On Thu, Aug 08, 2013 at 12:21:32PM +0200, Gerd Hoffmann wrote:
On 08/08/13 11:52, Michael S. Tsirkin wrote:
> On Thu, Aug 08, 2013 at 10:57:44AM +0200, Gerd Hoffmann wrote:
>> On 08/08/13 10:37, Michael S. Tsirkin wrote:
>>> On Thu, Aug 08, 2013 at 09:57:39AM +0200, Gerd Hoffmann wrote:
>>>> Also coreboot and seabios use different values for pmbase. coreboot on
>>>> q35 maps the pmbase below 0x1000. Which surely makes sense. When we
>>>> don't place chipset stuff at 0xb000 we can assign the
>>>> window to a pci bridge instead.
Re-reading this - if this has value, can't we generalize it
and make all firmware behave the same, getting values from QEMU?
>>> Yes, this might be useful. But I don't think it's required to
>>> use linker to patch ACPI tables for this - we can write ASL code to read
>>> the register back from device configuration, instead.
>> No, we can't, because the address is in the FADT.
> I see. Yes, PM base is there, this in fact makes it possible
> to patch it by linker in a sane way.
Exactly. Likewise the mmconf xbar config in the MCFG table.
> But to make addresses usable to devices they also need to be declared in
> the _CRS for the PCI root, correct? Which is code in DSDT.
Yes, the address ranges used for pci devices (aka 32bit + 64bit pci
window) need to be there. Well, placing in SSDT, then referencing from
DSDT works too, and this is what seabios does today to dynamically
adjust stuff. Fixing up the SSDT using the linker is probably easier as
we generate it anyway.
Yes but as I said, this makes things messy, since AML encoding for
numbers isn't fixed width.