Re: [PATCH v2] PCI: vmd: Enable Hotplug based on BIOS setting on VMD rootports

[Bjorn Helgaas <helgaas@kernel.org>]

On Tue, Jan 16, 2024 at 01:37:32PM -0700, Nirmal Patel wrote:
> On Fri, 2024-01-12 at 16:55 -0600, Bjorn Helgaas wrote:
> ...

> > Maybe it will help if we can make the
> > situation more concrete.  I'm basing this on the logs at
> > https://bugzilla.kernel.org/show_bug.cgi?id=215027.  I assume the
> > 10000:e0:06.0 Root Port and the 10000:e1:00.0 NVMe device are both
> > passed through to the guest.  I'm sure I got lots wrong here, so
> > please correct me :)
> > 
> >   Host OS sees:
> > 
> >     PNP0A08 host bridge to 0000 [bus 00-ff]
> >       _OSC applies to domain 0000
> >       OS owns [PCIeHotplug SHPCHotplug PME PCIeCapability LTR] in
> > domain 0000
> >     vmd 0000:00:0e.0: PCI host bridge to domain 10000 [bus e0-ff]
> >       no _OSC applies in domain 10000;
> >       host OS owns all PCIe features in domain 10000
> >     pci 10000:e0:06.0: [8086:464d]             # VMD root port
> >     pci 10000:e0:06.0: PCI bridge to [bus e1]
> >     pci 10000:e0:06.0: SltCap: HotPlug+        # Hotplug Capable
> >     pci 10000:e1:00.0: [144d:a80a]             # nvme
> > 
> >   Guest OS sees:
> > 
> >     PNP0A03 host bridge to 0000 [bus 00-ff]
> >       _OSC applies to domain 0000
> >       platform owns [PCIeHotplug ...]          # _OSC doesn't grant
> > to OS
> >     pci 0000:e0:06.0: [8086:464d]              # VMD root port
> >     pci 0000:e0:06.0: PCI bridge to [bus e1]
> >     pci 0000:e0:06.0: SltCap: HotPlug+         # Hotplug Capable
> >     pci 0000:e1:00.0: [144d:a80a]             # nvme
> > 
> > So the guest OS sees that the VMD Root Port supports hotplug, but
> > it can't use it because it was not granted ownership of the
> > feature?
>
> You are correct about _OSC not granting access in Guest.

I was assuming the VMD Endpoint itself was not visible in the guest
and the VMD Root Ports appeared in domain 0000 described by the guest
PNP0A03.  The PNP0A03 _OSC would then apply to the VMD Root Ports.
But my assumption appears to be wrong:

> This is what I see on my setup.
> 
>   Host OS:
> 
>     ACPI: PCI Root Bridge [PC11] (domain 0000 [bus e2-fa])
>     acpi PNP0A08:0b: _OSC: OS supports [ExtendedConfig ASPM ClockPM Segments MSI EDR HPX-Type3]
>     acpi PNP0A08:0b: _OSC: platform does not support [SHPCHotplug AER DPC]
>     acpi PNP0A08:0b: _OSC: OS now controls [PCIeHotplug PME PCIeCapability LTR]
>     PCI host bridge to bus 0000:e2
> 
>     vmd 0000:e2:00.5: PCI host bridge to bus 10007:00
>     vmd 0000:e2:00.5: Bound to PCI domain 10007
> 
>   Guest OS:
> 
>     ACPI: PCI Root Bridge [PC0G] (domain 0000 [bus 03])
>     acpi PNP0A08:01: _OSC: OS supports [ExtendedConfig ASPM ClockPM Segments MSI EDR HPX-Type3]
>     acpi PNP0A08:01: _OSC: platform does not support [PCIeHotplug SHPCHotplug PME AER LTR DPC]
>     acpi PNP0A08:01: _OSC: OS now controls [PCIeCapability]
> 
>     vmd 0000:03:00.0: Bound to PCI domain 10000
> 
>     SltCap: PwrCtrl- MRL- AttnInd- PwrInd- HotPlug- Surprise-

Your example above suggests that the guest has a PNP0A08 device for
domain 0000, with an _OSC, the guest sees the VMD Endpoint at
0000:03:00.0, and the VMD Root Ports and devices below them are in
domain 10000.  Right?

If we decide the _OSC that covers the VMD Endpoint does *not* apply to
the domain below the VMD bridge, the guest has no _OSC for domain
10000, so the guest OS should default to owning all the PCIe features
in that domain.

Bjorn