[libvirt] [PATCH 09/13] conf: new pci controller model "pcie-switch-upstream-port"

[Alex Williamson]

On Tue, 2015-06-23 at 13:47 -0400, Laine Stump wrote:
> On 06/23/2015 11:23 AM, Alex Williamson wrote:
> > On Mon, 2015-06-22 at 14:44 -0400, Laine Stump wrote:
> >> diff --git a/src/conf/domain_addr.c b/src/conf/domain_addr.c
> >> index 4b5e81e..59da745 100644
> >> --- a/src/conf/domain_addr.c
> >> +++ b/src/conf/domain_addr.c
> >> @@ -57,6 +57,9 @@ virDomainPCIAddressFlagsCompatible(virDevicePCIAddressPtr addr,
> >>          if (devFlags & VIR_PCI_CONNECT_HOTPLUGGABLE)
> >>              busFlags |= VIR_PCI_CONNECT_HOTPLUGGABLE;
> >>      }
> >> +    /* devices with PCIE_ONLY can't connect to PCI, even if fromConfig */
> >> +    if (devFlags & VIR_PCI_CONNECT_TYPE_PCIE_ONLY)
> >> +       devFlags |= VIR_PCI_CONNECT_TYPE_PCIE;
> >>  
> >>      /* If this bus doesn't allow the type of connection (PCI
> >>       * vs. PCIe) required by the device, or if the device requires
> >> @@ -199,8 +202,14 @@ virDomainPCIAddressBusSetModel(virDomainPCIAddressBusPtr bus,
> >>      case VIR_DOMAIN_CONTROLLER_MODEL_PCIE_ROOT_PORT:
> >>          /* provides one slot which is pcie and hotpluggable */
> >>          bus->flags = VIR_PCI_CONNECT_TYPE_PCIE | VIR_PCI_CONNECT_HOTPLUGGABLE;
> >> +        bus->minSlot = 0;
> >> +        bus->maxSlot = 0;
> > Shouldn't this have belonged to a previous patch?
> 
> Yes, and I'm pretty certain that it originally was. I must have messed
> it up when resolving merge conflicts during my myriads of rebases to
> change names.
> 
> >
> >> +        break;
> >> +    case VIR_DOMAIN_CONTROLLER_MODEL_PCIE_SWITCH_UPSTREAM_PORT:
> >> +        /* 31 slots, can only accept pcie-switch-port, no hotplug */
> >> +        bus->flags = VIR_PCI_CONNECT_TYPE_PCIE_SWITCH;
> >>          bus->minSlot = 1;
> >> -        bus->maxSlot = 1;
> >> +        bus->maxSlot = 31;
> > Why exactly are we reserving slot 0? 
> 
> 
> Uh, because I forgot to change it after talking to you? Thanks for
> catching it!
> 
> 
> >  I know we discussed on IRC that
> > slot 0 is reserved for the host bridge on root buses, but there are no
> > reserved slots downstream of an upstream switch port.
> 
> (Btw, I've forgotten - is slot 0 also reserved for pci-bridge and
> 82801b11-bridge? Or can I send a patch to allow slot 0 on those as well?)

I don't think there's anything special about slot 0 for either of those
bridges, a quick test would verify.  However, we're immediately into
your case of changing the auto-assignment parameters.  Have we always
stored the resulting auto-assigned bus addresses for devices so we don't
suddenly change the location of devices for existing domains?  Thanks,

Alex