3] acpi-build: allocate mcfg for multiple host bridges

[Marcel Apfelbaum <marcel.apfelbaum@gmail.com>]

On 05/23/2018 10:32 AM, Laszlo Ersek wrote:
> On 05/23/18 01:40, Michael S. Tsirkin wrote:
>> On Wed, May 23, 2018 at 12:42:09AM +0200, Laszlo Ersek wrote:
>>> Hold on,
>>>
>>> On 05/22/18 21:51, Laszlo Ersek wrote:
>>>
>>>> It had taken years until the edk2 core gained a universal
>>>> PciHostBridgeDxe driver with a well-defined platform customization
>>>> interface, and that interface doesn't support multiple domains /
>>>> segments.
>>> after doing a bit more research: I was wrong about this. What I
>>> remembered was not the current state. Edk2 does seem to support multiple
>>> domains, with different segment numbers, ECAM base addresses, and bus
>>> number ranges.

Good news!

>>>   If we figure out a placement strategy or an easy to
>>> consume representation of these data for the firmware, it might be
>>> possible for OVMF to hook them into the edk2 core (although not in the
>>> earliest firmware phases, such as SEC and PEI).

Can you please remind me how OVMF places the 64-bit PCI hotplug window?
We may do something similar.
Let me emphasize,  I am not implying you/anybody else should work on 
that :),
I just want to be on the same page if/when the time will come.
For the moment we are looking for a POC, nothing more.

>>> In retrospect, I'm honestly surprised that so much multi-segment support
>>> has been upstreamed to the edk2 core. Sorry about the FUD. (My general
>>> points remain valid, for the record... But perhaps they no longer matter
>>> for this discussion.)
>>>
>>> (I meant to send this message soon after
>>> <http://mid.mail-archive.com/fc603491-7c41-e862-a583-2bae6f165b5a@redhat.com>,
>>> but my internet connection had to die right then.)
>>>
>>> Thanks
>>> Laszlo
>> Is there support for any hardware which we could emulate?
> I don't see any actual hw support in the edk2 project, but I'll ask.

I think we may be able to succeed with "standard" APCI declarations of
the PCI segments + placing the extra MMCONFIG ranges before the 64-bit 
PCI hotplug area.

Thanks,
Marcel

> Thanks
> Laszlo