Re: [PATCH 2/2] PCI, ACPI, x86: update MMCFG information when hot-plugging PCI host bridges

[Jiang Liu <liuj97@gmail.com>]

On 04/10/2012 11:47 PM, Yinghai Lu wrote:
> On Tue, Apr 10, 2012 at 3:32 AM, Kenji Kaneshige
> <kaneshige.kenji@jp.fujitsu.com> wrote:
>> Additionally, I think there is a remaining issue even if we change
>> this check like below.
>>
>>
>>        if (!!(pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF))
>>                return 0;
>>
>> I think this check has an assumption that system has at least one
>> MCFG table entry and it has been initialized before
>> acpi_pci_root_add() is called. I think this doesn't work on the
>> system that doesn't have MCFG and all the pci root bridge have
>> _CBA (that is, all host bridges are hot-pluggable and BIOS is
>> implemented in the way PCI FW spec defines). As a result, MMCONFIG
>> would never be enabled on such systems. Could you double check this?
> 
> You are right.
> 
> We can just remove that checking.
> 
> But wonder if current x86_64 system support that.
> All peer root buses can be physically removed ?
It's theoretically possible that BIOS doesn't provide MMCFG or _CBA
for the legacy (non-removable bus 0) host bridge. 

> 
> Thanks
> 
> Yinghai