[libvirt] [PATCH] qemu: Don't special case mdevs when assigning PCI addresses

[Laine Stump]

On 05/26/2017 06:51 AM, Andrea Bolognani wrote:
> We can treat mdevs the same as all other PCI hostdevs
> and figure out whether they are PCI Express or legacy PCI
> by checking the size of their config space.
> 
> Signed-off-by: Andrea Bolognani <abologna at redhat.com>
> ---
>  src/qemu/qemu_domain_address.c | 3 ---
>  1 file changed, 3 deletions(-)
> 
> diff --git a/src/qemu/qemu_domain_address.c b/src/qemu/qemu_domain_address.c
> index 2106b34..3277d18 100644
> --- a/src/qemu/qemu_domain_address.c
> +++ b/src/qemu/qemu_domain_address.c
> @@ -645,9 +645,6 @@ qemuDomainDeviceCalculatePCIConnectFlags(virDomainDeviceDefPtr dev,
>              return pcieFlags;
>          }
>  
> -        if (hostdev->source.subsys.type == VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_MDEV)
> -            return pcieFlags;
> -
>          if (!(pciDev = virPCIDeviceNew(hostAddr->domain,
>                                         hostAddr->bus,
>                                         hostAddr->slot,
> 


This can't work. From the host's point of view, there is no PCI device
whose config space can be read, and from the code's point of view,
hostAddr->domain|bus|slot are invalid (they are in
hostdev->source.subsys.u.pci.addr, which is only valid when
(hostdev->source.subsys.type != VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_PCI).

Is there maybe something standard in the mdev's sysfs entry that could
be used to determine PCI vs PCIe? I think back when we were discussing
the implementation of this, Alex had said there wasn't.