[PATCHv2 1/3] pci: added pcie_get_speed_cap_mask function

[Bjorn Helgaas]

On Wed, Mar 27, 2013 at 9:25 AM, Benjamin Herrenschmidt
<benh at kernel.crashing.org> wrote:
> On Tue, 2013-03-26 at 12:39 -0600, Bjorn Helgaas wrote:
>> But we also know pdev is a PCIe device, and I think a PCIe device on a
>> root bus must be a "Root Complex Integrated Endpoint" (PCIe spec sec
>> 1.3.2.3).  Such a device does not have a link at all, so there's no
>> point in fiddling with its link speed.
>
> This is where our IBM hypervisor makes things murky. It doesn't expose
> the PCIe parents (basically somewhat makes PCIe look like PCI except we
> still have the PCIe caps on the child devices, just no access to the
> parent device).

Interesting.  I wonder if we'll trip over this anywhere else, e.g.,
ASPM or other link-related things.  I guess we'll just have to see if
anything else breaks.

> It's garbage but can't be fixed (would break AIX :-)
>
> However we might be able to populate the bus->max_bus_speed from some
> architecture specific quirk and have radeon use that.

That sounds like a good solution to me.  It seems like it's really an
arch-specific deviation from the spec, so it'd be nice to have an
arch-specific solution for it.

Bjorn